Datasets:
TomTBT
/
pmc_open_access_section

Dataset card Data Studio Files Community
1
accession_id
stringlengths
9
11
pmid
stringlengths
1
8
introduction
stringlengths
0
134k
methods
stringlengths
0
208k
results
stringlengths
0
357k
discussion
stringlengths
0
357k
conclusion
stringlengths
0
58.3k
front
stringlengths
0
30.9k
body
stringlengths
0
573k
back
stringlengths
0
126k
license
stringclasses
4 values
retracted
stringclasses
2 values
last_updated
stringlengths
19
19
citation
stringlengths
14
94
package_file
stringlengths
0
35
PMC10787788
38218939
Introduction High-throughput spatial omics technologies are at the forefront of modern molecular biology, and promise to provide topographic context to the wealth of available transcriptomic data. Recent breakthroughs in profiling technology have revolutionised our understanding of multicellular biological systems, and the collection of Subcellular Spatial Transcriptomics (SST) technologies (e.g. 10x Genomics Xenium 1 ; NanoString CosMx 2 ; BGI Stereo-seq 3 ; and Vizgen MERSCOPE) now offer the promise to tackle biological problems that were previously inaccessible and better understand intercellular communication by preserving tissue architecture. Depending on the commercial platforms, these ultra-high resolution, spatially resolved single-cell data contain mixtures of nuclear, cytoplasmic, and/or cell membrane signals, and create new data challenges in information extraction. More specifically, the aim is to ensure all available data can be capitalised to automatically and accurately distinguish the boundaries of individual cells, as the fundamental goal of SST technologies is to understand how single-cell transcriptomes behave in situ within a given tissue 4 . Limited attempts have been made to address these data challenges and to date, three conceptual categories have emerged. The first employs morphological operations originally designed for lower-resolution imaging technologies such as microscopy. Within this category, initial nuclei segmentation is accomplished with a nuclear marker, using thresholding or pretrained models such as Cellpose 5 and Mesmer 6 . Cell boundaries are then identified using either morphological expansion by a prespecified distance 1 or using a watershed algorithm on a mask of the cell bodies 3 . Chen et al. applied a global threshold to the density of all molecules in SST data to estimate the cell body mask. The limitation of Cellpose 5 and similar approaches is that they were primarily designed for microscopy modalities and fluorescent markers, so they may not always be suitable for SST due to dissimilar visual characteristics. Secondly, an alternative approach to cell segmentation does not identify cell boundaries directly, but classifies or clusters individual transcripts into distinct measurement categories that pertain to cells. These include segmentation-free and transcript-based methods, as exemplified by Baysor 7 , StereoCell 8 , pciSeq 9 , Sparcle 10 , and ClusterMap 11 . However, a key limitation of these approaches is their assumption that expression of all RNAs within a cell body are homogeneous, and in the case of Baysor, that cell shapes (morphologies) can be well approximated with a multivariate normal prior. This can result in visually unrealistic segmentations that do not correspond well to imaging data. Thirdly, more recent approaches have begun to leverage deep learning (DL) methods. DL models such as U-Net 12 have provided solutions for many image analysis challenges. However, they require ground truth to be generated for training. DL-based methods for SST cell segmentation include GeneSegNet 13 and SCS 14 , though supervision is still required in the form of initial cell labels or based on hard-coded rules. Further limitations of existing methods encountered during our benchmarking, such as lengthy code runtimes, are included in Supplementary Table 1 . The self-supervised learning (SSL) paradigm can provide a solution to overcome the requirement of annotations. While SSL-based methods have shown promise for other imaging modalities 15 , 16 , direct application to SST images remains challenging. SST data are considerably different from other cellular imaging modalities and natural images (e.g., regular RGB images), as they typically contain hundreds of channels, and there is a lack of clear visual cues that indicate cell boundaries. This creates new challenges such as (i) accurately delineating cohesive masks for cells in densely-packed regions, (ii) handling high sparsity within gene channels, and (iii) addressing the lack of contrast for cell instances. While these morphological and DL-based approaches have shown promise, they have not fully exploited the high-dimensional expression information contained within SST data. It has become increasingly clear that relying solely on imaging information may not be sufficient to accurately segment cells. There is growing interest in leveraging large, well-annotated scRNA-seq datasets 17 , as exemplified by JSTA 18 , which proposed a joint cell segmentation and cell type annotation strategy. While much of the literature has emphasised the importance of accounting for biological information such as transcriptional composition, cell type, and cell morphology, the impact of incorporating such information into segmentation approaches remains to be fully understood. Here, we present a biologically-informed deep learning-based cell segmentation (BIDCell) framework (Fig. 1 a), that addresses the challenges of cell body segmentation in SST images through key innovations in the framework and learning strategies. We introduce (a) biologically-informed loss functions with multiple synergistic components; and (b) explicitly incorporate prior knowledge from single-cell sequencing data to enable the estimation of different cell shapes. The combination of our losses and use of existing scRNA-seq data in supplement to subcellular imaging data improves performance, and BIDCell is generalisable across different SST platforms. Along with the development of our segmentation method, we created a comprehensive evaluation framework for cell segmentation, CellSPA, that assesses five complementary categories of criteria for identifying the optimal segmentation strategies. This framework aims to promote the adoption of new segmentation methods for novel biotechnological data.
Methods Datasets and preprocessing We used publicly available data resources from three different SST commercial platforms (10 × Genomics Xenium, NanoString CosMx, and Vizgen MERSCOPE), and sequencing data from Human Cell Atlas. Subcellular spatial transcriptomics data For all datasets and for each gene, detected transcripts were converted into a 2D image where the value of each pixel represents the number of detected transcripts at its location. The images were combined channel-wise, resulting in an image volume , where H is the height of the sample, W is the width of the sample, and n g e n e s is the number of genes in the panel. (i) Xenium-BreastCancer1 and Xenium-BreastCancer2 The Breast Cancer datasets included in this study were downloaded from https://www.10xgenomics.com/products/xenium-in-situ/preview-dataset-human-breast (accessed 9 Feb 2023), and included two replicates. Low-quality transcripts for 10 × Genomics Xenium data with a phred-scaled quality value score below 20 were removed, as suggested by the vendor 1 . Negative control transcripts, blanks, and antisense transcripts were also filtered out. This resulted in 313 unique genes with the overall pixel dimension of the images being 5475 × 7524 × 313 for Xenium breast cancer replicate 1 (Xenium-BreastCancer1) and 5474 × 7524 × 313 for Xenium breast cancer replicate 2 (Xenium-BreastCancer2). (ii) Xenium-MouseBrain The Mouse Brain data included in this study was downloaded from https://www.10xgenomics.com/resources/datasets/fresh-frozen-mouse-brain-replicates-1-standard (accessed 14 Feb 2023) and were processed following the steps in (i). There were 248 unique genes, and the resulting size of the image was 7038 × 10,277 × 248 pixels. (iii) CosMx-Lung The CosMx NSCLC Lung dataset included in this study was downloaded from https://nanostring.com/products/cosmx-spatial-molecular-imager/nsclc-ffpe-dataset/ (accessed 24 Mar 2023). We used data for Lung5-1, which comprised 30 fields of view. Transcripts containing “NegPrb” were removed, resulting in 960 unique genes and an overall image dimension of 7878 × 9850 × 960 pixels. (iv) MERSCOPE-Melanoma The MERSCOPE melanoma data included in this study were downloaded from https://info.vizgen.com/merscope-ffpe-solution (for patient 2, accessed 26 Mar 2023). Transcripts with “Blank-” were filtered out, resulting in 500 unique genes and an image with 6841 × 7849 × 500 pixels. (v) Stereo-seq-MouseEmbryo The Stereo-seq data used in this study, including the DAPI image and detected gene expressions (bin 1), were downloaded from https://db.cngb.org/stomics/mosta/download/ for sample E12.5_E1S3. Stereo-seq data contains a far greater number of genes compared to Xenium, CosMx, and MERSCOPE. For efficiency, we selected a panel of 275 highly variable genes (HVGs) as the input to BIDCell. The HVGs are the common genes of the top 1000 HVGs from both Stereo-seq data and the single-cell reference data. Nuclei segmentation DAPI images were directly downloaded from the websites of their respective datasets. In cases where the maximum intensity projection (MIP) DAPI image was not provided, we computed the MIP DAPI by finding the maximum intensity value for each (x,y) location for each stack of DAPI. DAPI images were resized to align with the lateral resolutions of spatial transciptomic maps using bilinear interpolation. Nuclei segmentation was performed on the MIP DAPI using the pretrained Cellpose model with automatic estimation of nuclei diameter 5 . We used the “cyto" model as we found the “nuclei" model to undersegment or omit a considerable number (e.g., 21k for Xenium-BreastCancer1) of nuclei given the same MIP DAPI image, which is consistent with another study 29 . Other nuclei segmentation methods may be used with BIDCell as our framework is not limited to Cellpose. Transcriptomics sequencing data We used five publicly available single-cell RNA-seq data collections as references to guide the cell segmentation in BIDCell and evaluation with CellSPA. For the reference data with multiple datasets, we constructed cell-type specific profiles by aggregating the gene expression by cell type per dataset. (i) TISCH-BRCA The reference for Xenium-BreastCancer used in BIDCell was based on 10 single-cell breast cancer datasets downloaded from The Tumor Immune Single Cell Hub 2 (TISCH2) 17 from http://tisch.comp-genomics.org/gallery/?cancer=BRCA&species=Human , which contains the gene by cell expressions and cell annotations of the data. We used the “celltype major lineage" as the cell type labels. We combined the “CD4Tconv" and “Treg" as “CD4Tconv/Treg" and “CD8T" and “CD8Tex" as “CD8T/CD8Tex", which results in 17 cell types in total. (ii) Chromium-BreastCancer To evaluate the performance of Xenium-BreastCancer, we downloaded the Chromium scFFPE-seq data from the same experiment from https://www.10xgenomics.com/products/xenium-in-situ/preview-dataset-human-breast (accessed 22 March 2023), which contains 30,365 cells and 18,082 expressed genes. We then performed Louvain clustering on the k-nearest neighbour graph with k = 20, based on the top 50 principal components (PCs) to obtain 22 clusters. We then annotated each cluster based on the markers and annotation provided in the original publication 1 . (iii) Allen Brain Map The reference for Xenium-MouseBrain data was based on Mouse Whole Cortex and Hippocampus SMART-seq data downloaded from https://portal.brain-map.org/atlases-and-data/rnaseq/mouse-whole-cortex-and-hippocampus-smart-seq , which contains both gene by cell expressions and cell annotations of the data. We used the cluster annotation from “cell_type_alias_label" as the cell type labels and combined some of the labels with a small number of cells. For example, we combined all “Sst" subtypes as “Sst" and all “Vip" subtypes as “Vip", which results in 59 cell types in total. (iv) HLCA and TISCH-NSCLC The reference for CosMx-Lung for both BIDCell and CellSPA was based on Human Lung Cell Atlas (HLCA) 30 , provided in the “HLCA_v1.h5ad" file from https://beta.fastgenomics.org/p/hlca , including both gene expressions and cell type annotations of the data. We used “ann_finest_level" as cell type labels, which contained 50 cell types in total. As HLCA only contains single-cell datasets from non-cancer lung tissue, we complemented the reference data with malignant cells provided in TISCH2, where we downloaded 6 single-cell NSCLC datasets with tumour samples from http://tisch.comp-genomics.org/gallery/?cancer=NSCLC&species=Human . We only included the cells labelled as malignant cells in the reference. (v) TISCH-SKCM The reference for MERSCOPE-Melanoma for both BIDCell and CellSPA was based on 10 single-cell melanoma datasets downloaded from TISCH2 from http://tisch.comp-genomics.org/gallery/?cancer=SKCM&species=Human , which contains the gene by cell expressions and cell annotations of the data. We used the “celltype major lineage” as the cell type labels. We combined the “CD4Tconv” and “Treg” as “CD4Tconv/Treg” and “CD8T” and “CD8Tex” as “CD8T/CD8Tex”, which resulted in 15 cell types in total. (vi) Mouse Embryo reference The reference for Stereo-seq-MouseEmbryo was downloaded from GEO database under accession code: GSE119945 31 , which contain both counts and cell type annotation data. The E12.5 data was then used as reference. Biologically-informed deep learning-based cell segmentation (BIDCell) overview BIDCell is a self-supervised deep learning framework that computes biologically-informed loss functions to optimise learnable parameters for the prediction of cell segmentation masks for spatial transcriptomic data. BIDCell uses three types of data: (i) spatial transcriptomic maps of genes, (ii) corresponding DAPI image, and (iii) average gene expression profiles of cell types from a reference dataset, such as the Human Cell Atlas. A major innovation in developing BIDCell is the use of biologically-informed prior knowledge via the SSL paradigm to enable DL models to learn complex structures in SST data, to derive cell segmentations that are visually more realistic and capture better expression profiles. The BIDCell framework has the following four key characteristics: BIDCell predicts diverse cell shapes for datasets containing various cell types to better capture cell expressions (see section Elongated and non-elongated shapes). BIDCell uses positive and negative markers from sequencing data to enhance the guidance for learning relationships between spatial gene expressions and cell morphology in the form of cell segmentations (see section Positive and negative cell-type markers). BIDCell is parameterised by a deep learning architecture that learns to segment cells from spatial transcriptomic images (see section Deep learning-based segmentation). BIDCell uses biologically-informed, self-supervised loss functions to train the deep learning architecture without the need for manual annotations and better capture cell expressions (see section BIDCell training and loss functions). Elongated and non-elongated shapes BIDCell is capable of generating cell segmentations that exhibit different morphologies for different cell types, rather than assume a generally circular profile for all cell types. In particular, BIDCell can distinguish between cell types that typically appear more elongated, such as fibroblasts and smooth muscle cells, and those that are typically more rounded or circular, such as B cells. Elongated cell types can be directly specified for each tissue sample as desired, based on existing biological knowledge. We used the expression within the nuclei (see section Nuclei segmentation) of cells to perform an initial classification of elongated and non-elongated cell types. Transcripts were mapped to nuclei using nuclei segmentations, and the Spearman correlation was computed between nuclei expression profiles and reference cell types of the Human Cell Atlas. Nuclei were classified as the cell type with which it was most highly correlated to. This initial classification coupled with the eccentricity of the nuclei were used to inform the cell-calling loss function (described in section Cell-calling loss) to produce segmentation morphologies with more variation that are more appropriate for different cell types. We considered epithelial cells, fibroblasts, myofibroblasts, and smooth muscle cells to be elongated for samples of breast cancer and melanoma. Endothelial cells, fibroblasts, myofibroblasts, fibromyocytes, and pericytes were deemed elongated for NSCLC. We considered all cell types in the mouse brain sample to be elongated. Positive and negative cell-type markers BIDCell learns relationships between the spatial distribution of gene expressions and cell morphology in the form of cell segmentations. This relationship can be enhanced by incorporating biological knowledge in the form of cell-type markers, specially, the genes that are typically more expressed (positive markers) and less expressed (negative markers) in different cell types, which allows BIDCell to predict segmentations that lead to more accurate cell expression profiles. Cell-type marker knowledge is drawn from the Human Cell Atlas, which allows BIDCell to be applied without requiring a matched single-cell reference for the same sample of interest. Markers were incorporated into BIDCell through our positive and negative marker losses (described in section Positive and negative marker losses). Deep learning-based segmentation BIDCell is parameterised by a set of learnable parameters θ of a deep learning segmentation model. We used the popular UNet 3+ 19 as the backbone of our framework to perform cell segmentation by predicting the probability of cell instances at each pixel. This architecture may be swapped out for other segmentation architectures. UNet 3+ was originally proposed for organ segmentation in computed tomography (CT) images. It was built on the original U-Net 12 and incorporated full-scale skip connections that combined low-level details with high-level features across different scales (resolutions). UNet 3+ comprised an encoding branch and decoding branch with five levels of feature scales. We did not adopt the deep supervision component proposed by UNet 3+, and instead only computed training losses at the lateral resolution of the original input. Input The input to the UNet 3+ model was a cropped multichannel spatial transcriptomic image , where n g e n e s represents the channel axis corresponding to the total number of genes in the dataset, h is the height of the input patch, and w is the width of the input patch. Prior to being fed into the first convolutional layer, the input was reshaped to [ n c e l l s , n g e n e s , h , w ], effectively placing n c e l l s in the batch size dimension. In this way, all the cells in a patch were processed simultaneously, and the model could flexibly support an arbitrary number of cells without requiring extra padding or preprocessing. n c e l l s was determined by the corresponding patch of nuclei to ensure consistency with predicted cell instances. Input volumes that were empty of nuclei were disregarded during training and yielded no cells during prediction. Output and segmentation prediction The softmax function was applied to the output of UNet 3+ to yield probabilities of foreground and background pixels for each cell instance. This produced multiple probabilities for background pixels (i.e., n c e l l s probabilities per pixel for a patch containing n c e l l s ), due to the placement of cell instances in the batch size dimension. These probabilities were aggregated by averaging across all the background predictions per pixel. The argmax function was applied pixel-wise to the foreground probabilities for all cells and averaged background probabilities. This produced a segmentation map corresponding to the object (cell instance or background) with the highest probability at each pixel. Morphological processing The initial segmentation output by the deep learning model was further refined to ensure pixel connectivity within each cell (i.e., all the sections of the cell were connected). The process involved standard morphological image processing techniques to each cell, including dilation, erosion, hole-filling, and removal of isolated islands, while ensuring that the nucleus was captured. First, dilation followed by erosion were applied using a 5 × 5 circular kernel with two iterations each. Hole-filling was then carried out on the cell section with the largest overlap with the nucleus. Any remaining pixels initially predicted for the cell that were still not connected to the main cell section were discarded. After morphological processing, the number of transcripts captured within each cell is slightly higher, while purity metrics and correlation with Chromium are the same or slightly higher (Supplementary Fig. 24 ). Mapping transcripts to predicted cells The detected transcripts were mapped to cells using the final predicted segmentations. The segmentation map was resized back to the original pixel resolution using nearest neighbour interpolation. Transcripts located in the mask of a cell were added to the expression profile of the cell. This produced a gene-cell matrix n c e l l s × n g e n e s , which was used for performance evaluation and downstream analysis. BIDCell training and loss functions The BIDCell framework combines several loss functions that automatically derive supervisory signals from the input data and/or predicted segmentations at each step of the training process. This approach to learning is a core aspect of SSL 32 . Furthermore, the modular and additive design of the loss functions allows each loss to be swapped out with alternative approaches to compute training signals. The SSL label describes the ability of the framework to automatically learn relationships between gene expressions and cell morphology from its inputs. Our approach for learning the parameters θ of the segmentation model relies on minimising a total of 6 loss functions that we propose with our framework. Some of the losses effectively increase the number of pixels predicted for a cell, while others reduce the size of its segmentation. The nuclei encapsulation, cell-calling, over-segmentation, and overlap losses guide the basic morphology of cells. The positive and negative marker losses refine the cell morphologies learned through the other loss functions, by further guiding the model to learn biologically-informed relationships between gene expressions and cell morphology. This is reminiscent of the pretext and downstream (fine-tuning) stages commonly encountered in SSL, where the pretext task aids the model to learn better representations or intermediate weights, while the fine-tuning task refines the weights and further improves performance for a particular prediction task. Taken together, the losses ensure that the segmentation model learns relationships between spatially-localised, high-dimensional gene expression information and the morphology of individual cells. (A) Nuclei encapsulation loss The segmentation of a cell must contain all the pixels of the cell’s nucleus. Additionally, the expressed genes in nuclei can guide the model to learn which genes should be predicted within cells. Hence, we included a loss function L n e that incentivises the model to learn to correctly predict nuclei pixels: where x nuc is the binary nucleus segmentation mask, and is the predicted segmentation for all cells of the corresponding training patch. (B) Cell-calling loss The aim of the cell-calling loss was to increase the number of transcripts assigned to cells. We also designed the cell-calling loss to allow BIDCell to capture cell-type specific morphologies. Unique expansion masks e c ∈ {0, 1} h × w were computed for each cell based on the shape of its nucleus and whether its nucleus expression profile was indicative of an elongated cell type. The expansion mask of a non-elongated cell was computed by applying a single iteration of the morphological dilation operator with a circular kernel of 20 × 20 pixels to its binary nucleus mask. The expansion mask of an elongated cell was computed based on the elongation of its nucleus, defined as the eccentricity of an ellipse fitted to its nucleus mask: where a represents the length of the major axis, and b is the length of the minor axis. We found that elongated cell types tended to have nuclei with higher eccentricity (Supplementary Fig. 1 ). Hence, the eccentricity of a nucleus could serve as a proxy for the shape of its cell via an elongated expansion mask. We computed each cell-specific elongated expansion mask using an elliptical dilation kernel applied to the nucleus. The horizontal and vertical lengths of the elliptical kernel were computed by: where α is a scaling factor set to 0.9, e c c n u c is the eccentricity of the nucleus, l t is the sum of l h and l v , which was set to 60 pixels, and l v m is the minimum vertical length, which was set to 3 pixels. These values were selected based on visual inspection (e.g., the cells appear reasonably sized), and were kept consistent across the different elongated cell types and datasets used in this study. The elliptical dilation kernel was rotated to align with the nucleus and applied to the nucleus mask to produce the elongated expansion mask of the cell. The expansion masks were used in our cell-calling loss function that was minimised during training: where e c is the expansion mask and is the predicted segmentation of cell c of M cells in an input patch. (C) Over-segmentation loss We introduced the over-segmentation loss to counter the cell size-increasing effects of the cell-calling loss to prevent the segmentations becoming too large and splitting into separate segments. This loss function elicited a penalty whenever the sum of cytoplasmic predictions exceeded the sum of nuclei predictions for a cell in a given patch: where for cell c at pixel ( i , j ), is the predicted foreground probability for cell c , x n u c , i j ∈ {0, 1} is the binary nucleus mask, and σ is the sigmoid function. L o s was normalised by number of cells M to aid smooth training. (D) Overlap loss Cells are often densely-packed together in samples of various human tissues. This poses a challenge to segmentation models in predicting clear boundaries and coherent segmentations for neighbouring cells without overlap. We introduced the overlap loss to penalise the prediction of multiple cells occurring at each pixel: L o v was normalised by number of cells M , and the lateral dimensions h and w of the input to aid smooth training. (E) Positive and negative marker losses The purposes of our positive and negative marker losses were to encourage the model to capture pixels that contained positive cell-type markers, and penalise the model when segmentations captured pixels that contained negative cell-type markers for each cell. The marker losses refine the initial morphology learned through the other loss functions, by further guiding the model to learn biologically-informed relationships between gene expressions and cell morphology. The positive and negative markers for the training loss were those with expressions in the highest and lowest 10 percentile for each cell type of a tissue sample. In our experiments, we found that a higher number of positive markers tended to increase the size of predicted cells as the model learns to capture more markers, and vice versa. We found that removing positive markers that were common to at least a third of cell types in each tissue type was appropriate across the different datasets for training. The one-hot encoded lists of positive and negative markers of the cell type for cell c were converted into sparse maps m pos, c ∈ {0, 1} h × w and m neg,c ∈ {0, 1} h × w . At each pixel, 0 indicated the absence of all markers, while 1 indicated the presence of any positive or negative marker for its respective map. m pos, c and m neg, c were then multiplied element-wise by the expansion mask e c to remove markers far away from the current cell. Each marker map was dilated by a 3 × 3 kernel, which was based on the assumption that pixels in a 3 × 3 region around each marker were most likely from the same cell. We found this dilation to improve training guidance and segmentation quality, as the maps tended to be quite sparse. The marker maps were then used to compute the positive and negative marker losses: Total loss The model was trained by minimising the sum of all the loss functions over N training patches: where each λ represents a hyperparameter that scaled its respective L . The value of λ for all loss functions was set to 1.0 (except for the ablation and lambdas studies); this ensured our losses were not fine-tuned to any particular datasets. Practical implementation Details To address computational efficiency concerns related to memory usage, we partitioned the spatial transcriptomic maps into patches of 48 × 48 × n g e n e s for input into UNet 3+. BIDCell has been verified for datasets containing up to 960 genes on a 12 GB GPU. It is also important to note that the number of genes primarily affects the weights of the first convolutional layer, thus having a minor impact on memory usage. The patch-based predictions could result in effects along the patch boundaries such as sharp or cut-off cells. When dividing the transcriptomic maps into patches, we create two sets of patches of the same lateral dimensions with an overlap equal to half the lateral size of the patches. The predictions for the patches were combined (see Supplementary Fig. 25 ), without additional operations to resolve potential disagreement between predictions of the two sets. Only patches from the first set (no overlaps) were selected during training, while all patches were used during inference. One image patch was input into the model at one time, though batch size was effectively n c e l l s due to reshaping (see section Deep learning-based segmentation-Input). Neither normalisation nor standardisation were applied to the input image patches, such that the pixels depicted raw detections of transcripts. The model was trained end-to-end from scratch for 4000 iterations (i.e., using 4000 training patches). This amounted to a maximum of 22% of the entire image, thereby leaving the rest of the image unseen by the model during inference. Weights of the convolutional layers were initialised using He et al.’s method 33 . We employed standard on-the-fly image data augmentation by randomly applying a flip (horizontal or vertical), rotation (of 90, 180, or 270 degrees) in the (x,y) plane. The order of training samples was randomised prior to training. We employed the Adam optimiser 34 to minimise the sum of all losses at a fixed learning rate of 0.00001, with a first moment estimate of 0.9, second moment estimate of 0.999, and weight decay of 0.0001. Time and system considerations We ran BIDCell on a Linux system with a 12GB NVIDIA GTX Titan V GPU, Intel(R) Core(TM) i9-9900K CPU @ 3.60GHz with 16 threads, and 64GB RAM. BIDCell was implemented in Python using PyTorch. For Xenium-BreastCancer1, which contained 109k detected nuclei, 41M pixels (x,y) , and 313 genes, training was completed after approximately 10 minutes for 4000 steps. Inference time was about 50 minutes for the complete image. Morphological processing required approximately 30 min to generate the final segmentation. A comparison of the runtimes between different methods is included in Supplementary Fig. 26 . Ablation study We performed an ablation study to determine the contributions from each loss function and effects of different hyperparameter values (Supplementary Figs. 4 , 5 ). We used Xenium-BreastCancer1 for these experiments. We evaluated BIDCell without each of the different loss functions by individually setting their corresponding weights λ to zero. Furthermore, we evaluated different parameterisations of the cell-calling loss. We experimented with different diameters for the dilation kernel for non-elongated cells, including 10, 20, and 30 pixels, and different total lengths of the minor and major axes l t of the dilation kernel for elongated cells, including 50, 60, and 70 pixels. We also ran BIDCell without shape-specific expansions, thereby assuming a non-elongated shape for all cells. Performance evaluation We compared our BIDCell framework to vendor-provided cell segmentations, and methods designed to identify cell bodies via cell segmentation. Table 2 provides a summary of all methods compared from adapting classical approaches including Voronoi expansion, nuclei dilation, and the watershed algorithm, to recently proposed approaches for SST images including Baysor, JSTA, and Cellpose. Methods that were excluded from the evaluations include those that focus on the assignment of transcripts to cells and do not consider the cell boundaries, underperformance on the public datasets, lack of code and instructions to prepare data into the required formats, and failure of the method to detect any cells (Supplementary Table 1 ). Settings used for other methods We used publicly available code for Baysor, JSTA, and Cellpose with default parameters unless stated otherwise. All comparison methods that required nuclei information used identical nuclei as BIDCell, which were detected using Cellpose (v2.1.1) (see Nuclei segmentation). Baysor - Version 0.5.2 was applied either without a prior, or with a prior nuclei segmentation with default prior segmentation confidence of 0.2. For both instances, we followed recommended settings 35 , including 15 for the minimum number of transcripts expected per cell, and not setting a scale value, since the sample contained cells of varying sizes. We found the scale parameter to have a considerable effect on segmentation predictions, and often resulted in cells with unrealistically uniform appearances if explicitly set. JSTA - default parameters were used. We encountered high CPU loading and issues with two regions of Xenium-BreastCancer1, which yielded empty predictions for those regions despite multiple attempts and efforts to reduce input size. Cellpose - Version 2.1.1 was applied to the channel-wise concatenated image comprising DAPI as the “nuclei” channel, and sum of spatial transcriptomic maps across all genes as the “cells” channel, using the pre-trained “cyto” model with automatic estimation of cell diameter. Voronoi - Classical Voronoi expansion was seeded on nuclei centroids and applied using the SciPy library (v1.9.3). Watershed - The watershed algorithm was performed on the sum of transcriptomic maps across all genes. Seeded watershed used nuclei centroids and was applied using OpenCV (v4.6.0). Cellpose nuclei dilation - we applied dilation to nuclei masks as a comparison segmentation method. Each nucleus was enlarged by about 1 micron in radius by applying morphological dilation using a 3 × 3 circular kernel for one iteration. Overlaps between adjacent cell expansions were permitted. Evaluation metrics and settings We introduce the CellSPA framework, that captures evaluation metrics across five complementary categories. A summary of this information is provided in Supplementary Table 2 . [A] Baseline metrics Overall characteristics Number of cells Proportion of transcripts assigned Cell-level QC metrics Proportion of cells expressing each gene Number of transcripts per cell Number of genes expressed per cell Cell area where ∑ i ∈ I n i represents the sum of all total transcripts over a set I , and A represents the cell area. Cell morphology metrics We evaluated multiple morphology-based metrics and provide diagrammatic illustrations in Supplementary Fig. 27 . • Elongation = where W bb represents the width of the bounding box, and H bb represents the height of the bounding box. Elongation measures the ratio of height versus the width of the bounding box (Supplementary Fig. 27 f). Elongation is insensitive to concave irregularities and holes present in the shape of the cell. The value of this metric will be 1 for a perfect square bounding box. As the cell becomes more elongated the value will either increase far above 1 or decrease far below 1, depending on whether the elongation occurs along the height or width of the bounding box. • Circularity = where A represents the area, and P convex represents the convex perimeter. Circularity measures the area to perimeter ratio while excluding local irregularities of the cell. We used the convex perimeter of the object as opposed to its true perimeter to avoid concave irregularities. The value will be 1 for a circle and decreases as a cell becomes less circular. • Sphericity = where R I represents the radius of the inscribing circle, and R C represents the radius of the circumscribing circle. Sphericity measures the rate at which an object approaches the shape of a sphere while accounting for the largest local irregularity of the cell by comparing the ratio of the radius largest circle that fits inside the cell (inscribing circle) to the radius of the smallest circle that contains the whole cell (circumscribing circle). The value is 1 for a sphere and decreases as the cell becomes less spherical. • Compactness = where A represents the area, and P cell represents the cell perimeter. Compactness measures the ratio of the area of an object to the area of a circle with the same perimeter. Compactness uses the perimeter of the cell thus it considers local irregularities in the cell perimeter. A circle will have a value of 1, and the less smooth or more irregular the perimeter of a cell, the smaller the value will be. For most cells the numerical values for compactness and circularity are expected to be similar. Identifying which cells have large differences between these metrics can identify cells with highly irregular perimeters which may be of interest for downstream analysis and quality control for segmentation. • Convexity = where P convex represents the convex perimeter and P cell represents the cell perimeter. Convexity measures the ratio of the convex perimeter of a cell to its perimeter. The value will be 1 for a circle and decrease the more irregular the perimeter of a cell becomes, similar to compactness. • Eccentricity = where L minor represents the length of the minor axis and L major represents the length of the major axis. Eccentricity (or ellipticity) measures the ratio of the major axis to the minor axis of a cell. The major axis is the longest possible line that can be drawn between the inner boundary of a cell without intersecting its boundary. The minor axis is the longest possible line can be drawn within the inner boundary of a cell while while also being perpendicular to the major axis. This gives a value of 1 for a circle and decreases the more flat the cell becomes. • Solidity = where A represents the area, and A convex represents the convex area. Solidity measures the ratio of the area of a cell to the convex area of a cell. This measures the density of a cell by detecting holes and irregular boundaries in the cell shape. The maximum value will be 1 for a cell with a perfectly convex and smooth boundary and will decrease as the cell shape becomes more concave and/or irregular. Gene-level QC characteristics • Proportion of cells expressing each gene [B] Segmented cell expression purity . We implemented two broad classes of statistics to capture (i) the concordance of expression profile with scRNA-seq data and (ii) the expression purity or homogeneity of cell type markers. The scRNA-seq data used are described in Section Datasets and preprocessing and listed in Table 1 . • Concordance with scRNA-seq data - We calculated the similarity of the expression pattern between the segmented cells and publicly available single-cell datasets. Here the similarity was measured by Pearson correlation of the average log-normalised gene expression for each cell type. We also calculated the concordance of the proportion of non-zero expression for each cell type between the segmented cells and scRNA-seq data. For data with paired Chromium data from the same experiment, i.e., Xenium-Brain, we also compared the cell type proportion and quantify the concordance using the Pearson correlation. We annotated the cell type annotation for segmented cells using scClassify 36 with scRNA-seq data as reference. • Purity of expression - We first curated a list of positive markers and negative markers from the scRNA-seq reference data. For each cell type, we selected the highest and lowest 10 percentile of the genes with difference of expression compared to other cell types. We also removed the positive markers that were common to more than 25% of cell types for a more pure positive marker list. For each segmented cell, we then consider the genes with the highest 10 percentile of expression as positive genes and lowest 10 percentile as negative markers. We then calculated the Precision, Recall and F1 score for both positive and negative markers. We further summarised the average positive marker F1 scores and negative marker F1 scores into one Purity F1 score for each method, where we first scaled the average positive and negative marker F1 scores into the range of [0, 1] and then calculated the F1 score of transformed metrics as the following: [C] Spatial characteristics In this category, we measured the association between cell type diversity in local spatial regions and all the cell-level baseline characteristics provided in [A]. We first divided each image into multiple small regions. Then, for each local spatial region, we calculated the cell type diversity using Shannon entropy with the R package ’entropy’, where a higher entropy indicates a more diverse cell type composition. Next, we assessed the variability of cell-level baseline characteristics within each local region using the coefficient of variation. Subsequently, for each of the cell-level baseline characteristics mentioned in [A], we calculated the Pearson correlation between the cell type diversity (measured using Shannon entropy) and the coefficient of variation of these characteristics across all local regions. Here, we anticipate that regions with more diverse cell type compositions will exhibit higher variability in cell-level characteristics, leading to a stronger correlation between these two metrics. [D] Neighbouring contamination This metric is designed for cell segmentation to ensure that the expression signals between neighboring cells are not contaminated. For a pair of cell types (e.g., cell type A and B), we computed the Euclidean distance from each cell in cell type A to its nearest neighbor belonging to cell type B. We then grouped the cells of cell type A based on a range of distances. Within each group, we calculated the proportion of cells expressing a selected negative marker, which is a cell type marker for cell type B. We anticipate that the method with less contamination will result in segmented cells expressing lower levels of the negative marker, even when the distance to a different cell type is minimal. [E] Replicability Our analysis involved assessing the agreement between the Xenium-BreastCancer1 and Xenium-BreastCancer2 datasets, which are closely related in terms of all the cell-level baseline characteristics provided in [A]. As these datasets are considered to be sister regions, we anticipated that the distribution of all the baseline characteristics, as well as the cell type composition, would be similar. We use Pearson correlation to quantify the degree of concordance. Statistics and reproducibility All analysis was done in R version version (4.3.0). No statistical method was used to predetermine sample size. No data were excluded from the analyses. All cells that passed quality control were included in the analyses. The experiments were not randomized. The Investigators were not blinded to allocation during experiments and outcome assessment. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Results BIDCell: Incorporating biological insights using deep learning to improve cell shape representation BIDCell is a DL-based cell segmentation method that identifies each individual cell and all its pixels as a cohesive mask. BIDCell uses subcellular spatial transcriptomic maps, corresponding DAPI images, and relevant average expression profiles of cell types from single-cell sequencing datasets; the latter is obtained from public repositories such as the Human Cell Atlas. Given the lack of ground truth and visual features that indicate cell boundaries in the SST images, BIDCell instead focuses on the relationships between the high-dimensional spatial gene expressions and cell morphology. The BIDCell framework automatically derives supervisory signals from the input data and/or predicted segmentations, which is an approach to learning that we borrow from SSL. To achieve this, we designed multiple loss functions that represent various criteria based on biological knowledge, that work synergistically to produce accurate segmentations (Fig. 1 a; see Methods and Supplementary Materials for a detailed description). BIDCell learns to use the locations of highly- and lowly-expressed marker genes to calibrate the segmentation to capture higher “cell expression purity”, thereby ensuring transcripts within each cell share the same profile. Furthermore, BIDCell captures local expression patterns using a data-driven, cell-type-informed morphology. We found that the eccentricity measure of nuclei could reveal diverse cell morphologies that correspond to established knowledge, such as elongated morphologies for fibroblasts (Supplementary Fig. 1 ). By capturing a diverse set of cell shapes and leveraging marker information from previous single-cell experiments (Table 1 ), BIDCell generates superior segmentations (Fig. 1b and Supplementary Figs. 2 and 3 ), and overcomes the limitations of many existing methods (Table 2 ) that rely primarily on SST image intensity values for cell segmentation. We further ensure the integrity of cell segmentations by proposing three other cooperative loss functions. Appropriate cell sizes are supported by capturing expression patterns local to nuclei using guidance from cell-type informed morphologies (cell-calling), while ensuring the cohesiveness of cell instances (over-segmentation) and enhancing segmentation in densely-populated regions (overlap loss). BIDCell also leverages expression patterns within nuclei to guide the identification of cell body pixels. We investigated removing individual losses in an ablation study with Xenium-BreastCancer1 data (Supplementary Figs. 4 , 5 ). Our investigation shows that the losses work synergistically; e.g., there was a marked increase in purity F1 relative to the amount of captured transcripts when the losses were combined. With the inclusion of single-cell data (which informs the positive and negative losses, and contributes to the ability to predict elongated cell shapes), performance improved considerably, particularly in purity metrics and correlation to Chromium data. The use of single-cell data helped the model to better capture transcripts that are more biologically meaningful within cells. By default, the weights of the losses are all 1.0 and do not need to be tuned for BIDCell to perform well, though further fine-tuning is possible (Supplementary Fig. 6 ). The popular UNet 3+ 19 serves as the segmentation backbone architecture in BIDCell, though this is not a requirement and it may be replaced with alternative architectures (Supplementary Fig. 7 ). CellSPA comprehensive evaluation framework captures diverse sets of metrics of segmentation aspects across five complementary categories To ensure an unbiased comparison, we introduce a Cell Segmentation Performance Assessment (CellSPA) framework (Fig. 2 a) that captures cell segmentation metrics across five complementary categories. These categories, detailed in Fig. 2 a and Supplementary Table 2 , include (i) baseline characteristics at both the cell and gene levels; (ii) measures of segmented cell expression, where we assess the “ expression purity” of our assigned segmented cells based on how well transcripts within the segmented cell share a similar expression profile; (iii) measures of baseline cell characteristics in its spatial environment , including spatial region diversity and corresponding diversity in morphology; (iv) a measure of contamination between nearest neighbours (Supplementary Fig. 8 ); and (v) measures of replicability . Using CellSPA, we compared the performance of BIDCell with several recently developed methods for the segmentation of SST data. These methods included classical segmentation-based approaches such as simple dilation, watershed, and Voroni; and transcript-based approaches including Baysor. Additionally, we evaluated JSTA 18 , which attempts to jointly determine cell (sub) types and cell segmentation based on an extension from the traditional watershed approach. In all comparisons, we limited the computational time to within 72 h, which we deemed a practical requirement for the solutions provided by each approach (see Discussion). To ensure the minimal appropriateness of segmented cells, we examine a series of quality control (QC) statistics. As an illustrative example using Xenium-BreastCancer1 data, we segmented cells using BIDCell, generating 100,000 number of cells, with 53.4% of transcripts assigned (Fig. 2 b). We first confirm that the total number of transcripts per cell and the number of genes per cell were greater in the whole cell (cell body + nuclei) compared to just the nuclei (Fig. 2 c and Supplementary Fig. 9 ). Similarly, using the percentage of cells expressing each gene between the nuclei vs. the cell body, we further evaluate the level of information presented in the nuclei and the cell body from the gene level (Fig. 2 d). We find that the segmented cells of some of the methods (e.g. Baysor) did not yield any additional transcript information beyond that of the nuclei, where we see a tight concordance (lying on a 45-degree line) between the segmented cell body and the cell nuclei. However, BIDCell, 10x, Cellpose, and JSTA are all able to capture additional transcript information. Moving forward, we will focus on methods that provide “additional" information to the nuclei, with an emphasis on the ability to better capture cell boundaries. Lastly, we examine the cell morphology of the segmented cells against the segmented nuclei, including cell area, elongation, compactness, sphericity, convexity, eccentricity, solidity and circularity (See Methods and Supplementary Fig. 10 ). Through these metrics, we are able to identify the outliers of the segmented cells, such as cells with extremely large areas in JSTA, Voronoi and Watershed in the sparse areas (Supplementary Fig. 11 ). We illustrate that as intended from our cell-mask, BIDCell has cell morphology that is highly correlated with the nuclei morphology (Fig. 2 e). Furthermore, we find that segmented cells from BIDCell exhibit more diverse cell morphology characteristics compared to other methods (Supplementary Fig. 12 ). BIDCell captures improved purity of cell expression, leading to less contamination from neighbouring cells To determine whether various cell segmentation methods can improve spatial resolution without sacrificing detection efficiency, we first compare the correlation between cell type signatures in the Xenium and Chromium V2 platforms for Xenium-BreastCancer1 data (Fig. 3 a). We observed that the performance of correlation for average expression between the spatial and sequencing profile ranges between 0.72 and 0.8 across all methods. Interestingly, we observe a trade-off between the size of the cell (average total transcript per cell) and the level of correlation. Fig. 3 a demonstrates the importance of employing two metrics to quantify segmentation performance. While Cellpose achieved the highest Pearson correlation overall, BIDCell achieved the highest Pearson correlation among methods that detect a similar number of transcripts as Chromium data (i.e., cell sizes that are more similar to segmentation of the cell body as opposed to the nuclei). Similar results are shown in the average percentage of expressed genes (Fig. 3 b). Furthermore, Fig. 3 c highlights a high level of consistency in cell type proportion between the segmented cells generated by BIDCell and Chromium (cor = 0.95). BIDCell also has a higher presence of positive markers and a lower presence of negative markers in large cells (Fig. 3 d and Supplementary Fig. 13 ), demonstrating an improvement in the expression purity of segmentation. In category III of CellSPA, we investigate the potential contamination between neighbouring cells by comparing the percentage of B cells that expressed negative markers, such as CD3D and CD3E, which are positive T cell markers but are considered negative markers in B cells. The presence of T cell marker genes in B cells suggests potential contamination during the cell segmentation process. Fig. 3 e and Supplementary Fig. 14 indicate that BIDCell showed the smallest percentage of contamination cells, indicating its ability to reduce contamination in a densely populated region. Lastly, we investigate the spatial diversity by examining the association between the cell type composition and the various cell level characteristics of spatial local regions. Here, we expect the region with a diverse composition of cell types would have a high variety of cell sizes and morphologies. We first divide the image into several local regions and then quantify the diversity of the cell type composition of a region using entropy (Fig. 3 f). As shown in Fig. 3 g and Supplementary Fig. 15 , we find that BIDCell achieves a higher correlation of the coefficient of variation of the cell-level characteristics (the total transcripts, the total genes expressed and cell area) with the cell type entropy compared to the other methods. Similarly, we observe that the variety of cell elongation in BIDCell is highly correlated with the proportion of fibroblasts, one of the dominant cell types in the data (Fig. 3 h). Together, with a comprehensive benchmarking using CellSPA, we demonstrate that the BIDCell segmentation achieves a better balance between high cell expression purity and a large cell body compared to the other state-of-the-art methods, which capture a more diverse range of cell morphologies and provide the potential for a more accurate representation of the topographic context of neighbouring cellular interactions. BIDCell is replicable and generalisable to multiple SST platforms As an additional sensitivity analysis to the ablation study, we evaluated the replicability of BIDCell. We compared the results between the two replicated studies (Xenium-BreastCancer1 and Xenium-BreastCancer2). Figure 3 i displays images of the two replicates, with corresponding cell types highlighted in Fig. 3 j (left panel). The results are very similar, demonstrating that BIDCell is replicable. The tSNE plot in Fig. 3 j (right panel) shows a well-mixed population of cells between the two replicated studies. The high correlation of the cell morphology metrics of segmented cells from BIDCell between the two replicates further confirm the replicability of our method (Supplementary Fig. 16 ). We demonstrate the generalisability of BIDCell to other SST platforms and tissue types by applying BIDCell to data generated by CosMx from NanoString (Fig. 4 a–c, Supplementary Figs. 17 , 18 ) and MERSCOPE data from Vizgen (Fig. 4 d–f, a–c, Supplementary Figs. 19 , 20 ). In particular, we observed that BIDCell had a lower percentage of B cells expressing negative markers (markers indicating contamination) for the CosMx-Lung data (Fig. 4 c), suggesting more accurate cell segmentation. Additionally, in MERSCOPE-Melanoma data, regions with more diverse cell types corresponded to more diverse cell type characteristics (Fig. 4 f). Furthermore, we also applied BIDCell to Stereo-seq from BGI (Supplementary Fig. 21 ). We have now demonstrated the applicability of BIDCell on data from four major platforms, and from five different tissue types. We believe that our method has the flexibility and generalisability to other data from other SST platforms and tissues. Accurate cell segmentation can reveal region-specific subtypes among neuronal cells To further assess the performance of BIDCell in accurately segmenting closely packed cells, we performed an evaluation on another case study from Xenium-MouseBrain data. The hippocampus is critical for learning and memory 20 , and the tripartite synapses formed between the dentate gyrus and cornu ammonis (CA) have been well studied 21 . Because of the density of pyramidal neurons within the CA region, we asked whether or not BIDCell could accurately distinguish CA1, 2, and 3 from one another. Figure 5 a, b show the spatial image and highlight the neuronal cell type and neuronal regions using scClassify trained existing sequencing data (Table 1) . Fig. 5 c compares the segmentation pattern obtained using 10x vs. BIDCell. Note that BIDCell generates a more finely textured and tighter pattern of cells than 10x, and the output more closely resembles the pattern seen in Fig. 5 a. The superior performance of BIDCell is further confirmed by the evaluation metrics. With similar size of the segmented cells with 10x (Supplementary Fig. 22 ), BIDCell achieves a higher similarity with scRNA-seq and expression purity score (Fig. 5 d–e, Supplementary Fig. 23 ). BIDCell can identify neuronal subtype markers that distinguish granule neurons in the dentate gyrus ( Prox1 ) from pyramidal neurons in CA1-3 ( Neurod6 ) ( 22 ; Fig. 5 f). Furthermore, it is able to spatially subdivide pyramidal neurons in the CA region despite their close proximity to one another. Fig. 5 f shows the expression patterns of Wfs1 in CA1 23 , Necab2 in CA2 24 and Slit2 in CA3 25 , consistent with prior studies. Interestingly, we found a new gene ( Cpne8 ) that is enriched in CA1, consistent with in situ data from the Allen Brain Atlas and illustrates BIDCell’s capacity for biological discovery.
Discussion Here we presented BIDCell, a method for cell segmentation in subcellular spatially resolved transcriptomics data. BIDCell leverages DL with its biologically-informed loss functions that allow the model to self-learn and capture both cell type and cell shape information, while optimising for cell expression purity. Its default components (such as the backbone architecture and use of cell type profiles) may be exchanged for other architectures and Atlas datasets. We have demonstrated the effectiveness of BIDCell by comparing it to state-of-the-art methods and have shown that BIDCell provides better cell body delineation. Moreover, our flexible approach can be applied to different technology platforms, and different gene panels. Our study highlights the potential of BIDCell for accurate cell segmentation and its potential impact on the field of subcellular spatially resolved transcriptomics. The typical approach to leverage advancements in DL relies on ground truth to guide models to learn relationships between inputs and outputs. However, manual annotation of individual pixels is unattainable for SST that contain hundreds of molecular units per pixel, given the time and effort of manual labour. Further, we have shown (e.g., with Cellpose) that models pretrained on other imaging modalities do not transfer well to SST images. BIDCell innovates through its integrated loss functions that inject biological knowledge of cell morphology and expressions, to allow the model to self-learn from the given spatial transcriptomic and DAPI images, and produce superior visual and quantitative performance compared to previous methods. Our loss functions also allow BIDCell to be broadly applicable across diverse tissue types and various SST platforms. Therefore, BIDCell can facilitate faster research outputs and new discoveries. Establishing an easy-to-use evaluation system is crucial for promoting reproducible science and transparency, as well as facilitating further methods development. In CellSPA, we have extended beyond a single accuracy metric and introduced metrics that represent important downstream properties or biological characteristics recognised by scientists. This concept of evaluation by human-recognised criteria is also discussed by the computer vision community as “empirical evaluation” 26 . Another aspect that is often overlooked is related to the practical establishment of benchmarking studies. As benchmarking studies gain recognition, they can be time-consuming due to challenges with software versioning and different operating systems, and different methods may require varying degrees of ease of use and time to adjust the code for comparison. The CellSPA tool is available as a R package with all necessary dependencies, simplifying its installation and usage on local systems, and promoting reproducible science and transparency. Rather than generating a comprehensive comparison of existing methods, which can quickly become outdated, evaluation metrics are generated to allow new methods to be compared to a database of existing methods, without the need to re-implement a large collection of methods. This approach reduces redundancy, allows for direct comparison with state-of-the-art methods, and saves time and effort. Examples of this approach include those for cell deconvolution 27 and simulation methods 28 . A comprehensive evaluation framework is vital when comparing diverse segmentation approaches in the absence of a ground truth. It is important to recognise that different segmentation approaches may purposefully have different priorities and outcomes. For example, a segmentation approach such as a seeded Voronoi tessellation will identify larger cells than a fixed expansion around the nuclei, such as Cellpose cell. The former will typically assign more transcripts and produce a denser map of which cells are touching. In contrast, the latter may produce more homogenous profiles of the cells with fewer assigned molecules and tighter cell boundaries, limiting its capability to estimate physical cell interactions. While achieving more homogenous cell bodies is desired, it can also result from the arbitrary over-segmentation of nuclei. This emphasises that the use of employing a variety of metrics to quantify segmentation performance enables a systematic assessment and reveals the desirable properties of each approach. Cells have a three-dimensional structure, thus analyses in a two-dimensional perspective may achieve limited representation. BIDCell can be further adapted (e.g., via its cell-calling loss) to incorporate cell membrane markers to enhance segmentation. In MERSCOPE data that display cell membrane markers, there is a percentage (25%) of cells that lack nuclei in their segmentation, likely due to being elongated melanocytes or fibroblasts in a section without a nucleus. While platforms like MERSCOPE can utilise cell membrane markers as cell masks to perform cell segmentation, it is necessary to conduct further research to understand whether a cell’s slicing affects the measurement of expression in tissues. Similarly, in the nervous system, a future challenge will be to accurately identify and segment dendritic and axon morphologies. Like melanocytes and fibroblasts, the varied and elongated nature of these cell morphologies will make it challenging to accurately identify cell boundaries in the absence of nearby nuclei. Because of these difficulties, most approaches may instead generate similar results between the segmentation of the whole cell and the corresponding segmentation of the cell nuclei. In conclusion, the development of subcellular spatial transcriptomics technologies is revolutionising molecular biology. We have introduced a deep learning approach that does not require ground truth supervision and incorporates prior biological knowledge by leveraging the myriad of single-cell datasets in Atlas databases. We illustrate that our BIDCell method outperforms the current state-of-the-art cell segmentation methods, and we are able to uncover region-specific subtypes in the brain with explicit highlighting of cell bodies and boundaries. Furthermore, recognising the importance of evaluation, we developed CellSPA, a Cell Segmentation Performance Assessment framework, that covers a wide variety of metrics across five complementary categories of cell segmentation characteristics.
Recent advances in subcellular imaging transcriptomics platforms have enabled high-resolution spatial mapping of gene expression, while also introducing significant analytical challenges in accurately identifying cells and assigning transcripts. Existing methods grapple with cell segmentation, frequently leading to fragmented cells or oversized cells that capture contaminated expression. To this end, we present BIDCell, a self-supervised deep learning-based framework with biologically-informed loss functions that learn relationships between spatially resolved gene expression and cell morphology. BIDCell incorporates cell-type data, including single-cell transcriptomics data from public repositories, with cell morphology information. Using a comprehensive evaluation framework consisting of metrics in five complementary categories for cell segmentation performance, we demonstrate that BIDCell outperforms other state-of-the-art methods according to many metrics across a variety of tissue types and technology platforms. Our findings underscore the potential of BIDCell to significantly enhance single-cell spatial expression analyses, enabling great potential in biological discovery. Subcellular in situ spatial transcriptomics offers the promise to address biological problems that were previously inaccessible but requires accurate cell segmentation to uncover insights. Here, authors present BIDCell, a biologically informed, deep learning-based cell segmentation framework. Subject terms
Supplementary information Source data
Supplementary information The online version contains supplementary material available at 10.1038/s41467-023-44560-w. Acknowledgements The authors thank all their colleagues, particularly at the Sydney Precision Data Science Centre and Charles Perkins Centre for their support and intellectual engagement. Special thanks to Yue Cao, Lijia Yu, Andy Tran, and Bárbara Zita Peters Couto for their contributions in weekly discussions, and to Nick Robertson for his contribution to the BIDCell package. Thanks also go to Brett Kennedy and Daniel Dlugolenski from the 10x Genomics team in Australia for providing the initial motivation in discussions. This work is supported by the AIR@innoHK programme of the Innovation and Technology Commission of Hong Kong to J.Y.H.Y., J.K., E.P., X.F., Y.L. The work is also supported by Judith and David Coffey funding to J.Y.H.Y. and Y.L.; NHMRC Investigator APP2017023 to J.Y.H.Y. and D.M. Australian Research Council Discovery project (DP200103748) to J.K.; Discovery Early Career Researcher Awards (DE220100964) to S.G. and (DE200100944) to E.P. Research Training Program Tuition Fee Offset and Stipend Scholarship to F.A.; Chan Zuckerberg Initiative Single Cell Biology Data Insights grant (2022-249319) to S.G.; and USyd-Cornell Partnership Collaboration Awards to S.G. and D.L. The funding source had no role in the study design, in the collection, analysis, and interpretation of data, in the writing of the manuscript, or in the decision to submit the manuscript for publication. Author contributions J.Y.H.Y. conceived and led the study with design input from E.P. and S.G. X.F. led the development of the method with input and guidance from J.K., J.Y.H.Y., E.P., and Y.L. Y.L. led the development and interpretation of the evaluation framework with input from E.P., S.G., J.Y.H.Y., D.M., and X.F. D.L. performed the data analysis and interpretation of the mouse brain data with input from J.Y.H.Y. and Y.L. Y.L. and X.F. performed all data curation and processing. D.M., C.W., and F.A. contributed to the refinement of the code and evaluation framework with guidance from Y.L. and X.F. All authors contributed to the writing, editing, and approval of the manuscript. Peer review Peer review information Nature Communications thanks Jordao Bragantini, Qinghua Jiang and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Data availability All datasets used in this study are publicly available and were downloaded from the following links (more details including accession codes are provided in Table 1 ). 10x Genomics Xenium breast cancer replicates 1 and 2: https://www.10xgenomics.com/products/xenium-in-situ/preview-dataset-human-breast . 10x Genomics Xenium mouse brain: https://www.10xgenomics.com/resources/datasets/fresh-frozen-mouse-brain-replicates-1-standard . NanoString CosMx NSCLC: https://nanostring.com/products/cosmx-spatial-molecular-imager/nsclc-ffpe-dataset/ . Vizgen MERSCOPE melanoma2: https://info.vizgen.com/merscope-ffpe-solution (requires filling in the form to access). The Stereo-seq E12.5_E1S3 data were downloaded from https://db.cngb.org/stomics/mosta/download/ . Tumor Immune Single Cell Hub 2 (TISCH2) BRCA: http://tisch.comp-genomics.org/gallery/?cancer=BRCA&species=Human . 10x Chromium breast cancer: https://www.10xgenomics.com/products/xenium-in-situ/preview-dataset-human-breast . Allen Brain Map Mouse Whole Cortex and Hippocampus SMART-seq: https://portal.brain-map.org/atlases-and-data/rnaseq/mouse-whole-cortex-and-hippocampus-smart-seq . Human Lung Cell Atlas: https://beta.fastgenomics.org/p/hlca . TISCH-NSCLC: http://tisch.comp-genomics.org/gallery/?cancer=NSCLC&species=Human . TISCH-SKCM: http://tisch.comp-genomics.org/gallery/?cancer=SKCM&species=Human . The mouse embryo reference was downloaded from GEO database under accession code [ GSE119945 ]. The TISCH-BRCA datasets were downloaded from GEO database under accession codes [ GSE110686 ], [ GSE114727 ], [ GSE138536 ], [ GSE143423 ], [ GSE176078 ], [ GSE148673 ], [ GSE150660 ]; from EBI database under accession code [ E-MTAB-8107 ]; and from SRA under accession code [ SRP114962 ]. The original published datasets of HLCA can be accessed under GEO accession number [ GSE135893 ] for Banovich_Kropski_2020; URL [ https://www.synapse.org/#!Synapse:syn21041850 ] for Krasnow_2020; [ GSE128033 ] for Lafyatis_Rojas_2019; URL [ https://explore.data.humancellatlas.org/projects/c4077b3c-5c98-4d26-a614-246d12c2e5d7 ] for Meyer_2019; [ GSE158127 ] for Misharin_2021; [ GSE122960 ] and [ GSE121611 ] for Misharin_Budinger_2018; European Genome-phenome Archive study ID [ EGAD00001005065 ] for Teichmann_Meyer_2019. The TISCH-NSCLC datasets were downloaded from GEO database under accession codes [ GSE117570 ], [ GSE127465 ], [ GSE143423 ], [ GSE148071 ], [ GSE150660 ]; and from EBI database under accession code [ E-MTAB-6149 ]. The SKCM datasets were downloaded from GEO database under accession codes [ GSE115978 ], [ GSE120575 ], [ GSE123139 ], [ GSE139249 ], [ GSE148190 ], [ GSE72056 ], [ GSE134388 ], [ GSE159251 ], [ GSE166181 ], and [ GSE179373 ]. Source data are provided with this paper. Code availability We provide our code for data pre-processing, BIDCell training and inference in https://github.com/SydneyBioX/BIDCell , 10.5281/zenodo.10070794 37 . We provide our CellSPA framework in https://github.com/SydneyBioX/CellSPA , 10.5281/zenodo.10295991 38 . Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:41:59
Nat Commun. 2024 Jan 13; 15:509
oa_package/8d/28/PMC10787788.tar.gz
PMC10787789
38218986
Introduction The surface faulting rupture site will rupture under the appropriate bedrock dislocation and damage the tunnel structure on it. With the building and operation of the cross-tunnel structure in an active fault-intensive area, tunnel designs that avoid or only conceptually cross the active fault have been unable to match the rapid development of tunnel engineering transportation. As a result, it is critical to investigate the site rupture mechanism of a cross-fault tunnel, as well as the failure characteristics of the tunnel construction and its anti-rupture measures. The location of the upper breakpoint of the overburden soil and how to destroy the tunnel structure at the location of the earthquake rupture dislocation have become important practical issues of concern to geological engineering, geotechnical engineering, and tunnel engineering researchers. Currently, the research approaches for studying the failure of cross-fault tunnel structures and their failure mechanisms on site include analyzing earthquake damage cases 1 – 4 , conducting model tests 5 – 12 , and doing numerical simulations 13 , 15 . He et al. conducted an analysis of the data on damage to surrounding rock tunnels in fault rupture zones during the Wenchuan earthquake "5.12" and other significant earthquakes 1 . Their findings indicate that tunnels passing through fault rupture zones are prone to damage during earthquakes. The failure of tunnels in fault zones is primarily caused by the disparity in displacement between the surrounding rock in the fault zone and the more stable portion of the surrounding rock. Zhu et al. 2 conducted a comprehensive review of the research conducted by the Tunnel Fault Disaster Mechanism Research Institute. They examined various models, including the engineering geological model, the mechanical model, and the dislocation displacement model. They also analyzed different methods, such as analytical methods, numerical simulation, and physical simulation. Additionally, the existing fault dislocation prevention and control measures, as well as the design and defense principles underlying them, are summarized. Through induction and classification, Li et al. 5 compared and analyzed the key technologies in the current tunnel engineering physical simulation test system and proposed a test system classification method based on the size of the model. Guo et al. 6 performed a 1:80 scale test using a dislocation model test instrument they designed themselves, together with digital image correlation technology. The objective was to investigate the process of shear rupture in the overburden resulting from the dislocation of a 60° dip-reverse fault. Shen 7 used the design method of flexible joint segmented lining to conduct a shaking table test of a cross-fault tunnel in a hilly environment with a geological similarity ratio of 1:30. It is confirmed that it is applied to the structural design of cross-fault tunnels by improving the deformation adaptability of the tunnel structure. Sabagh et al. 8 conducted a tunnel model test with a similarity ratio of 1:60 on a centrifuge for a reverse fault sand overburden site with a bedrock dislocation angle of 60°. The damage state of the tunnel was evaluated by the gradual increase in permanent ground displacement (PGD). Large-scale box test based on cross-fault tunnel seismic damage. Researchers can simulate tunnel stress under seismic fault dislocation and vibration conditions, investigate tunnel failure characteristics and causes, and devise effective fault prevention and control strategies. Model test analysis and equipment for simulating cross-fault tunnels have garnered attention because of limited on-site seismic damage data, restrictions in research techniques, and funding. However, the test methods are mostly based on the shaking table model test, which simulates the input of near-fault ground motion. However, the shaking table model test cannot simulate the site rock and soil mass rupture caused by the sudden bedrock dislocation of the straight-down fault and the large deformation of the soil near the tunnel foundation or the dislocation of the surrounding rock, and the large deformation of the rock and soil around the tunnel is the main cause of the earthquake damage to the cross-fault tunnel. It is difficult to reasonably simulate the actual cross-fault tunnel structure and its overburden site by using the constant gravity dislocation loading device due to load reasons or the centrifuge model due to the small scale of the test due to the size of the hanging basket box. As a result, we urgently need It is critical to develop a large-scale constant gravity model test device to simulate the dislocation loading mode of seism genic fault bedrock, analyze the response of tunnels across dip-slip faults under different site soil types, investigate the rupture distribution of overburden soil under dip-slip faults and the failure characteristics of internal tunnel structures, and provide basic data and a foundation for reasonably determining the damage and failure mechanisms of tunnels 16 . So, this paper combines the large-scale test device platform of the cross-fault tunnel model developed by the research group to conduct research on the failure mechanism of the cross-fault segmented tunnel structure, as well as its site failure characteristics and anti-rupture measures, providing a certain reference value for the design and operation of the cross-fault tunnel structure.
Conclusion This article adopts the design concept of "fuse", which is to use segmented flexible joint tunnel segments during fault displacement, based on the principle of "sacrificing small for large", to reduce the overall damage of soil and surrounding rock to the length of the tunnel body. It’s analyzed that the failure mechanism of segmented tunnel structures in sandy and cohesive soil sites during low-dip reverse fault displacement. The tunnel structure's failure characteristics in the soil field across the fault cover layer are determined not only by the degree of bedrock dislocation but also by the failure range of soil with different properties under the action of dislocation. As a result, the position and shape of the overburden soil rupture range induced by bedrock dislocation at different angles should be obvious in the design, particularly the longitudinal soil rupture range in the buried depth where the tunnel is placed. Compared with the free field of cohesive soil overburden, the uneven deformation of the surface of the cohesive soil site with tunnel structure becomes slower as a whole. With the increase in bedrock dislocation, the downward deformation at the tunnel structure at the reverse position of the bedrock dislocation dip angle is significant. The variation of internal earth pressure in the hanging wall area is about 2–4 times that of its free field. In the clay, the damage to the arch bottom of the tunnel structure is more serious, and the strain change is about twice that of the vault. From the longitudinal length of the tunnel, the strain range at the bottom of the arch is about 2 times the length of the top of the arch. Compared with the free field of sand overburden, the surface uneven deformation of the sand site with tunnel structure does not slow down as a whole. With the increase in bedrock dislocation, the downward deformation of the tunnel structure in the V-shaped extrusion shear rupture zone composed of bedrock dislocation dip angle and its reverse position is significant. The length of the influence range is about 5 times that of the free field, and the maximum inclination angle is about 1.8 times that of the free field. The internal earth pressure change in the hanging wall area is about 3–5 times that of the free field. For the ground main rupture zone region across the tunnel length direction, the peak acceleration of the tunnel structure is significantly changed to about 3–5 times of the free field, and the long-period effect of the response spectrum in this region is obvious, and the characteristic period can reach 1 s faster. In sandy soil, the damage to the arch bottom of the tunnel structure is more serious, and the strain change is about three times that of the vault. However, from the longitudinal length of the tunnel, the strain variation range at the top of the arch is about 2 times the length of the bottom of the arch. Compared with the tunnel structure site of cohesive soil, the strain change of the tunnel structure in the sandy soil site is more significant, and the arch top is about 2 times and the arch bottom is about 6 times. In addition, the uplift range of the sand cover layer increases, the rupture zone migrates to the right, the lifting height of the soil is higher, and the uneven deformation is obvious. Therefore, the sandy soil site plays an " adding seismic " role in the cross-fault tunnel structure. In the area spanning the ground fault rupture zone, the tunnel structure is affected by soil compression and shear failure. The main structure of the tunnel and the pipeline interface are vulnerable to serious threats. Therefore, it is necessary to take anti-rupture measures, such as over-excavation design, segmented flexible connection, damping joint, damping layer, etc 17 – 20 . This paper utilizes a substantial tunnel model test apparatus to carry out field testing on various soil cover layers over low-dip reverse faults. The experimental data offer valuable insights for the structural design and construction of tunnels across fault rupture.
The fortification system of the tunnel structure spanning the active fault, such as the failure mechanism and fault-resistant design (measures), has not been thoroughly established. In this study, the self-developed cross-fault large-scale bedrock dislocation loading device platform is utilized to carry out the model test of the tunnel structure and soil site of sand and cohesive soil when the low-angle reverse fault dislocation occurs, based on the earthquake damage. The results demonstrate that: (1) When the fault is staggered, the segmented flexible joint tunnel segment is more favorable in the cohesive soil site. (2) When compared to the cohesive soil tunnel structure site, the strain change of the tunnel structure in the sandy soil site is greater, with the vault increasing by roughly two times and the arch bottom increasing by nearly six times. After the tunnel is buried, the uplift range of the sand cover layer grows, revealing uneven deformation, and the rupture zone migrates to the footwall; hence, the sand site plays a “add seismic” role in the cross-fault tunnel structure. (3) Knowing the location and shape of the rupture range of the overburden soil caused by bedrock dislocation under different inclination angles and soil properties is required in the design in order to place the buried depth and segment length of the tunnel reasonably and take fault-resistant measures. Subject terms
Design of a large-scale box test apparatus and scheme for cross-fault rupture tunnel models Large-scale model box test apparatus and instrumentation arrangement The device adopts a self-developed large-scale experimental box platform, including a fault dislocation model platform (soil box, connecting device, base, actuator, and angle support); a bedrock fault dislocation input is simulated using an oil pressure loading device. The test chamber measures 4.96 m long, 1.85 m wide, 1.85 m high, and 1.4 m high. On the front and back of the chamber, a high-strength organic transparent glass with a thickness of 0.025 m is fitted, and three vertical rigid ribs are positioned on the exterior side. The chamber's side is made of 0.015-m-high-strength steel plate. The box's bottom is made up of two totally rigid steel plates, an L-shaped movable steel plate and a fixed steel plate, which are used to model the dislocation of the top and lower plates of bedrock beneath the overburden dirt. All the steel plates and the box are connected by a soft canvas around them, and the loading device is placed on the L-shaped movable steel plate. The right side of the soil box is provided with a strip-through hole for inserting the transverse partition board. The partition board is used to bear the gravity caused by the soil compaction in the upper part of the model to avoid damage to the tunnel model. After the soil is rammed, the partition board is drawn out to better simulate the surrounding rock condition. The loading device adjusts the loading direction through the diagonal brace support on the base to simulate the tunnel structure failure under the fault dislocation at different tilt angles (Figs. 1 , 2 ). The advantage of this test device is that different test schemes can be formulated based on the tunnel's buried depth. By controlling the loading direction of the loading device, the failure of the tunnel structure under different dip angles and different fault dislocations of the normal and reverse faults is simulated in order to study the tunnel structure's failure mechanism under various working conditions. Pilot program design The investigation primarily conducted four iterations of low-dip reverse fault dislocation simulation tests (Table 1 ), including two in a free field and two in a field with a tunnel structure. In this experiment, the tunnel model in the field test of the tunnel structure was constructed by assembling four sections of models with the same standard. The construction process followed the conventional procedures typically used for building tunnels. The steps of each trial can be summarized as follows: Complete the overall installation of the model box, adjust the actuator angle to 45°, and fix it. After the inspection is completed, lay a thin film at the joint between the base plate and the box to prevent soil leakage during the backfilling process. Once everything is in place, start the backfilling process. Layer the soil and compact it. Fill 15 cm of soil each time and compact it to 10 cm. While compacting, bury sensors such as soil pressure gauges and accelerometers in the soil according to the layout of the testing instruments. After compacting to a depth of 100 cm (measured from the surface of the soil), install a top bar displacement meter on the surface of the soil, and place cameras and other equipment above and in front of the soil box. During the tunnel site testing, when compacting to a depth of 70 cm, according to the testing plan for the tunnel site, dig a hole in the soil with a depth of 40 cm and dimensions of 330 × 30 cm. Install strain gauges, displacement meters, and other sensors in the soil to set up the tunnel model. Lift the tunnel model and place it in the dug hole, connect the sections together, and secure them. Install partition boards and backfill the soil, compacting it to a depth of 100 cm. Control the actuator to jack 10 mm for each sub-stage to record the data once. Each test is conducted under 11 different working conditions. In each working condition, ensure that the data acquisition instrument collects data according to the requirements of the sensor, and make sure to film the test using a camera. Model-making Carry out a 1:30 similarity ratio modeling based on the developed multifunctional fault dislocation loading test device platform (Table 2 ). Two types of soil, clay and sand, were used for the tests. Determining the similarity of the soils in the scaled-down modeling tests was challenging. Therefore, typical natural cohesive and sandy soils obtained from the field were used as soil samples. The tunnel structural modeling materials include gypsum, cement, coarse sand, and fine sand. After conducting several proportioning tests based on a 1:30 similarity ratio, the appropriate ratio of these materials for the tunnel structure can be selected (Fig. 3 ). The coefficient of inhomogeneity, Cu, of the sandy soil samples was determined to be 2 through sieve testing and the particle grading curve (Fig. 4 ). The final selection of the cement: gypsum: fine sand: water ratio is 5:5:30:11. The similarity ratio between the reinforcing mesh used in the test and the actual reinforcement in the project is 1:30, and the reinforcing mesh used is a 25 × 25 mm galvanized iron wire mesh with a diameter of 1 mm. The length of each section of the model is 800 mm. Pour the tunnel model according to the specified ratios and ensure that it is adequately cured until it reaches its required strength (Fig. 5 ). Twist together the protruding main bars and fill them with silicone weather-resistant sealant until the desired strength is achieved. Sensor deployment programme The test data acquisition system is based on the system provided by Jiangsu Dong Hua Company. The primary sensors used in the system include the top-bar displacement gauge, earth pressure gauge, and accelerometer. Accelerometers and earth pressure gauges were installed at 30 cm, 55 cm, 90 cm, 100 cm, and the bottom plate of the soil body. Additionally, 11 top bar displacement gauges were positioned on the soil surface. The model strain gauges are primarily arranged in sections C1 to C13. Among them, 8 strain gauges are arranged in sections C9 to C11, 4 strain gauges are placed in sections C7 and C8, and the remaining sections have 2 strain gauges each, positioned at the model's arch top and arch waist. Based on the prediction that soil rupture is concentrated in the V-shaped rupture zone, four top-bar displacement gauges were arranged inside the segmental model interface on both sides. The specific arrangement of the sensor locations can be determined through a combination of pre-tests and analysis conducted by the subject group (Fig. 6 , 7 , 8 ). Comparative Analysis of Cross-Fault Tunnel Site Tests Comparative analysis of sites with a 45° inclination clay overburden on reverse faults and sites containing tunnels on them The clay site and its corresponding tunnel site tests, i.e., test numbers 1 and 3.These included high-precision top-mounted displacement measurements for surface deformation, internal soil pressure monitoring using miniature pressure gauges, testing of active and passive discs for rock foundations, large dynamic acceleration measurement of the box-soil mass, and the tunnel model's strain test employing a small grid large range sensor. To this end, a systematic analysis of monitoring data was carried out. The specific analysis is as follows (Figs. 9 and 10 ): The typical loading levels of 30 mm and 70 mm are chosen for analysis by comparing the front views and top views of the free clay site test 1 and the containing tunnel site test 3. Based on the rupture traces, it is observed that the soil cracks in Test 1 are more numerous and wider. When the loading amount reaches 30 mm, the soil on the east side of the front view bulges, and cracks appear on the surface of the soil on the west side. Additionally, as the loading amount increases, a crack emerges on the right side of the original crack at approximately 100 mm in the top view (Fig. 11 a). During Test 3, when loaded to 30 mm, significant cracks appeared above the soil body on the west side in the front view. Additionally, two noticeable cracks appeared on the west side of the box in the top view, and there was substantial soil deformation (Fig. 11 b). When loaded to 70 mm, two cracks emerged in the soil above the east side of both Test 1 and Test 3. The cracks in Test 1 were wider, while those in Test 3 were longer and developed at a larger angle. On closer observation from the top view, it was noticed that both Test 1 and Test 3 exhibited two rupture zones. The widths of the rupture zones were similar, but in Test 1, there were more, wider, and more prominent tiny cracks on both sides of the main rupture zone. Furthermore, the uplifted area of the rupture zone in Test 3 was higher. In Test 1, cracks started to appear on the ground surface at a loading amount of 10 mm. However, in Test 3, cracks appeared on the ground surface only when the loading amount reached 20 mm. As a result, when compared to tests 1 and 3, test 1 damage earlier (Fig. 11 c,d). In order to conduct a comparative analysis of the soil pressure on the east and west sides of each layer in Test 1 and Test 3 (Fig. 9 ), it is observed that in Test 1, the soil pressure increases with the loading amount. Notably, at a depth of 300 mm, there is a significant change in the soil pressure. In Test 3, the soil pressure undergoes changes at depths of 300 mm and 550 mm, with the largest change occurring at 550 mm. In Test 1, significant changes in earth pressure are observed in the range of -300 ~ 300 mm from the center of the rupture zone at the position of 300 mm from the bedrock. However, the earth pressure undergoes little change at other positions. In Test 3, at a distance of 550 mm and 300 mm from the bedrock on the west side, the earth pressure exhibits significant changes with the increase in loading. On the east side, the earth pressure changes are not as pronounced with increasing loading. However, at a loading amount of 30 mm and a distance of 300 mm from the bedrock, both the 300 mm and 550 mm positions show higher values of earth pressure. Based on the analysis, it can be concluded that in Experiment 3, the tunnel structure is indeed under a significant threat at depths of 550 mm and 300 mm from the bedrock. Specifically, within the range of -550 mm to 550 mm and -300 mm to 300 mm from the center of the rupture zone, there is a serious risk to the integrity and stability of the tunnel structure. Based on the comparison between Test 1 and Test 3, it is observed that the change in soil pressure at the center in Test 3 is greater than that in Test 1. This suggests that the area surrounding the tunnel structure, especially at the bedrock dislocation, needs to be strengthened in order to enhance the stability of the tunnel. Reinforcing the soil (surrounding rock) around the tunnel structure would be necessary in this case to ensure the safety and integrity of the tunnel under the increased soil pressure. Based on the comparative analysis of high-precision top bar displacement and surface deformation, the displacement difference between Test 1 and Test 3 at a loading amount of 30 mm does not show significant changes. However, as the loading amount increases to 70 mm, the soil body in Test 3 exhibits a larger lifting height compared to Test 1. This shows that the overall uneven deformation in test 3 is relatively large (Fig. 10 , Table 3 ). According to the test 3 comparison diagram of the overall structural damage of the tunnel, the tunnel structure 2/3 and 3/4 pipe interface damage is serious. The western tunnel has minimal damage, while the eastern tunnel is severely damaged. The range of inhomogeneous deformation in test 3 is larger than that in test 1, and the rupture zone region is wider, so the range of soils for the rupture-resistant design of shallow tunnel structures is larger than their free fields. At the same time, this can be analyzed in the structural damage map of the tunnel in Test 3 (Fig. 12 ). Comparative analysis of the reverse fault 45° dip sand overburden site and its tunnel-bearing site The sandy soil site and its corresponding tunnel site tests, i.e., test numbers 2 and 4. These included high-precision top-mounted displacement measurements for surface deformation, internal soil pressure monitoring using miniature pressure gauges, testing of active and passive discs for rock foundations, large dynamic acceleration measurement of the box-soil mass, and the tunnel model's strain test employing a small grid large range sensor. To this end, a systematic analysis of monitoring data was carried out. The specific analysis is as follows: Compared with test 2, the rupture zone area of test 4 obviously moved to the east. The scope of the soil uplift containing the tunnel site increased, the rupture zone migrated to the east side, the height of the soil uplift was higher, and the inhomogeneous deformation was obvious, which had the effect of "adding seismic" to the tunnel structure (Fig. 18 ). Test 2 When the loading amount was 30 mm, the soil body was lifted about 30 mm, but there was no soil rupture in the top view (Fig. 18 a); when the loading amount of 70 mm, the sand soil was on the east side of the soil above a transverse crack. The top view found that the rupture was mainly concentrated on the east side, with the emergence of horizontal and vertical cracks and the soil bulging obvious (Fig. 18 c). Test 4 When the loading amount was 30 mm the obvious soil bulged, top view found from the east side of the box 800 mm from the development of fine cracks to the middle position but did not penetrate the soil; near the edge of the box body, there is a crack that develops obliquely to the east side (Fig. 18 b); loading to 70 mm did not appear obvious cracks; top view found that there are two cracks, the east side of the top of the crack is close to the edge of the box, and the east side of the bottom of the cracks away from the edge of the box of about 800 mm (Fig. 18 d). Cracks appeared on the surface of test 2 at a loading amount of 40 mm, and cracks appeared on the surface of test 4 at a loading amount of 30 mm. The surface of test 2 was damaged before that of test 4. Comparative analysis of soil pressure in tests 2 and 4(Fig. 13 ) shows that the soil pressure in test 2 mainly changes in the middle of the soil body, with the most obvious change in the position of 300 mm and not much change in the other positions; test 4, after being put into the tunnel, shows that the change of soil pressure in the middle of the soil body is more significant; the variation of soil pressure on the west side of test 4 is greater than that on the east side; and the change in the position of 550 mm is more than that in the position of 300 mm, and the change in the position of the top of 900 mm basically does not have any change. In Test 2, the change in earth pressure is significant within the range of − 300 mm–300 mm from the center of the rupture zone at the 300 mm position (Fig. 13 ). In Test 4, the change in earth pressure is again evident within the range of − 300 mm–300 mm from the center of the rupture zone at the 300 mm position. Additionally, there is a noticeable change in earth pressure within the range of − 550 mm–550 mm from the center of the rupture zone at the 550 mm position. Thus, both of these ranges suggest potential serious damage to the tunnel structure (Fig. 13 ). Comparing the two tests, Test 4 demonstrates a greater change in earth pressure in the middle of the soil body compared to Test 2. Based on the analysis of the valid data measured by accelerometers, which were placed along the east and west sides of the boundary of the "V" range of influence of 300 mm, 550 mm, and 900 mm, it was found that there was no significant change in acceleration on the east and west sides of tests 2 and 4. Consequently, the peak acceleration in the middle of the soil layer remained relatively unchanged in both tests 2 and 4 (Fig. 14 ). The reaction spectrum is further analyzed. The results show that the platform value of the west side of the soil surface in both Test 2 and Test 4 is larger, with a value of 2.0. The platform value of the bedrock active disk is 1.5 (Fig. 15 ). With an increase in the loading amount, the platform value of the west side of the bedrock active disk increases to 2.5 when the loading reaches 70 mm. However, there is no significant change in the platform value for the east side of the soil surface and the bedrock active disk as the loading amount increases (Figs. 16 , 17 ). Therefore, it can be concluded that in the influence zone of the rupture, the tunnel outside this zone can be designed using the reaction spectrum theory in the conventional seismic design method, as there are no significant changes observed in the platform values for the east side of the soil surface and the bedrock active disk with the increase in the loading amount (Fig. 18 ). In Test 2 and Test 4, the top bar displacement is indeed approximately the same when the loading amount is 30 mm. However, as the loading amount increases, Test 2 shows an increased deformation difference in the rupture zone from the center position, ranging from -900 mm to 600 mm. Furthermore, in the region of 600 mm to 1200 mm, both tests exhibit significant and uneven deformation (Fig. 15 , Table 4 ). Based on the comparative diagram of the overall structural damage of the tunnel in Test 4, it is evident that severe damage occurred at the interface of tubes 2/3 and 3/4 of the tunnel structure. Damage to the western tunnels is less severe, while damage to the eastern tunnels remains severe. The range of inhomogeneous deformation in Test 4 is larger than that in Test 2, and the rupture zone region is wider, so the range of soils designed for rupture resistance of shallow tunnel structures is larger than their free fields. Also, this can be analyzed in the damage map of the tunnel structure in test 4 (Fig. 19 ). The comparison of tests 3 and 4 (Figs. 12 and 19 ), analyzed under the same inclination angle at the sandy soil site and the clay site, reveals a larger rupture range in the sandy soil site tunnel structure. At the tunnel interface, they are all severely damaged. Therefore, when the reverse fault is the same as the fault dip, the sandy soil site is more dangerous to the underground tunnel structure, and both sites need to strengthen the seismic design of the main tunnel structure and joints.
Acknowledgements The authors would like to express their profound gratitude to my supervisor, Jian Yi Zhang, for his invaluable direction and counsel during the study process. My supervisor gave me tolerance and support when I ran into problems, as well as insightful inquiries regarding the direction of my study. My academic and professional progress has benefited immensely from my supervisor's knowledge and instruction, which has been a major factor in this. The authors are appreciative of the financial assistance provided by Science for Earthquake Resilience (XH22021A) and National Natural Science Foundation of China (51608118), which allowed me to successfully finish my study. It assisted in acquiring crucial resources, including equipment, internships, and experimental materials, which greatly enhanced the development and outcomes of my study. Author contributions Z.J.Y Funding acquisition, Supervision, Writing—Review & Editing, Conceptualization. S.Y.J Visualization, Writing—Original Draft, Data Curation, Formal analysis. L.Z.H Project administration, Resources, Data Curation, Methodology. Q.S.H and W.S Resources, Visualization. All authors reviewed the manuscript. Funding National Natural Science Foundation of China,51608118,Earthquake Science and Technology Spark Plan Project, XH22021A. Data availability The datasets used and/or analyzed during the current study available from the corresponding author on reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1281
oa_package/d0/58/PMC10787789.tar.gz
PMC10787790
38218887
Introduction Neurological illnesses are characterized by the progressive and permanent loss of neurons in particular brain areas 1 . Many neurodegenerative disorders are the result of several causes, most likely involving a variety of mechanistic pathways 2 . Monoamine oxidase-B (MAO-B) represents such a pathway and may play a significant role in neurological disorders such as Alzheimer's disease (AD) and Parkinson's disease (PD) 3 . The mitochondrial outer membranes of neurons, glia, and other mammalian cells are closely related to the C-terminal transmembrane polypeptide components of MAOs, also known as flavin adenine dinucleotide (FAD)-carrying enzymes 4 , 5 . For synaptic connections to function appropriately, xenobiotic and biogenic amine oxidation must be stimulated 6 . The three-dimensional shapes of the MAO isoforms MAO-A and MAO-B share 70% identical amino acid residues. With only a change of six amino acids between the 16 active-site residues of the two MAOs, their active-site geometries are also similar 7 – 10 . The exact location of MAO isoforms in the brain is not yet fully elucidated. In contrast to studies that employed cell cultures and suggested MAO-A localization in glial cells, tests in both primate and non-primate species demonstrated that the glial enzyme is primarily present as type B in the intact brain 11 , 12 . MAO-B catalyzes phenyl ethylamine and phenyl methylamine disintegration, whereas MAO-A catalyzes noradrenaline, adrenaline, and serotonin deamination. MAO-A and MAO-B also metabolize dopamine, tryptamine, and tyramine. While MAO-B inhibition increases dopamine levels in the Parkinsonian brain, partially depletes dopaminergic neurons in the substantia nigra pars compacta, and has anti-Parkinsonian effects, selective MAO-A inhibition increases neurotransmitter levels in central nervous system (CNS) noradrenergic and 5-hydroxytryptaminergic neurons 13 – 15 . The mechanism-based inhibitors, selegiline and rasagiline (both MAO-B inhibitors) and clorgyline (a MAO-A inhibitor), are among the isoform-specific inhibitors described. Both MAO isoforms are inhibited by pargyline, a different propargylamine molecule. Other notable irreversible MAO inhibitors include the nonspecific inhibitors phenelzine and tranylcypromine 16 , 17 . The MAO-A inhibitors, toloxatone and moclobemide, and the MAO-B inhibitor, safinamide, are well-known examples of isoform-specific reversible inhibitors 18 – 20 . Dry mouth, nausea, diarrhea, constipation, drowsiness, sleeplessness, dizziness, and light-headedness are the most frequently reported side effects of the current medications used for treatment. When using a patch, skin irritation may also develop at the patch site. The search for novel MAO-A and MAO-B inhibitors has extensively used a variety of heterocycle families as scaffolds, including pyrazolines, chromones, chalcones, xanthines, benzyloxy, thiazoles, coumarins, and their precursors, isatin congeners, thiazolidiniones, and betacarboline 21 – 26 . As a result, isatin was identified as an effective MAO inhibitor. Isatin (Fig. 1 ) is an endogenous small molecule with an indole-containing moiety and exhibits a broad range of biological and pharmacological activities. It comprises a nitrogen atom at position 1 and two carbonyl groups at positions 2 and 3. It further contains two rings: a six-membered aromatic ring and a five-membered antiaromatic ring 27 , 28 . It is widely distributed in the body fluids and different tissues of mammals and occurs naturally in plants 29 . In addition to clinical studies on the anticancer medications Toceranib, Semaxinib, and Orantinib, the Food and Drug Administration (FDA) has approved isatin-based therapies, such as Sunitinib (anti-tumor) and Nintedanib (anti-tumor) 30 – 32 (Fig. 1 ). Isatin reversibly inhibits human MAO-A and MAO-B, with K i values of 15 and 3 μM, respectively 33 . According to previous studies, isatin is located close to the FAD cofactor in the MAO-B substrate cavity. The entrance cavity of the enzyme is free as isatin binds to its substrate cavity 34 , 35 . We hypothesized that the C-3 position could be exploited with hydrophobic moieties to improve MAO efficacy 26 . Therefore, we selected the C-3 position and replaced it with an acyl hydrazone linker and a halogenated phenyl (hydrophobic) moiety. The structural cores of acyl hydrazones, which include two distinctly connected nitrogen atoms, are generally responsible for the physical and chemical properties of these compounds. Therefore, acyl hydrazones are frequently used to develop novel molecules with various functions. Hydrazone derivatives have also been linked to MAO inhibition. Recently, Vishnu et al. synthesized piperonylic hydrazone-based isatin derivatives, however, insignificant interactions were observed in 3,4-methylenedioxy groups with MAO-B binding pocket 36 . Therefore, we replaced piperonylic with phenyl moiety and designed new approach toward acylhydrazone-based isatin derivatives (Fig. 2 ) to get a new family of effective MAO inhibitors in this study.
Materials and methods Chemicals For synthesis, isatin derivatives, hydrazine hydrate, and benzoic acid were purchased from Sigma-Aldrich (St. Louis, MO, USA) and TCI Chemical (Toshima, Tokyo, Japan). Substrates (benzylamine and kynuramine) and reference inhibitors (clorgyline, lazabemide, pargyline, and toloxatone) as well as recombinant human MAO-A and MAO-B were purchased from Sigma-Aldrich. Reversibility test was performed by using Dialyzer DiaEasyTM (6–8 kDa, BioVision, St. Grove, MA, USA). Synthesis Benzoic acid (1 eq.) and hydrazine hydrate (2.5 eq.) were combined, and the reaction was conducted by using a microwave synthesizer (Monowave 50 Synthesizer, Anton-Paar, Graz, Austria) at 200 °C for 10–20 min. Upon completion of the reaction, the product benzohydrazide was recrystallized from methanol. Then, the mixture of isatin or substituted isatin (0.001 mol) and benzohydrazide (0.001 mol) 37 , 38 in methanol, by adding a catalytic amount of acetic acid, was placed in a reaction vial and subjected to the microwave synthesizer at 100–120 °C for 5–10 min. The reaction progress was monitored by using thin-layer chromatography (TLC) with an eluent of ethyl acetate and hexane (50:50). Cold ethanol was used to wash the reaction mixture upon completion, and the resulting product was dried to obtain acylhydrazone-based isatin derivatives (76–96% yield). The synthetic scheme of the isatin derivatives is illustrated in Scheme 1 . MAO-A and MAO-B inhibition studies MAO-A and MAO-B activities were determined using 0.06 mM kynuramine and 0.3 mM benzylamine, respectively 39 . In preliminary kinetic study, the K m values of MAO-A and MAO-B were about 0.039 and 0.20 mM, respectively. Concentrations of the substrates used were around 2.0 times of their K m values for the enzyme assay. The absorbance was measured using the continuous assay method described previously 40 . Compound inhibitions were compared to those of the reference inhibitors of MAO-A (toloxatone and clorgyline) and MAO-B (lazabemide and pargyline). Enzyme kinetics The compounds IC 50 values were calculated using the GraphPad Prism software 5 41 . The selectivity index (SI) values of the compounds were calculated as follows: (IC 50 of MAO-A)/(IC 50 of MAO-B) 42 . The type of enzyme inhibition was determined at five different substrate concentrations (0.0075–0.12 μM of MAO-A and 0.0375–0.6 μM of MAO-B). The inhibitor was used at three concentrations (approximately 0.5, 1.0, and 1.5) times its IC 50 value 41 . Enzyme inhibition patterns and K i values were determined by comparing Lineweaver–Burk (LB) plots and their secondary plots, respectively 40 . Reversibility studies The lead compounds reversibility for MAO-A and MAO-B inhibition was evaluated by comparing the undialyzed and dialyzed values at a concentration of 1.5 × the IC 50 value after incubation for 30 min prior to measurement, as previously described 39 , 43 . The restored activities of the compounds were compared to those of the reference compounds toloxatone and clorgyline (reversible and irreversible inhibitors, respectively) of MAO-A, and lazabemide and pargyline (reversible and irreversible inhibitors, respectively) of MAO-B. Reversibility patterns were determined by comparing the activities of undialyzed (A U ) and dialyzed (A D ) compounds 41 , 43 . Parallel artificial membrane permeability assay (PAMPA) The blood–brain barrier (BBB) permeation abilities of the four lead molecules were analyzed using the PAMPA method 44 , 45 . The detailed procedure is explained in the Supplementary Data. Cytotoxicity, reactive oxygen species (ROS), anti-oxidant, and anti-inflammatory characteristics of the cell line-based assay Cell culture and treatments The National Center for Cell Science (NCCS), Pune, India, supplied the SH-SY5Y-human bone marrow neuroblastoma cell line, which was maintained in accordance with the recommended protocol in DMEM-high-glucose medium (Cat No. AL111, Himedia) supplemented with 10% fetal bovine serum (FBS) and a 1% antibiotic–antimycotic solution at 37 °C in a CO 2 incubator. Subcultures were performed every two days. Briefly, 5 × 10 5 cells/mL were cultured on a plate and incubated for 24 h to promote cell attachment and reach the required cell density. Neuroinflammation was induced in the SH-SY5Y cells with 1 ug/mL for 2 h followed by cell treatment with different concentrations of the test molecules ( IS6 , IS7 , IS13 , and IS15 ) and incubation for 24 h. LPS-treated cells served as positive controls, and untreated cells served as controls. 3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay The cytotoxicity of synthetic compounds IS6, IS7, and IS13 was assessed using the common MTT test using SHSY-5Y cells. The cells were grown in 96-well plates with 20,000 cells per well. Different compound concentrations were applied to the cells, and the treated cells were then left to incubate for 24 h. After applying 50 μL of MTT (0.5 mg/mL) to each well, the plates were incubated for 3 h, then 100 μL of DMSO was added to dissolve the purple formazan crystal, and an absorbance was measured at 570 nm using a microplate reader (ELX-800, BioTek, CA, USA). Compounds' growth inhibitory concentration (IC 50 ) values were computed. Superoxide dismutase (SOD) activity Using a superoxide dismutase assay kit (KrishGen Biosystems, India), which measures the concentration of formazan crystals using a colorimetric assay, SOD activity was determined. In this test, a tetrazolium salt was used to detect the superoxide radicals produced by xanthine oxidase and hypoxanthine. After treatment, the medium was removed, centrifuged at 2000 rpm for 5 min at room temperature, and placed on ice. A diluted radical detector and lysed cell supernatant or a standard were applied to each well of a 96-well plate to evaluate SOD activity using an ELISA kit, according to the manufacturer’s instructions. The absorbance of the wells was then determined after 5 min at a wavelength of 450 nm using a microplate reader (Safire2, Tecan Group Ltd., Maennedorf, Switzerland). The results were displayed as ng/mL 46 – 49 . Glutathione (GSH) activity A glutathione assay kit (KrishGen Biosystems) was used to determine the GSH levels. The assay kit was based on the disulfide dimer-oxidized GSH reductase recycling method for 5,5'-dithiobis-2-(nitrobenzoic acid) (DTNB). After treatment, adherent cells were removed by scraping the media from the wells. Following suspension in 50 mM phosphate solution (0.5 mL) with a pH of 6.5 and 1 mM ethylenediaminetetraacetic acid, the cells were chilled. The lysed cell supernatant was used to test GSH levels using an ELISA kit. The absorbance of the yellow product was determined at a wavelength of 450 nm. The total GSH activity was estimated using a GSH standard curve. The results were obtained as ng/mL 46 – 49 . Glutathione peroxidase (GPx) activity A GPx assay kit (KrishGen Biosystems) was used to evaluate the GPx activity. The kit uses a colorimetric assay to determine the quantity of GPx. Glutathione reductase (GR) mediates GPx activity Oxidized glutathione (GSSG) is produced via hydroperoxide reduction by GPx. This glutathione is recycled back to its reduced state by GR and NADPH. The NADPH to NADP + oxidation was accompanied by a decrease in absorbance at 450 nm (A450). When GPx activity was rate-limiting, the rate of decline in A450 was directly correlated with GPx activity. Following treatment, adherent cells were removed from the wells, suspended in cold PBS, sonicated, and frozen. In accordance with the ELISA kit’s instructions, the lysed cell supernatant, or standard, was applied to all 96 wells of a plate together with a diluted radical detector to assay the activity of GPx. A microplate reader was used to measure the absorbance of the wells after 5 min 46 – 49 . Results were obtained in ng/mL. Catalase (CAT) activity CAT activity was determined according to Aebi 50 . A human CAT ELISA kit was purchased commercially (KrishGen Biosystems). Following treatment, the adherent cells were scraped off, suspended in cold PBS, sonicated, and placed on ice. The medium was then removed from each well. The 3 mL CAT assay combination comprised of extract (0.05 mL), phosphate buffer (1.5 mL, 100 mM buffer, pH 7.0), H 2 O 2 (0.5 mL), and distilled water (0.95 mL). The absorbance decreased at 450 nm. CAT activity was reported in terms of ng/mL of H 2 O 2 oxidized per min per gram 46 – 49 . ROS assays The OxiSelect Intracellular ROS Assay Kit (Cell Biolabs Inc., San Diego, CA, USA) was used to quantify the levels of the fluorescent probe 20,70-dichlorodihydrofluorescin diacetate (DCFH-DA). In a microplate reader, fluorescence was measured using excitation and emission filters at wavelengths of 488 and 535 nm, respectively 51 . IL-6, TNF-α, and NF-kB expression TNF-α, IL-6, and NF-kB expression in cell lysates was assessed using the respective antibodies (PerCP-Cy5.5, PE, and p65–FITC) according to the manufacturer’s protocol. Briefly, the spent medium was aspirated, and the cells were treated with LPS (1 μg/mL) for 2 h. Then, the required concentrations of experimental compounds and controls were added and incubated for 24 h. The cells were harvested into polystyrene tubes and centrifuged at 25 °C, washed with PBS, and 70% cold ethanol was added drop wise to create a cell pellet while vortexing. The mixture was then incubated at − 20 °C. The cells were pelleted at a high speed, washed twice with PBS, antibodies added (10 μL), mixed thoroughly, and incubated for 30 min in the dark at 20–25 °C. PBS (500 μL) was added and mixed thoroughly, and the reaction was analyzed using BD FACS—Cell Quest pro software 46 – 49 , 52 . Statistical analysis Statistical significance was determined by one-way ANOVA followed by Dunnett-t test using Graph Pad Prism Version 8.0.2. Computational studies Molecular docking The Schrödinger suite 53 was used to perform the molecular docking investigation of IS3 , IS6 , IS7 and IS13 , and IS15 . The human MAO-A (hMAO-A, 2Z5Z) and MAO-B (hMAO-B, 2V5Z) X-ray solved structure was obtained from the Protein Data Bank 9 , 54 . The both crystal structures were improved and optimized using the protein preparation wizard included in the Schrödinger suite, which performed energy minimization, hydrogen atom addition, protonation-state correction, and protonation-state addition. The LigPrep tool was used to construct the ligand structures. The co-crystallized ligands served as the automated center of the grid box. For docking simulations, the force Field OPLS4 default settings and extra precision (XP) docking protocol default settings were used 55 , 56 . Molecular dynamic simulation Schrödinger LLC’s Desmond simulation program was used to run the molecular dynamics (MD) simulations 52 . The protein–ligand combination was initially created for the Desmond system builder panel using compound IS7 against MAO-B in an aqueous solvent system. For complete protein–ligand simulations and stability trajectory analysis (RMSD, RMSF, and protein–ligand contact), the simulation parameters were 100 ns at 300 K, 1.01325 bar pressure, and 1000 frames 55 , 56 . MM-GBSA The Generalized Born and Surface Area (MM-GBSA) solvation technique in Molecular Mechanics was utilized to compute the free binding energies of the ligands to the proteins. In this case, we used several postures from MD simulations of the docked complex to evaluate macromolecular stability and protein–ligand binding affinity. The free energy was calculated using the following formula at the post-processing stage that comes after the MD studies. The contributions of the internal, electrostatic, and van der Waals energies to molecular mechanics are denoted by the symbols E int , E ele , and E vdw , respectively. Within the equation, the free energy contributions of the polar and non-polar solvation systems are denoted by G pol and G np , respectively. S is an estimate of the entropy, and T is the absolute temperature. The following formula was used to estimate the binding free energy, or ΔG Bind, between the ligand and the protein. The protein, ligand, and protein–ligand complex are denoted by the letters P, L, and PL, respectively. The equation above expressed the free energy for each of these entities. We used the solvated systems that we obtained prior to performing MD calculations to determine free binding energies. In this case, solvent molecules more than 5 Å away from the bound ligand were replaced with an implicit model via the GB approach, in their post-processing stages 57 , 58 .
Results and discussion Synthesis The target molecules were synthesized in two steps. In the first step, an intermediate acylhydrazide molecule was synthesized by reacting benzoic acid with hydrazine hydrate. This intermediate was then reacted with isatin and halogenated substituted isatins to obtain the final molecules (substituted acylhydrazone-based isatin derivatives: ( IS1 – IS16 ) via an acid-catalyzed nucleophilic addition reaction. All the procedures were performed using the microwave reactor. The structures of all synthesized compounds were confirmed by 1 H and 13 C nuclear magnetic resonance ((Bruker Advance Neo 400 MHz NMR spectrometer). The de-shielded protons in all compounds were NH atoms from isatin, and the hydrazone linker exhibited ranges of 11.5–11.0 δ and 12.50–14.0 δ, respectively. Sharp de-shielded Sp 2 carbonyl carbons of the isatin and hydrazone linkers were observed at 163.60 δ and 141.10 δ, respectively (Supporting Information Figs. S1 – S48 ). MAO-A and MAO-B inhibition studies Of the 16 compounds, IS7 most potently inhibited MAO-B with an IC 50 value of 0.082 μM, followed by IS13 (IC 50 = 0.104 μM) (Table 1 , Fig. S52 ). Compounds IS7 and IS6 ( para -Br and –Cl in the B-ring, respectively) showed higher MAO-B inhibition than the basic compound IS5 (–H in B-ring, IC 50 = 4.136 μM), i.e., –Br > –Cl > –H > –F in order. In contrast, MAO-B inhibition decreased in the order of meta -position substitution in the A-ring, that is, IS5 (–H) > IS13 (–Cl) > IS1 (–Br) > IS9 (–F), suggesting that the meta -F substituent of the A-ring contributed to a decrease in MAO-B inhibition. These IC 50 values were lower than those of the aldoxime- and hydroxy-functionalized chalcones ACE7 and HC6 (IC 50 = 0.012 and 0.0046 μM, respectively) 59 , but higher than those of the dimethoxy-halogenated chalcone DM2 (IC 50 = 0.067 μM) 60 . In contrast, compound IS15 most inhibited MAO-A with an IC 50 value of 1.852 μM, followed by IS3 (IC 50 = 2.385 μM). These values are more efficient than those of the halogenated pyrazoline EH8 (IC 50 = 4.31 μM). Compound IS6 had the highest selectivity index (SI) value (263.8); however, compounds IS7 and IS13 showed similar SI values (SI = 233.85 and 212.57, respectively) and high MAO-B inhibition. These SI values indicated that compounds IS6 , IS7, and IS13 are selective MAO-B inhibitors (Table 1 ). Structurally, compound IS7 (–Br in the B-ring) showed higher MAO-B inhibition than IS6 (–Cl in the B-ring), and both compounds showed 50.4 × and 33.4 × , higher inhibition than the basic compound IS5 (–H in the B-ring), respectively. In the subseries, MAO-B inhibition increased in the following order: Br > Cl > H > F at the para -position of the B-ring. In contrast, in the sub-series containing –Br in the A-ring, IS2 (–Cl in the B-ring, IC 50 = 0.269 μM) showed a higher MAO-B inhibition than the sub-parental compound IS1 (–H in B-ring, IC 50 = 0.420 μM), and the inhibition increased with the substituents of –Cl > –H > –Br > –F at para -position in the B-ring in order). In the other sub-series containing –F in the A ring, IS10 (–Cl in B-ring, IC 50 = 3.995 μM) showed higher MAO-B inhibition than the sub-parental compound IS9 (–H in B), and MAO-B inhibition increased with the substituents of –Cl > –F > –H > –Br at para -position in B-ring in order. In the sub-series containing –Cl in the A-ring, IS13 (–H in the B-ring, IC 50 = 0.104 μM) showed the highest MAO-B inhibition, which increased with the substituents of –H > –Cl > –Br > –F at para -position in the B-ring. In comparing substituents in A ring, MAO-B inhibition increased in order by –Cl ( IS13 , IC 50 = 0.104 μM) > –Br ( IS1 , IC 50 = 0.420 μM) > –H ( IS5 , 4.136 μM) > –F ( IS9 , 9.094 μM), and by –H ( IS7 , 0.082 μM) > –Cl ( IS15 , 0.337 μM) > –Br ( IS3 , 0.514 μM) > –F ( IS11 , 10.586 μM). Overall, most compounds with F substituents showed low MAO-B inhibition (Table 1 , Fig. 3 ). IS7 , IS6 , and IS13 were more selective (SI = 233.85, 263.80, and 212.57, respectively) towards MAO-B. The lead molecules ( IS7 , IS6 , and IS13 ) were comparable to lazabemide and pargyline. In MAO-A inhibition, compound IS15 (–Cl in the A-ring and –Br in the B-ring) was the highest (IC 50 = 1.852 μM) (Table 1 , Fig. S53 ) and showed 11.94-times higher MAO-A inhibition than IS13 (–Cl in the A-ring and –H in the B-ring), and 10.35-times higher than IS7 (–H in the A-ring and –Br in the B-ring). This indicates that the –Cl substituent in the A-ring contributed to an increase in MAO-A inhibition (Table 1 , Fig. 3 ). These results suggest that compounds IS6 , IS7 , and IS13 are potent selective MAO-B inhibitors and that compound IS15 is a selective MAO-A inhibitor. Reversibility studies Reversibility tests were performed using the dialysis method. In this study, the concentration of compound IS15 used for MAO-A was 1.5 × that of the IC 50 (3.00 μM), and those of compounds IS6 , IS7 , and IS13 used for MAO-B were 1.5 × that of the IC 50 (0.18, 0.12, and 0.15 μM, respectively). Recovery patterns were compared using undialyzed (A U ) and dialyzed (A D ) relative activity after 30 min of pre-incubation. For MAO-A inhibition, compound IS15 recovered from 47.16 to 78.73% (Fig. 4 ). The recovery of the compound was similar to that of toloxatone (from 33.76 to 87.22%), and it could be distinguished from clorgyline (from 32.32 to 39.23%). For MAO-B inhibition, compounds IS6 , IS7, and IS13 recovered from 42.81 to 79.52%, 28.65–72.89%, and 31.45–80.12%, respectively (Fig. 5 ). The recovery values of the compounds were similar to those of lazabemide (from 41.48 to 77.71%) and could be distinguished from those of pargyline (from 41.04 to 34.34%). These results indicate that IS15 is a reversible inhibitor of MAO-A, whereas IS6 , IS7 , and IS15 are reversible inhibitors of MAO-B. Enzyme kinetics The enzyme kinetics and inhibition types were analyzed at five substrate concentrations and three inhibitor concentrations. In the LB plot, IS15 showed was a competitive MAO-A inhibitor (Fig. 6 A), and the secondary plot revealed that the K i value was 1.004 ± 0.171 μM (Fig. 6 B). In contrast, IS6 , IS7 , and IS13 LB plots indicated competitive MAO-B inhibitors (Fig. 7 A, C, and E), and the secondary plots showed that their K i values were 0.068 ± 0.022, 0.044 ± 0.002, and 0.061 ± 0.001 μM, respectively (Fig. 7 B, D, and F). The K i value of the inhibitor was calculated by the secondary plot constructed with each slope vs. inhibitor concentration in LB plot. The minus value of X-axis of the plot means − K i . Though IS6 and IS7 were not exactly intercepted on one point of Y-axis, V max values in the presence of the inhibitors were almost same within the experimental error range, indicating both also were competitive inhibitors. In the presence of the inhibitors IS6 , IS7 , IS13 , and IS15 , K m values were increased and V max values were the same as the control. These results suggest that IS15 is a competitive MAO-A inhibitor, whereas IS6 , IS7 , and IS13 are competitive MAO-B inhibitors. PAMPA assay The PAMPA assay demonstrated that isatin-based hydrazone derivatives ( IS6 , IS7 , IS13 , and IS15 ) had high permeability and CNS bioavailability, with P e values of > 4.00 × 10 –6 cm/s (Table 2 ). Brain penetration is crucial for the efficient administration of CNS medication 61 . The effective permeability of the chemical and the equation were used to calculate the penetration rate (Log Pe). A compound is categorized as potentially permeable (CNS+), if its P e value is > 4.00 × 10 –6 cm/s, and perhaps non-BBB permeable (CNS-), if < 2.00 × 10 –6 cm/s. This study showed that while halogenated isatin has BBB permeability, the substitution of the phenyl ring results in greater penetration. Chloro substitution resulted in higher BBB permeability, as revealed in this study. MTT, ROS, anti-oxidant, and anti-inflammatory characteristics of the cell line-based assay Cytotoxic activity of IS6 , IS7 , and IS13 was tested by MTT assay using the human neuroblastoma SH-SY5Y cells. The cytotoxicity level was assessed by how well the live cells converted the tetrazolium dye into formazan crystals, with the untreated cells serving as the control group. IS6 , IS7 , and I S13 remarkably decreased cell viability at 100 μM. However, IC 50 values of IS6 , IS7 , and 1S13 were 75.72, 97.15, and 85.30 μM, respectively, indicating that these were cell- proliferative and non-cytotoxic in nature or working low concentration to SH-SY5Y cells (Fig. 8 ). The preliminary in vitro neuroprotective activity against LPS-induced inflammatory events was tested using the synthesized IS6 , IS7 , IS13 , and IS15 in SH-SY5Y neuroblastoma cell lines. Cell viability assays and ELISA measurements of the intracellular pro-inflammatory cytokines TNF-alpha, IL-6, and NF-kB were used to evaluate neuroprotective activity. A common technique is to incubate SH-SY5Y cell lines with LPS (10 ng/mL) in minimum necessary medium at 37 °C for 24 h to induce neuroinflammation. We previously discovered that a 2 h incubation period with 1 ug/ml of LPS is adequate to trigger inflammatory responses. LPS treatment significantly raised the levels of IL-6, TNF-alpha, and NF-kB in LPS-intoxicated SH-SY5Y cell lines compared to control SH-SY5Y cell lines, demonstrating the magnitude of inflammatory reactions mediated by LPS toxicity (Fig. 9 ). IS6 , IS7 , IS13 , and IS15 pretreatment significantly ( p < 0.0001) reduced TNF-alpha (Fig. 9 A), IL-6 (Fig. 9 B), and NF-kB (Fig. 9 C) levels in comparison to the LPS-treated group, demonstrating the anti-inflammatory potential of isatin derivatives (Figs. S49 , S50 ). Intriguingly, in LPS-treated cell lines, IS6 and IS7 significantly decreased the levels of TNF-alpha and IL-6 compared to IS13 and IS15 . When compared to IS6 and IS7 , IS15 and IS13 have significantly lower NF-kB expression. The lead compounds confirmed the anti-inflammatory potential of all the compounds in the human neuroblastoma model by inhibiting LPS-induced pro-inflammatory cytokine expression (IL-6, TNF-alpha, and NF-kB). To further confirm the neuroprotective and anti-oxidant effects of the lead compounds, the effect of the lead compounds on decrease of ROS production using the LPS-treated SH-SY5Y cells (Fig. 10 ). The compounds IS6 , IS7 , IS13, and IS15 significantly inhibited ( p < 0.0001) 2′,7′-dichlorofluorescin (DCF) expression in the LPS-induced model compared to cells treated with LPS alone. The maximum concentration of test compounds that significantly inhibited LPS inflammation was considered in the ROS study. The lead compounds, IS15 and IS7 , at 10 μM/mL concentrations, exhibited an effective DCF intensity decrease compared to the LPS-induced cells, whereas IS6 and IS13 , at 10 μM concentrations, exhibited moderate DCF intensity suppression (Fig. S51 ). LPS alone induced 65% of DCF expression. The cellular anti-oxidant assay results suggested that the lead compounds showed significant neuroprotective activity by enhancing the cellular oxidant enzymes CAT, GPx, GSH, and SOD (Figs. 11 , 12 , 13 and 14 ) in the LPS-induced model, while cells treated with LPS alone effectively expressed SOD, CAT, GSH, and GPx activities. The obtained values confirmed the promising neuroprotective and neuro-inflammatory activity of IS6 , IS7 , IS13, and IS15 compounds (at 10 μM/mL) ( p < 0.0001) in relation to the neuroprotective effect in the LPS-induced human neuroblastoma model by enhancing the enzyme activity and inhibiting the oxidative stress-induced apoptosis caused by LPS by determining DCF intensity. Molecular docking Molecular docking studies were performed to better understand the binding processes of lead compounds. Interactions occurred between the molecules IS3 , IS6 , IS7 , IS13 , and IS15 , with MAO-B (2V5Z) and MAO-A (2Z5X). Native ligands were used to confirm docking 57 . The compounds were mentioned in Table 3 , docking scores (XP mode) ranges were − 6.55 to − 10.85 kcal/mol. The scores of the best lead molecules IS6 , IS7 , and IS13 , through biological evaluation, were − 9.47, − 9.88, and − 9.72 kcal/mol, respectively, while safinamide had a score that was comparable (− 10.85 kcal/mol). The docking scores showed similar affinity like biological activity (IC 50 ). When looking into the orientations, the fluorobenzyl group of safinamide was placed towards the opening of the cavity, whereas the amide side chain pointed in the direction of the FAD molecule. A similar orientation was present in the compounds IS3 , IS6 , IS7 , IS13 , IS15 , and isatin (Fig. 15 A), where the phenyl group was positioned towards the cavity opening and the variable isatin moiety was oriented towards FAD. The entrance and substrate cavities of the MAO-B-binding pocket (Fig. 15 B) were completely occupied with all inhibitors. The best lead molecule ( IS7 ) interacted with Tyr60, Pro102, Pro104, Trp119, Phe168, Leu171, Cys172, Tyr326, and Tyr435 were primarily hydrophobic, whereas Gln206 was in polar contact. The isatin moiety demonstrated that Pi-Pi stacking with Tyr398 provided the IS7 -MAO-B protein with complex stability. For MAO-A, the docking score range is − 3.96 to − 8.00 kcal/mol (Table 3 ). The IS15 and Harmine docking scores (XP mode) were − 8.00 and − 6.04 kcal/mol, respectively. Comparing the docking score to each compound’s IC 50 , IS15 had the best profile, followed by IS3 (− 7.65 kcal/mol), IS13 (− 4.93 kcal/mol), IS7 (− 4.32 kcal/mol), and IS6 (− 3.96 kcal/mol). Every molecule was oriented in the same way as its native ligand, and its isatin moiety was always closed to the substrate cavity (FAD) (Fig. 16 A). Thorough analysis of compound IS15 in the MAO-A active site revealed that it was present at the following positions: Leu97, Ala111, Ile180, Asn181, Tyr197, Ile207, Phe208, Ser209, Val210, Gln215, Cys323, Ile325, Thr326, Ile335, Leu337, Tyr407, and Tyr444 (Fig. 16 B). Through hydrophobic contacts, the phenyl ring of IS15 interacted with Leu97, Ala111, Val210, Ile325, and Cys323. Residues Ile180, Asn181, Tyr197, Tyr407, and Tyr444 interacted with the isatin ring through hydrophobic and polar interactions. Its similar interaction with IS15 and harmine 62 during the binding contact demonstrated its potential to inhibit MAO-A. Molecular dynamic simulation Desmond MD simulations were used to follow the binding mode of IS7 in the inhibitor-binding cavity (IBC) of MAO-B. Protein C-alpha and ligand were tracked within an acceptable range for a long simulation duration (100 ns) according to root mean square deviation (RMSD) analysis. In contrast to the protein RMSD, the ligand (red) RMSD remained steady after 25 ns. The protein RMSD ranged between 1.2 and 3.6 Å with an average of 2.54 ± 0.01 Å (Fig. 17 A). The protein-specific RMSD for the simulation was constant, with the exception of a slight variation, reaching a maximum of 3.6 Å at 68–70 ns, where after it stabilized. The simulation evaluated the flexibility of the protein system by computing the RMSF for each amino acid residue of the protein. The 480–498 residues of MAO-B showed a larger fluctuation. The atoms in the benzoyl ring of the RMSF ligand (Fig. 17 B) showed slight fluctuations during the binding process. The 21 amino acid residues that interacted with the ligand were Tyr60 (0.541 Å),Gly101 (1.13 Å), Pro102(1.117 Å), Pro104 (0.981 Å), Trp119 (1.15 Å), Leu167 (0.955 Å), Phe168 (0.892 Å), Leu171 (0.638 Å), Cys172 (0.706 Å), Ile198 (0.689 Å), Ile199 (0.833 Å), Ser200 (0.88 Å), Thr201 (0.94 Å), Gln206 (0.618 Å), Ile316 (0.604 Å), Tyr326 (0.544 Å), Leu328 (0.572 Å), Met341 (0.493 Å), Phe343 (0.642 Å), Tyr398 (0.97 Å), and Tyr435 (0.497 Å). Hydrogen bonds, hydrophobic contacts, and water bridges were identified in the interaction histograms of IS7 and MAO-B (Fig. 17 C and D). Over a trajectory of 100 ns, the number of individual interactions between the amino acids and ligand was normalized. Several significant amino acids, including Tyr326 (hydrogen bond, water bridge, and hydrophobic), Tyr398 (hydrogen bond), Leu171 (hydrophobic), Cys172 (hydrogen bond and water bridge), and Ile199 (water bridge and hydrophobic), interact with IS7 . The measured fraction of interactions with Tyr326 was > 1.0. As previously observed 63 , 64 , the hydrophobic interaction of Tyr326 at the active site of MAO-B was significant. Figure 17 C and D depict hydrogen bonding, water bridges, and hydrophobic stability in the ligand–protein complexes. Cys172 forms an 86% hydrogen bond with the carbonyl and NH atoms in the linker between the isatin and benzoyl rings. Tyr398 contributed 49% via hydrogen bonding with the NH atom of the isatin ring. With carbonyl and water molecules, Tyr326 is a 33% active participant in hydrogen bonding. Overall, it is estimated from the trajectory analysis and full MD simulation that the lead compound IS7 will inhibit MAO-B. MM-GBSA From their MD simulation frames, the free binding energy was estimated for the best molecule IS7 with the highest docking energy and activity value prediction. Total average energies of ΔG Bind, ΔG Bind H-bond, ΔG Bind Lipo, and ΔG Bind vdW was − 190.04, − 12.26, − 45.94, − 140.23 for 0–100 ns MD snapshot, respectively. Across all interactions, the ΔG Bind vdW and ΔG Bind Lipo energies exerted the most significant impact on the average binding energy (Table 4 ). The values ΔG Bind vdW for the interactions of IS7 with protein complexes indicated the presence of stable van der Waals interaction with amino acid residues. Consequently, the MM-GBSA calculations, derived from MD simulation trajectories, aligned well with the binding energies computed from the docking results. The molecule exhibited very low free binding energy, indicating its higher binding affinity towards the receptor. Consequently, it can be inferred that IS7 compound exhibited a strong affinity for the MAO-B protein.
Results and discussion Synthesis The target molecules were synthesized in two steps. In the first step, an intermediate acylhydrazide molecule was synthesized by reacting benzoic acid with hydrazine hydrate. This intermediate was then reacted with isatin and halogenated substituted isatins to obtain the final molecules (substituted acylhydrazone-based isatin derivatives: ( IS1 – IS16 ) via an acid-catalyzed nucleophilic addition reaction. All the procedures were performed using the microwave reactor. The structures of all synthesized compounds were confirmed by 1 H and 13 C nuclear magnetic resonance ((Bruker Advance Neo 400 MHz NMR spectrometer). The de-shielded protons in all compounds were NH atoms from isatin, and the hydrazone linker exhibited ranges of 11.5–11.0 δ and 12.50–14.0 δ, respectively. Sharp de-shielded Sp 2 carbonyl carbons of the isatin and hydrazone linkers were observed at 163.60 δ and 141.10 δ, respectively (Supporting Information Figs. S1 – S48 ). MAO-A and MAO-B inhibition studies Of the 16 compounds, IS7 most potently inhibited MAO-B with an IC 50 value of 0.082 μM, followed by IS13 (IC 50 = 0.104 μM) (Table 1 , Fig. S52 ). Compounds IS7 and IS6 ( para -Br and –Cl in the B-ring, respectively) showed higher MAO-B inhibition than the basic compound IS5 (–H in B-ring, IC 50 = 4.136 μM), i.e., –Br > –Cl > –H > –F in order. In contrast, MAO-B inhibition decreased in the order of meta -position substitution in the A-ring, that is, IS5 (–H) > IS13 (–Cl) > IS1 (–Br) > IS9 (–F), suggesting that the meta -F substituent of the A-ring contributed to a decrease in MAO-B inhibition. These IC 50 values were lower than those of the aldoxime- and hydroxy-functionalized chalcones ACE7 and HC6 (IC 50 = 0.012 and 0.0046 μM, respectively) 59 , but higher than those of the dimethoxy-halogenated chalcone DM2 (IC 50 = 0.067 μM) 60 . In contrast, compound IS15 most inhibited MAO-A with an IC 50 value of 1.852 μM, followed by IS3 (IC 50 = 2.385 μM). These values are more efficient than those of the halogenated pyrazoline EH8 (IC 50 = 4.31 μM). Compound IS6 had the highest selectivity index (SI) value (263.8); however, compounds IS7 and IS13 showed similar SI values (SI = 233.85 and 212.57, respectively) and high MAO-B inhibition. These SI values indicated that compounds IS6 , IS7, and IS13 are selective MAO-B inhibitors (Table 1 ). Structurally, compound IS7 (–Br in the B-ring) showed higher MAO-B inhibition than IS6 (–Cl in the B-ring), and both compounds showed 50.4 × and 33.4 × , higher inhibition than the basic compound IS5 (–H in the B-ring), respectively. In the subseries, MAO-B inhibition increased in the following order: Br > Cl > H > F at the para -position of the B-ring. In contrast, in the sub-series containing –Br in the A-ring, IS2 (–Cl in the B-ring, IC 50 = 0.269 μM) showed a higher MAO-B inhibition than the sub-parental compound IS1 (–H in B-ring, IC 50 = 0.420 μM), and the inhibition increased with the substituents of –Cl > –H > –Br > –F at para -position in the B-ring in order). In the other sub-series containing –F in the A ring, IS10 (–Cl in B-ring, IC 50 = 3.995 μM) showed higher MAO-B inhibition than the sub-parental compound IS9 (–H in B), and MAO-B inhibition increased with the substituents of –Cl > –F > –H > –Br at para -position in B-ring in order. In the sub-series containing –Cl in the A-ring, IS13 (–H in the B-ring, IC 50 = 0.104 μM) showed the highest MAO-B inhibition, which increased with the substituents of –H > –Cl > –Br > –F at para -position in the B-ring. In comparing substituents in A ring, MAO-B inhibition increased in order by –Cl ( IS13 , IC 50 = 0.104 μM) > –Br ( IS1 , IC 50 = 0.420 μM) > –H ( IS5 , 4.136 μM) > –F ( IS9 , 9.094 μM), and by –H ( IS7 , 0.082 μM) > –Cl ( IS15 , 0.337 μM) > –Br ( IS3 , 0.514 μM) > –F ( IS11 , 10.586 μM). Overall, most compounds with F substituents showed low MAO-B inhibition (Table 1 , Fig. 3 ). IS7 , IS6 , and IS13 were more selective (SI = 233.85, 263.80, and 212.57, respectively) towards MAO-B. The lead molecules ( IS7 , IS6 , and IS13 ) were comparable to lazabemide and pargyline. In MAO-A inhibition, compound IS15 (–Cl in the A-ring and –Br in the B-ring) was the highest (IC 50 = 1.852 μM) (Table 1 , Fig. S53 ) and showed 11.94-times higher MAO-A inhibition than IS13 (–Cl in the A-ring and –H in the B-ring), and 10.35-times higher than IS7 (–H in the A-ring and –Br in the B-ring). This indicates that the –Cl substituent in the A-ring contributed to an increase in MAO-A inhibition (Table 1 , Fig. 3 ). These results suggest that compounds IS6 , IS7 , and IS13 are potent selective MAO-B inhibitors and that compound IS15 is a selective MAO-A inhibitor. Reversibility studies Reversibility tests were performed using the dialysis method. In this study, the concentration of compound IS15 used for MAO-A was 1.5 × that of the IC 50 (3.00 μM), and those of compounds IS6 , IS7 , and IS13 used for MAO-B were 1.5 × that of the IC 50 (0.18, 0.12, and 0.15 μM, respectively). Recovery patterns were compared using undialyzed (A U ) and dialyzed (A D ) relative activity after 30 min of pre-incubation. For MAO-A inhibition, compound IS15 recovered from 47.16 to 78.73% (Fig. 4 ). The recovery of the compound was similar to that of toloxatone (from 33.76 to 87.22%), and it could be distinguished from clorgyline (from 32.32 to 39.23%). For MAO-B inhibition, compounds IS6 , IS7, and IS13 recovered from 42.81 to 79.52%, 28.65–72.89%, and 31.45–80.12%, respectively (Fig. 5 ). The recovery values of the compounds were similar to those of lazabemide (from 41.48 to 77.71%) and could be distinguished from those of pargyline (from 41.04 to 34.34%). These results indicate that IS15 is a reversible inhibitor of MAO-A, whereas IS6 , IS7 , and IS15 are reversible inhibitors of MAO-B. Enzyme kinetics The enzyme kinetics and inhibition types were analyzed at five substrate concentrations and three inhibitor concentrations. In the LB plot, IS15 showed was a competitive MAO-A inhibitor (Fig. 6 A), and the secondary plot revealed that the K i value was 1.004 ± 0.171 μM (Fig. 6 B). In contrast, IS6 , IS7 , and IS13 LB plots indicated competitive MAO-B inhibitors (Fig. 7 A, C, and E), and the secondary plots showed that their K i values were 0.068 ± 0.022, 0.044 ± 0.002, and 0.061 ± 0.001 μM, respectively (Fig. 7 B, D, and F). The K i value of the inhibitor was calculated by the secondary plot constructed with each slope vs. inhibitor concentration in LB plot. The minus value of X-axis of the plot means − K i . Though IS6 and IS7 were not exactly intercepted on one point of Y-axis, V max values in the presence of the inhibitors were almost same within the experimental error range, indicating both also were competitive inhibitors. In the presence of the inhibitors IS6 , IS7 , IS13 , and IS15 , K m values were increased and V max values were the same as the control. These results suggest that IS15 is a competitive MAO-A inhibitor, whereas IS6 , IS7 , and IS13 are competitive MAO-B inhibitors. PAMPA assay The PAMPA assay demonstrated that isatin-based hydrazone derivatives ( IS6 , IS7 , IS13 , and IS15 ) had high permeability and CNS bioavailability, with P e values of > 4.00 × 10 –6 cm/s (Table 2 ). Brain penetration is crucial for the efficient administration of CNS medication 61 . The effective permeability of the chemical and the equation were used to calculate the penetration rate (Log Pe). A compound is categorized as potentially permeable (CNS+), if its P e value is > 4.00 × 10 –6 cm/s, and perhaps non-BBB permeable (CNS-), if < 2.00 × 10 –6 cm/s. This study showed that while halogenated isatin has BBB permeability, the substitution of the phenyl ring results in greater penetration. Chloro substitution resulted in higher BBB permeability, as revealed in this study. MTT, ROS, anti-oxidant, and anti-inflammatory characteristics of the cell line-based assay Cytotoxic activity of IS6 , IS7 , and IS13 was tested by MTT assay using the human neuroblastoma SH-SY5Y cells. The cytotoxicity level was assessed by how well the live cells converted the tetrazolium dye into formazan crystals, with the untreated cells serving as the control group. IS6 , IS7 , and I S13 remarkably decreased cell viability at 100 μM. However, IC 50 values of IS6 , IS7 , and 1S13 were 75.72, 97.15, and 85.30 μM, respectively, indicating that these were cell- proliferative and non-cytotoxic in nature or working low concentration to SH-SY5Y cells (Fig. 8 ). The preliminary in vitro neuroprotective activity against LPS-induced inflammatory events was tested using the synthesized IS6 , IS7 , IS13 , and IS15 in SH-SY5Y neuroblastoma cell lines. Cell viability assays and ELISA measurements of the intracellular pro-inflammatory cytokines TNF-alpha, IL-6, and NF-kB were used to evaluate neuroprotective activity. A common technique is to incubate SH-SY5Y cell lines with LPS (10 ng/mL) in minimum necessary medium at 37 °C for 24 h to induce neuroinflammation. We previously discovered that a 2 h incubation period with 1 ug/ml of LPS is adequate to trigger inflammatory responses. LPS treatment significantly raised the levels of IL-6, TNF-alpha, and NF-kB in LPS-intoxicated SH-SY5Y cell lines compared to control SH-SY5Y cell lines, demonstrating the magnitude of inflammatory reactions mediated by LPS toxicity (Fig. 9 ). IS6 , IS7 , IS13 , and IS15 pretreatment significantly ( p < 0.0001) reduced TNF-alpha (Fig. 9 A), IL-6 (Fig. 9 B), and NF-kB (Fig. 9 C) levels in comparison to the LPS-treated group, demonstrating the anti-inflammatory potential of isatin derivatives (Figs. S49 , S50 ). Intriguingly, in LPS-treated cell lines, IS6 and IS7 significantly decreased the levels of TNF-alpha and IL-6 compared to IS13 and IS15 . When compared to IS6 and IS7 , IS15 and IS13 have significantly lower NF-kB expression. The lead compounds confirmed the anti-inflammatory potential of all the compounds in the human neuroblastoma model by inhibiting LPS-induced pro-inflammatory cytokine expression (IL-6, TNF-alpha, and NF-kB). To further confirm the neuroprotective and anti-oxidant effects of the lead compounds, the effect of the lead compounds on decrease of ROS production using the LPS-treated SH-SY5Y cells (Fig. 10 ). The compounds IS6 , IS7 , IS13, and IS15 significantly inhibited ( p < 0.0001) 2′,7′-dichlorofluorescin (DCF) expression in the LPS-induced model compared to cells treated with LPS alone. The maximum concentration of test compounds that significantly inhibited LPS inflammation was considered in the ROS study. The lead compounds, IS15 and IS7 , at 10 μM/mL concentrations, exhibited an effective DCF intensity decrease compared to the LPS-induced cells, whereas IS6 and IS13 , at 10 μM concentrations, exhibited moderate DCF intensity suppression (Fig. S51 ). LPS alone induced 65% of DCF expression. The cellular anti-oxidant assay results suggested that the lead compounds showed significant neuroprotective activity by enhancing the cellular oxidant enzymes CAT, GPx, GSH, and SOD (Figs. 11 , 12 , 13 and 14 ) in the LPS-induced model, while cells treated with LPS alone effectively expressed SOD, CAT, GSH, and GPx activities. The obtained values confirmed the promising neuroprotective and neuro-inflammatory activity of IS6 , IS7 , IS13, and IS15 compounds (at 10 μM/mL) ( p < 0.0001) in relation to the neuroprotective effect in the LPS-induced human neuroblastoma model by enhancing the enzyme activity and inhibiting the oxidative stress-induced apoptosis caused by LPS by determining DCF intensity. Molecular docking Molecular docking studies were performed to better understand the binding processes of lead compounds. Interactions occurred between the molecules IS3 , IS6 , IS7 , IS13 , and IS15 , with MAO-B (2V5Z) and MAO-A (2Z5X). Native ligands were used to confirm docking 57 . The compounds were mentioned in Table 3 , docking scores (XP mode) ranges were − 6.55 to − 10.85 kcal/mol. The scores of the best lead molecules IS6 , IS7 , and IS13 , through biological evaluation, were − 9.47, − 9.88, and − 9.72 kcal/mol, respectively, while safinamide had a score that was comparable (− 10.85 kcal/mol). The docking scores showed similar affinity like biological activity (IC 50 ). When looking into the orientations, the fluorobenzyl group of safinamide was placed towards the opening of the cavity, whereas the amide side chain pointed in the direction of the FAD molecule. A similar orientation was present in the compounds IS3 , IS6 , IS7 , IS13 , IS15 , and isatin (Fig. 15 A), where the phenyl group was positioned towards the cavity opening and the variable isatin moiety was oriented towards FAD. The entrance and substrate cavities of the MAO-B-binding pocket (Fig. 15 B) were completely occupied with all inhibitors. The best lead molecule ( IS7 ) interacted with Tyr60, Pro102, Pro104, Trp119, Phe168, Leu171, Cys172, Tyr326, and Tyr435 were primarily hydrophobic, whereas Gln206 was in polar contact. The isatin moiety demonstrated that Pi-Pi stacking with Tyr398 provided the IS7 -MAO-B protein with complex stability. For MAO-A, the docking score range is − 3.96 to − 8.00 kcal/mol (Table 3 ). The IS15 and Harmine docking scores (XP mode) were − 8.00 and − 6.04 kcal/mol, respectively. Comparing the docking score to each compound’s IC 50 , IS15 had the best profile, followed by IS3 (− 7.65 kcal/mol), IS13 (− 4.93 kcal/mol), IS7 (− 4.32 kcal/mol), and IS6 (− 3.96 kcal/mol). Every molecule was oriented in the same way as its native ligand, and its isatin moiety was always closed to the substrate cavity (FAD) (Fig. 16 A). Thorough analysis of compound IS15 in the MAO-A active site revealed that it was present at the following positions: Leu97, Ala111, Ile180, Asn181, Tyr197, Ile207, Phe208, Ser209, Val210, Gln215, Cys323, Ile325, Thr326, Ile335, Leu337, Tyr407, and Tyr444 (Fig. 16 B). Through hydrophobic contacts, the phenyl ring of IS15 interacted with Leu97, Ala111, Val210, Ile325, and Cys323. Residues Ile180, Asn181, Tyr197, Tyr407, and Tyr444 interacted with the isatin ring through hydrophobic and polar interactions. Its similar interaction with IS15 and harmine 62 during the binding contact demonstrated its potential to inhibit MAO-A. Molecular dynamic simulation Desmond MD simulations were used to follow the binding mode of IS7 in the inhibitor-binding cavity (IBC) of MAO-B. Protein C-alpha and ligand were tracked within an acceptable range for a long simulation duration (100 ns) according to root mean square deviation (RMSD) analysis. In contrast to the protein RMSD, the ligand (red) RMSD remained steady after 25 ns. The protein RMSD ranged between 1.2 and 3.6 Å with an average of 2.54 ± 0.01 Å (Fig. 17 A). The protein-specific RMSD for the simulation was constant, with the exception of a slight variation, reaching a maximum of 3.6 Å at 68–70 ns, where after it stabilized. The simulation evaluated the flexibility of the protein system by computing the RMSF for each amino acid residue of the protein. The 480–498 residues of MAO-B showed a larger fluctuation. The atoms in the benzoyl ring of the RMSF ligand (Fig. 17 B) showed slight fluctuations during the binding process. The 21 amino acid residues that interacted with the ligand were Tyr60 (0.541 Å),Gly101 (1.13 Å), Pro102(1.117 Å), Pro104 (0.981 Å), Trp119 (1.15 Å), Leu167 (0.955 Å), Phe168 (0.892 Å), Leu171 (0.638 Å), Cys172 (0.706 Å), Ile198 (0.689 Å), Ile199 (0.833 Å), Ser200 (0.88 Å), Thr201 (0.94 Å), Gln206 (0.618 Å), Ile316 (0.604 Å), Tyr326 (0.544 Å), Leu328 (0.572 Å), Met341 (0.493 Å), Phe343 (0.642 Å), Tyr398 (0.97 Å), and Tyr435 (0.497 Å). Hydrogen bonds, hydrophobic contacts, and water bridges were identified in the interaction histograms of IS7 and MAO-B (Fig. 17 C and D). Over a trajectory of 100 ns, the number of individual interactions between the amino acids and ligand was normalized. Several significant amino acids, including Tyr326 (hydrogen bond, water bridge, and hydrophobic), Tyr398 (hydrogen bond), Leu171 (hydrophobic), Cys172 (hydrogen bond and water bridge), and Ile199 (water bridge and hydrophobic), interact with IS7 . The measured fraction of interactions with Tyr326 was > 1.0. As previously observed 63 , 64 , the hydrophobic interaction of Tyr326 at the active site of MAO-B was significant. Figure 17 C and D depict hydrogen bonding, water bridges, and hydrophobic stability in the ligand–protein complexes. Cys172 forms an 86% hydrogen bond with the carbonyl and NH atoms in the linker between the isatin and benzoyl rings. Tyr398 contributed 49% via hydrogen bonding with the NH atom of the isatin ring. With carbonyl and water molecules, Tyr326 is a 33% active participant in hydrogen bonding. Overall, it is estimated from the trajectory analysis and full MD simulation that the lead compound IS7 will inhibit MAO-B. MM-GBSA From their MD simulation frames, the free binding energy was estimated for the best molecule IS7 with the highest docking energy and activity value prediction. Total average energies of ΔG Bind, ΔG Bind H-bond, ΔG Bind Lipo, and ΔG Bind vdW was − 190.04, − 12.26, − 45.94, − 140.23 for 0–100 ns MD snapshot, respectively. Across all interactions, the ΔG Bind vdW and ΔG Bind Lipo energies exerted the most significant impact on the average binding energy (Table 4 ). The values ΔG Bind vdW for the interactions of IS7 with protein complexes indicated the presence of stable van der Waals interaction with amino acid residues. Consequently, the MM-GBSA calculations, derived from MD simulation trajectories, aligned well with the binding energies computed from the docking results. The molecule exhibited very low free binding energy, indicating its higher binding affinity towards the receptor. Consequently, it can be inferred that IS7 compound exhibited a strong affinity for the MAO-B protein.
Conclusion We synthesized acylhydrazone-based isatin compounds and evaluated their ability to inhibit MAOs. IS15 was a potent competitive reversible MAO-A inhibitor, whereas IS6 , IS7 , and IS13 were potent competitive reversible and selective MAO-B inhibitors. A CNS permeability study using a PAMPA assay revealed that the lead compounds were BBB-permeable. The lead compounds also exhibit non-cytotoxic, neuroprotective and anti-inflammatory effects. The lead compounds (at a concentration of 10 μM/mL) effectively reduced DCF intensity. Additionally, a docking analysis of MAO-B and IS7 revealed the stability of the complex due to the pi–pi stacking of Tyr326. The Cys172 residue participated in the interaction with the ligand at 86% during dynamic examination. Finally, MM-GBSA energy binding revealed that IS7 provided strong stability to MAO-B protein. Overall, the results of this investigation suggest that the lead compounds, IS7 , IS6 , IS13 , and IS15, may be viable therapeutic agents for the treatment of neurological disorders such as PD.
Sixteen isatin-based hydrazone derivatives ( IS1 – IS16 ) were synthesized and assessed for their ability to inhibit monoamine oxidases (MAOs). All the molecules showed improved inhibitory MAO-B activity compared to MAO-A. Compound IS7 most potently inhibited MAO-B with an IC 50 value of 0.082 μM, followed by IS13 and IS6 (IC 50 = 0.104 and 0.124 μM, respectively). Compound IS15 most potently inhibited MAO-A with an IC 50 value of 1.852 μM, followed by IS3 (IC 50 = 2.385 μM). Compound IS6 had the highest selectivity index (SI) value of 263.80, followed by IS7 and IS13 (233.85 and 212.57, respectively). In the kinetic study, the K i values of IS6 , IS7 , and IS13 for MAO-B were 0.068 ± 0.022, 0.044 ± 0.002, and 0.061 ± 0.001 μM, respectively, and that of IS15 for MAO-A was 1.004 ± 0.171 μM, and the compounds were reversible-type inhibitors. The lead compounds were central nervous system (CNS) permeable, as per parallel artificial membrane permeability assay (PAMPA) test results. The lead compounds were examined for their cytotoxicity and potential neuroprotective benefits in hazardous lipopolysaccharide (LPS)-exposed SH-SY5Y neuroblastoma cells. Pre-treatment with lead compounds enhanced anti-oxidant levels (SOD, CAT, GSH, and GPx) and decreased ROS and pro-inflammatory cytokine (IL-6, TNF-alpha, and NF-kB) production in LPS-intoxicated SH-SY5Y cells. To confirm the promising effects of the compound, molecular docking, dynamics, and MM-GBSA binding energy were used to examine the molecular basis of the IS7 -MAO-B interaction. Our findings indicate that lead compounds are potential therapeutic agents to treat neurological illnesses, such as Parkinson's disease. Subject terms
Supplementary Information
Supplementary Information The online version contains supplementary material available at 10.1038/s41598-024-51728-x. Acknowledgements The authors thank the Indian Council of Social Science Research under the Grant number (File No. 02/90/2022-23/RP/MN). Author contributions B.M. and S.K. planned and designed the study. S.K., J.M.O., A.A. and P.P. carried out the experiment and collected the data. B.M., H.K., and P.P. analyzed the data. H.K. provided technical support. B.M. and H.K. revised the manuscript. All authors approved the final version of the manuscript. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1264
oa_package/c8/1a/PMC10787790.tar.gz
PMC10787791
38218976
Introduction Mucosal pentraxin 2 ( Mptx2 ) has been proposed to be a member of the pentraxin family due to the high homology of its sequences ( ~ 83% identity in amino acid sequences) to C-reactive protein ( CRP ) and serum amyloid P component protein ( SAP ) in other family members 1 . CRP and SAP are involved in defending against pathogenic bacteria 2 – 4 . Mptx2 is strongly regulated by dietary heme and calcium 5 , 6 . Recently, a single-cell ribonucleic acid (RNA) sequencing of small-intestinal epithelium indicates Mptx2 is a novel Paneth cell marker 7 , but its function is still unknown. Paneth cells are specialized intestinal epithelial cells (IECs) that reside at the bases of small-intestinal crypts and are important maintainers of intestinal homeostasis 8 – 10 . Impaired autophagy in Paneth cells alters the expression and secretion of antimicrobial factors and intestinal immune responses 11 – 13 . Therefore, understanding regulatory factors in Paneth cell function is critical to the design of new therapeutic approaches to diseases featuring mucosal inflammation. Autophagy is the process to degrade intracellular entities, such as damaged mitochondria, nuclear fragments, viruses, and bacteria 14 . In the intestinal tract, autophagy is essential to engulf and degrade the invading bacteria, thereby being importantly involved in regulating the intestinal immune response 15 . Indeed, the autophagy-associated genes, including the nucleotide-binding oligomerization domain containing 2 ( NOD2 ), and autophagy-related 16 like 1 ( ATG16L1 ), have been linked to pathogenesis of inflammatory bowel disease (IBD) 16 , 17 . In the current study, to reveal the roles and mechanisms of Mptx2 in intestinal homeostasis, we began by demonstrating that Mptx2 was specifically expressed in Paneth cells. We then generated Mptx2 knockout ( Mptx2 −/− ) mice to study the precise effects of this gene in intestinal inflammation and injury, and to test whether autophagy is involved in mechanisms. Our results showed that Mptx2 deficiency impaired functions of Paneth cells and thus to cause intestinal inflammation and injury.
Materials and methods Generation of Mptx2 knockout mice Mptx2 −/− mice (Δexon 1–2) were generated as in our previous study 18 via genome engineering mediated by clustered regularly interspaced short palindromic repeats (CRISPRs) and CRISPR-associated protein 9 (Cas9) in C57BL/6 J mice. All procedures involving mice were approved by the Institutional Animal Care and Use Committee of Xinhua Hospital School of Medicine, Shanghai Jiao Tong University (Shanghai, China; No. XHEC-C-F-2022-010). We have complied with all relevant ethical regulations for animal use. Production of recombinant Mptx2 protein We produced recombinant Mptx2 protein via the methods described in our previous study 18 . Briefly, the coding sequence of Mptx2 (NM_001205011) was cloned into the pET-28 vector with an N-terminal 6-histone (6-His) tag (Genechem Co., Ltd, Shanghai, China). We induced Mptx2 protein via supplementation of 1 mM isopropyl-β-d-thiogalactopyranoside (IPTG) in BL21 (DE3)–competent cells. Generation of antibody against Mptx2 The protein was purified using a nickel–nitrilotriacetic acid (Ni-NTA) column, a PD MidiTrap G-25 column (GE Healthcare, Chicago, IL, USA), and a Vivaspin 20 centrifugal concentrator (GE Healthcare) according to a published protocol 18 . The purified protein was assessed using Coomassie Brilliant Blue staining. We purified the antibody against Mptx2 via antigen immunoaffinity. Reactivity was assessed using an enzyme-linked immunosorbent assay (ELISA). Dextran sulfate sodium–induced colitis We used 6-week-old Mptx2 −/− mice (female, n = 6; male, n = 10) and their wild-type ( Wt ) (female, n = 9; male, n = 6) littermates for dextran sulfate sodium (DSS)–induced colitis experiments. Mptx2 −/− (female, n = 8; male, n = 7) mice and Wt (female, n = 8; male, n = 8) mice were untreated as controls. Acute colitis was induced by administration of 2% DSS (36–50 kDa; MP Biomedicals, Solon, OH, USA) in drinking water for 7 days. We monitored changes in mouse body weight (BW) daily. To construct the DSS-induced colitis and recovery mouse model, C57BL/6 mice were induced by 3.5% DSS for 7 days and allowed to recover for 2 weeks (day 0, female, n = 8; male, n = 7), (day 1, female, n = 5; male, n = 5), (day 3, female, n = 5; male, n = 5), (day 7, female, n = 5; male, n = 5), (recovery 1 week, female, n = 5; male, n = 5), and (recovery 2 week, female, n = 6; male, n = 5). Lipopolysaccharide–induced systemic inflammation To induce an acute systemic inflammatory response, we injected C57BL/6 mice ~6 weeks old with lipopolysaccharide (LPS) intraperitoneally (i.p.; 5 mg/kg; #G5032; Wuhan Servicebio Technology Co., Ltd., Wuhan, China). Mice were sacrificed at the time points of 0 h (female, n = 8; male, n = 7), 3 h (female, n = 5; male, n = 5), 6 h (female, n = 5; male, n = 5), 18 h (female, n = 5; male, n = 5), 24 h (female, n = 5; male, n = 5), and 48 h (female, n = 5; male, n = 5)after LPS injection. Control mice received normal saline. Methicillin-resistant Staphylococcus aureus peritoneal infection We housed Wt and Mptx2 −/− mice ~6 weeks old in a specific-pathogen–free (SPF) unit with access to tap water and pelleted food ad libitum . The murine-peritonitis model was established according to the previously described protocol 18 , 31 . In brief, we infected Wt (female, n = 5; male, n = 7) and Mptx2 −/− mice (female, n = 4; male, n = 3) i.p. with methicillin-resistant Staphylococcus aureus (MRSA; 1.5 × 10 7 CFU). 2 h post-infection, Wt mice received a single dose of 10 μg/ml Mptx2 recombinant protein (female, n = 7; male, n = 7). Mice were sacrificed 24 h after infection, and tissues were collected for further analysis. Antibiotic treatments We treated mice with gentamicin (GM; 2 g L − 1 ; Shandong LuKang Pharmaceutical, Jining, China) or vancomycin (VCM; 500 mg L − 1 ; Eli Lilly Japan, Kobe, Japan) dissolved in autoclaved drinking water with 2% DSS for 7 days. BW was monitored afterward. GM-DSS-treated mice ( Mptx2 −/− , female, n = 9; male, n = 5; Wt , female, n = 4; male, n = 5); VCM-DSS-treated mice ( Mptx2 −/− , female, n = 6; male, n = 6; Wt , female, n = 6; male, n = 6). Intestinal characterization Intestinal tissues were fixed in 4% paraformaldehyde (PFA) for 24 h and sectioned (4 μm) for hematoxylin and eosin (H&E) staining. We determined villus height and crypt depth using National Institutes of Health (NIH) Image software (NIH, Bethesda, MD, USA) with a microscope (Nikon, Tokyo, Japan). Villus height was measured from five well-oriented villi on each slide; five fields were analyzed per section. We counted goblet cells and quantified mucous secretions using Alcian blue/periodic acid–Schiff (AB/PAS) staining. The count of goblet cells in each villus was calculated from 10 well-oriented villi. Histological score We graded histological changes in the intestinal mucosa as previously described 32 , 33 . Briefly, histological scores were determined blindly based on the sum of epithelial and infiltration scores. Epithelial scores were as follows: 0 = normal; 1 = loss of goblet cells in small areas; 2 = loss of goblet cells in large areas; 3 = loss of crypts in small areas; and 4 = loss of crypts in large areas. Infiltration scores were as follows: 0 = normal; 1= infiltrate around crypt base; 2 = moderate infiltrate reaching the muscularis mucosae; 3 = extensive infiltration reaching the muscularis mucosae; and 4 = infiltration of the submucosa. 5-bromo-2′-deoxyuridine (BrdU) assay We performed a 5-bromo-2′-deoxyuridine (BrdU) assay according to previously described protocols 34 , 35 . Briefly, mice were injected with BrdU (50 mg/kg; Solarbio, Beijing, China) 18 h after LPS treatment and euthanized by CO 2 inhalation 1 h afterward. The entire small intestine was removed, flushed with cold saline, fixed with 4% PFA, and embedded in paraffin. Each group mice count is four ( Mptx2 −/− , female, n = 2; male, n = 2; Wt , female, n = 2; male, n = 2). We stained tissue sections (4 μm) via immunofluorescence (IF) using anti-BrdU antibody followed by examination under a fluorescence microscope (Leica, Wetzlar, Germany). BrdU + cells per crypt were counted for 10 random fields per mouse and then averaged. Transmission electron microscopy We prepared intestinal tissues for transmission electron microscopy (TEM) examination followed protocols 36 , 37 . Briefly, tissues from 6-week-old Mptx2 −/− mice and their Wt littermates were fixed with 2.5% glutaraldehyde (GLUT) at room temperature (RT). We then washed the tissues, postfixed them with 1% osmium tetroxide in 0.05 mol/L sodium cacodylate buffer (pH 7.4) at 4 °C for 2 h, stained them with saturated uranyl acetate for 3.5 h at RT, dehydrated them in graded alcohol, and embedded them in Eponate 12 resin (Ted Pella, Inc., Redding, CA, USA). Sections were then cut with a diamond knife and stained with a saturated solution of uranyl acetate in 50% ethanol and lead citrate. We examined and photographed the sections under a Philips CM120 transmission electron microscope (Philips Healthcare, Bothell, WA, USA) at 80 kV. Scanning electron microscopy We cut ~5 mm 2 gut mucosa from Mptx2 −/− and Wt littermate mice and fixed them with 2.5% GLUT overnight at 4 °C. The tissues were rinsed, dehydrated in ethyl alcohol, dried with carbon dioxide, coated with gold, and examined under a Hitachi S-4800 field emission scanning electron microscope (SEM; Hitachi, Tokyo, Japan). 16S rRNA gene sequencing We extracted total microbial genomic-DNA samples from gut mucosa and feces using a DNeasy PowerSoil Kit (QIAGEN, Inc., Venlo, the Netherlands) per the manufacturer’s instructions. Polymerase chain reaction (PCR) amplification of the bacterial 16S ribosomal-RNA (rRNA) gene V4–V5 region was performed using the forward primer 515 F (5′-GTGCCAGCMGCCGCGGTAA-3′) and the reverse primer 907 R (5′-CCGTCAATTCMTTTRAGTTT-3′). We incorporated sample-specific 7-bp barcodes into primers for multiplex sequencing. PCR amplicons were purified using Agencourt AMPure Beads (Beckman Coulter, Indianapolis, IN, USA) and quantified using a PicoGreen Double-stranded Deoxyribonucleic Acid (dsDNA) Assay Kit (Invitrogen, Carlsbad, CA, USA). After the individual-quantification step, amplicons were pooled in equal amounts, and paired-end (PE) 2 × 300 bp sequencing was performed on an Illumina MiSeq platform with MiSeq Reagent Kit version 3 (Illumina, Inc., San Diego, CA, USA) at Shanghai Personal Biotechnology Co., Ltd. (Shanghai, China). We used the Quantitative Insights into Microbial Ecology (QIIME; https://qiime.org ) version 1.8.0) pipeline to process sequencing data, as previously described 38 . Sequence data were mainly analyzed using QIIME and R software version 3.2.0 (R Foundation for Statistical Computing, Vienna, Austria). Quantitative real-time polymerase chain reaction amplification of 16S rRNA genes After recording weights, we extracted bacterial DNA from colonic content using a QIAamp Fast DNA Mini Kit (QIAGEN). Quantitative real-time PCR (qRT-PCR) was performed on an ABIViiA 7 system (Applied Biosystems [Thermo Fisher Scientific, Waltham, MA, USA]) using an SYBR Green Universal Master Mix Kit (Thermo Fisher). We used the following primers, which were modified from a previous study’s sets 39 : all bacteria , F-5′-CGGTGAATACGTTCCCGG-3′ and R-5′-TACGGCTACCTTGTTACGACTT-3′; Bifidobacterium , F-5′-CTCCTGGAAACGGGTGG-3′ and R-5′-GGTGTTCTTCCCGATATCTACA-3′; Lactobacillus , F-5′-TGGAAACAGRTGCTAATACCG-3′ and R-5′-GTCCATTGTGGAAGATTCCC-3′; Bacteroides , F-5′-GAGAGGAAGGTCCCCCAC-3′ and R-5′-CGCTACTTGGCTGGTTCAG-3′; Prevotella , F-5′-CACRGTAAACGATGGATGCC-3′ and R-5′-GGTCGGGTTGCAGACC-3′; Escherichia/Shigella , F-5′-GAGTAAAGTTAATACCTTTGCTCATTG-3′ and R-5′-GAGACTCAAGCTKRCCAGTATCAG-3′; Helicobacter , F-5′-CTATGACGGGTATCCGGCC-3′ and R-5′-TCGCCTTCGCAATGAGTATT-3′; Staphylococcus , F-5′-TTTGGGCTACACACGTGCTACAATGGACAA-3′ and R-5′-AACAACTTTATGGGATTTGCWTGA-3′; Commensal segmented filamentous bacteria (Com-SFB), F-5′-AGGAGGAGTCTGCGGCACATTAGC-3′ and R-5′-CGCATCCTTTACGCCCAGTTATTC-3′; and murine SFB (Mus-SFB), F-5′-TGAGCGGAGATATATGGAGC-3’ and R-5′-CATGCAACTATATAGCTATATGCGG-3′. Quantitative real-time polymerase chain reaction Total RNA was extracted from intestinal-mucosal tissues using an RNeasy kit (QIAGEN) per the manufacturer’s protocol. We quantified RNA using a NanoDrop spectrophotometer (Applied Biosystems). A High Capacity Complementary DNA (cDNA) Reverse Transcription Kit (Applied Biosystems) was employed for reverse transcription using 2 μg RNA. Subsequently, we performed real-time PCR reactions using a ViiA 7 Real-Time PCR System with PowerUp SYBR Green Master Mix Kit (both, Applied Biosystems). PCR reactions were incubated in a 384-well plate at 95 °C for 10 min, followed by 40 cycles at 95 °C for 15 s and 60 °C for 1 min. All samples were assayed in triplicate, and data were normalized to endogenous control β-actin. We calculated relative RNA expression levels using the ΔΔ Ct method. Primers, which were modified from previous studies 40 – 42 and synthesized by Invitrogen (Shanghai, China), were as follows in Supplementary Table 1 . Western blotting For Western blotting (WB), we homogenized ~50 mg tissue in 500 μL radioimmunoprecipitation assay (RIPA) buffer (Invitrogen, Carlsbad, CA, USA) supplemented with a protease inhibitor cocktail (Servicebio). Bicinchoninic acid (BCA) reagent (Pierce Biotechnology [Thermo Fisher]) was used to determine the protein concentration. Next, we separated equal amounts of protein onto 10% NuPAGE Bis-Tris gels (Invitrogen) and transferred them to polyvinylidene difluoride (PVDF) membranes (MilliporeSigma, Burlington, MA, USA). After blocking in 5% nonfat milk, membranes were incubated with primary antibodies overnight at 4 °C. Information for primary antibodies used in this study was showed Supplementary Table 2 . We then washed the membranes three times with tris-buffered saline containing 0.1% Polysorbate 20 (TBST) and incubated them with secondary antibodies. After final washes of the tissues with TBST, we detected signals using an Electrochemiluminescence (ECL) Reagent Kit (Pierce). All of original WB bands were provided in Supplementary information . Immunofluorescence (IF) assay Immunofluorescence (IF) assay was performed as we described previously 43 . Briefly, the intestinal tissues from were immediately fixed in 4% paraformaldehyde for 24 h and went through dehydration, clearing and paraffin embedding. Sections were mounted on positively charged slides after cutting at 4 μm thick, were then incubated with xylol and descending concentrations of ethanol. After antigen retrieval, blocking was performed using 5% bovine serum albumin for 30 min at room temperature. The primary antibodies were incubated in a humid chamber for overnight at 4 °C. The following day, the slides were incubated with the secondary antibody for 50 min at room temperature away from light after washing with phosphate-buffered saline (PBS). The information for primary antibodies used is listed in Supplementary Table 2 . Statistics and reproducibility Numerical source data for all charts are provided in Methods and Figure legends. Statistical tests were performed using GraphPad Prism 8 Software (GraphPad, San Diego, CA) via two–tailed unpaired t test between two groups and one-way ANOVA for multiple comparisons. Each mouse was assessed as an individual sample. All data were obtained by performing at least 3 independent experiments with representative data shown and expressed as the mean ± standard error of the mean (SEM). P values < 0.05 were considered statistically significant. Significance levels were split further as to ** P < 0.01, *** P < 0.001, **** P < 0.0001. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Results Mptx2 was specifically expressed in Paneth cells As shown in Fig. 1a , Mptx2 messenger RNA (mRNA) was synthesized in the intestines of Wt mice from embryonic day 12.5 (E12.5). The levels Mptx2 mRNA increased significantly starting on postnatal day 0 (P0; Fig. 1a ). Similarly, expression of the Paneth cells marker Lyz1 followed an analogous pattern at the indicated time (Fig. 1a ). Under normal conditions, expression of Mptx2 and Lysozyme was higher in the mucosa of mouse middle (mid) and distal (dis) small intestine than in that of proximal (pro) small intestine or colon, but it was not for intestinal stem cell marker Lgr5 (Fig. 1b, c and Supplementary Fig. 1 ). Consistently, immunofluorescence (IF) staining showed that Mptx2 protein was exclusively expressed in the intestinal mucosa, especially in those of the mid and dis small intestine (Fig. 1d, e ). IF analysis also showed that Mptx2 protein was mainly co-localized with the Paneth cell marker Lyz1 in crypt basements (Fig. 1d, e ). Mptx2 −/− mice were predisposed to intestinal inflammation To examine whether Mptx2 directly affected intestinal homeostasis, we first generated mice lacking the Mptx2 gene ( Mptx2 −/− , Supplementary Fig. 2a ). Average villus height and crypt depth in Mptx2 −/− mice did not differ from those of their Wt littermates (Fig. 2a and Supplementary Fig. 2b ). Mptx2 −/− mice had an elevated number of lymphoid structures at the distal part of the small intestines compared to that of Wt mice (Fig. 2a and Supplementary Fig. 2c ). In agreement with histological findings, inflammatory genes, including Il1b , Ifng , Cxcl2 , Cxcl3 , Cxcr3 , and Cxcl12 , were significantly increased in the small-intestinal mucosa of Mptx2 −/− mice compared with their Wt littermates (Fig. 2b ). Transmission electron microscope (TEM) analysis showed abnormalities in intestinal epithelial intercellular junctions and irregular distribution of microvilli in Mptx2 −/− mice compared with their Wt littermates (Fig. 2c ). As shown Fig. 3 , we observed the autophagy-associated molecules Atg12 , Atg16l1 , and Becn1 mRNA levels significantly decreased in pro small intestines of Mptx2 −/− mice in relation to Wt littermates, but the difference of autophagy marker LC3 ( Map1lc3a ) did not arrive at significant level. (Fig. 3 ). Similarly, ER stress sensor activating transcription factor 4 and 6 ( ATF4 , ATF6 ) mRNA levels especially decreased in the pro small intestines of Mptx2 −/− mice compared with their Wt littermates (Fig. 3 ). Subsequently, downstream markers, including spliced Xbp1 ( sXbp1 ), Chop ( Ditt3 ), ERdj4 , and BiP , were also reduced evidently in the pro small intestines of Mptx2 −/− mice compared with their Wt littermates (Fig. 3 ). Interestingly, the levels of these mRNAs did not altered evidently in the mid, dis small intestines or in the colon (Fig. 3 ). Mptx2 deficiency altered intestinal bacteria composition in mice Scanning electron microscopy (SEM) analysis showed that an increased number of invading bacteria attached to and aggregated over the epithelial surface in the small intestines and colons of Mptx2 −/− mice (Fig. 4a ). We next used 16S rRNA sequencing analysis to explore intestinal bacteria composition in Mptx2 −/− and Wt mice (Supplementary Fig. 3 ). In comparison with their Wt littermates, Mptx2 −/− mice showed greater numbers of operational taxonomic units (OTUs) in the genera Lactobacillus , Bifidobacterium, Akkermansia, Bacteroides , and Prevotella in intestines, but reduced abundances of Staphylococcus , Megasphaera , Coprococcus , and Pseudomonas in feces (Fig. 4b ). Numbers of OTUs in the genera Staphylococcus, Bacteroides, Lactobacillus , and increased in the small-intestinal and colonic mucosa in Mptx2 −/− mice (Fig. 4b ). We further isolated DNA from the small-intestinal mucosa, colonic mucosa, and fecal content and then analyzed it for the presence of bacteria using PCR with bacterial-genus–specific primers. It showed that the numbers of several bacterial genera increased in the intestinal-mucosa of Mptx2 −/− mice compared to that of Wt mice, particularly all bacteria , Bacteroides , and Prevotella (Supplementary Fig. 4 ). Mptx2 knockout aggravated MRSA infection with impairing autophagy in Paneth cell Alcian blue-periodic acid Schiff (AB-PAS) staining showed that Mptx2 −/− mice presented Paneth cell loss compared with Wt littermates (Fig. 5a ). Immunofluorescence (IF) staining-based detection of Lyz1-expressing cells confirmed that Paneth cell count was reduced in the small-intestinal mucosa of Mptx2 −/− mice compared with those of their Wt littermates (Fig. 5b ). Additionally, quantitative teal-time PCR (qRT-PCR) analysis indicated that representative Paneth cell antimicrobial peptides (AMPs), including Lyz1 and Reg3g , decreased in dis small intestinal mucosa of Mptx2 −/− mice, but it did not reach significant level (Fig. 5c ). Representative electron microscope images further showed that Mptx2 −/− mice had more eosinophilic granules (Fig. 5d ). We previously found that administration of recombinant Mptx2 protein ( rMptx2 ) could directly reduce methicillin-resistant Staphylococcus aureus ( MRSA ) load in the bloodstream, peritoneal lavage, liver, kidney, spleen, and ileum 18 . In the current study, we showed that Mptx2 −/− mice were more vulnerable to the MRSA infection in distal small intestine than Wt mice (Fig. 6a, b ). Messenger RNA expression of Paneth cell–derived AMPs Lyz1 , Reg3g , and Defa was significantly reduced in the distal small intestinal mucosa of Mptx2 −/− mice compared with those of Wt mice following MRSA infection (Fig. 6c ). Lyz1 and PCNA protein levels were also reduced in the distal-small intestinal mucosa of Mptx2 −/− mice versus those of Wt mice (Fig. 6d, e ). MAP1LC3 (LC3) is an autophagy marker that can capture and eliminate invading bacteria 19 . IF staining indicated that MRSA infection increased LC3 puncta in distal-small intestinal mucosa. LC3/Lysozyme colocalization puncta was lower in Mptx2 −/− mice compared to those of Wt mice (Fig. 6f, g ). Western blot (WB) confirmed that expression of LC3 was significantly reduced in the distal-small intestinal mucosa of Mptx2 −/− mice compared to those of Wt mice (Fig. 6h, i ). P62/sequestosome 1 (SQSTM1), a autophagosomal cargo for degradation 20 , increased in distal-small intestinal mucosa of Mptx2 −/− mice (Fig. 6h, i ). Mptx2 −/− mice increased susceptibility to lipopolysaccharide (LPS)-induced intestinal injury In our LPS-induced mouse sepsis model, Mptx2 mRNA peaked at 18 h and was gradually silenced by 48 h in the small-intestinal mucosa (Supplementary Fig. 5a ). Nineteen hours after LPS injection, we observed more-severe mucosal injury in the small intestines of Mptx2 −/− mice than in those of their Wt littermates (Supplementary Fig. 5b, c ). In addition, we observed that crypt proliferation was impaired in Mptx2 −/− mice versus their Wt littermates after LPS administration (Supplementary Fig. 6a, b ). Proliferative marker Yap1 , but not Lgr5 , decreased in the small-intestinal mucosa of Mptx2 −/− mice (Supplementary Fig. 6c ). WB analysis first showed the tight junction proteins E-cadherin and ZO-1 reduced in the pro and dis small intestines of Mptx2 −/− mice compared to their control littermates, but it failed to reach a significant difference (Fig. 7a, b and Supplementary Fig. 7 ). The apoptotic marker cleaved caspase-3 increased in the pro and dis small intestines of Mptx2 −/− mice compared to their littermates (Fig. 7a, b ). It also showed that loss of Mptx2 impaired the process of the autophagy featured with Atg5 , Atg12-Atg5 , and LC3 proteins were reduced in the pro and dis small intestinal mucosa of Mptx2 −/− mice compared to those of Wt mice after LPS-treatment, but the p62/ SQSTM1 increased in that of Mptx2 −/− mice (Fig. 7a, b ). Loss of Mptx2 worsened dextran sulfate sodium (DSS)-induced colitis in mice In our DSS-induced colitis and recovery mouse model, high Mptx2 expression occurred in colonic mucosa during the acute colitis and recovery phases (Fig. 8a ). During the process of DSS-induced colitis, Mptx2 −/− mice slightly greater body weight loss than their Wt littermates (Supplementary Fig. 8a ). The length of colons did not altered evidently between Mptx2 −/− mice and their Wt littermates (Supplementary Fig. 8b, c ). Histologically, Mptx2 −/− mice had greater colonic-mucosal damage and more inflammatory infiltration than DSS-treated Wt mice (Fig. 8b, c ). In agreement with histological findings, expression of inflammatory genes, including Ifng and Cxcl2 , was increased in the colonic mucosa of Mptx2 −/− mice after DSS treatment versus their DSS-treated Wt littermates, but the difference of neither Tnfa nor Cxcl12 arrive at significant level (Supplementary Fig. 9 ). Moreover, Mptx2 −/− mice had fewer goblet cells in their colons than Wt mice in the presence of DSS-treatment (Supplementary Fig. 10 ). We also found that oral antibiotics (gentamicin, GM or vancomycin, VCM) affected colonic injury and inflammation in Mptx2 −/− mice after DSS treatment (Fig. 8b, c and Supplementary Fig. 10a, b ).
Discussion To the best of our knowledge, this study confirms Mptx2 acts as a novel marker of Paneth cells, implying that it might play important roles in intestinal inflammation and homeostasis. We firstly showed that Mptx2 protein was selectively expressed in Paneth cells in the normal crypt base and that it increased in response to LPS treatment or MRSA infection. Mptx2 deficiency increased susceptibility to intestinal inflammation and injury might via impairing the autophagy process in Paneth cells. In the normal intestine, we found that Mptx2 was exclusively expressed in the mucosa and at higher levels in the small intestine than in the colon. We recently found that Mptx2 mRNA is also expressed in bone marrow and the spleen 18 . Taken together, these findings suggested that Mptx2 might be involved in the immune response. In a previous study, a single-cell survey of the small-intestinal epithelium revealed that Mptx2 might be a new marker for Paneth cells 7 . Indeed, in this study, we confirmed Mptx2 protein was mainly localized in Paneth cells. Paneth cells are specialized small-intestinal epithelial cells that reside at the bases of crypts and protect the small intestine from enteropathogens by constitutively secreting a broad spectrum of AMPs and bactericidal proteins 8 , 21 . It is reported that autophagy deficiency within the intestinal leads to an aberrant morphology of Paneth cells 11 , 12 , 22 . Our study indicated Mptx2 loss resulted in susceptibility to intestinal inflammation may via the Paneth cells in mice. Salmonella infection has been found to induce expression of AMPs and Mptx2 in Paneth cells 7 . In the current study, Mptx2 mRNA increased in the small-intestinal mucosa shortly after LPS treatment. Therefore, Mptx2 might defend the gut from bacterial infection by modulating Paneth cell. Indeed, we found that loss of Mptx2 not only reduced Paneth cell count but also inhibited expression of AMPs such as Lyz1 and Reg3g . Subsequently, SEM indicated that Mptx2 −/− mice had greater numbers of invading bacteria that attached to and aggregated over the epithelial surface of the intestine. 16S rRNA sequencing showed that loss of Mptx2 altered bacteria composition and caused bacteria including the Staphylococcus , Bacteroides , and Enterococcus increased in the intestinal mucosa. Therefore, it is very likely that Mptx2 defends against invading bacteria via its bactericidal activity and/or by modulating Paneth cell functions. Indeed, we previously found that Mptx2 exerted bactericidal activity against methicillin-resistant Staphylococcus aureus ( MRSA ) both in vitro and in vivo 18 . In the colon, Mptx2 −/− mice aggravated DSS-induced colitis that was ameliorated by GM or VCM treatment, indicating Mptx2 maintain bacteria homeostasis may via its bactericidal activity. It has been reported that Paneth cells secrete lysozyme to counteract bacterial infection via secretory autophagy 23 . In this study, we found that loss of Mptx2 aggravated MRSA infection with inhibiting the autophagy process in Paneth cells. An impaired ER stress/autophagy crosstalk has been strongly linked to inflammatory bowel disease (IBD) 24 – 27 . Conditional deletion of intestinal ER stress-marker Xbp1 leads to a spontaneous enteritis in mice 25 . We observed in the current study that Mptx2 deficiency exaggerated LPS-induced intestinal injury with reducing Xbp1 expression and autophagy process. Paneth cells are intercalated between active intestinal stem cells (ISCs) in the small intestine (SI) of mice and humans 28 . Other studies suggest that Paneth cells can constitute a niche for intestinal stem cells in crypts and modulate the regeneration of the intestinal epithelium 29 , 30 . Therefore, Paneth cells might produce Mptx2 to promote the regeneration of the intestinal epithelium via constituting the niche. Therefore, we suggest that Mptx2 might maintain intestinal homeostasis in three ways: (1) acting as an AMP to kill invading bacteria directly, (2) regulating the secretory functions of beyond regulating the microbiota via secretion of AMPs, and (3) contributing to the intestinal repaired via regulating the autophagy/ER-stress.
Conclusions Our findings revealed that Mptx2 was a novel marker of Paneth cells. Mptx2 deficiency triggered microbiota dysbiosis and increased epithelial invasion by bacteria, leading to greater susceptibility to intestinal inflammation with reducing secretory autophagy. In addition, Paneth cells, which can produce Mptx2 , contributed to the regeneration of the intestinal epithelium. These findings suggested that Mptx2 was essential to the functions of Paneth cells and maintained intestinal homeostasis.
A recent single-cell survey of the small-intestinal epithelium suggests that mucosal pentraxin 2 ( Mptx2 ) is a new Paneth cell marker, but its function and involved mechanism in the Paneth cell are still unknown. Therefore, we create Mptx2 knockout ( Mptx2 −/− ) mice to investigate its precise effects on intestinal homeostasis using models of lipopolysaccharide (LPS), methicillin-resistant Staphylococcus aureus ( MRSA ) peritoneal infection, and dextran sulfate sodium (DSS)–induced intestinal injury and inflammation. We here find that Mptx2 is selectively expressed in Paneth cells in the small intestines of mice. Mptx2 −/− mice have increased susceptibility to intestinal inflammation and injured. Mptx2 deficiency reduces Paneth cell count and expression of antimicrobial factors, leading to altered intestinal bacteria composition. Loss of Mptx2 aggravates MRSA infection–induced damage in the intestine while decreasing autophagy in Paneth cells. Mptx2 −/− mice are more vulnerable to LPS-induced intestinal possibly due to inhibition of the autophagy/endoplasmic reticulum (ER) stress pathway. Mptx2 −/− mice are susceptible to DSS-induced colitis that could be ameliorated by treatment with gentamicin or vancomycin antibiotics. In conclusion, Mptx2 is essential to maintain intestinal homeostasis potentially via regulation of autophagy in Paneth cells. Mucosal pentraxin 2 (Mptx2) is identified as a marker of Paneth cells, and its deficiency is shown to trigger microbiota dysbiosis and increased epithelial invasion by bacteria while also reducing secretory autophagy. Subject terms
Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s42003-024-05785-7. Acknowledgements This work was supported by the Natural Science Foundation of Shanghai (22ZR1480600), Shanghai Key Laboratory of Pediatric Gastroenterology and Nutrition (17DZ2272000). We thank Drs. Hui Cai, Jie Zhou, Jing Zhu and Xinbei Tian from Xin Hua hospital, School medicine of Shanghai Jiao Tong University for their technical supports. Author contributions Y.W., C.S., L.Y., Y.Y., and W.W.: Investigation, Methodology, Visualization. W.Y., W. B., P.S., and D.J.: Software, and Validation. C. W. and X.Y.: Conceptualization, Data curation, Project administration, Resources, Funding acquisition, Supervision, Writing—original draft, Writing—review & editing. Peer review Peer review information Communications Biology thanks Relber Gonçales and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Si Ming Man and David Favero. Data availability The data generated or analyzed during this study are available from the corresponding author upon reasonable request. Source data, as well as statistical analysis for all graphs, are provided in the Excel file Supplementary Data 1 . Source images for representative Western blots shown in figures are provided in Supplementary Figure 11 in Supplementary information . Source data of 16S rRNA gene sequencing is deposited in National Center for Biotechnology Information BioProject (Accession: PRJNA1051373). Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Commun Biol. 2024 Jan 13; 7:94
oa_package/e2/64/PMC10787791.tar.gz
PMC10787792
38218905
Introduction The recent development of mapping technologies such as Hi–C 1 that probes the 3D genome organization reveals that a chromosome is divided into topologically associating domains (TADs) 2 , 3 . TADs are genomic regions where chromatin loci more frequently interact with other chromatin loci within the TAD than with those from outside of the TAD. TADs are functional units for transcriptional regulation by constraining interactions between enhancers and promoters 4 , for example. Although TADs are stable between cell types as revealed by earlier studies 2 , 5 , there is growing evidence for TAD reorganization in diseases 6 – 8 , cell differentiation 5 , 9 , 10 , somatic cellular reprogramming 11 , between neuronal cell types 12 , and between species 13 , 14 . For example, extensive reorganizations of TADs are observed during somatic cell reprogramming, associating with dynamics of transcriptional regulation and changes in cellular identity 11 . TADs are also variable among individual cells, as revealed by single-cell studies 15 – 20 and live-cell imaging 21 . Thus, it is important to identify reorganized TADs through comparative analysis to further understand the functional relevance of 3D genome organization, a major priority of current work in the field 22 . The majority of current methods call a reorganized TAD if at least one of its two boundaries changed between two conditions 11 , 12 , 23 – 28 . These methods enable easy integration with other analysis pipelines and identify reorganized TADs with clear biological interpretation. However, they fail to identify reorganized TADs without changes in boundaries, in addition to lacking statistical tests to differentiate random perturbations and significant structural reorganization of a TAD. Only a few nonparametric statistical methods are proposed to call TAD reorganization 29 – 32 . These methods define the structural similarity of a TAD by statistics from two Hi–C contact matrices, such as the stratum-adjusted correlation coefficient used by DiffGR 30 . Distributions of the statistics on pairs of simulated Hi–C matrices are then used to compute empirical P values. However, these nonparametric statistical methods are conservative (see our own comparison later). TADs in high-resolution Hi–C data are relatively small. The median size of TADs is 185 kb 33 . The small size feature of TADs poses another computational challenge for identifying structurally rewired TADs using low-resolution Hi–C data. Importantly, identifying reorganized TADs using emerging single-cell Hi–C (scHi–C) data is largely under-explored. Other methods are developed for comparing Hi–C matrices at different scales and for different purposes: quantifying similarities of genome-wide Hi–C contact matrices 34 , 35 , identifying differential A/B compartments 36 , and identifying differential chromatin interactions 37 – 39 . However, these methods are not tailored to compare Hi–C contact matrices at the TAD level, which is not optimal for identifying reorganized TADs (see our own comparison later). Therefore, new algorithms are needed to fill these gaps. Here, we develop DiffDomain, a new parametric statistical method for identifying reorganized TADs. Its inputs are two Hi–C contact matrices from two biological conditions and a set of TADs called in biological condition 1. This setting enables straightforward integration of DiffDomain with other analysis pipelines of Hi–C data, such as TAD calling and integrative analysis of multi-omics data. For each TAD, DiffDomain directly computes a difference matrix and then normalizes it properly, skipping the challenging normalization steps for individual Hi–C contact matrices. DiffDomain then borrows well-established theoretical results in random matrix theory to compute a P value. We show that the assumptions of DiffDomain are reasonable. Method comparisons on real data reveal that DiffDomain has substantial advantages over alternative methods in false positive rates and accuracy in identifying truly reorganized TADs. Reorganized TADs identified by DiffDomain are biologically relevant in different human cell lines and disease states. Application to scHi–C data reveals that DiffDomain can identify reorganized TADs between cell types and TADs with differential variabilities among individual cells within the same cell type. Moreover, DiffDomain can quantify cell-to-cell variability of TADs between individual cells. Together, these analyses demonstrate the power of DiffDomoain for better identification of structurally reorganized TADs using both bulk Hi–C and single-cell Hi–C data.
Methods In this section, we first introduce the first part of DiffDomain: a model-based method to identify reorganized TADs. We state the model assumptions and their verification using real Hi–C data before reporting the second part of DiffDomain: classification of reorganized TADs into six subtypes. Last is missing value imputation. Model-based method to identify reorganized TADs In this paper, our aim is to identify a subset of TADs that are reorganized between two biological conditions, such as a pair of healthy and diseased cell lines/tissues. Specifically, given a set of TADs identified in one biological condition, we aim to identify the subset of TADs that are reorganized in another biological condition. To achieve this goal, we develop DiffDomain that takes a set of TADs and their Hi–C contact matrices as the input. The TADs are identified using the Arrowhead method 33 (Supplementary Method 1), and Hi–C contact matrices specific to each TAD region are extracted from the genome-wide KR-normalized Hi–C contact maps unless specified otherwise. The core of DiffDomain is converting the comparison of Hi–C contact matrices into a hypothesis-testing problem on their difference matrix. This difference matrix is modeled as a symmetric random matrix, enabling DiffDomain to borrow well-established theoretical results in high-dimensional random matrix theory. Before explaining the hypothesis testing problem, we first introduce some mathematical notations and normalization operations. For each TAD in biological condition 1, let N denote the number of consecutive and equal-length chromosome bins within the genomic region covered by the TAD. Let represent the symmetric KR-normalized Hi–C contact matrix, where represents the non-negative Hi–C contact frequency between chromosome bins i and j (1 ≤ i , j ≤ N ) in the TAD region in condition 1. In other words, A 1 serves as the Hi–C contact matrix specific to the TAD region in biological condition 1, forming a submatrix within the genome-wide Hi–C contact matrix. Similarly, denotes the KR-normalized Hi–C contact matrix corresponding to the same TAD region but in biological condition 2. It is well-known that the Hi–C contact frequency A i j exponentially decreases with an increased linear distance between bins i and j . We first log-transform the Hi–C contact matrices A 1 and A 2 and compute their entry-wise differences, denoted by D , as shown in Eq. ( 2 ). Values in Hi–C contact matrices A 1 and A 2 could have large differences because of variations in reading depths. For example, the GM12878 Hi–C experiment has 4.76 times more Hi–C contacts than the K562 Hi–C experiment (Supplementary Table 2) . Among the 889 GM12878 TADs on Chromosome 1, the averages in the 889 D s range from 1.479 to 2.611, with a median at 2.229. To adjust for the differences due to variations in read depths, we normalize by standardizing each of its k -off diagonal blocks by where k = j − i , 2 − N ≤ k ≤ N − 2, and are the sample mean and standard deviation of . Here, without abuse of notations, we continue to use D to denote the resulted normalized difference matrix. Note that the normalization is TAD-specific because two different TADs most likely have different and , 2 − N ≤ k N − 2. Besides visualization in Fig. 1 a–c, the effects of the above procedures for both bulk and single-cell Hi–C matrices from the same TAD are also visualized in Supplementary Fig. 1 . Intuitively, if a TAD does not undergo structural reorganization from biological condition 1 to biological condition 2, the differences between A 1 and A 2 are caused by multiple factors, including variations in read depths and random perturbations of 3D genome organization. Thus, we assume that entries in D follow a standard Gaussian distribution, resulting in D being a symmetric random noise matrix with entries that follow a standard Gaussian distribution. Scaling D by , where N represents the number of bins in the TAD, results in exhibiting characteristics typical of a well-studied random matrix known as a generalized Wigner matrix. With the justifications presented in the next subsection, we assume that is a generalized Wigner matrix. The problem of identifying reorganized TADs is reformulated as the following hypothesis testing problem: The largest eigenvalue of , denoted by λ N , converges to 2 with increased N 67 . This result helps us to reformulate the hypothesis testing problem as the following: Under H 0 , θ N = N 2/3 ( λ N − 2), a normalized λ N , converges in distribution to Tracy-Widom distribution with index β = 1, denoted as T W 1 68 . In other words, under H 0 , a TAD does not undergo structural reorganization in biological condition 2. Then, the fluctuations of θ N is governed by Tracy–Widom distribution T W 1 . Thus, we choose θ N as the test statistic and compute a one-sided P value by A smaller P value means that the TAD is more likely to be reorganized in condition 2. For a set of TADs, P values are adjusted for multiple comparisons by a few methods, with the BH method as the default. The pseudocode of our DiffDomain algorithm is presented in Supplementary Method 2 . Model assumptions and their verifications Given two KR-normalized Hi–C contact matrices, DiffDomain computes the normalized difference matrix D , bypassing complicated further normalization of individual Hi–C contact matrices 69 . Thus, DiffDomain makes no explicit assumptions on the individual Hi–C contact matrices. DiffDomain only makes assumptions on the normalized difference matrix D . First, under H 0 , DiffDomain assumes that is a generalized Wigner matrix: a symmetric random matrix with independent mean zero upper diagonal entries. Symmetry is satisfied by because Hi–C contact matrices are symmetric. The independence assumption on the upper diagonal entries is violated by considering the well-known fact that Hi–C contact frequencies positively correlate with each other among nearby chromosome bins. However, the violation of the independence assumption does not substantially alter the properties of and DiffDomain (Supplementary Note 1) . Briefly, the empirical properties of and DiffDomain resemble the following theoretical properties: (i) empirical spectral distribution of a generalized Wigner matrix converging to the well-established semicircle law 70 , (ii) λ N → 2 67 , (iii) unadjusted P values following a uniform distribution when H 0 is true and model assumptions are satisfied (Supplementary Note 1 , Supplementary Fig. 3) . The key result ( 5 ) requires one more assumption. It holds under the condition that the distributions of entries in generalized Wigner matrices have vanishing third-moments as N tends to infinity 71 . After the standardization procedure ( 3 ), approximately follows a Gaussian distribution N (0, 1/ N ) whose third moment is 0, satisfying the vanishing third-moment assumption. Taken together, assumptions of DiffDomain are appropriate. For additional references on the generalized Wigner matrix, please refer to the comprehensive books authored by Bai and Silverstein 72 and Couillet and Liao 73 . Reorganized TAD classification Once a subset of reorganized TADs is identified, the classification of reorganized TADs is critical to interpreting TAD reorganization and linking them to the dynamics of genome functions. Motivated by classifications in previous studies 10 , 30 , 31 , DiffDomain classifies reorganized TADs into six subtypes: strength-change , loss , split , merge , zoom , and complex . TADs are hierarchically organized, as identified by methods such as Arrowhead. Large TADs can subdivide into smaller TADs, and a genomic region may belong to multiple TADs, complicating reorganized TAD classification. To address this, we compare the TAD list in condition 1 with the TAD list in condition 2, utilizing combinations of identical TADs and overlapping TADs between the two conditions to distinguish the distinct reorganized TAD subtypes (Fig. 1 f, Supplementary Method 3) . A brief description of the subtypes is provided below. Strength-change represents that the boundaries of the reorganized TAD are the same in both conditions. Specifically, the reorganized TAD in condition 1 has a one-to-one identical relationship with a TAD in condition 2. Loss represents that the reorganized TAD in condition 1 does not overlap with or be identical to any TADs in condition 2. Split represents that a reorganized TAD in condition 1 is split into at least two TADs in biological condition 2. Specifically, the reorganized TAD has either a one-to-many identical relationship or a one-to-many overlapping relationship with TADs in condition 2. Merge represents that the reorganized TAD has a many-to-one identical or a many-to-one overlapping relationship with a TAD in condition 2. Specifically, the reorganized TAD and at least one of its adjacent/overlapping TADs in condition 1 are identical to or overlap with a single TAD in condition 2. Merge is the opposite of split when condition 2 is treated as condition 1. Zoom represents that the reorganized TAD in condition 1 has a one-to-one overlapping relationship with a TAD in condition 2, addition to not being identical to any TADs in condition 2. Complex represents the other reorganized TADs. After the classification of reorganized TADs into the six subtypes, the strength-change TADs are further subdivided into two categories. Within a strength-change TAD, the Hi–C contact frequencies either increase or decrease in biological condition 2, after proper normalization on the differences in total sequenced reads. Subsequently, a strength-change TAD can be classified into a strength-change up TAD or a strength-change down TAD. Before explaining the classification, a few mathematical notions are introduced. Given a strength-change TAD, let m 1 be the median value of KR-normalized Hi–C contact frequencies within the strength-change TAD in condition 1, m 2 be the median value of the KR-normalized Hi–C contact frequencies within the same TAD region but in condition 2. Let s 1 be the sum of the KR-normalized Hi–C contact frequencies across all condition 1 TADs, s 2 be the sum of the KR-normalized Hi–C contact frequencies across all condition 2 TADs. If the strength-change TAD satisfies , it is classified as a strength-change up TAD. Otherwise, the strength-change TAD is classified as a strength-change down TAD. Missing value imputation Missing values may exist in Hi–C contact matrices A 1 or A 2 for a specific TAD region, and their origin can vary. DiffDomain distinguishes between missing values caused by SVs and those caused by other factors, such as low sequencing depth. When SVs are present, say in condition 2, DiffDomain first checks if an m × m submatrix, m ≥3, of A 2 contains exclusively missing values. In such cases, the m × m submatrix is imputed with a constant, with the default value of 1. Otherwise, if any row and column with a proportion of missing values greater than a given threshold, with a default value of 0.5, DiffDomain removes the corresponding row/column from both A 1 and A 2 . Subsequently, the remaining missing values are imputed by the median contact frequency of interactions at the same distance within the corresponding contact matrix. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Results Overview of DiffDomain The workflow of DiffDomain is illustrated in Fig. 1 . Its input is a set of TADs called in biological condition 1 and their corresponding Hi–C contact matrices from biological conditions 1 and 2 (Fig. 1 a). In this paper, TADs are called by Arrowhead 33 and Hi–C contact matrices are KR-normalized, unless otherwise stated (Supplementary Method 1 ). Our goal is to test if each TAD identified in biological condition 1 has significant structural reorganization in biological condition 2. The core of DiffDomain is formulating the problem as a hypothesis testing problem, where the null hypothesis is that the TAD doesn’t undergo significant structural reorganization in condition 2. To achieve this goal, for each TAD with N bins, DiffDomain extracts the N × N KR-normalized Hi–C contact matrices specific to the TAD region from the two biological conditions, which are denoted as A 1 and A 2 (Fig. 1 a). Note that A 1 and A 2 are N × N submatrices of the genome-wide Hi–C contact matrices. DiffDomain first log-transform them to adjust for the exponential decay of Hi–C contacts with increased 1D distances between chromosome bins. Their difference is calculated and denoted by D (Fig. 1 b). D is further normalized by a 1D distance-stratified standardization procedure, similar to the procedures in HiC-DC+ 38 and SnapHiC 40 . Specifically, each d -off diagonal part of D is subtracted by its sample mean and divided by its sample standard deviation (Fig. 1 c), − N + 2 ≤ d ≤ N − 2, reducing 1D distance-dependence of values in D and differences caused by variation in read depths between two biological conditions (see Supplementary Fig. 1 for two more detailed visualization). Intuitively, if a TAD is not significantly reorganized, normalized D would resemble a white noise random matrix, enabling us to borrow theoretical results in random matrix theory. Under the null hypothesis, DiffDomain assumes that is a generalized Wigner matrix (Fig. 1 d), a well-studied random matrix model. Its largest eigenvalue λ N is proved to be fluctuating around 2. Armed with this fact, DiffDomain reformulates the reorganized TAD identification problem into the hypothesis testing problem: The key theoretical results empowering DiffDomain is that θ N = N 2/3 ( λ N − 2), a normalized λ N , asymptotically follows a Tracy-Widom distribution with β = 1, denoted as T W 1 . Thus, θ N is chosen as the test statistic, and the one-sided P value is calculated as . H 0 is rejected if the P value is less than a predefined significant level α , which is 0.05 in this paper (Fig. 1 e). The pseudocode is shown in Supplementary Method 2 . For a set of TADs, P values are adjusted for multiple comparisons using the Benjamini–Hochberg (BH) method as the default. Once DiffDomain identifies the subset of reorganized TADs, it further classifies them into six subtypes based on changes in their boundaries, which is beneficial for downstream biological analyses and interpretations (Fig. 1 f, Supplementary Method 3) . TAD reorganization subtypes are verified by aggregation peak analyses (APA) on multiple real datasets (Supplementary Fig. 2) . A few reorganized TADs in real Hi–C data are shown in Fig. 1 g. Details are described in the Methods section. Note that although each biological condition may have multiple Hi–C replicates, DiffDomain takes the combined Hi–C contact matrix from the replicates as the input, which is a common practice to generate a large number of Hi–C interactions 33 . Correlations among Hi–C interactions lead to correlations among entries in the matrix, violating the independent assumption among upper diagonal entries of generalized Wigner matrix. However, the violation of the independence assumption does not substantially alter the properties of and DiffDomain based on empirical analysis results, suggesting that assumptions of DiffDomain are appropriate (see Methods section, Supplementary Note 1 , Supplementary Fig. 3) . When N = 10, the Tracy-Widom distribution T W 1 is an adequate approximation of the exact distribution of θ N 41 . Under common 10 kb resolution Hi–C data, N = 10 refers to TADs with 100 kb in length, much smaller than the median TAD length of 185 kb 33 . Thus DiffDomain only computes the P value for TADs with at least 10 chromosome bins, a practical constraint. DiffDomain is robust to a varied number of sequencing reads, Hi–C resolution, and different TAD callers (Supplementary Note 2 , Supplementary Figs. 4 and 5) . DiffDomain consistently outperforms alternative methods in multiple aspects First, we assess the false positive rate (FPR) which is the ratio of the number of false positives to the number of true negatives. A smaller FPR means that the identified significantly reorganized TADs are more likely to be true. Due to the lack of gold-standard data, we resort to analyzing the proportions of significantly reorganized TADs between five different Hi–C replicates from the GM12878 cell line. These Hi–C replicates are generated by different experimental procedures and have a highly varied total number of Hi–C contacts (Supplementary Table 1) . However, the TADs are expected to have few structural changes between these Hi–C replicates. The proportion of identified reorganized TADs is treated as an estimate of FPR (Supplementary Method 4) and is expected to be small. The Hi–C resolution is chosen as 10 kb. Comparing GM12878 Hi–C replicates primary and replicate , we find that DiffDomain, TADCompare 31 , and HiCcompare 42 have FPRs that are close to the given significant level of 0.05, suggesting good controls of FPR. In contrast, DiffGR 30 , DiffTAD 29 , and HiC−DC+ 38 have inflated FPRs (more than two-fold higher than 0.05), indicating poor controls of FPR (Fig. 2 a). Similar results are observed by repeating the above analysis to other GM12878 Hi–C replicates and 25 kb resolution Hi–C data (Supplementary Fig. 6) . Good control in FPR does not necessarily represent high power in detecting reorganized TADs between biological conditions. To investigate this, we compare TADs between two blood cell lines, GM12878 and K562. DiffDomain identifies that 30.771% of GM12878 TADs are reorganized in K562. In contrast, TADCompare, HiCcompare, and HiC-DC+ only identify ≤8.256% of GM12878 TADs that are reorganized in K562 (Fig 2 b), suggesting that they are too conservative in identifying reorganized TADs between biological conditions. Similar results are observed by repeating the above analysis to other human cell lines and 25 kb resolution Hi–C data (Supplementary Table 2 , Supplementary Fig. 7) , demonstrating the robustness of the observations. Conservation of TADCompare is because it is designed to scan every chromatin loci for potential reorganized TAD boundaries. But this analysis uses a given list of TADs, a common practice in Hi–C data analysis, which sharply narrows down the search space of TADCompare. Conservations of HiCcompare and HiC−DC+ are because they are designed for detecting differential chromatin interactions, not specifically tailored for identifying reorganized TADs. Compared with TADsplimer which specifically identifies split and merge TADs 25 , DiffDomain identifies similar numbers of split and merge TADs between multiple pairs of human cell lines (Supplementary Fig. 8) . Importantly, DiffDomain identifies that the majority (minimum 43.137%, median 81.357%, maximum 98.022%) of the identified reorganized TADs are the other four subtypes (Supplementary Fig. 9) , which can not be detected by TADsplimer. For example, among the GM12878 TADs that are identified as reorganized in K562 by DiffDomain, strength-change is the leading subtype of reorganized TADs, consistent with the fact that both GM12878 and K562 are blood cell lines (Fig 2 c). These results demonstrate that DiffDomain has substantial improvements over TADsplimer. We next investigate the true positive rate (TPR). A higher TPR means that more truly reorganized TADs are correctly identified as reorganized TADs. Through an extensive literature search, we collect 65 TADs that are reorganized between 146 pairs of biological conditions in 15 published papers (Supplementary Table 3 , Supplementary Method 5) . We use these TADs as the gold standard data to compute the TPR (Supplementary Method 4) and also call these TADs truly reorganized TADs. Four truly reorganized TADs, either correctly identified or missed by DiffDomain, are shown in Fig. 2 d. HiCcompare and HiC−DC+, designed for identifying differential chromatin interactions, are not directly applicable to the only testing reorganization of one single TAD and thus are excluded from the analysis. We find that the TPR of DiffDomain is 68.493%, which is 1.639, 2.703, and 16.665 times higher than that of alternative methods DiffGR, DiffTAD, and TADCompare, respectively (Fig. 2 e). Compared with DiffDomain, DiffGR, DiffTAD, and TADCompare only uniquely identify 11, 10, and 1 truly reorganized TADs, respectively (Supplementary Fig. 10) . Closer examination shows that DiffDomain has much smaller P values than other methods (Wilcoxon rank-sum test, P ≤ 2.31 × 10 −8 , Fig. 2 f), demonstrating that DiffDomain has stronger statistical evidence in favor of truly reorganized TADs. Based on the depictions of TAD changes reported in the publications, the truly reorganized TADs are broadly categorized into three groups: domain-level change, boundary-level change, and loop-level change (Supplementary Method 5) . These groups have decreased reorganization levels with increased stratum-adjusted correlation coefficient (SCC) scores 34 between biological conditions (Supplementary Fig. 11a) . Across the groups, DiffDomain consistently achieves the highest TPRs, while the second-best method varies (Supplementary Fig. 11b) , further demonstrating the advantages of DiffDomain over alternative methods. DiffDomain still misses 31.507% of possible pairwise comparisons of truly reorganized TADs. One reason is that some of the missed truly reorganized TADs have highly similar Hi–C contact matrices between biological conditions. For example, the missed truly reorganized TAD, chr11:1500000–2200000 (Fig. 2 d), has the SCC score at 0.998. Generally, missed truly reorganized TADs have significantly ( P = 8.26 × 10 −6 ) higher SCC scores than those correctly identified reorganized TADs by DiffDomain (Fig. 2 g). Similar results are observed when stratifying by the groups of truly reorganized TADs (Supplementary Fig. 11c) . Because DiffGR uses SCC as the test statistic, these results also partially explain the low TPR (41.781%) of DiffGR and highlight that SCC alone is not optimal for identifying reorganized TADs. Another reason is that the resolutions of some Hi–C data are low since P values are moderately negatively associated with the maximum values of Hi–C contact matrices (Spearman’s rank correlation coefficient ρ = − 0.534). Additionally, DiffDomain is efficient in memory usage and acceptable in computation time compared with alternative methods (Supplementary Note 3 , Supplementary Fig. 12) . In summary, compared with alternative methods, DiffDomain has multiple improvements, including FPRs, proportions of identified reorganized TADs between different biological conditions, subtypes of reorganized TADs, and TPRs. Reorganized TADs are associated with epigenomic changes Armed with the advantages of DiffDomain over alternative methods, we explore the connections between TAD reorganization and epigenomic dynamics. We first showcase a GM12878 TAD that is significantly reorganized in K562 and classified as a strength-change TAD by DiffDomain (Fig. 3 ). The TAD covers a 445 kb region on chromosome 6. The TAD structural changes involve the vascular endothelial growth factor gene VEGFA , which is a major tumor angiogenic gene that is over-expressed in leukemia (see reviews 43 , 44 for more details), consistent with the fact that K562 cells are chronic myelogenous leukemia cells. We find that the reorganized TAD has K562-specific functional annotations. The genomic region covered by the TAD is more accessible (1.71 times higher DNase peak coverage) in K562 than in GM12878 (Fig. 3 b). The H3K27ac and H3K4me1 peak coverages of the TAD region in K562 are 3.24 and 3.28 times higher than the coverages in GM12878, respectively. In contrast, the H3K4me3 and H3K36me3 peak coverages of the TAD in K562 are only 1.31 and 1.25 times higher than the coverages in GM12878, respectively. Four regions in the TAD are annotated as super-enhancers 45 only in K562 (Fig. 3 b). Note that the TAD region is in A compartments in both cell types, suggesting that the A/B compartments switch is not the reason for the gain in accessibility and histone modifications that are associated with gene activation. The normalized difference matrix D between Hi–C contact matrices of the TAD highlights that super-enhancer SE2 has increased Hi–C contacts with the VEGFA gene in K562 (Fig. 3 c). To gain further insights into structural differences of the TAD in the two cell types, we compare the 3D structural representations of the TAD region. We run Chrom3D 46 100 times to construct 100 possible 3D structures in each cell type for statistical comparisons. Two possible 3D structures with each per cell type illustrate the 3D structural differences of the TAD between GM12878 and K562 (Fig. 3 d, e). Overall, the super-enhancer SE2, but not SE3 (Fig. 3 b), is much spatially closer ( P < 2.22 × 10 −16 ) to VEGFA in K562 than in GM12878 (Supplementary Fig. 13) . These results show that the reorganized GM12878 TAD in K562 has K562-specific chromatin organization and potential biological functions. Generally, comparative analyses across multiple pairs of human normal and disease cell lines reveal that strength-change reorganized TADs with increased contact frequencies have significant increase in the number of CTCF binding sites at TAD boundaries compared with other TADs, whereas lost TAD boundaries associated with loss , zoom , and merge TADs have significantly fewer number of CTCF binding sites, consisting with enrichment of CTCF binding sites in TAD boundaries (Supplementary Note 4 , Supplementary Fig. 14) . Across diverse human cell lines, while TADs remain relatively stable, their proportions of reorganized TADs vary depending on the cell type, and these variations can cluster cell lines with similar cell identities (Supplementary Note 5 , Supplementary Figs. 7 and 15) . GM12878 (normal lymphoblastoid cell line) TADs that are reorganized in K562 (chronic myeloid leukemia cell lines) are enriched ( P = 0.01) in disease genes in chronic myelogenous leukemia. Across pairs of cell types, reorganized TADs tend to gain in chromatin accessibility and active transcription signals H3K27ac and K3K4me1. Particularly, TAD reorganization subtypes have distinct associations with chromatin accessibility as well as histone modifications. Specifically, TAD reorganization subtypes strength-change-up , zoom , split , and complex are associated with increased chromatin accessibility and histone modification signals marking active transcription activities. Conversely, TAD reorganization subtypes loss , strength-change-down , and merge are associated with decreased histone modifications signals marking active transcription activities, emphasizing the importance of TAD reorganization subtypes in investigating genome activity and functionality (Supplementary Note 6 , Supplementary Figs. 16 – 18) . Compared to normal human astrocytes (NHA), patient-derived diffuse intrinsic pontine glioma cell lines DIPG007 and DIPGXIII share a substantial proportion (73.46%) of reorganized TADs, some harboring potential oncogenes and super-enhancers, while dBET6 treatment demonstrates a stronger effect on TAD reorganization than BRD4 inhibition (Supplementary Note 7 , Supplementary Figs. 19 – 21 , Supplementary Table 4) . Together, these results demonstrate the functional relevance of reorganized TADs in multiple human normal and disease cell lines. Reorganized TADs are enriched with structural variations (SVs) SVs can contribute to diseases by rewiring 3D genome organization. To further demonstrate the biological relevance of reorganized TADs, we systematically investigate the associations between SVs and reorganized TADs. High-resolution SVs, including deletions and duplications, from erythroleukemia (K562 cell line) and pediatric high-grade glioma (DIPG007 and DIPGXIII cell lines) are downloaded from Wang et al. 47 . Because Arrowhead TADs does not necessarily cover the whole genome, SVs are filtered by keeping only those with their genomic regions overlapping with TADs (illustrated by two examples in Fig. 4 a). The number of SVs and paired normal Hi–C data are summarized in Supplementary Table 5 . If an SV region overlaps with one reorganized TAD, we consider the SV to have an associated reorganized TADs. We find that the majority (72.2%) of the SVs have such associations (Fig 4 b), with proportions significantly higher than those of randomly sampled, equal-numbered reorganized TADs (Supplementary Fig. 22) . SVs are associated with distinct abnormal patterns in Hi–C contact maps and are categorized into four types: deletions and duplications with 5′ to 3′ fusion, 5′ to 5′ fusion, and 3′ to 3′ fusion 47 . When stratified by SV types, the majority (>55%) of SVs with the same type also have associated reorganized TADs (Fig 4 b). However, each type of SVs has distinct associations with the subtypes of reorganized TADs. For example, in the comparison between GM12878 and K562 cell lines, the reorganized TADs associated with the four types of K562 SVs have differential distributions over their subtypes (Fig. 4 c). The leading subtype of reorganized TADs, strength-change , is consistently observed across the four types of SVs. However, the second leading subtype of reorganized TADs varies among the four types of SVs (Fig. 4 c). This observation is further emphasized by evident differences in the APA plots (Fig. 4 d). Importantly, the association between SV type and TAD reorganization subtype is disease-specific, supported by the clear distinctions in both the APA plots and the subtype distributions of reorganized TADs across K562, DIPG007, and DIPGXIII cell lines (Fig. 4 c, d). Particularly, leading reorganized TAD subtypes associated with SVs vary among cell types, with strength-change and loss in K562; strength-change , zoom , and split in DIPG007; and loss in DIPGXIII (Fig. 4 c). This variability may be due to the substantial differences in SV lengths among these cell types (Fig. 4 e). Notably, small proportions (17.4–27.7%, 8 in K562, 10 in DIPG007, and 4 in DIPGXIII) of SVs lack associated reorganized TADs (Fig. 4 b). Upon visual examination through the Nucleome Browser, in total, 13 SVs have reorganized TADs that are not detected by DiffDomain, implying that the remaining 9 SVs in the three cell types may lack associated TAD reorganization (Supplementary Figs. 23 – 25) . Nevertheless, these results significantly enhance our understanding of the relationship between SVs and TADs compared to the previous study 47 , further highlighting the biological relevance of reorganized TADs. DiffDomain improves profiling of TAD reorganization related to SARS-CoV-2 infection Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) caused over 640 million confirmed coronavirus disease 2019 (COVID-19) cases, including over 6.6 million deaths, worldwide as of December 2, 2022 , posing a huge burden to global public health. Wang et al. 48 is the first Hi–C study into the effects of SARS-CoV-2 infection on host 3D genome organization, finding a global pattern of TAD weakening after SARS-CoV-2 infection. However, the analysis uses aggregation domain analyses which cannot directly identify individual weakened TADs, in addition to missing other subtypes of TAD reorganization. To further demonstrate the biological applications of DiffDomain, we reanalyze the data. We find that 20.58% (840 in 4082) mock-infected A549-ACE2 TADs are reorganized in SARS-CoV-2-infected A549-ACE2 cells. Among the reorganized TADs, strength-change TADs are the leading subtype (64.64%) (Fig. 5 a), which is consistent with the global pattern of TAD weakening 48 , verifying the reorganized TADs identified by DiffDomain. The following most frequent subtypes are merge TADs and complex TADs (18.45% and 10.95%, Fig. 5 a), refining the characterization of TAD reorganization after SARS-CoV-2 infection. These reorganized TADs also enable refined profiling of transcriptional regulation in response to SARS-CoV-2 infection. Compared with the other TADs, the reorganized TADs have significantly higher numbers of upregulated genes and downregulated genes ( P ≤ 1.27 × 10 −4 , Fig. 5 b, Supplementary Fig. 26a) . Similar significant patterns are observed comparing strength-change TADs and split TADs with the other TADs ( P ≤ 8.07 × 10 −3 , Supplementary Fig. 26b) , highlighting that the two subtypes have stronger connections with differentially expressed genes than other subtypes of reorganized TADs. In contrast, compared to the other TADs, the six subtypes of reorganized TADs have significantly higher numbers of both enhanced and weakened peaks of H3K27ac, SMC3, and RAD21 where H3K27ac is a marker for active enhancers and SMC3 and RAD21 are two critical cohesin subunits that regulate 3D genome organization (Supplementary Note 8 , Fig. 5 c–e, Supplementary Fig. 26c–e) . Gene-centric analysis shows that differentially expressed genes in reorganized TADs have stronger connections with differential chromatin interactions than in other TADs. In particular, the strength-change subtype has a 3-fold higher proportion (9.73%) of downregulated genes with both enhanced and weakened chromatin interactions compared to other TADs (Supplementary Note 9 , Supplementary Fig. 27) . These results suggest that, after SARS-CoV-2 infection, the subtypes of reorganized TADs all have strong associations with epigenome reprogram, and strength-change TADs and split TADs have strong associations with deregulation of gene expression, highlighting the importance of subtypes of reorganized TADs identified by DiffDomain. DiffDomain characterizes cell-to-population and cell-to-cell variability of TADs using scHi–C data Recent advances in scHi–C sequencing methods enable profiling of 3D genome organization in individual cells, revealing intrinsic cell-to-cell variability of TADs among individual cells. However, quantifying the variability is challenging due to the properties of scHi–C data, such as high sparsity, low genome coverage, and heterogeneity 49 , 50 . As a proof-of-concept, we apply DiffDomain to a moderate-sized scHi–C dataset from mouse neuronal development (median number of contacts per cell at 400,000) 51 . We first ask how many individual cells are sufficient to identify reorganized TADs between cell types with high reproducibility using pseudo-bulk Hi–C data (Supplementary Method 6.1) . To do this, we design a sampling experiment to gradually increase the number of used individual cells, and the reproducibility in identified reorganized TADs is quantified using the Jaccard index (Supplementary Method 6.2) . We find that DiffDomain can identify reorganized TADs between cell types with reasonable reproducibility (average Jaccard index ≥ 0.104) using as few as one hundred sampled cells (Fig. 6 b, c). For example, DiffDomain consistently identifies that a neuronal TAD, harboring neuronal marker genes GM24071 , LRFN2 , MOCS1 , and 1700008K24RIK 51 , is reorganized in oligodendrocytes with numbers of cells starting at 100 (Fig. 6 a). Consistent of DiffDomain on other example genomic regions are shown in Supplementary Fig. 28 . On average, DiffDomain identifies that 19.25% neuronal TADs are reorganized in oligodendrocytes using only 250 sampled cells from each cell type, consistent (average Jaccard index at 0.49) with the identified reorganized TADs using all available cells in both cell types (Fig. 6 b). Similar results are observed when identifying neonatal neuron 1 (the youngest structure type) TADs that are reorganized in cortical L2–5 pyramidal cells (adult type) (Fig. 6 c, d). Jointly increasing the numbers of sampled cells in both cell types improves the performance of DiffDomain, as expected (Fig. 6 b). In contrast, only increasing the number of sampled cells in one cell type has a limited boost in performance. For example, oligodendrocytes have only 257 cells, but neurons have 1380 cells. Further increasing the number of sampled neurons from 250 to 500 has a slight performance improvement (Fig. 6 b). The observation is further confirmed when comparing neuronal subtypes neonatal neuron 1 and neonatal cortical L2–5 pyramidal cells, in which the number of sampled cells in the latter subtype is no more than 150 (Fig. 6 c) or 228 (Fig. 6 d). Repeating the analysis by using bulk Hi–C data 52 , 53 to create gold-standard reorganized TADs, we observed similar patterns in neuronal TADs that are reorganized in astrocytes. For example, sampling 150 cells in each cell type identifies 12.40% neuronal TADs that are reorganized in astrocytes on average. Among the reorganized TADs, 62.55% are also identified as reorganized TADs when bulk Hi–C data are used (Fig. 6 e). Considering the median number of contacts per cell at 400000, the merged Hi–C data from hundreds of cells are ultra-sparse pseudo-bulk Hi–C data. These results demonstrate that DiffDomain can work with ultra-sparse Hi–C data. Next, we move to quantify the cell-to-population variability of TADs, that is, comparing TADs in individual cells to the population average. To do this, scHi–C data with 50 kb resolution from neonatal neuron 1 cells is imputed by scHiCluster 54 . For each TAD, DiffDomain compares the imputed Hi–C contact map of the TAD in each cell to the pseudo-bulk Hi–C contact map. Resulted P values reflect cell-to-population variability of TADs and thus are used by hierarchical clustering to divide TADs into three categories: high, median, and low cell-to-population variational TADs (Fig. 7 a). We find that TADs have clear differential cell-to-population variability. One example high cell-to-population variational TAD and its adjacent median cell-to-population variational TADs in 9 cells are shown in Fig. 7 c. Among the 2146 neonatal neuron 1 TADs, 8.90% (191) are high cell-to-population variational TADs, 7.50% (161) and 83.60% (1794) are median and low cell-to-population variational TADs (Fig. 7 b). They are distributed across chromosomes (Fig. 7 d). Similar results are observed in other cell types (Supplementary Fig. 29) . These results demonstrate that TADs have clear differential variability between individual cells and the population average, consistent with earlier observations 26 , 55 . Next, we move to investigate the cell-to-cell variability of TADs. Requiring only one Hi–C contact matrix from each condition, DiffDomain can directly quantify cell-to-cell variability of TADs between individual cells using imputed scHi–C data. Note that, similar to other methods, pairwise comparison of TADs using scHi–C data from thousands of individual cells leads to exponential growth in runtime and thus is computationally expensive 49 . As a proof of concept, we apply DiffDomain to scHi–C data from randomly selected 50 cortical L2–5 pyramidal cells and 50 adult astrocytes. We find that the cell-to-cell variability of TADs is heterogeneous. The heterogeneity is consistent among the pairwise comparisons but quite different from those from random scenarios in which equal-numbered reorganized TADs are randomly assigned in pairwise comparisons. The proportion of reorganized TADs is consistent among the 2500 pairs of individual cells, ranging from 46.0% to 75.7% (Fig 7 e). Across the 50 cortical L2–5 pyramidal cells, the proportions of TADs that are reorganized in a varied number of adult astrocytes are fairly consistent (Fig. 7 f). Moreover, the proportions of TADs that are either low in cell-to-cell variability (reorganized in no more than 10 adult astrocytes) or high in cell-to-cell variability (reorganized in more than 40 adult astrocytes) are much higher than those from random scenarios (Fig. 7 f, Supplementary Fig. 30) , consistent with differential cell-to-population variability of TADs as reported in the previous paragraph. This observation is also in concordance with the randomized placement of TAD-like blocks in individual cells but with a strong preference for TAD boundaries observed in bulk Hi–C data 18 , further demonstrating the utilization of DiffDomain. In summary, DiffDomain works on scHi–C data to identify reorganized TADs between cell types, identify TADs with differential cell-to-population variability, and characterize cell-to-cell variability of TADs.
Discussion In this work, we present a statistical method, DiffDomain, for comparative analysis of TADs using a pair of Hi–C datasets. Extensive evaluation using real Hi–C datasets demonstrates clear advantages of DiffDomain over alternative methods for controlling false positive rates and identifying truly reorganized TADs with much higher accuracy. Applications of DiffDomain to Hi–C datasets from different cell lines and disease states demonstrate that reorganized TADs are enriched with structural variations and associated with CTCF binding site changes and other epigenomic changes, revealing their condition-specific biological relevance. By applying it to a scHi–C dataset from mouse neuronal development, DiffDomain can identify reorganized TADs between cell types with considerable reproducibility using pseudo-bulk Hi–C data from as few as a hundred cells. Moreover, DiffDomain can reliably characterize the cell-to-population and cell-to-cell variability of TADs using scHi–C data. The major methodological contribution of DiffDomain is directly characterizing the differences between Hi–C contact matrices using high-dimensional random matrix theory. First, DiffDomain makes no explicit assumption on the input chromatin contact matrices, directly applicable to both bulk and single-cell Hi–C data. Second, DiffDomain computes the largest eigenvalue λ N of a properly normalized difference contact matrix D , enabling the quantification of the differences of a TAD using all chromatin interactions within the TAD. Third, leveraging the asymptotic distribution of λ N , DiffDomain computes theoretical P values, which is much faster in computation than simulation methods used in alternative methods. The model assumptions are realistic (Methods). Last but not least, the normalized difference contact matrix D can help pinpoint genomic regions with increased or decreased chromatin interactions within the reorganized TAD, enabling model interpretation and refined integrative analysis with other genomic and epigenomic data. There is room for improvement. First, DiffDomain has the highest accuracy in detecting truly reorganized TADs, but it misses some truly reorganized TADs that only show subtle structural changes (see example in Fig. 2 d, g). Developing more powerful model-based methods is future work. The manually created list of gold-standard reorganized TADs is deposited in the GitHub repository that hosts the source code, which would benefit the research community for better method development. Second, because of the hierarchy of TADs and sub-TADs 2 , 33 , generalizations of DiffDomain to explicitly consider dependencies among TADs to further refine reorganized TADs identification and classification is future work. Third, it would be desirable to generalize DiffDomain to compare other TAD-like domains 56 – 61 . scHi–C data is imputed by scHiCluster 54 for the characterization of cell-to-population and cell-to-cell variability of TADs. Benchmarking the effects of different imputation algorithms, including Higashi 26 , 62 and scVI-3D 63 , on quantifying cell-to-population and cell-to-cell variability of TADs is future work. As a subset of TADs, the identified reorganized TADs could be a critical unit for refined integrative analyses of multi-omics data. We demonstrate this type of application on different human cell lines and disease states, including SARS-CoV-2-infected A549-ACE2 cells. Applying DiffDomain to investigate the connections between TAD reorganization and changes in H3K27me3 modification, a marker recently implicated in development and disease 64 , 65 , is a valuable future work. Notably, future work integrating multiple types of omics data and functional perturbation experiments 66 is necessary to elucidate the causal relationships between TAD reorganization and disease. DiffDomain is an interpretable statistical method for enhanced comparative analysis of TADs and it works for both bulk and single-cell Hi–C data. The accelerated application of Hi–C and scHi–C mapping technologies would generate ever-growing numbers of bulk and single-cell Hi–C data from different health and disease states. DiffDomain and its future generalizations would be an essential part of the Hi–C analysis toolkit for the emerging comparative analysis of TADs, which in turn would advance understanding of the genome’s structure-function relationship in health and disease.
Topologically associating domains (TADs) are critical structural units in three-dimensional genome organization of mammalian genome. Dynamic reorganizations of TADs between health and disease states are associated with essential genome functions. However, computational methods for identifying reorganized TADs are still in the early stages of development. Here, we present DiffDomain, an algorithm leveraging high-dimensional random matrix theory to identify structurally reorganized TADs using high-throughput chromosome conformation capture (Hi–C) contact maps. Method comparison using multiple real Hi–C datasets reveals that DiffDomain outperforms alternative methods for false positive rates, true positive rates, and identifying a new subtype of reorganized TADs. Applying DiffDomain to Hi–C data from different cell types and disease states demonstrates its biological relevance. Identified reorganized TADs are associated with structural variations and epigenomic changes such as changes in CTCF binding sites. By applying to a single-cell Hi–C data from mouse neuronal development, DiffDomain can identify reorganized TADs between cell types with reasonable reproducibility using pseudo-bulk Hi–C data from as few as 100 cells per condition. Moreover, DiffDomain reveals differential cell-to-population variability and heterogeneous cell-to-cell variability in TADs. Therefore, DiffDomain is a statistically sound method for better comparative analysis of TADs using both Hi–C and single-cell Hi–C data. Topologically associating domains (TADs) are critical structural units in 3D genome organization, and their reorganization between health and disease states is associated with essential genome functions. However, computational methods for identifying reorganized TADs are still in the early stages of development. Here, the authors present an algorithm leveraging random matrix theory to identify reorganized TADs. Subject terms
Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s41467-024-44782-6. Acknowledgements This work was supported by the National Natural Science Foundation of China grant 12271536 (D.T. and Z.B.), National Key Research and Development Program of China grant 2021YFC2300102 (D.T.), GuangDong Basic and Applied Basic Research Foundation grant 2022A1515010043 (D.T.), Shenzhen Sustainable Research grant KCXFZ20211020172545006 (D.T.), National Natural Science Foundation of China grant 12171198 (Z.B.), and Jilin Provincial Foundation grant 20210101147JC (Z.B.). We thank Jiang Hu for the helpful discussion on the theoretical properties of the proposed method, Jun Ding and Yang Zhang for the helpful discussion that improved the paper Jian Ma for helpful comments to improve the paper. Author contributions Conceptualization: X. Zhu and D.T.; Methodology: M.G., X. Zhang, and D.T.; Software: M.G., D.H., X. Zhang, and D.T.; Investigation: D.H., M.G., X. Zhang, Y.D., H.X., and D.T.; Writing-Original Draft: D.T.; Writing-Review and Editing: L.Q., X.D., Z.B., X. Zhu, and D.T.; Funding Acquisition: Z.B. and D.T. Peer review Peer review information Nature Communications thanks Daniele Tavernari and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Data availability All datasets used in this study are publicly available. Hi–C data of multiple human cell lines and replicates of GM12878 cell line are downloaded from the Gene Expression Omnibus (GEO) database under accession code GSE63525 [ https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE63525 ] 33 . Hi–C data of patient-derived DIPG, NHA, and GBM cell lines and DIPG frozen tissue specimens are downloaded from the GEO database under accession code GSE162976 [ https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE162976 ] 74 . Hi–C data of mock-infected and SARS-CoV-2-infected A549-ACE2 cells are downloaded from the GEO database under accession code GSE179184 [ https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE179184 ] 48 . Processed single-cell Dip-C data of multiple cell types in mouse brains are downloaded from the GEO database under accession code GSE162511 [ https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE162511 ] 51 . The other Hi–C data, DNase-seq data, super-enhancers, and cancer genes are downloaded from the GEO, 4DN, OncoKB, and GeneCards databases, and the data sources are listed in Supplementary Method 5 . All relevant analyzed data is available upon request. Code availability The software is published under the GNU GPL v3.0 license. The source code of DiffDomain is available at https://github.com/Tian-Dechao/diffDomain or at this 10.5281/zenodo.10205208 75 . Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Nat Commun. 2024 Jan 13; 15:502
oa_package/30/94/PMC10787792.tar.gz
PMC10787793
38218975
Background & Summary GRACE (Gravity Recovery And Climate Experiment) satellites are designed to monitor spatiotemporal variations of the Earth’s gravitational field to improve our understanding of the changes in the global climate system, with the primary goal of properly mapping mass variations, including terrestrial water cycle, ice sheet and glacier mass balance, sea level change, and ocean bottom pressure variations. Data have been acquired for fifteen years, exceeding the anticipated five-year mission span from March 17 2002 through October 2017 1 , 2 . The collection of science mission data ended in October 2017 because of the age-related battery issue on GRACE-B in September 2017. GRACE-FO was launched in May 2018 as a successor mission to GRACE in order to ensure the mission’s continuity 3 , 4 . 11 consecutive months of data gap exist between GRACE and GRACE-FO missions. In addition, some of the monthly solutions are missing due to improper retracked orbit issues throughout the lifetime of satellites 5 . In recent years, there have been studies using different methods to fill this gap. While mostly focusing only on terrestrial water storage and excluding mass changes over oceans, few studies have also reconstructed long-term simulations of total water storage anomaly, i.e., the climate-induced mass anomaly, before the GRACE-era using different approaches and spaceborne data. For instance, Humphrey and Gudmundsson 6 simulated six different forms of mass anomaly from 1901 to 2019 using a statistical approach with three different land surface temperature (TEMP) and two different precipitation (PPT) data products as meteorological forcing datasets, and two different GRACE mascon solutions. Li et al . 7 , first separated both input (PPT, TEMP, Sea Surface Temperature (SST), and 17 other climate indices) and output (GRACE mascon mass anomaly) into spatial patterns and temporal modes using independent component /principal component analysis techniques. Then, the temporal modes were further decomposed using least squares and seasonal-trend decomposition to obtain trend, seasonal, inter-annual and residual components. Excluding the trend, each decomposed component is used in Artificial Neural Networks (ANN), AutoRegressive Exogenous (ARX) and Multiple Linear Regression (MLR) approaches independently to simulate/predict temporal modes of each component at grid cell scale. Finally, GRACE-estimated trend and spatial patterns are restored by adding them back to simulated modes. Thus, the long-term mass anomaly simulations from 1979 to 2020 are obtained. Differently from the two studies mentioned above, Löcher and Kusche 8 calculated the monthly global gravity field by combining the low-degree gravity solution estimated from Satellite Laser Ranging (SLR) observations with the decomposed spatial patterns retrieved from the available monthly GRACE gravity field solutions using Empirical Orthogonal Functions (EOF). In this way, the hybrid monthly spherical harmonic gravity field models with GRACE-like spatial resolution, i.e., degree/order (d/o) 60 models from 1992 to 2019 are obtained though the solutions before 1994 are dominated by very large noise due to worse constellation of the SLR satellites prior to 1994. In this study, we used an in-house developed hybrid deep learning architecture, namely Residual Deep Convolutional Autoencoder (ResDCAE), to simulate long-term high resolution (at monthly temporal and 1° × 1° spatial resolution) mass anomaly from 1994 to 2021. ResDCAE is based on the concept of residual learning and utilizes stacked autoencoders to increase learning efficiency and is developed considering the TensorFlow 9 and Keras 10 libraries. No prior detrending, deseasoning, or decomposing processes either to the input or to the output datasets are applied. Thus, the simulations avoid possible biasing or aliasing of long-term climate signals. In order to successfully simulate trend, interannual, and seasonal signals, we included both SLR-based coarse resolution mass anomaly and normalized Day of Year (nDOY) as additional input, where the latter is computed by dividing the DOY of the mid-day of that month by 365 (or 366). For this purpose, the monthly SLR-only spherical harmonic gravity field models (up to d/o 10) 8 are used to effectively simulate the long-term trend, since the long-wavelength component of gravitational signals can be derived from SLR-only temporal gravity solutions. Interannual and seasonal signals, on the other hand, are simulated more accurately, thanks to nDOY. Because all geophysical signals are functions of time, using time epoch (nDOY) as an input acts as a constraint to obtain more realistic simulations. These novel ideas have already been tackled in recent study by Uz et al . 11 comprehensively, but Swarm-derived mass anomaly instead of SLR mass anomaly is used to obtain the long-wavelength component of gravitational signals between January 2014 and January 2021. Here, we focused on longer-term simulation considering a similar strategy and provided global simulations including both continents and oceans. The simulated mass anomalies are validated using the internal and external validation data. Furthermore, each monthly mass anomaly simulations are also converted to global geopotential field models expressed in spherical harmonics complete to degree and order 200.
Methods Residual deep convolutional autoencoders Convolutional Neural Networks (CNNs) are special types of neural networks and are useful for processing data with a grid-like architecture, such as images or time series data 12 , 13 . In particular, CNNs utilize the convolutional layers, which are linear operators and convolve the input with the set of filters. Therefore, the CNNs can be considered as spatial feature extractors by their layered structure. For this reason, CNNs have been widely used for problems such as filling the data gap in remote sensing 14 , land surface temperature reconstruction 15 , etc. In this manner, the relationship between the output vectors a of the consecutive layers of CNNs is represented as: where ∗ denotes convolution operator, σ(·) is the activation function, with weight matrix W and bias vector b while the superscript indicates the layer ID. In addition, CNNs mainly comprise three types of layers: convolutional layer, pooling layer, and fully connected layer. Deep Convolutional AutoEncoder (DCAE) is a deep learning architecture that may be considered as a combination of two neural networks, namely encoder and decoder 16 – 18 . In particular, the encoder maps the input space into a lower-dimensional latent space by h = f ( x ), while the decoder maps the latent space into the reconstruction space by r = g (h) and here x is the input vector. By this way, the network learns the representations of input data by reducing the dimensionality of data in either a supervised or an unsupervised manner. Basically, the high-abstraction features are learned while mapping through an internal representation, or code, h, in the intermediate layer. In addition, the distinction between standard AutoEncoder (AE) and DCAE is the utilization of convolutional layers. Accordingly, DCAE takes input data and maps it to h, where σ denotes the activation function, W and b are the weight matrix and bias vector of the encoder. Accordingly, the output of the decoder is given as follows: One of the common themes in deep learning architectures is that the deeper the network, the more advantageous and the better the modelling performance. However, there is the problem of vanishing/exploding gradients due to the successive calculation of gradients with respect to the gradient from the previous layer. To overcome this problem, a residual neural network was proposed by He et al . 19 , which led to an effective strategy for developing deeper neural networks. The main reason for this is the fact that instead of calculating gradients over F(x) which represents the mapping, gradients are calculated over F(x)+x, by introducing the skip connections between layers. Accordingly, output y is obtained by the combination of the input and output of the earlier layer as follows. where F ( x,W i ) is the residual mapping and W i corresponds the i-th weight matrix in the hidden layer weighted value of the layer. It should also be noted that the dimensions of the x and y must be equal. The residual learning strategy is successfully applied to problems such as classification 20 , the spatiotemporal estimation of citywide crowd flows 21 and influenza trends 22 . The proposed architecture is given in Supplementary Fig. S1 and based on the combination of DCAEs and CNNs with the concept of residual learning. The main reason for this implementation is that it improves learning efficiency by developing deeper structures with the help of residual learning. Accordingly, the structure of the proposed network may be divided into three parts as follows: Convolutional building block: The developed architecture is based on the use of the CNN which consists of two convolutional layers with 126 and 63 filters in each layer, respectively, followed by a dense layer with 21 neurons. It should also be emphasized that the size of the filters and neurons is chosen according to the channel size of the input. In addition, for each convolutional and dense layer, an Exponential Linear Unit (ELU) activation function is used throughout the network, and each convolutional layer also has a regularizer to prevent overfitting. Besides, a single convolution layer with 21 filter sizes is utilized as the residual connection. In the stage of selecting hyperparameters of CNN, we have considered the lowest generalization error subject to runtime and memory constraints. In addition, elastic net is utilized as a regularization method which combines the lasso and ridge regulators 23 , with penalty value of 10 −4 . It is also worth noting that the reason for adopting the ELU as an activation function is that it allows to negative outputs which leads to adjusting weights and biases in the correct direction during the iterative optimization process. DCAE building block: The structure of the DCAE model consists of five convolutional layers with a gradually increasing filter size from 21 to 316. Each convolutional layer is followed by a maxpooling layer with a pool size 2 × 2 and 3 × 3 respectively. In the intermediate layer, which is also known as latent space, a flattened layer and a fully connected layer with 400 neurons are used. In this manner, in the decoding part, the fully connected layer of 400 neurons is followed by the four transposed convolutional layers, symmetric to the encoder part. In addition, the ELU activation function is used in each layer of the network. Regression building block: To complete the end-to-end image-to-image regression task, a CNN-based structure is employed, which consists of two convolutional layers with 21 and 14 filters, respectively, followed by a dense layer with 1 neuron. Furthermore, the ELU activation function is applied to each convolutional layer except the dense layer. Accordingly, the input features first pass through the DCAE building block and are concatenated to the output features of this block. Further, this serves as the input features to the convolutional building block, and before the concatenation, input features are convolved. In order to complete image-to-image regression, regression building blocks are employed as the final layers of the model. Therefore, the model has 6 consecutive DCAE and convolutional blocks and 1 regression block. Regarding the implementation of the network, the Adamax optimizer is utilized with the initial learning rate equal to 10 −3 and a batch size of 27. In addition, the learning rate is reduced by a factor of 0.8 when learning stagnates by monitoring the validation loss. Furthermore, early-stopping is implemented to mitigate overwriting. According to this, validation loss is selected as the monitored metric, and the training procedure is stopped if no improvement is seen for 25 epochs. The Huber function has been selected as the training loss function since it is robust against outliers and has fast convergence to near negligible loss. The main reason for this implementation is that the input consists of various earth observations that have different characteristics. The training is performed on a single NVIDIA Tesla P100-PCIE-16GB GPU. It should also be noted that the full memory capacity of the graphic card is used, and the running time of the model is about 45 minutes. Mitigating trend error in backwards extrapolation Our objective is to provide monthly gravity field data products similar to those within the GRACE era, even for the pre-GRACE period. Achieving this goal involves extrapolating data backwards in time, which is a complex task. Extrapolation should be performed with caution, especially when dealing with non-stationary processes, as is the case in our study. Non-stationarity in Earth and environmental systems, such as spatiotemporal changes in Earth’s water mass, primarily results from the inherent secular trend signal. This signal alters the mean rather than the signal variance and may or may not follow a linear pattern. Factors like climate change, human interventions, and low-frequency internal variability, such as the Atlantic multidecadal oscillation, affected by the slow dynamics of ice sheets and the ocean, contribute to this non-stationarity 24 . This introduces the challenge such that the behaviour of the signal outside the training data period (in our case, the GRACE and GRACE-FO period) may differ from that within the training data time span, even if the seasonal amplitudes remain relatively consistent 25 . Like all data-driven approaches, deep learning (DL)-based methods adjust their parameters through optimization algorithms to find the best fit to available output data based on the corresponding input data. This optimization aims to establish a statistically optimal mapping from input to output, limited to the training data period. Consequently, deep learning models typically perform well when producing or predicting outputs for new input values falling within the range of input data used for training. However, predicting outcomes outside the range of the training data, referred to as ‘out of sample prediction,’ can be challenging. In other words, a deep learning model can provide reasonable results within the hyperspace defined by the boundary of the training data set, which can be seen as a high-dimensional interpolation, as long as the number of input data variables is fewer than 100 26 . In the case of mass anomaly, an efficient extrapolation requires a priori knowledge about the signal behaviour in the extrapolation regime, i.e., in the time span out of the training data period. Unfortunately, this kind of information is usually not available, at least globally and at grid cell scale. However, the main differences of the mass anomaly signal as well as of the input climate data signal in the extrapolation regime from those in the interpolation regime (i.e., training data span) are in the long-term trends while the seasonal amplitudes do not vary much (see e.g., Supplementary Fig. S2 ). Therefore, the main errors of extrapolation are due to mismodeling of the trend component which is retrieved from the training data. Some studies 6 , 7 estimate and remove a linear trend using the data in the GRACE and GRACE-FO mission spans before calibrating their models based on residual signal and then extend and restore this trend to the extrapolation regime by assuming that their estimated linear trends also hold out of the training data period. Such an assumption is too optimistic and may not be valid globally. A typical example is the surface mass balance estimates at polar regions, e.g., Greenland and Antarctica (see The IMBIE team 27 , 28 ) exhibit relatively lower mass change rates until late 1990s followed with an onset of dramatic increase in mass loss after 2000 due to the accelerated ocean-driven melting of the ice sheets. Similar extrapolation errors are also reported 29 for the long-term static gravity field solutions with co-estimated (TVC) time-variable coefficients (secular and seasonal periodic components), e.g. GOCO06S 30 , by evaluating the differences of mass anomaly from the monthly GRACE-FO solutions and those extrapolated from the static field with TVC computed from the data solely within the GRACE era, suggesting that the static gravity models with TVC cannot be used for long-term (>2-3 years) extrapolation and at least should be frequently updated with the newly available GRACE-FO data, e.g. for mass change studies as well as for improved precise orbit determination of low earth orbiting satellites. Also, Mouginot et al . 31 and Rignot et al . 32 discuss these mass changes for the last 40 years. Mouginot et al . 31 reported variability in the mass balance of the Greenland Ice Sheet since the 1980s, along with a sixfold increase in mass loss. This has resulted in a significant 13.7 mm contribution to global sea level rise since 1972, with half of this effect occurring during the period from 2010 to 2018. According to Rignot et al . 32 , the primary cause of mass loss in Antarctica is the glacier flow near warm, saline circumpolar deep-water regions, particularly in East Antarctica, with significant implications for future sea-level rise. In addition to these, Caceres et al . 33 studied the land water storage except for the glaciers mass change. They reveal that from 1948 to 2016, continents contributed to a sea-level rise of 34–41 mm, with glacier mass loss responsible for 81% of the cumulative loss and land water storage anomalies accounting for the remaining 19%. Climate-driven land water storage anomalies are notably influenced by precipitation and linked to El Niño Southern Oscillation, although uncertainties persist in modelling these anomalies, particularly in relation to irrigation water use and artificial reservoirs. The power of DL, besides the computational resources, is attributed to the number of training data, that is, higher the number of data higher the accuracy can be achieved by large neural networks whose parameters are updated through deep learning algorithms (Aggarwal 34 , pp. 3). After training the deep learning model with initial training and test data (see section Data Architecture for details of the training and test data) within the GRACE/-FO era and simulating the a priori monthly mass anomaly of 3-years backwards in time (i.e. from April 1999 to March 2002), here we applied a step-by-step piecewise trend correction approach where at each step the number of training data is incrementally increased backwards in time. The overall approach can be summarized as follows. We start with retraining our deep neural network after removing the first three years of the initial training data within the GRACE/-FO era (i.e., 3 years of data starting from April 2002 to March 2005) without altering the network architecture or any of the hyperparameters or the learning algorithm. The retrained model was then used to simulate global 1° × 1° monthly mass anomaly grids for the period coinciding with that of the removed 3-years of initial training data. For each grid cell, the linear trends from both the simulated and the corresponding original monthly mass anomaly data in the initial training set in these 3-years were estimated by least squares fit, and the difference between the two trends was computed. The computed trend difference at each grid cell was then applied to correct the simulated mass anomaly of the earlier 3-years (i.e., the monthly mass anomaly from April 1999 to March 2002). These new corrected simulations were added to the initial training data set which now constitutes the extended training data set for the next iteration. With the extended training data, the procedure above was repeated with removing the first 3-years (i.e., this time April 1999 to March 2002) and computing trend-corrected mass anomaly simulations for the previous 3-years (i.e., April 1996 to March 1999). This step-by-step correction process was applied once again with the updated training data from previous iteration so that the trend-corrected mass anomaly simulations from January 1994 to the beginning of GRACE/-FO era (i.e., April 2002) were completely obtained. The efficiency of the above procedure and the adequacy of the chosen step-size, i.e., 3-years have been verified with results shown in Technical Validation section. Descriptions of the input and output data of our DL-based simulation model Our DL architecture possess the multichannel input consisting of seven variables and a single output variable. Input are monthly coarse resolution SLR-only mass anomaly and five different Hydroclimatic/meteorological Variables (HV) from ERA5 (European Centre for Medium-Range Weather Forecast-ECMWF Reanalysis-5) as well as normalized (Day-of-Year) time epoch of these monthly dataset, i.e., nDOY. The five HV from ERA5 are PPT, TEMP, SST, Cumulative Water Storages Changes (CWSC) and mass anomaly retrieved from ERA5 model data while the single output is the monthly CSR RL06 Mascon (CSRM) mass anomaly solutions. The details of both input and output are given in the following. GRACE mass anomaly data The monthly mass anomaly of GRACE/-FO is derived from CSR RL06 Mascon (CSRM) solutions 35 , 36 . The time span of this dataset is fragmented into two main parts that are April 2002–June 2017 for GRACE and May 2018–to 2021 for GRACE-FO missions. There is an 11 successive months of so-called intermission data gap between GRACE and GRACE-FO, but there are also missing months during the operational time period of each mission. All the standard post-processing corrections have been applied to CSRM models, i.e., degree 1 correction 37 , replacement of C 20 and C 30 coefficients 38 , Glacial Isostatic Adjustment (GIA) 39 and Ellipsoidal correction 40 . The monthly mass anomalies are calculated considering the mean baseline between 2004.0 and 2009.9999. Besides, while the temporal resolution is monthly for CSRM, the spatial coverage is global, with a spatial sampling resolution of 0.25° × 0.25°. We resampled the original CSRM to 1.0° × 1.0° grids considering the native resolution of CSRM which determines the spatial resolution of our target mass anomaly simulations. SLR mass anomaly data The monthly SLR-only spherical harmonic gravity field models 8 up to d/o 10 are provided by Dr. Anno Löcher from the Astronomical, Physical, and Mathematical Geodesy Institute of Bonn University via personal communication. The coarse resolution SLR-only monthly gravity field solutions are available from November 1992 to January 2021. The monthly mass anomalies from SLR-only models are calculated by applying post-processing to the spherical harmonic coefficients of those models after removing the same mean baseline (2004.0–2009.9999) with CSRM for consistency. A 1500 km Gaussian smoothing filter is applied, i.e., the filter radius is decided after applying both 1000 and 2000 km, but the optimum results are obtained from using the 1500 km radius showing a compromise between signal loss and noise reduction. No de-striping filter was applied as suggested by the data provider (Dr. Anno Löcher, pers. comm.) Mass anomalies are directly calculated for each 1.0° × 1.0° grids on the globe including both ocean and land. ERA5 data The ERA5 dataset is released by the European Centre for Medium-Range Weather Forecast (ECMWF - https://cds.climate.copernicus.eu/ ) and consists of both monthly averaged and hourly sub datasets. Besides, ERA5 has two different parameter levels, i.e., single and pressure levels. We used the monthly averaged single level from 1979 to present; 41 a subset of ERA5 considering the timespan of the available SLR-only input data. Therefore, ERA5 was downloaded and used from November 1992 to January 2021. The basic input HV data are chosen in order to be used in the DL algorithm which are PPT, TEMP, SST, RunOff (RO), evapotranspiration (ET), snow water storage (SnWS), soil moisture storage (SMS) and Canopy Water Storage (CnWS). The input ERA5 mass anomaly and cumulative water storage changes (CWSC) are calculated using these downloaded variables applying the equations given in e.g., Uz et al . 11 , and Mo et al . 42 . Similar to the CSRM and SLR-only mass anomaly, each input data from ERA5 is referenced to the mean baseline, i.e., 2004.0–2009.9999, by removing its mean within this baseline period. Finally, all HV data are resampled from 0.25° × 0.25° to 1.0° × 1.0° to ensure consistency between all input and output data. Data architecture There are two considerations while designing the DL architecture which are addressed with the questions; (i) how do the temporal patterns of input and output data for each grid vary throughout their own time-spans? and (ii) how do the temporal correlations change with respect to time lags? The answers to these questions could give information about how many additional layers should be considered for the input data variables. In other words, how many channels should be used for the input data. We answered these questions by inspecting the Partial AutoCorrelations (PAC) computed for varying time lags in Amazon River basin. The Amazon River basin was chosen because it is a good example of reserving climate change signatures 43 . The grid closest to the centre of the basin, according to the basin boundaries from Total Runoff Integrating Pathway (TRIP) database 44 , was selected, and the PACs for each input and output were computed and plot. The time series of the input and output data and their corresponding PACs are given in Supplementary Fig. S2 and, except for SLR-only mass anomaly, all other input and output data show more or less a certain annual signal pattern. The SLR-only mass anomaly between December 1992 and December 1993 has higher variations with respect to those during the rest of the time span. Similar results are also reported in Löcher and Kusche 8 . The obvious reason for this is that the SLR-only models in the first couple years were computed without using the Stella satellite, which joined the constellation in September 1993 and is an important part of SLR-only temporal gravity field recovery due to its low polar orbit and hence providing more redundant sets of normal equations for gravity inversion. Thus, the time series of SLR-only mass anomaly includes even higher noise until September 1993 (Dr. Anna Löcher, pers. comm.). On the other hand, PACs are calculated from the time series of input and output, starting from zero lag up to a 12-month time lags. All correlations are computed throughout the GRACE/-FO time period and are illustrated in Supplementary Fig. S2e . It is clearly seen that almost all correlations reduce to below zero at a two-month lag and are almost zero beyond the two-months. Thus, we decided to set the number of additional layers to 2, i.e., the successive two months of relevant time epoch of input, i.e., t and t-1 . According to the pre-analyses above, DL architecture is constituted considering both the time-span of SLR-only models with lower noise level and the computed temporal correlations. In addition, our initial training and testing data sets are randomly selected from the GRACE/-FO time period. While the total number of initial training months is 135, the number of testing months which is kept unchanged when applying the trend correction procedure is 57. The entire time series of predicted/simulated monthly mass anomaly cover both the pre-GRACE era as well as the existing data gaps within the GRACE/-FO time period after the step-by-step trend correction was applied. The number of predictions where no GRACE/-FO data is available is 136, starting from January 1994 due to a higher noise issue in SLR-only solutions between 1992 and 1994.
Since April 2002, Gravity Recovery and Climate Experiment (GRACE) and GRACE-FO (FollowOn) satellite gravimetry missions have provided precious data for monitoring mass variations within the hydrosphere, cryosphere, and oceans with unprecedented accuracy and resolution. However, the long-term products of mass variations prior to GRACE-era may allow for a better understanding of spatio-temporal changes in climate-induced geophysical phenomena, e.g., terrestrial water cycle, ice sheet and glacier mass balance, sea level change and ocean bottom pressure (OBP). Here, climate-driven mass anomalies are simulated globally at 1.0° × 1.0° spatial and monthly temporal resolutions from January 1994 to January 2021 using an in-house developed hybrid Deep Learning architecture considering GRACE/-FO mascon and SLR-inferred gravimetry, ECMWF Reanalysis-5 data, and normalized time tag information as training datasets. Internally, we consider mathematical metrics such as RMSE, NSE and comparisons to previous studies, and externally, we compare our simulations to GRACE-independent datasets such as El-Nino and La-Nina indexes, Global Mean Sea Level, Earth Orientation Parameters-derived low-degree spherical harmonic coefficients, and in-situ OBP measurements for validation. Subject terms
Data Records The simulated monthly dataset is released both in the form of gridded mass anomalies with and spherical harmonic coefficients in accordance with the ICGEM (International Center for Global Earth Models) format. These datasets are available in figshare repository 45 . The datasets cover the time span from January 1994 to December 2021, hence provides simulations for 324 months in total. The gridded dataset is available to users with data format identical to official CSR mascon data products in figshare as netcdf file with four variables: lat, lon, time and mass anomaly. Lat and lon are latitude and longitude vectors of dimension 180 × 1 and 360 × 1, representing the positions of the centre of 1.0° × 1.0° grid cells on the surface of the Earth. The mass anomaly represents the relative change in the water mass with respect to the mean baseline (2004.0-2009.9999) in terms of cm Equivalent Water Height (EWH); thus, the associated netcdf file has a dimension of 324 × 180 × 360 while the time is a column vector of dimension 324 × 1 with days since 1994 01 01 00 00 00.0 UTC. The monthly mass anomaly in the form of spherical harmonic coefficients from degree 2 up to degree/order 200 were released as ASCII files. The file naming convention similar to the official GRACE data processing centers was adopted, i.e., determined by the year and the day of the year corresponding to the first and last day of the respective month. Each file has a header that contains information regarding constant values. Technical Validation This section evaluates the simulation performance of our models and summarizes the findings in two categories: internal and external validation. Internal validation is performed by comparing our simulations to CSRM mass anomaly solutions and to those from previous studies in terms of common mathematical goodness-of-fit metrics such as Root Mean Square Error (RMSE), Nash - Sutcliffe efficiency (NSE 46 ), and Pearson Correlation Coefficient (PCC). On the contrary to internal validation, the simulated mass anomalies (or spherical harmonic coefficients derived from them) are validated externally by comparisons to GRACE-independent datasets, such as the long-term surface mass balance estimates of Greenland, the El Niño/La Niña SST index, global barystatic mean sea level changes, degree 2 order 1 spherical harmonic coefficients (a.k.a. C 21 , S 21 ) retrieved from daily Earth Orientation Parameters (EOP) series, degree 2 order zero spherical harmonic coefficient (i.e., C 20 ) from SLR and in situ Ocean Bottom Pressure observations. Internal validation Comparison of different input scenarios The first step in our internal validation process is to assess the GRACE-like mass anomaly simulations generated by our deep learning algorithm, namely the ResDCAE model. Our objective is to determine how the inclusion of SLR and nDOY inputs in the model affects these simulations. To do so, we have devised four distinct simulation scenarios, which we refer to as DL models. These scenarios are grouped based on whether they incorporate SLR and/or nDOY inputs, while all DL models consistently include all four ERA5 layers. To make it easier to distinguish between these various combinations, we have assigned specific names to them: Sol1 (which includes both SLR and nDOY), Sol2 (which includes only SLR), Sol3 (which includes only nDOY), and Sol4 (which excludes both SLR and nDOY). We then compare the mass anomaly simulations from these four solutions with the reference mass anomaly data from original CSR Mascon solutions. In each of these comparisons, we calculate two commonly used metrics, namely RMSE (Root Mean Square Error) and NSE (Nash-Sutcliffe Efficiency), for all test months. These metrics allow us to effectively assess the performance of the models and gain insights into the impact of different input configurations. From April 2002 to August 2020, RMSE and NSE metrics for each of the randomly chosen 57 test months and their overall mean are computed and shown in Fig. 1 . The overall mean values of the metrics are also listed in Table 1 . Furthermore, the metrics are separately computed over (i) entire globe (land + ocean), (ii) land-only, and (iii) ocean-only areas. All three scenarios (i-iii) exhibit a level of accuracy in retrieving the missing test months, with RMSE of 4 cm, 5 cm, and 3 cm, respectively. The corresponding NSE metrics are computed as 0.86, 0.91, and 0.68. A similar comparison was made by Uz et al . 11 between April 2014 and September 2020 for the thirteen test months but only over land areas. The simulative performance of test months that either encompass the ACC (accelerometer) transplanted 47 time period or those which uses the piled data from two successive months to solve for the corresponding CSRM mass anomaly is worse than those within the other time spans (see Fig. 1 of Uz et al . 11 ). The same is also reported over oceans (see Fig. 2 of Chen et al . 48 ). From this point of view, while the CSRM products of GRACE after November 2016 are calculated using transplanted ACC data to mitigate the effect of the GRACE-B battery issue 49 , the ACC transplantation has been carried out since the start of the GRACE-FO mission because the standard ACC data derivation procedure from Level-1A (L1A) to Level-1B (L1B) does not ensure sufficient accuracy for gravity field recovery 50 , 51 . Thus, the Science Data System (SDS) produces and distributes the transplanted and calibrated ACC data product on a regular basis. Additionally, the specifics of Level-2 (L2) data products, metadata including whether the ACC transplantation is applied or piled data from consecutive months used for gravity inversion, are listed in Table 2 of the SDS Newsletter( https://isdc.gfz-potsdam.de/grace-isdc/grace-gravity-data-and-documentation/ ). As expected, the RMSE of all solutions over land are greater than those over the oceans. This is because the mass anomaly signal over land is stronger while those over ocean has lower magnitude which is dominated by noise. Therefore, the computed NSE over ocean are lower due to the low signal-to-noise ratio of CSRM over ocean at grid cell scale. The lowest RMSE are found in time series between 2004 and 2010, which is attributed to the better orbit configuration and availability of telemetry data with minor gaps during this time span (Fig. 1 ). Additionally, the NSE over oceans is at its lowest level within this time period while the RMSE is still minimum (~2 cm), implying that the DL algorithm successfully mitigated the high frequency spatiotemporal ocean mass change errors of CSRM at grid cell scale (Fig. 1c ). There is a significant jump in RMSE of all simulations, which is clearly seen in the comparison over land-only areas (Fig. 1b ), around August 2014. According to the August 2014 SDS Newsletter, the swap manoeuvre, during when the satellite twins exchange positions 52 , 53 , was carried out in July 2014. It may have impacted the satellite observations in August 2014 and in the following few months. The overall metrics in Table 1 reveal the following. Over land, the RMSE and NSE of Sol1 and Sol2 are higher and lower, respectively, than those of Sol3 and Sol4. However, the metrics of Sol3 and Sol4 are nearly identical. The primary distinction between Sol1-2 and Sol3-4 is the presence of SLR-only mass anomaly as a training input. It is expected that SLR-only mass anomaly would be noisier and will propagate to the Sol1 and Sol2 simulated models. Thus, mass anomalies that incorporate the spatiotemporal variations of the SLR-only mass anomaly through input are noisier than those that do not. On the other hand, the relationships of the metrics with the simulation models over oceans are different from those over land. While Sol2 metrics continue to have the greatest RMSE and the lowest NSE values over the ocean, Sol1 metrics reveal the opposite. Differently from Sol2, Sol1 uses normalized time (nDOY) as an additional training input. Although Sol1 exhibits propagated noise from input SLR-only mass anomaly, the simulated model is also associated with the temporal changes of nDOY input. On Earth, the oceanic regions have more complicated dynamics, and gravitational signals are influenced by the higher-frequency mass redistributions at oceans. Thus, the temporal variations retrieved by the aid of nDOY parameter may guarantee to provide more accurate description of the signal throughout the ocean. The metrics computed considering both land and ocean (global) grids are used to assess which simulation model has the best mathematical fit. The simulation models are further evaluated based on the spatial distribution of RMSE, NSE, and PCC by the illustration given in Fig. 2 . These metrics are calculated for each 1.0° × 1.0° grid cell from monthly differences between DL-based simulations and corresponding CSRM mass anomaly in the 57 testing months. While the rows of Fig. 2 correspond to RMSE (a, b, c, and d), NSE (d, e, f, and g), and PCC (h, i, j, and k), the columns correspond to simulations Sol1 to Sol4, from left to right. The highest RMSE is seen in the same regions for all simulations, i.e., the Amazon, Ganges, Greenland, and Gulf of Alaska. This result is also observed by previous studies (e.g., see Fig. 2a of Humphrey and Gudmundsson 6 , Fig. 2d of Li et al . 7 , Fig. 4j-1 of Mo et al . 42 and Fig. 3 of Uz et al . 11 ). These basins are hydrologically active in terms of signal variations and are the main contributors to ice sheet melting areas on Earth. For example, the Amazon is the largest drainage basin and is subject to the largest seasonal changes that can be surpassed by variations as much as 1 m of EWH in Total Water Storage (TWS) globally 54 . The Ganges basin is under the coupled effects of groundwater depletion due to human intervention for irrigation 55 and the melting of ice sheets in High Mountain Asia glaciers. The ice sheet mass loss in Greenland and the Gulf of Alaska, moreover, contributes to global sea-level rise 27 , 56 . The mass anomaly signals and variations in these regions are higher than those in other basins on Earth. Thus, the discrepancy between CSRM and simulated mass anomaly is sourced from this outcome, and systematically worse or higher RMSE are calculated at these regions. In addition, the spatial distribution of RMSE for Sol1 and Sol2 is more intense than for Sol3 and Sol4 due to the propagation of the noise of the SLR-only input into the simulations. The Empirical Cumulative Distribution Functions (ECDF) of RMSE over land are illustrated in Fig. 2m . While 80% of RMSE are below 5 cm for all simulations, there is a significant difference between SLR-only mass anomaly included simulations, i.e., Sol1, Sol2, and not included ones, i.e., Sol3, Sol4. Similarly, NSE and PCC values over land can be evaluated by considering the spatial distributions of these metrics as shown in Fig. 2 . Both NSE and PCC of all simulations are almost zero throughout arid regions on Earth, e.g., North Africa. These results are similar to those of Uz et al . 11 . Although the RMSE at arid regions are almost the same and having the lowest values, there is also very little correlation between CSRM and simulated mass anomaly which can be explained by the low signal-to-noise ratio of CSRM at these regions. The simulative performances of the DL models over oceans are also demonstrated spatially with the same metrics. The geophysical dynamics are more complex in oceans. This complexity is sourced from the oceanographic variables and their temporal variations. Thus, the variations of these variables are more difficult to model when compared to land and seem to degrade the simulation performances over oceans in all four simulations. As shown in Fig. 2a–d , the RMSE at high latitude regions are even higher compared to the other parts. These regions may be influenced more by polar climatic characteristics. On the other hand, NSE and PCC over the ocean possess the lowest values not only in areas that are close to polar regions but also in different parts of the ocean, i.e., the Atlantic Ocean. There is a significant difference between Sol1 and other simulations, based on the ECDF of RMSE and NSE scores computed over the oceans (see Fig. 2m,n and Table 1 ). The simulations of Sol1 clearly exceeds other simulations in terms of metrics meaning that the oceans are better modelled using the input combination adopted for Sol1, i.e., when both SLR-only mass anomaly and nDOY are included as additional input data. For example, among all only the NSE of Sol1 is above zero, which indicates that only Sol1 modelled the mass change over oceans realistically. In general, if the NSE of a simulation model is below zero it means that the mean of observation is better than the simulation results 57 . According to our analyses so far, Sol1 and Sol2 suffer from propagation of noise in SLR-only input mass anomaly over land, but Sol1 provides better simulations over oceans. Comparison with previous studies Based on the comparisons among the four DL simulation models in the previous section we pick the simulation results of Sol1 as our final reconstructed data products. Thus, the simulations of Sol1 are compared to the previous similar studies of Humphrey and Gudmundsson 6 , Li et al . 7 , and Löcher and Kusche 8 (which are called Humphrey, Li, and Löcher in the rest of the paper, respectively) regarding the performance metrics. The chosen reconstruction of Humphrey uses both JPL RL06 mascon and ERA5 HV, and the spatial resolution is 0.5° × 0.5° covering the time span from January 1979 to July 2019. The simulated mass anomaly of Li is calculated based on the CSR RL06 mascon with a spatial resolution of 0.5° × 0.5° and covers the time span between July 1979 and June 2020. On the other hand, hybrid models of Löcher were released as spherical harmonic coefficients up to degree and order 60 from November 1992 to January 2021. Thus, mass anomaly is calculated from the model coefficients with spatial resolution of 1.0° × 1.0°, removing the mean-field between 2004.0 and 2009.9999. To ensure consistency, Humphrey and Li’s mass anomalies are also calculated by removing this mean-field and up-sampled to 1.0° × 1.0° grids. The time period of comparison of all studies was chosen to be within the GRACE/-FO time period. In total, comparisons based on overlapping 175 months are made, and the illustration of the metrics is given in Fig. 3 . The columns of Fig. 3 are represented by Sol1, Li, Humphrey, and Löcher from left to right, respectively. Similar to the results of the four DL simulations as shown in the previous section, the RMSE of all models are higher in hydrologically dominant regions on Earth. The order of RMSE performances of the models from best to worst with respect to CSRM mass anomaly are Sol1, Li, Humphrey, and Löcher. A similar performance order is also clearly seen for NSE and PCC. The illustrated ECDF in Fig. 3m,n , o also represent the model accuracies prominently. Throughout the GRACE/-FO era, the Sol1 simulation has the lowest spatial RMSE and the highest NSE and PCC with CSRM. In contrast to our DL-based simulation, the Li, Humphrey, and Löcher used different approaches or estimation strategies, therefore the long-term trend of each study may be different from the other. For consistency, the RMSE, NSE and PCC metrics are recalculated using detrended and detrended-deseasoned mass anomalies and are given in Supplementary Figs. S3 and S4 , respectively. These results reveal that the Sol1 provides significantly the best simulation and outperforms the previous studies when compared to CSRM within the GRACE/-FO period. For simplicity, we will use the name ‘ResDCAE’ to represent the DL model ‘Sol1’ in the rest of the paper. External Validation Comparison with Greenland long-term surface mass balance estimates In order to validate our simulation results, we performed a comparison with independent surface mass balance estimates of Greenland. Greenland is particularly chosen as a test bed because of the following three main reasons, among others. First, Greenland surface mass balance data is a unique independent data set which has a long temporal coverage, e.g., the IMBIE (Ice sheet Mass Balance Inter-comparison Exercise) surface mass balance data record starts ten years before the GRACE era. Second, the beginning of a dramatic increase in the ice mass loss trend was observed in ~2002 58 , almost right before the launch of GRACE mission. Thus, Greenland is the most challenging region on the Earth with a perfect data set to test the performance of the strategy adopted to mitigate the trend error in backwards extrapolation in this study. Third, as a consequence of the exacerbated global climate change, the melting of the Greenland ice sheets, and its peripheral glaciers and ice caps is the major contributor to contemporary sea level rise 59 . Here, we use the most recent Greenland surface mass balance data from the IMBIE as the reference for comparison. The IMBIE data has been produced using 26 estimates of ice sheet mass balance derived from satellite altimetry (9 datasets), satellite gravimetry (14 datasets) and the input–output method (3 datasets) to assess changes in the Greenland Ice Sheet mass balance 27 . Prior to 2003, the estimates are solely from input-output method consistency of which with the estimates from satellite altimetry and gravimetry has been shown for common spatial and temporal domains after 2003 (see Fig. 2 of The IMBIE team 27 for the number of individual mass balance estimates and the temporal coverage of the ach measurement type used). The final IMBIE surface mass balance data is available as a reconciled time series of cumulative total mass changes between 1992 and 2018 along with the uncertainty estimates and can be downloaded from http://imbie.org/data-downloads/ . The cumulative total mass change has been produced by integrating the rates of mass changes computed at annual intervals from time series of relative mass change using a 3-yr window. Therefore, for a fair comparison, we applied a 13-months moving average filter to our monthly mass change simulations from ResDCAE. The cumulative mass changes from the IMBIE surface mass balance and from ResDCAE (ResDCAE Sm °° thed ) starting from 1994 are shown in Fig. 4 . Note that the IMBIE data is referred to the mean baseline between 2004.000–2009.9999 to be consistent with the ResDCAE in the plot. The time series of original monthly ResDCAE mass change simulations as well as the estimated 1-σ uncertainties of the IMBIE are also presented in Fig. 4 . Figure 4 shows an almost excellent agreement between IMBIE and the ResDCAE simulations throughout the entire time span. The slight differences between ResDCAE and IMBIE cumulative mass change time series are all within the 1-σ envelope of the IMBIE. The standard deviation is computed as ± 90 Gt based on these differences. The results indicate that our simulations are not only accurate within the GRACE/-FO era in which the training of the DL model was performed, but also provide good predictions of mass anomaly for the pre-GRACE era; the trend error mitigation strategy seems to work reasonably well even if not perfect. Validation with ENSO events The most active climate variability on the interannual timescale affecting long-term mass anomaly values is the El Niño–Southern Oscillation (ENSO), which results from large-scale ocean–atmosphere interactions over the equatorial Pacific 60 – 63 . Positive Sea Surface Temperature Anomalies (SSTA) in the eastern or central equatorial Pacific Ocean, as well as a weakening of equatorial trade, define the first phase of ENSO events, which occur every 2–7 years on average. El Niño events produce several severe droughts 64 , 65 in the western Pacific and floods 66 , 67 in the eastern Pacific, affecting climate globally. Besides that, the negative phase of ENSO is La Niña, which is a phenomenon in the tropical Pacific causing exceptional cooling of SSTs. The Southern Oscillation is the other part of ENSO, and it is a large-scale see-saw trend in the sea level pressure between the eastern and western tropical Pacific. El Niño results in low sea-level pressure in the eastern Pacific and higher pressure in the western Pacific, whereas La Niña has the reverse effect. Statistical models are frequently employed to determine ENSO evolution, with the SSTA index of Nino3.4 (120°W–170°W, 5°N–5°S, as given in Supplementary Fig. S5 ). These anomalies are the deviation of monthly SSTs from their long-term mean. When the Nino3.4 index surpasses + 0.5 °C and −0.5 °C for at least five consecutive months, it is considered an El Niño and a La Niña event, respectively. The 2015/16 El Niño is the most powerful event in recorded El Niño history. It exceeded the previous two extreme occurrences in 1997/98 and 1982/83 68 , 69 . The ENSO has been demonstrated to have a significant impact on precipitation and air temperature in a variety of regions 70 – 72 . Mass anomaly is highly dependent on the integrated water mass changes due to precipitation, evapotranspiration, and runoff. It is also heavily influenced by regional meteorological circumstances such as droughts, flooding, and extended periods of high temperatures. All these components, particularly precipitation, are linked to ENSO. As a result, it is possible to conclude that ENSO and mass anomaly are related, as well. Several studies 66 , 70 , 73 – 75 have demonstrated the impact of ENSO on mass anomaly in different basins of the Earth. For instance, Chen et al . 66 explored the relationship between interannual mass anomaly variations and ENSO occurrences in the Amazon basin. Furthermore, Ni et al . 75 analysed this phenomenon globally and discovered that ENSO occurrences have a significant impact on local Precipitation Anomalies (PPTA) and interannual TWS variations. As the first external validation of our reconstruction, ResDCAE, a thorough assessment focused on the relationship between interannual mass anomaly and ENSO was carried out. It was also compared to previous studies which utilized the TRIP database for major river basin boundaries 44 . The mass anomaly reconstruction models are validated both with GRACE/-FO mass anomaly data and through examining their relationship with precipitation anomalies from ERA5, the National Oceanic and Atmospheric Administration (NOAA) Climate Prediction Center (CPC) 76 and the Global Precipitation Climatology Center (GPCC) 77 global precipitation dataset. The three different precipitation data above is chosen in order to ensure fair comparison among simulated mass anomaly from different studies. This is because ERA5 precipitation is an input dataset in both our ResDCAE and Humphrey while Li used CPC as an input for reconstruction. Furthermore, GPCC was also chosen due to its independence, regardless of the relationship between the input and simulations. Before the dataset used in validation, both CPC and GPCC were resampled to monthly temporal and 1.0° × 1.0° spatial resolutions. First, the average signals of reconstructions and precipitation datasets over river basins are calculated considering TRIP basin boundary data, which are also illustrated in Supplementary Fig. S5 . In order to ensure consistency between these time series, 5-month moving average filters are applied as in the study by Ni et al . 75 , and then these smoothed time series are temporally matched to obtain the same time coverage for all datasets. According to Ni et al . 75 , the Amazon basin exhibits the highest correlation between mass anomaly and ENSO events, according to their global analyses, and it takes 5 months for the influence to appear in the basin. The illustration of average basin signals for the Amazon basin is given in Fig. 5 . In Fig. 5a ,the Nino3.4 index ( http://www.cpc.ncep.noaa.gov/data/indices/ ) is employed as a measure of ENSO activity in a comparison to simulated mass anomaly in the Amazon basin. A similar comparison of simulated mass anomaly versus precipitation anomalies from three different precipitation data set is also given in Fig. 5b . Figure 5a,b show that interannual mass anomaly variations are strongly linked to ENSO and precipitation with several months of time lags over the Amazon basin, particularly during massive El Niño events in 1997/98 and 2015/16. Furthermore, when the time series of the models are compared to CSRM, it is found that the ResDCAE model distinguishes rapid changes more easily and matches well with the CSRM mass anomaly. The GRACE and pre-GRACE correlations between the time series of all investigated mass anomaly solutions and the SSTA time series are calculated separately. The calculated correlations are all maximal values with specific phase lags. The correlations for the comparison between ResDCAE and SSTA in the GRACE-period were −0.57 (6-month lags) and −0.85 (4-month lags) in the pre-GRACE period. Similarly, the correlations for the Li, Humphrey, Löcher, and CSRM time series were calculated as −0.65 (5), −0.73 (6), −0.60 (7), and −0.57 (5) for the GRACE period, whereas they (excluding CSRM) were calculated as −0.87 (6), −0.84 (7), and −0.63 (5), respectively for the pre-GRACE period. During the GRACE period, our ResDCAE solution has nearly the same correlation as the CSRM time series. This results in a significant convergence to the simulated CSRM data. On the other hand, our simulation is also consistent with other studies in both time periods. The time series of all three precipitation anomaly data are in good agreement with the mass anomaly simulations. In order to quantify the Cross-Correlations (CCR) between these precipitation anomalies and mass anomaly simulations at all river basins in the TRIP database, the CCR coherence spectrum is calculated by determining and considering time lags between all signals separately. Totally 176 river basins are considered in this comparison. Each comparison, e.g., ResDCAE vs ERA5 PPTA for all basins, has its own defined time lag, which is determined by the maximum correlation computed between the compared time series. Thus, both cross-correlations and time lags are calculated for all compared time series pairs at each river basin. CCR metrics are given in Fig. 6a–l for the comparison between mass anomaly simulations and ERA5, CPC, and GPCC datasets. The lags between the time series of mass anomaly simulations and precipitation are also given in Supplementary Fig. S6 . According to Fig. 6 , almost similar CCR are calculated between all mass anomaly simulations and precipitation datasets, except for Löcher’s. The discrepancies between precipitation and mass anomaly simulations of Löcher is most likely due to the fact that, contrary to ResDCAE, Humprey and Li, no precipitation data was considered when Löcher’s spherical harmonic coefficient models are calculated. On the other hand, as expected, Humphrey’s simulations have slightly higher correlations with PPTA in all basins, because the precipitation data has been directly used as the dominant input in a linear water store model (see Eq. 1 of Humprey and Gudmundsson 6 ). Nevertheless, all simulations have similar CCR in almost all river basins. As it can be seen in Fig. 6a–l , the Amazon River basin has the highest CCR (>0.75) in all comparisons. Validation with independent degree-2 spherical harmonic coefficient estimates The long-wavelength components of gravity change due to variations of mass redistribution on Earth can also be recovered from SLR tracking measurements or EOP, independently 48 , 78 . Thanks to advancements in satellite geodetic techniques, EOP- and SLR-derived low-degree spherical harmonic coefficients, i.e., ∆C 20 , ∆C 21 and ∆S 21 , can be determined with higher accuracy than GRACE/-FO observations 48 . These degree-2 coefficients are related to the different geophysical dynamics of mass redistribution on Earth 1 , 48 . These relationships could be exemplified by the fact that while SLR-derived ∆C 20 provides information about mass variations due to the oblateness of the Earth, ∆C 21 and ∆S 21 are related to the variations of the Earth’s rotational axis 79 , 80 . Therefore, degree-2 spherical harmonic coefficients recovered from GRACE/-FO could be validated independently using EOP- and SLR-derived counterparts. In order to validate our simulations, global 1.0° × 1.0° gridded mass anomaly from ResDCAE were first converted to spherical harmonic coefficients using Eq. 35 of Wahr et al . 81 . The maximum degree and order of the spherical harmonic expansion were chosen as 96. Note that these spherical harmonic coefficients represent the relative anomalies w.r.t the 2004.0–2009.9999 mean baseline, i.e., the coefficients are ∆C nm and ∆S nm . Similarly, CSRM mass anomaly were also converted to spherical harmonic coefficients (CSRM spherical harmonic coefficients). On the other hand, monthly C 20 coefficients estimated from SLR were taken from https://grace.jpl.nasa.gov/data/get-data/oblateness/ and the 2004.0–2009.9999 mean was removed from the entire time series to calculate the SLR ∆C 20 series consistent with those of ResDCAE and CSRM. Note that the background models adopted within CSRM and CSR GRACE/-FO RL06 processing chain have also been used for the recovery of SLR-only gravity field models for comparison in this study 78 , 79 . Further, EOP-derived, ∆C 21 , ∆S 21 , and ∆C 20 coefficients were calculated from the mass term of the Earth Rotation excitations using Eq. 2 of Chen et al . 82 . Mass excitations are obtained by subtracting motion terms from observed excitations of polar motion components (X, Y), and Length of Day (LOD), respectively. While the mass excitations are due to the mass load variations, the motion excitations arise from the angular momentum exchange between the Earth’s crust and the atmosphere, i.e., due to the frictions of atmospheric wind and ocean current fields on the Earth 82 , 83 . Daily Polar motion (X, Y) and LOD observations are taken from the International Earth Rotation and Reference Systems (IERS) EOP 14 C04 series 84 . In order to calculate mass terms, the daily observed and motion excitations were computed using interactive tools of the IERS ( https://hpiers.obspm.fr/eop-pc/analysis/excitactive.html ) with the Chandler period set at 433 days and quality factor at 100. The motion terms are also computed considering the angular momentum series, namely ECWMF and Max-Planck-Institute for Meteorology Ocean Model (MPIOM), that are provided by GFZ 85 which also serve as a basis for atmosphere and ocean de-alising (AOD1B RL06) model in GRACE/-FO data processing 83 . Thus, the consistency between ResDCAE and EOP-derived also ensured using these models as angular momentum components of motion terms. The daily mass excitations are obtained by removing motion terms from observed excitations and these datasets comprise different periodic signals, such as the 5.8-yr oscillation in the observed LOD series, which is sourced from core-mantle interaction and is not related to gravity change 86 . Thus, first a zero phase-shift Butterworth high-pass filter with a cut-off frequency of 1/4 cpy is applied to remove this oscillation and other long-period signals from the LOD series. Besides, the linear trends were removed from all mass excitations using unweighted least squares trend estimation because the long-term variation was in good agreement with the EOP- or SLR-derived series at seasonal time scales. After that, the daily degree-2 coefficients were computed using these detrended mass excitations, and a low-pass filter with a cut-off frequency of 6 cpy was applied to remove from the every signal shorter than 2 months. Finally, the monthly ∆C 21 , ∆S 21 , and ∆C 20 were computed by averaging the daily ones in each month. In order to make a fair comparison, the corresponding time series of ResDCAE, CSRM and SLR were also detrended using unweighted least squares. All the detrended series are presented in Fig. 7 . All degree-2 coefficient time series are in a good agreement of ± 2e-10 for both GRACE and GRACE FO period (Fig. 7 ). In contrast to ∆C 21 and ∆S 21 , EOP-derived ∆C 20 has a phase-shift of about 4-months that may be sourced from high-pass filtering applied to the remove the long-period signals from ∆LOD, and therefore not shown in Fig. 7a . On the other hand, the seasonal amplitude of ∆S 21 series is higher than that of ∆C 21 . This is because ∆S 21 is more sensitive to mass changes over land while ∆C 21 is more sensitive to mass changes over the oceans. Similar to the results in e.g., Meyrath et al . 86 and Chen et al . 82 , while GRACE/-FO-derived (ResDCAE and CSRM) ∆C 20 coefficients are more consistent to those estimated from SLR ones, the ∆C 21 and ∆S 21 from ResDCAE and CSRM are both closer to each other and the EOP-derived estimates. Zero phase-lag correlations, after removing annual/semiannual variations and linear trends using unweighted least squares fit, between every pair of estimates of ∆C 20 , ΔC 21 and ΔS 21 shown in Fig. 7 are listed in Table 2 . As expected, for all three coefficients, the highest correlations (0.94-0.96) are observed between ResDCAE and CSRM estimates when compared to others within the GRACE/-FO era. The correlations between the ∆C 20 from ResDCAE and from SLR within and pre-GRACE era are still as high as 0.80 and 0.70, respectively. The correlations for ∆C 21 are slightly smaller than those for ∆S 21 in all comparisons, which may be due to the lower signal-to-noise ratio of ∆C 21 as it is more relevant to the mass change signal over the oceans. The correlations given in Table 2 within GRACE/-FO era are all comparable to the results from earlier studies, e.g., Chen et al . 83 , Meyrath et al . 86 although they used the coefficient estimates from GRACE/-FO Level-2 GSM data rather than mascon solutions. This, as well as the different time spans used to compute correlations explain the slight differences between their results and the results in this study. It is worth noting that the degree-2 coefficients from our ResDCAE model and from GRACE/-FO are converted from mascon type mass grids on Earth. The coefficients from ResDCAE model simulations also reveal reasonably high correlations (0.58 for ∆C 21 and 0.65 for ∆S 21 ) with EOP-derived ones before the GRACE era which shows the efficiency of the methodology (DL + backwards trend error mitigation strategy) used in this study. Validation with global barystatic mean sea level change data Barystatic mean sea level change data is another independent data source used to validate GRACE/-FO temporal gravity solutions 48 on global scale. Satellite altimetry has been a well-established space geodetic technique for accurately measuring global sea level change for about three decades. Thus, we compared our simulation results over the oceans with the altimeter-observed Global Mean Sea Level (GMSL) change time series after appropriate preprocessing steps were applied. To this end, the time series of GMSL anomalies (from altimetry) and associated steric components are calculated and compared to mean ocean mass change retrieved from ResDCAE simulations and CSRM. GMSL records consist of both ocean mass and steric components and can be expressed with the sea level-budget equation as GMSL altimetry = GMSL steric + GMSL oceanmass . While GMSL steric represents the contributions of oceans’ thermal expansion and salinity to sea-level variations, GMSL oceanmass represents the change in ocean mass 87 , 88 . Before comparing the time series of GMSL to ocean mass anomalies from ResDCAE and CSRM, GMSL steric must be removed from the GMSL altimetry anomalies to obtain GMSL oceanmass in order to make physically consistent comparison. The GMSL altimetry dataset is provided by Copernicus Marine Environment Monitoring Service (CMEMS) and the Copernicus Climate Change Service (C3S) and derived from ECWMF database 89 , 90 , which includes the gridded (with daily and 0.25° × 0.25° spatial resolutions) merged satellite altimetry sea level anomaly data products combining different satellite altimetry observations and covering the time span from January 1993 to present. We downloaded the gridded GMSL altimetry dataset from ECWMF from January 1994 to December 2017. We confined our comparison between this time span as there is no reliable data available for computation of steric contribution beyond December 2017 91 . Some pre-processing steps such as upsampling and some corrections, were applied before the calculating GMSL altimetry time series from gridded dataset. First, GMSL dataset were upsampled to monthly 1.0° × 1.0° grids by averaging from its own native resolutions to ensure consistency with our simulations. After that, the so-called TOPEX-A instrumental drift corrections, which is sourced from the instrumental problems of satellite and spanning the period from January 1993 to December 1998 87 , were added to each grid using provided correction values along with the dataset. On the other hand, GMSL steric component was calculated from gridded steric-height anomalies that are retrieved from Camargo et al . 92 , which is the ensemble mean derived from the 10 different temperature and salinity data sets and has monthly sampling with 1.0° × 1.0° spatial resolution covering oceans between 66° N–66° S latitudes from January 1993 to December 2017. GMSL altimetry and GMSL steric anomalies are calculated considering the mean baseline between 2004.0 and 2009.9999 to ensure consistency with ocean mass change from our simulations as well as from CSRM. Then the time series are obtained by averaging grids over the oceans, excluding a 300-km buffer zone along the coasts to avoid any signal leakage from land hydrology. The average monthly sampling of time series was obtained from the weighted ocean mass change grids. The weights were determined considering the surface area of each grid cell over the oceans within 65° N–65° S latitudes. In addition, GIA correction was applied to GMSL time series by adding a constant value of −0.23 mm/year derived from the ICE-6G_D VM5a model 39 , which also was used as GIA correction CSRM. GMSL oceanmass time series was then calculated by removing GMSL steric from GMSL altimetry time series to compare with ocean mass change from ResDCAE simulations and CSRM. The same averaging procedure was also applied to the time series of ResDCAE simulations (from January 1994 to December 2017) and CSRM (from April 2002 to December 2017). The GIA and GAD corrections are readily included in our simulation as it uses the corrected version of CSRM as output data for training the DL model. Finally, the seasonal (annual/semi-annual) signals were removed using unweighted least-squares from all time series and a moving average filter with a window length of 400 days was applied to each of the time series before comparison. Ocean mass change time series both from our ResDCAE simulation and CSRM as well as the altimetry derived GMSL oceanmass are given in Fig. 8 . While trend values are calculated for GRACE period as 2.11, 2.15, and 2.21 mm/yr, they are calculated for pre-GRACE period only for ResDCAE and GMSL oceanmass as 0.13 and 0.77 mm/yr shown in Fig. 8 . The long-term linear trends estimated from ResDCAE, CSRM, GMSL altimetry , GMSL steric and GMSL oceanmass time series are 1.47, 2.15, 2.82, 1.47 and 1.34 mm/yr, respectively. The deseasoned time series of ocean mass change are consistent to each other especially after 2004. This improvement can be attributed to the accurate Argo-based steric height models developed in early 2005 87 , 93 , 94 . Dieng et al . 87 has shown that the ensemble members of steric heights used in various studies show significant differences between 1993 and 2004. However, Argo data from January 2005 to end of 2015 significantly reduce the uncertainties of the steric sea level change data products 87 , 94 . We re-estimated the linear trends from all time series but for the time period from January 2005 to December 2017 and obtained 2.26, 2.26, 3.58, 0.88 and 2.70 mm/yr, respectively for ResDCAE, CSRM, GMSL altimetry , GMSL steric and GMSL oceanmass . The GRACE-based (ResDCAE simulations and CSRM) ocean mass change time series and altimetry-steric (GMSL oceanmass ) are all in excellent agreement throughout the Argo data time-span. We also computed the correlations of the original (non-detrended/non-deseasoned) ResDCAE simulations as well as of the CSRM with the Altimetry-Steric (barystatic) sea level change time series. Within the GRACE era (April 2002 – December 2017) a correlation coefficient of 0.86 is obtained both with ResDCAE and CSRM. The correlation coefficient computed between ResDCAE and the Altimeter-derived barystatic sea level for the pre-GRACE era (January 1994 – March 2002) is still as high as 0.79, indicating reasonably well simulation performance and the effectiveness of the trend error mitigation strategy (c.f. section Methods) adopted in this study. Validation with in situ ocean bottom pressure data The simulated mass changes over oceans can be compared to in situ ocean bottom pressure (OBP) observations for qualitative validation. The detection of spatiotemporal mass variations over oceans from GRACE/-FO is more challenging than those of the continental hydrology signal since the detectable variations of gravity signal over the ocean are weaker 95 . Moreover, the comparison of these variations to independent point-wise OBP variations is much more challenging due to the differences between spatial and temporal resolutions, the irregular distribution of OBP stations over the globe, or the necessity of isolating signals that are related to ocean circulation and sea level changes 96 . Considering these aspects, in situ OBP observations that are publicly available in the Permanent Service for Mean Sea Level (PSMSL) database ( https://www.psmsl.org/data/bottom_pressure/ ) were used to compare to the gridded mass change time series from our simulation and CSRM. To this end, we chose two stations from PSMSL database considering the temporal coverage of data records (stations with data records available for longer time) and taking into account the proximity and distance from the land to test any possible leakage effect from the land. Thus, the stations Dark Passage South – DPS (60.9° S–54.7° W) and NDBC 51406 – Central South Pacific (8.5° S–125.0° W) were chosen the start and end deployments of which are November 1992 – June 2011 and September 2001 – February 2013, respectively. The daily sampled mean OBP data constructed from hourly observations at each deployment are readily available after removal of diurnal and shorter period tides by averaging 24 hourly values and sensor drift corrections. The long-term trends as well as the remaining drifts were first removed with a quadratic fit from daily OBP time series at each deployment as suggested by Poropat et al . 96 in situ . In order to generate monthly sampled OBP time series, first the low-pass Butterworth filter was applied with 12 cpy cut-off frequency to remove any remaining signal with sub-monthly periods from the daily time series. The monthly time series were then computed by averaging these filtered daily samples. On the other hand, the monthly mass change time series from ResDCAE and CSRM at the two 1.0° × 1.0° grids which contain the selected OBP stations were first detrended. The monthly time series of in situ OBP and mass change usually need to be compared for long-wavelength signal content due to the complexity between observed mass change by GRACE and OBP induced by their differences in spatial and/or temporal resolutions. Therefore, six-months moving average filter was further applied to monthly sampled ResDCAE and CSRM time series. The resulting time series are given with a dual axis plot in Fig. 9 . The agreement between in situ OBP and mass change time series is better at NDBC station than at DPS station as shown in Fig. 9a,b . This is most likely due to the signal leakage from land and sea ice at northern Antarctica to the mass change signal at DPS station. The NDBC station, on the other hand, is in the open ocean and thus almost no signal leakage from land hydrology exists. The comparison of GRACE-based mass change solutions to in situ OBP observations is highly dependent on both oceanographic priors and applied post-processing as well as the basin size adopted for averaging while generating the mass change time series 96 , 97 . Despite the fact that the general signal preprocessing tools such as filtering and smoothing were applied to obtain time series of ResDCAE, CSRM, and OBP observations, the agreements with in situ OBP records at these two different locations over the oceans are comparable with those in previous studies 98 , 99 . Summary and Future Perspectives In this study we employed a hybrid deep learning architecture called resDCAE to simulate mass anomalies at a spatial resolution of 1.0 degree by 1.0 degree and a monthly temporal resolution from January 1994 to January 2021. We proposed and successfully performed a strategy to reduce the error of the trend component in the simulations during the pre-GRACE period (1994 to 2002). The primary objective was to achieve a better understanding and characterization of various climate-induced geophysical phenomena, including the terrestrial water cycle, ice sheet and glacier mass balance, sea level changes, and variations in ocean bottom pressure by providing longer time series of global water storage changes both over continents and oceans. The research demonstrated that the use of a combination of ERA5 and SLR datasets, along with time channel information, provided better, if not the best, solution for simulations. This study contributes to the monitoring and comprehension of long-term global gravity field changes, offering valuable insights into climate change and other significant geophysical events. Such research is advancing our understanding of climate changes and their impacts on Earth’s water cycle. With the new data sets as well as advanced satellite gravity missions and developments in deep learning era and algorithms, improved simulations of the water mass change with enhanced resolutions will be possible in the future. Usage Notes The simulated data is available with no gaps from January 1994 to end of December 2020 both in the form of monthly 1.0° × 1.0° mass anomaly grids and spherical harmonic coefficients similar to official GSM data products but with a much higher resolution up to degree and order 200. The user should note that the data set may not include seismic signal and thus is not proper for e.g. earthquake signal detection. For conversion of mass anomaly grids to spherical harmonic coefficients, Equations 6–8, and load Love numbers in Table 1 of Wahr et al . 81 were used with ρ ave = 5517 kg/m 3 as the average density of the solid Earth. Both data sets represent the anomalies relative to 2004.0–2009.9999 mean baseline similar to CSR mascon solutions. When using spherical harmonic coefficients data, no further destriping or smoothing filter is required. Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s41597-023-02887-5. Acknowledgements This work is partially supported by Scientific and Technological Research Council of Turkey - TÜBİTAK (Contract number 119Y176), by the United States Agency for International Development (USAID)/Indian Partnerships Program (Cooperative Agreement: 72038621CA00002), by National Geospatial-Intelligence Agency (NGA) GEO-ESCON Program (No. HM157522D0009, Task 8.8), by NSF Partnerships for Innovation Program (2044704) and by NASA Earth Surface Interior Focused Area Program (80NSSC20K0494). Dr. Anno Löcher from University of Bonn is gratefully acknowledged for providing monthly gravity field solutions from SLR data. Special thanks to the editors and the two anonymous reviewers who significantly improved the quality of the manuscript with constructive comments. Author contributions This work is a part of M.U.’s PhD dissertation supervised by O.A. at Istanbul Technical University. M.U. conceived the idea with O.A. and C.K.S., and developed the initial framework. K.G.A. and M.U. designed and numerically implemented the deep learning models adopted in the study. S.O. and Ö.G. contributed in the data curation and preprocessing. M.U., O.A., C.K.S. and K.G.A. wrote the manuscript. All authors commented on the manuscript. Code availability There is no customized code in generation or processing of datasets. For setting up and training the Deep Learning Models, the publicly available codes in Python language from TensorFlow 9 and Keras 10 libraries were used. The trend error mitigation and all the figure plots in the paper were implemented using the existing routines/functions in MATLAB software. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Data. 2024 Jan 13; 11:71
oa_package/a2/10/PMC10787793.tar.gz
PMC10787794
38218858
Introduction Neural networks have made significant contributions to the field of Artificial Intelligence, serving as both a tool for mathematical modeling and a means to understand brain function. The Hopfield paradigm 1 , 2 has played a crucial role in this domain, utilizing a synaptic matrix to represent the interconnections between neurons. This matrix possesses the remarkable ability to store and recognize patterns, and serve as a fundamental framework for the realization of future content-addressable memory (CAM) 3 , 4 . To store a memory consisting of N elements, the widely adopted approach is to employ Hebb’s rule 5 . This rule entails constructing a synaptic matrix, denoted as T , by taking the tensorial product of the vector φ * (representing the pattern to be stored) and its conjugate transpose ( φ *† ): However, there is a fundamental limit to the number of memories that can be reliably stored using Hebbian-based approaches. As the network becomes more densely populated, the interactions between different memory elements can lead to the emergence of unintended and uncontrolled memory states 2 . To address this limitation, recent research has explored various methods to enhance the capacity of neural networks: dilution 6 – 9 , autapses 10 , 11 , and convex probability flow 12 , 13 . Recently, it was proposed to leverage the interaction among stored patterns in a constructive way: an emergent archetype may be stored by proposing to the network multiple prototypes that closely resemble the target pattern but are intentionally corrupted or filled with errors. The interaction between these prototypes serves to strengthen the emergence of the desired memory 14 . This paradigm is connected to the prototype concept developed in hierarchical clustering, in which prototypes are elements of the dataset representative of each cluster 15 . In this study, we introduce a novel learning strategy called Stochastic Emergent Storage (SES). SES taps into the abundance of natural randomness to construct an emergent representation of the desired memory. Capitalizing a vast database of fully random patterns freely produced by a disordered, self-assembled structure, we select a set of prototypes that bear resemblance to the target memory through a similarity-based criterion. Subsequently, by performing a weighted sum of the synaptic matrices corresponding to these selected prototypes, we are able to effectively generate the desired pattern in an emergent fashion. Given the inherent advantages of photonic computation, such as ultra-fast wavefront transformation and parallel operation, it results that optics is the ideal domain to explore the SES paradigm. The convergence of photonics, artificial intelligence, and machine learning represents a highly active and promising area of research 3 , 16 , leading to novel interdisciplinary paradigms such as Diffractive Deep neural networks 17 , 18 photonic Ising machines 19 and photonic Boltzmann computing Machines 20 . However, these approaches typically rely on direct control over optical properties of millions of scattering elements, which can be challenging and costly both with microfabrication or adaptive optical elements. In a departure from traditional approaches, disordered scattering structures have been proposed as a radically different avenue for optical computation in various applications: classification 21 , vector-matrix multiplication 22 , computation of statistical mechanics ensembles dynamics 23 , and others 24 . Here, we propose to employ the scattering intrinsic patterns, the optical transmission matrices, to realize a SES-based optical hardware, the disordered classifier. This device is capable of efficiently performing pattern storage, and subsequent pattern retrieval. It is able to simultaneously compare an input pattern with thousands of stored elements, and it enables a two-layer architecture, providing categorical (deep) classification, which allows for more complex tasks.
Methods Background In our experiment (similarly to a typical wavefront shaping experiment), light from a coherent source is controlled by a spatial light modulator and transmitted after propagation through a disordered medium into the mode ν . The field transmitted at ν is described as where the index n runs on the controlled segments at the input of the disordered medium, is the field resulting from laser field from the n t h segment transformed by the transmission matrix element on the sensor ν and is the phase value from the wavefront modulator. In our experiment, we consider the simplified configuration in which . The field at ν can be separated in its two components: the field-at-the-segment A n and transmission matrix elment Indeed the are Gaussianly distributed complex numbers: In the case in which just two segments n and m are active and in the + 1 configuration, we can ignore the φ n : In absence of modulation, intensity is written as the modulus square of E ν we recognize that thus or In general for N segments in an arbitrary configuration of the modulator the argument of the sum can be written in matrix form defining the matrix V ν also named optical coupling matrix: Matrix V ν is a bi-dyadic matrix and it can be rewritten in matricial notation: where the notation indicates a vector on lowercase Greek letters and a matrix on uppercase, while † is the conjugate transpose operator. Being bi-dyadic the matrix possesses the eigenvectors ξ ν and η ν by construction. Note that , V and is Hermitian. When modulation is present with an input modulation pattern φ Note that even if V ν is a complex matrix, being Hermitian, the double scalar product produces a real scalar because inverted sign imaginary contributions from above and below the diagonal result eliminated reciprocally thus producing a positive real intensity. The optical operator V ν , associates thus the pattern/array φ to the scalar I ν which is a measure of the degree of similitude of φ to the first eigenvector of V ν , EIG( V ν ) = ξ ν . Note that to simplify the realization of the experiment, we operate in the configuration in which each mode ν corresponds to a single sensor. As we employ a camera to measure I ν , e the one-mode-per-pixel configuration is obtained by properly tuning the optical magnification. Stochastic Hebb’s storage protocols details By summing intensity measured at two modes ν 1 and ν 2 , and considering linearity of the process: Generalizing, i.e. summing intensity at an arbitrary number M of modes pertaining to the set , we retrieve with Thus, the optical operator textbfJ associates a pattern/array φ to the scalar , the transformed intensity , which is a proxy of the degree of similitude of φ to the first eigenvector of J : EIG( J )= . To deliver an user-designed arbitrary optical operator, we introduce the tailored attenuation coefficients λ ν ∈ [0, 1]. These can be both obtained in “software version” (multiplying each I ν by an attenuation coefficient λ ν ) or by realizing a mode-specific hardware optical attenuator (such as proposed in the sketch in Fig. 1 , fuchsia windows, see below). Transformed intensity with the addition of the attenuation coefficients reads as: In SHS, the absorption coefficients λ are the free parameters which enable to design the arbitrary optical operator J . For example, to replicate the dyadic matrix constructed with he Hebb’s rule T and capable to store the pattern φ (see Fig 1 c of the main paper) one has to select λ so that the function is minimized. We name the artificial optical synaptic matrix in which λ have been optimized to deliver the optical operator T , and the relative transformed intensity . This approach employs the random, naturally-occurring optical synaptic matrices from the set as a random basis on which to build the target optical operator. Its effectiveness is thus dependent on the number of free parameters with respect to the constraints. The constraints are the number of independent elements that have to be tailored on T . These are as T is symmetric. Indeed as shown in Fig. 2 of the main paper (inset of panel b) for the N = 81 case, it is possible to replicate almost identically T when M > Π, that is when the number of free parameters (the λ ) is comparable with the constraints. Storage error probability In our storage paradigm, the stored pattern corresponds to the eigenvector of the T . As we are employing binary patterns, the sign operation is needed. The stored pattern is thus SEIG( )= ξ Σ , where the S E I G () operator retrieves the first eigenvalue of a matrix and applies the sign operation to it. The Storage Error Probability reported in Figs. 3 and 2 the storage process effectiveness. First, we calculate the number of elements of ξ Σ which differ from the target memory φ * , S _ E R R . The value of S _ E R R can be seen as the number of error pixels in the stored pattern image. Then we compute For storage purposes, obviously the lower, the Storage Error Probability the better. Recognition error probability The optical operator J associates the transformed intensity scalar to each input pattern φ : we can thus employ the experimentally measured transformed intensity to recognize patterns. We employed a repository of P = 5000 patterns containing digits with random orientation ( https://it.mathworks.com/help/deeplearning/ug/data-sets-for-deep-learning.html ), labeling as recognized patterns, the ones producing a transformed intensity above 10 standard deviations from the values obtained probing randomly generated binary patterns. The value R _ E R R is the number of wrongly identified patterns experimentally. Indeed, the transformed intensity is obtained experimentally optically presenting the pattern to our disordered classifier. The step-by-step presentation procedure is the following: i) the probe pattern φ is printed onto a propagating laser beam employing a DMD in binary phase modulation mode (see experimental section), ii) light scattered by the disordered medium is retrieved for the relevant mode/pixel set , iii) the transformed intensity measured by the selected sensors/camera pixels is obtained with Eq. ( 30 ), iv) a pattern is defined as recognized if the trained transformed intensity results higher than the threshold. The Recognition Error Probability is then obtained as The Supplementary Fig. 4 visualizes for the classification/recognition process. Note that Storage Error Probability ( S _ E R R ) and Recognition error probability ( R _ E R R ) provide insights on two very different aspects of our storage platform performance. S _ E R R is essentially a storage fidelity observable, counting the ratio of wrong/correct pixels in the pattern to be stored which differ from the target memory to be stored φ * , and accounts for the efficiency of our approach (the emergent storage) to instantiate a target memory in a memory repository. R _ E R R retrieves recognition efficiency, thus reports on the ratio of memory retrieval tests providing wrong memory addresses, when different input patterns from a repository are proposed as stimuli. The S _ E R R influences R _ E R R : i.e. if many error are present in the pattern injected in a repository the recognition fails. However R _ E R R is also affected by other features such as for example the order of nolinearity (we use intensity do appreciate differences in the field thus we employ a second order nonlinearity) which influences the capability to differentiate similar patterns and also the structure of the repository (if the repository contains very similar patterns then the recognition task is more difficult). Thus the relation is not a simple proportionality, while the two observable look at two very different aspects of the memory process i.e. storage fidelity and recognition efficiency. In Deep-SES instead, a single probe pattern is compared with many memories. We performed this task with digital data analysis but all the processes can be realized analogically, by performing pixel selection and weighting with DMDs. In such a case the probe pattern is directly tested against many memories: all the ones composing the training set. For the 9 class digit classification reported in the Fig, 3969 individual memories (441 per class) have been used. Employing a DMD with 33 kHz frame rate would mean essentially performing optical classification in 0.1 seconds. Stochastic emergent storage protocol details In SES (see code and data at 35 ) we exploit the fact that any optical coupling matrix V ν is a bi-dyadic thus hosting two intrinsic but random patterns: thus the optical coupling matrix at location ν , V ν , hosts the two random memory vectors ξ ν and η ν . To employ these disorder-embedded structures as memories we resorted to the following multi step strategy. i. We start measuring the transmission matrices from a large set of modes employing the Complete Couplings Mapping Method (CCMM, see below). We monitor M L = 65536 modes employing a region of interest for the camera of 256 × 256 pixels in the one-mode-per-pixel configuration. The retrieved transmission matrices are saved into a computer memory and compose our starting random structures repository . ii. We computationally find the first eigenvector ξ ν for each measured matrix V ν iii. The user, designs a target memory pattern to be stored φ * and a number M * of modes to be employed. iv. The target pattern φ * is compared with all the eigenvectors in by computing the similitude degree : with the symbol indicating vector normalization: . v. The set of modes is similarity-decimated to the set , i.e. we select the M * modes with the higher values to be part of the new, reduced repository . Once is realized, we need to “train” the attenuation coefficients λ . The attenuation values are selected between 16 values degrees of absorption in the ∈ [0, 1] range, so that they are identified with a 4 bits number. After initializing the lambda and computing the initial configuration optical operator the λ are optimized with a Monte Carlo algorithm. At each optimization step a single λ ν is modified and the change is accepted if the eigenvector similarity function decreases. Note that in Eq. ( 37 ), ξ Σ is the first eigenvector of J . After a sufficiently large number of steps is minimized and form the final configuration of λ we obtain the final version of the optical operator: . Note that the previous procedure can be cast in a computationally lighter version replacing some digital operations with optical measurements. The similarity selection can be substituted with intensity measurement. Indeed intensity itself is a direct measure (see Eq. ( 24 )) of the degree of similarity of the probe pattern with the correspondent t ν vector, thus similarity selection can be substituted by an optical operation with cost M L . Experimental setup and CCMM The same experimental setup is employed for two tasks. The first is the measurement of the optical synaptic matrices V ν , the second is to perform classification, presenting to the disordered classifier a test pattern φ and retrieving the transformed intensities for each trained memory. A sketch of the experimental setup is provided in Supplementary Fig. 1 in supplementary information file. We employ a single mode laser (AzurLight 532, 0,5W) with beam to about 1 cm. Then it is fragmented into N individually modulated light rays controlled by a Digital Micromirror Device (DMD) 36 composed by 1024 × 768 (Vialux, V-7000, pixel Pitch 13.68 μm, 22 kHz max frame rate) flipping mirrors which can be tuned into two configurations (on or off). Phase modulation is obtained employing the super-pixel method (see refs. 23 , 37 ) which require a spatial filtering to isolate the selected diffraction orders. DMD pixels are organized into N 4-elements super-pixels (segments) capable to produce a 0 or π phase pre-factors equivalent to field multiplication by φ n = ∈ { − 1, 1}. The bundle of light rays is then scrambled by a diffusive, multiple scattering medium (60 μ layer of ZnO obtained from ZnO powder from Sigma Aldrich item 544906-50g, transport mean free path 8 μm 38 ). The N super-pixels are organized on the DMD in a square array, which is illuminated by an expanded laser Gaussian beam (diameter of about 1 cm). Indeed, the DMD surface is imaged onto the Diffusive medium (0.3 × de-magnification). This de-magnification is required to ensure the diffused image to fit into the selected detection camera ROI. Then, the back layer of the disordered structure is imaged on the detection camera (11 × magnification). This magnification has been chosen to minimize the speckle grain size in order to work in the one-mode-per-pixel configuration (one-pixel-per-speckle-grain). The optical collection apparatus, does not require a particular performance, indeed we employed a commercial, low-cost 25.45 mm focal bi-convex lens for the light collection from the far side of the sample. Several constraints have to be considered in the experimental design. When light from a DMD super-pixel emerges from the disordered medium, it is diffused into a larger disk-shaped area. For this reason, we have to ensure that each these light disks are interfering with all the disks generated by other super-ixels in the detection camera ROI, and this introduces a constraint on the maximum ROI size ( M L ). The size of these diffusion disks is regulated by the thickness of the disordered scattering medium. Nevertheless, note that increasing the scattered thickness also decreases the light intensity on the camera and the stability of the speckle pattern thus a trade-off between thickness and signal-stability should be found at the experimental design step. Superpixel method is obtained thanks to 2.66 mm aperture iris in front of the DMD. As shown in Eq. ( 19 ) when two DMD mirrors are activated: while if a single segment is activated Thus putting together Eq. ( 38 ) and Eq. ( 39 ) one obtains Thus to determine one single element of the optical synaptic matrix, one has to perform three intensity measurements. The total number of measurement to reconstruct the full synaptic matrix is Π = N ( N − 1)/2 (as i.e. the optical synaptic matrix is symmetric then just the above-the-diagonal elements need to be measured). For N = 256 this means that 32896 measurements are required, which can be obtained in maximum 5 minutes employing our DMD-Camera experimental setup (speed bottleneck from the camera sensor which works at ~ 150 frames per second). At each measurement we take a image from a Region Of Interest (ROI) of 256 × 256 pixels, thus collecting info for M L = 65536 modes. This measurement realizes modes, random memories and optical synaptic matrix for the database . Experimental data are organized into a 32896 × 256 × 256 matrix. In our case increasing the size of N or M L is limited by the size of the Random Access Memory size of the computing workstation. Satistics and reproducibility In error bars in Figs. 2 – 4 represent standard error, obtained realizing 10 different target matrices T for each M/N value, and measuring standard deviation σ for each dataset and calculating standard error as ERR . T matrices where blindly and randomly extracted at each measure from a 5000 elements pattern repository. No statistical method was used to predetermine sample size. No data were excluded from the analyses.
Results The idea stems from the fact that intensity scattered by a disordered medium into a mode ν resulting from an input pattern φ ) may be written as: with the scattering process driven by the matrix V : generated from the tensorial product of the transmission matrix row (transmission vector) ξ ν ( )with its conjugate transpose ξ ν † . Indeed I ν ( φ ) is maximized if φ ∥ ξ ν : this paradigm is at the basis of the wavefront shaping techniques 25 , 26 , in which the input pattern is adapted to the transmission matrix elements. Thus, scattering into a mode (corresponding to one of our camera pixels, see methods) is described by the same mathematics of the Hopfield Hamiltonian and a pattern is “recognized” (produces maximal intensity) if it matches the ξ ν vector. Given this mapping, V ν may be named an optical synaptic matrix relative to the ξ ν memory. In naturally occurring scattering, one has no control over the pattern ξ ν and the relative optical synaptic matrix V ν because it results from a multitude of subsequent scattering events with micro-nano particles of unknown shape, optical properties, and location. Here, we propose to store an arbitrary, user-defined, memory (or pattern) in naturally occurring scattering media, by exploiting the fact that a scattering process generated billions of output modes, each with a unique and random embedded memory pattern ξ ν and the relative V ν . Thus we propose a new method to realize a photonic linear combination of V ν to generate an artificial, (user-designed) optical synaptic matrix. This method is based on the realization of a sensor collecting the transformed intensity resulting from the incoherent sum of M intensities realized from that many transmitted optical modes from which is a subset of all the modes monitored . Coefficient λ ν ( ∈ {0 − 1} and identified by a 4 bit positive real number) represent attenuation coefficients realized by mode-specific neutral density filters. Then employing the Eq. ( 2 ) in Eq. ( 4 ) we obtain Then, we propose two techniques to design the optical operator J : 1) the Stochastic Hebb’s Storage (SHS) which enables to realize an arbitrary optical operator, 2) the Stochastic Emergent Storage (SES) which instead aimed to the realization of optical memories. Stochastic Hebb’s storage First, we will employ this to realize an optical equivalent of the Hebbs rule: the stochastic Hebbs storage (SHS). Then we will show how the storage and recognition performance is greatly improved if SES is exploited. With the SHS we want to generate a synaptic optical matrix J equivalent to an Hebb’s matrix T with the aim to store the pattern φ * . To do this, we rely on a linear combination of a set of random optical synaptic matrices resulting from uncontrolled scattering: Thus given Eq. ( 5 ), the transformed intensity with the optical operator emulates the Hamiltonian function associated to Hebb’s synaptic matrix T . Indeed the matrix is connected to the intensities of the modes pertaining to the set with the following equation: The values for coefficients λ ν are obtained by a Monte Carlo algorithm, (see methods) minimizing the difference between the target matrix and J . Each coefficient may be then realized in hardware (mode-specific neutral density filters) or software fashion. Employing SHS we can design any arbitrary optical operator if the two following ingredients are available: i) the access to the intensity I ν ( φ ) produced by a sufficiently large number of modes and ii) the correspondent optical synaptic matrix V ν for each mode. This is now possible with the Complete Couplings Mapping Method (CCMM, see methods), which enables the measurement of the intrinsic (no interference with a reference) V ν with a Digital Micromirror Device (DMD). With the CCMM, and the experimental apparatus shown in Fig. 1 see methods we are able to gather a repository of tens of thousands ( M L = 65536) of optical synaptic matrices in minutes from which we sample a random subset (with M random samples) which we use as bases to construct our target artificial synaptic matrix. The performance of this optical learning approach is shown in Fig. 2 , in which we realized an Hebb’s dyadic-like optical synaptic matrix (see insets of Fig. 2 ) from a ZnO scattering layer. The memory pattern stored in our system is , with SEIG( H ) the operator that finds the eigenvector correspondent to the largest eigenvalue of H and then produces a binary vector with its elements’ sign. Performance in storage and recognition for SHS are reported respectively in Fig. 2 a, b (see methods). There we report the Storage Error Probability (the lower the better, indicates the average number of pixels differing between the stored and the target pattern, full definition in the, methods) and Recognition Error (the lower the better, the percentage of wrongly recognized memory elements out of a repository of 5000 presented patterns, full definition in the, methods). SHS is basis hungry, requiring a large number of random optical synaptic matrices (which means modes/sensors/pixels) to successfully construct a memory element. his is connected to the fact that the target matrix T is constructed on N × N /2 parameters (is symmetrical) acting as constraints, while we have M free parameters to emulate it. A full emulation of T is expected thus to be successful for M > N × N /2 which is consistent with what we retrieve in Fig. 2 (Data for M = 4096 are out of scale as storage and recognition error is negligible). Stochastic emergent storage For the remainder of the paper, we will discuss how the performance drastically improves with SES. We recognize that each optical synaptic matrix contains the strongest of two memories ξ ν = SEIG( V ν ) then (instead of randomly extracting modes) we perform a similarity selection (see Fig. 1 a and Supplementary Figs. 1 and 2 in the supplementary information file) in which we extract a set M * whose intrinsic memories are the closest possible to the target pattern φ * (see insets of Fig. 1 ). The fact that in a mesoscopic laser scattering process, billions of independent modes can be produced and millions of them can be measured at once with modern cameras, is strategically employed in SES to boost the performance. To perform the similarity selection with the optical modes the target pattern φ * is compared with the eigenvectors of all the modes in the repository of characterized modes . The comparison is driven by the parameter that quantifies the degree of similarity between the first eigenvector of mode ν , ξ ν , and φ * . The modes ν providing the higher are selected to feed a restricted repository of modes . The correspondent eigenvectors ξ ν can be seen as prototypes of the target archetype, i.e. imperfect representations of the pattern to be stored (such as the one in Fig. 1 c). In SES, these prototypes interact constructively, generating a representation of the memory φ * in an emergent fashion 14 . The interaction is obtained by the incoherent sum of the intensity of several pixels/modes with proper attenuation coefficients/weights λ . The attenuation coefficients λ are found by minimizing the distance between the archetype pattern to be stored φ * and the matrix first eigenvector SEIG (see methods). Thus substituting in Eq. ( 5 ) the transformed intensity in SES reads : The potential of SES is clarified in Fig. 3 : the panels on the top left represent the stored pattern (target pattern is reported in Fig. 1 c) for various sizes M * of the restricted repository. Note that SES greatly outperforms the random selection approach where emergent storage is absent (panel on the right). Figure 3 a shows the storage capability of the system. Blue triangles are relative to patterns with N = 81 elements, while for golden diamonds N = 256. The Storage Error Probability (Fig. 3 c) improves more than an order of magnitude with respect to random selection (red circles). Recognition Error Probability (Fig. 3 b) is three to four orders of magnitude better with respect to the randomly selected database. Note that the SES enormously outperforms SHS, indeed it is possible to perform recognition in the M < < N configuration, i.e. employing a number of camera pixels( M ) much smaller than the elements composing the pattern N . Thus, in summary, SHS enables to create an optical operator of arbitrary nature, which can effectively execute diverse tasks. This versatility arises from its capability to construct an artificial optical synaptic matrix designed by the user, effectively emulating a matricial operator T. Conversely, SES focuses its functionality on generating an operator designed primarily for memory storage, excelling in this singular aspect. Consequently, it demands significantly less computational power and a smaller optical hardware setup (with a smaller M * , see below), and enables lossy data compression (see supplementary information file and Supplementary Fig. 3) . This distinction influences the optimization procedure: SHS optimization relies on distances between matrices (measuring such distance computational cost scales as N × N ), while SES optimization is driven by distances between vectors (measuring such distance computational cost scales as N ). Secondly, SES leverages preliminary similarity selection to identify the most relevant pixels/modes, a feature absent in SHS. As a result, the modes chosen for SES provide higher contrast in the classification task, especially in the M < N regime. In contrast, in the M > N regime (more degrees of freedom than constraints), both approaches achieve essentially the same level of efficiency. Figure 3 c, d shows a recognition process example. The emergent learning process has been employed to store the pattern φ with index j = 241 from a repository of 5000 patterns. Figure 3 c reports transformed intensity for the first 600 repository elements: a clear peak is distinguishable at j = 241 this implies that the pattern is recognized). The same graph is shown for the case of the random basis case (no similarity selection), in which recognition is more noisy. Photonic disordered classifier Our disordered classifier can work in parallel, simultaneously comparing an input with all memories stored, effectively working as a content addressable memory 4 . The experimentally retrieved transformed intensity for 4096 different memory elements φ * is reported in Fig. 4 a (organized in a camera-like 64 × 64 pixels diagram) for the proposed pattern φ . Each value of represent the degree of similitude of φ to φ * . The patterns to the right side of the panel report the proposed pattern φ and the stored patterns relative to each arrow-indicated pixel. The pixel indicated with a red circle contains the pattern most similar to φ thus as expected produces the highest intensity. The system effectively works as a CAM in which an input query φ is tested in parallel against a list of stored patterns (the φ * ) identifying the matching memory as the most intense transformed intensity pixel. The interplay between Hopfield networks and Deep learning has been recently proposed and investigated 27 , 28 . In this framework here we demonstrate a new approach to perform higher rank categorical classification employing the cashed memories as features 29 , 30 : the deep-SES. We tested it on a 4500 randomly tilted digits images repository which is organized into 9 categories (digits from 1 to 9). We stored 3969 patterns/features in the disordered classifier ( m = 441 per each digit), leaving 59 patterns per category for validation. In the camera-like diagrams (Fig. 4 b, d) each category is found in the correspondent quadrant of the image. The two panels show the response of the disordered classifier to the inputs on the left for which the correspondent quadrants show a high number of intense pixels. Figure 4 c shows integrated intensity after threshold. Figure 4 e reports the confusion matrix for all labels, demonstrating categorical recognition efficiency above 90% which eventually may be enhanced employing error correction algorithms 31 . This result demonstrates the possibility to generate deeper optical machine learning achitectures and perform training by simply grouping memories. The potential of Deep-SES is further demonstrated by Fig. 4 f, where we report a figure of merit comparing the efficiency of Deep-SES with Ridge Regression with Speckles (RRS) 21 (simulated). Note, while Deep-SES reaches an efficiency 90% for M * = 40, the RRS suprasses this threshold for M * = 1600. As M * represent the number of physical camera pixel employed in the classification, SES is capable of delivering a classifier with a much smaller hardware and computational complexity. The origin of this advantage, emerges form the fact that our memory writing process, selects pixels/modes which are the most correlated with the pattern to be recognized thus outperform with respect to randomly chosen ones. Moreover deep-SES enables thus to reorganize memories into new classes (reshuffling of classes) with almost no computational cost, a task which typically requires a new training in standard digital or optical architectures (see methods and supplementary information file, “Comparison with other platforms” section and Supplemenatary Table 1) .
Discussion In summary, the Stochastic Emergent Storage (SES) paradigm enables classification with a significantly smaller number of sensors/pixels/modes compared to the elements composing the pattern. This opens up the possibility of fabricating complex pattern classifiers with only a few detecting elements, eliminating the fabrication processes. Deep-SES offers a new paradigm for network training, enabling to generate classes just by grouping memories, and it opens the way to a computation-free rearrangement of classes. The paradigms presented here can be potentially exported to other disordered systems, such as biological neural networks or neuromorphic computer architectures while exploring the emergent learning process in these systems can also provide valuable insights into the memory formation process in the brain. The results presented in this study contributes to the ongoing challenge of understanding the biological memory formation process. There are currently two major hypotheses that are the subject of debate, the connectionist hypothesis 5 , which suggests that neural networks form new links or adjust existing ones when storing new patterns, and the innate hypothesis 32 , which posits that patterns are stored using pre-existing neural assemblies with fixed connectivity. One central aspect in this ongoing debate pertains to the ’efficiency’ of the network, a facet that, in both artificial and natural networks, immediately invokes considerations related to energy consumption. On one side, it has long been established that in Hebbian networks, the number of memories (W) scales linearly with the number of nodes (N), expressed as W = α N . For this reason, many research efforts are dedicated to optimizing the proportionality constant α , which however appears to be upper bounded to two. On the other side, it has been recently demonstrated, both numerically 7 , 9 and theoretically 33 , 34 , that in the stochastic innate approach, the number of memory increases exponentially with the number of nodes: W ∝ e a N . In other words, for larger system sizes, the innate approach predicts a significantly greater number of memories compared to the connectionist perspective. The “complexity” of the system (artificial neural network or brain), denoted as , tends to zero for the connectionists, whereas it remains non-zero for the innatists. SES introduces a fresh perspective to the problem by leveraging the Hebbian structure of the synaptic matrix, with a foundation of the connectionist hypothesis. However, SES goes beyond by exploring the potential of a stochastic innate network in which, pre-existent random synaptic structures are combined to generate memory elements in an emergent manner. Whether the SES could bring a new point of view, lumping together the innatism and connectivism, is a fascinating hypothesis, that must be explored in the future.
Disorder is a pervasive characteristic of natural systems, offering a wealth of non-repeating patterns. In this study, we present a novel storage method that harnesses naturally-occurring random structures to store an arbitrary pattern in a memory device. This method, the Stochastic Emergent Storage (SES), builds upon the concept of emergent archetypes, where a training set of imperfect examples (prototypes) is employed to instantiate an archetype in a Hopfield-like network through emergent processes. We demonstrate this non-Hebbian paradigm in the photonic domain by utilizing random transmission matrices, which govern light scattering in a white-paint turbid medium, as prototypes. Through the implementation of programmable hardware, we successfully realize and experimentally validate the capability to store an arbitrary archetype and perform classification at the speed of light. Leveraging the vast number of modes excited by mesoscopic diffusion, our approach enables the simultaneous storage of thousands of memories without requiring any additional fabrication efforts. Similar to a content addressable memory, all stored memories can be collectively assessed against a given pattern to identify the matching element. Furthermore, by organizing memories spatially into distinct classes, they become features within a higher-level categorical (deeper) optical classification layer. Photonic Stochastic Emergent Storage is a neuromorphic photonic device for image storage and classification based on scattering-intrinsic patterns. Here, the authors show emergent storage employs stochastic prototype scattering-induced light patterns to generate categories corresponding to emergent archetypes. Subject terms
Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s41467-023-44498-z. Acknowledgements This research was funded by: Regione Lazio, Project LOCALSCENT, Grant PROT. A0375-2020-36549, Call POR-FESR “Gruppi di Ricerca 2020” (to M.L.); ERC-2019-Synergy Grant (ASTRA, n. 855923; to GR); EIC-2022-PathfinderOpen (ivBM-4PAP, n. 101098989; to G.R.); Project “National Center for Gene Therapy and Drugs based on RNA Technology” (CN00000041) financed by NextGeneration EU PNRR MUR - M4C2 - Action 1.4 - Call “Potenziamento strutture di ricerca e creazione di campioni nazionali di R&S” (CUP J33C22001130001) (to G.R.). The authors Acknowledge Enrico Ventura, and Luigi Loreti for fruitful discussions. Author contributions M.L. Conceived the Idea, Designed and Realized the experiments, Analyzed the data. G.G. analyzed the data, developed the geometrical interpretation and performed numerical simulations of RRS. G.R. conceived the mapping with the innate and connectionist conjectures. All authors contributed to data interpretation and writing the manuscript. Peer review Peer review information Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Data availability Experimental and generated data related to the generated in this study are deposited in the GitHub repository at the address 10.5281/zenodo.10222344 35 . Code availability Code realized in this study are deposited in the GitHub repository at the address 10.5281/zenodo.10222344 35 . Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Nat Commun. 2024 Jan 13; 15:505
oa_package/fc/45/PMC10787794.tar.gz
PMC10787795
38218740
Background & Summary Decapterus maruadsi , a pelagic fish in the family Carangidae, lives widely in distributed warm offshore waters of East and Southeast Asia 1 . And it is especially abundant along the coasts of the South China Sea 2 . From last century, D. maruadsi has become one of the most commercially valuable marine fishery species in Chinese aquaculture. It is also one of the main species captured by pelagic trawls and light-luring fishing vessels 3 . A short lifespan and fast growth and reproduction rates are the most notable features of D. maruadsi 4 . Meanwhile, as an r-selection strategy species, it is vulnerable to the environmental deterioration and fishing intensity including those unregulated fishing methods and advanced technologies 5 , 6 . In recent decades, under the multiple stresses of continuous high-intensity fishing, increasing temperature and feed structure changes caused by global climate change, the population of D. maruadsi has been subjected to strong selection pressure, which gradually showing adaptive evolution phenomena such as miniaturization, sex precocity, and the population size has also been decreasing year by year 7 , 8 . To deal with this dilemma, artificial cultivation of juvenile fish of D. maruadsi has been gradually realized in the offshore area of Dongshan island at present. To ensure the preservation of economically significant species, it is crucial to safeguard their germplasm resources and prevent any potential decline 9 . Genomic data are essential tools for investigating species germplasm resources and assessing population genetic structure and diversity. These resources are of great significance for managing fishery resources and promoting their sustainable use. High-quality reference genomes are essential basic genetic data, and their application value to the aquaculture is also very important. Furthermore, the value of long Nanopore reads which includes low cost, high-throughput sequencing, and high-quality assembly of genomes has been reported by many researches 10 , 11 . By combining third-generation sequencing and high-through chromosome conformation capture (Hi-C) 12 technologies, we can assemble the chromosome-level genome rapidly, efficiently and accurately. On this basis, the annotation of D. maruadsi genome can be completed. In this report, we provided a high-quality genome assembly of D. maruadsi using Illumina short-reads sequencing, Nanopore sequencing, and Hi-C technologies. We obtained a total of 47.61 Gb clean reads by Illumina platform, and through K-mer frequency distribution analysis, the genome size of D. maruadsi was about 720.70 Mb. For Nanopore genome sequencing, we assembled a total genome length of 723.69 Mb, which includes a total of 169 contigs. In addition, N50 and N90 lengths of filtered reads were respectively 24.67 Mb and 2.78 Mb, and contigs with a length of 2Kb accounted for 100%. We generated 76.37 Gb of Hi-C filtered data, after chromosome-level scaffolding, there are 23 chromosomes with a total length of 713.58 Mb, resulting in a scaffold N50 of 32.35 Mb. The reference genome of D. maruadsi can assist in subsequent population genomics and adaptive genome microevolution studies 13 .
Methods Ethics statement The D. maruadsi in our experiments were collected from Dongshan, Zhangzhou City, Fujian Province, China. Furthermore, the methods used in this work are strictly in accordance with the Guidelines for The Care and Use of Laboratory Animals and followed by the Laboratory Animal Laboratory Committee School of Ocean and Earth Sciences, Xiamen University. Sample collection and nucleic acid preparation A healthy alive female D. maruadsi was collected from the Dongshan Pacific Ocean Observation and Experiment Station, Xiamen University. Ten fresh tissue samples, including muscle, eye, skin, gill, kidney, liver, intestine, spleen, heart and stomach, were frozen in liquid nitrogen immediately and then stored in −80 °C. Following the standard protocol of QIAGEN DNeasy Blood & Tissue Kit (Qiagen, Shanghai, China), genomic DNA (gDNA) of muscle was extracted. Total RNA was extracted from ten tissues by a TRIzoL kit (Invitrogen, Shanghai, China) and mixed equally for RNA-seq. The quality of nucleic acid was detected by 1.0% agarose gel electrophoresis and quantified by a Qubit 4.0 fluorometer (Thermo Fisher Scientific, Waltham, MA). Library construction and sequencing For Illumina data, a pair-end sequencing library with 350 bp insert size was constructed using the Illumina TruSeq Nano DNA Library Prep Kit (Illumina, San Diego, CA, USA) and sequenced on the Illumina HiSeq X Ten platform with the 2 × 150 bp read strategy at Novogene company (Tianjin, China). A total of 47.85 Gb raw data were obtained and 47.61 Gb clean data were retained after quality filtering by fastp (V.0.23.1) 14 software (Table 1 ). For the Nanopore sequencing, the frozen muscle sample was lysed in SDS digestion buffer, and then the lysate was purified with AMPure XP microbeads (Beckman Coulter, High Wycombe, UK) to obtain High-Molecular-Weight(HMW) gDNA. DNA fragment sizes were selected with the BluePippin system (Sage Science, Beverly, MA, USA) and fragments larger than 20 kb were used for subsequent Nanopore sequencing. The Nanopore libraries were prepared using the Ligation Sequencing Kit (SQK-LSK109, Oxford Nanopore Technologies, Oxford, UK) according to the manufacturer’s instructions and sequenced on the flow cells of the PromethION sequencer at Novogene company (Tianjin, China). Finally, we obtained 86.39 Gb Nanopore data, which average and N50 read length were 22.07Kb and 26.23Kb, respectively. The Nanopore data were further screened before assembly to remove reads less than 1500 bp in length. For Hi-C sequencing, DNA fixed with formaldehyde was digested with the restriction enzyme ( DpnII ), and after being repaired by 5’overhangs biotinylated and blunt-end ligation, these fragments are connected in situ , the DNA is cross-linked and purified 15 . In the end, the Hi-C sequencing library was performed on the Illumina HiSeq X Ten platform with a strategy of 2 × 150 bp and generated 76.37 Gb raw reads overall. The RNA-seq library was constructed using Illumina standard protocol (San Diego, CA, USA) and sequenced on the Illumina HiSeq X Ten platform. Finally, we obtained 33.65 Gb paired-end raw reads and 32.61 Gb paired-end clean reads for the following gene prediction (Table 1 ). De novo genome assembly The Illumina clean reads were used for further assembly and estimation of genome size using 17-kmer analysis. With K-mer numbers of 434,759,917,857 and a dominant peak depth of 47.24, the genome size was approximately 720.70 Mb, which was similar to the species in Genus Decapterus and the heterozygosity and repetitive sequence content were about 0.69 and 32.6%, respectively 16 (Supplementary Table 1 & Supplementary Fig. 1 ). NextDenovo was used for genome assembly based on the overlap layout-consensus algorithm with default parameters. To obtain the contig-level genome, we utilized Racon 17 for three iterations of polish using the three-generation Nanopore data. Nextpolish 18 was then employed to correct the genome based on the Illumina data. Lastly, we utilized Purge_Dups 19 (v.1.25) to de-redundant the genome, resulting in the final contig-level genome. The assembled genome size was 723.69 Mb, including 169 contigs in total, with a contig N50 of 24.67 Mb (Table 2 ). The assembled genome size almost matched the estimated results of genome survey, which reflected the high assembly integrity. Hi-C sequencing data was used for chromosome assembly of D. maruadsi . Firstly, we filtered out Hi-C raw reads(low-quality and duplicated reads) using HiC-Pro 20 . Juicer 21 was used to map Hi-C clean reads to the reference genome. Subsequently, we used the genomic proximity signal in the Hi-C data sets to get chromosome-level scaffolds. Then, the 3D-DNA pipeline 22 was used to scaffold the D. maruadsi genome. Afterwards, scaffolds were fine-tuned to correct the misassemblies by Juicebox 23 assembly tools. Finally, we generated a chromosome-level genome assembly of 724.05 Mb and scaffold N50 is up to 32.35 Mb (Table 2 ). The genome assembly contained 23 chromosomes, with a total length of 713.58 Mb (98.6% of the total length of all contigs). Chromosome sizes ranged from 21.36 to 45.1 Mb, with an average chromosome length of 31.03 Mb (Fig. 1A , B & Table 3 ). Anotation of repeat sequences Both homology-based and de novo methods were used to annotate repeat sequences in the D. maruadsi genome. RepeatModeler 24 (v.2.0.1) and LTR_Finder 25 (v.1.07) were utilized to detect repetitive sequences in the D. maruadsi genome and generate a de novo repeat library. Combined with Repbase 26 , the final repeat library was constructed. RepeatMasker 27 (v.4.1.0) was used to search and classify repeats based on this library. Unclassified repeats were further annotated using TEclass 28 (v.2.1.3). Transposable Elements (TEs) annotation results were summarized by adopting the buildSummary.pl of RepeatMasker. Moreover, calcDivergenceFromalign.pl was used to calculate the Kimura divergence value of TEs and createRepeatLandscape.pl was used to draw TEs landscapes 29 . To estimate the insertion age, we compared the nucleotide distances between all copies of each TE using the Kimura two-parameter method 29 . We identified tandem repeats using the Tandem Repeats Finder 30 (v.4.0.9) and soft-masked all repetitive regions except for tandem repeats in the process of protein-coding gene annotation. Finally, a total of 199.49 Mb (27.57% in genome) of consistent and non-redundant repeat sequences were obtained by combining novel, known and tandem repeats. The most abundant repetitive elements were DNA transposons, which spanned more than 102.57 Mb, accounting for 14.17% of the genome of D. maruadsi . Besides, the repetitive sequences were also composed of long interspersed elements (LINE) in 37.62 Mb (5.20% in genome), short interspersed nuclear elements (SINEs) in 2.82 Mb (0.39% in genome) and long terminal repeats (LTRs) in 40.99 Mb (5.66% in genome) (Fig. 2A & Table 4 ). Prediction and functional annotation of protein-coding genes For non-coding RNA (ncRNA) annotation, the programs tRNAScan-SE 31 (v.1.3.1) and RNAmmer 32 (v.1.2) were used to predict tRNA and rRNA respectively. The other ncRNAs were predicted by searching the Rfam database 33 ( http://eggnogdb.embl.de/ ). As a result, we annotated four types of non-coding RNAs, including 1,285 miRNAs, 3,820 tRNAs, 1,592 rRNAs and 762 snRNAs (Table 4 ). For gene structure prediction, ab-initio strategies, homologous searching and transcriptome-assisted approaches were used to predict protein-coding genes in the D. maruadsi genome after soft-masking all repeat sequences. In homology-based prediction, the genetically proximal coding sequences of related species, containing Oryzias latipes , Seriola lalandi , Seriola dumerili , Oreochromis niloticus , and Trachinotus ovatus were downloaded from European Nucleotide Archive and provided to GenomeTreader 34 (v.1.7.0) (Supplementary Table 4 ). Additionally, the RNA-seq data was subjected to the assembly using Trinity 35 (v.2.10.0). The ab-initio gene prediction was performed using the transcripts assembled from RNA-seq and known genes of O. latipes , S. lalandi , S. dumerili , O. niloticus , and T. ovatus by Braker2 36 . After two rounds of model training, the optimal parameters are determined. Another gene prediction method involved aligning RNA-seq data to the D. maruadsi genome to assemble the transcriptome using Hisat2 37 and StringTie 38 (v.2.1.4). Then, the open reading frame (ORF) regions were predicted using TransDecoder (v.5.5.0). Ultimately, EvidenceModeler was utilized to create a thorough gene set, which was then further annotated for protein-coding gene structure via PASA 39 (v.2.4.1). For functional annotation of predicted gene, Diamond 40 (v.2.0.6) was applied to align protein-coding genes to the Swiss-Prot ( http://www.uniprot.org/ ), InterPro( https://www.ebi.ac.uk/interpro/ ) and NR protein databases with E-values < 1*10 −5 . Additionally, GO and KEGG pathway annotations were performed by InterProScan 41 (v.4.8) ( https://www.ebi.ac.uk/interpro/ ) and KEGG Automatic Annotation Server (KAAS 42 , https://www.genome.jp/tools/kaas/ ) (Table 5 ). In this study, a high-quality reference genome of D. maruadsi was generated, which could provide a solid foundation for species diversity and population genetic studies in the future. Nowadays, genomics is gradually being applied in every stage of large-scale aquaculture production and domestication. As an important aquatic economic fish, D. maruadsi is necessary to identify genetic diversity under phenotypic traits by a high-precision chromosome-scale genome to improve the economic benefits of aquaculture species. In addition, high-quality genome-wide maps are important as essential basic genetic data for industrial and scientific research applications, providing a genetic basis and more accurate genetic evaluation tools for the management and sustainable use of D. maruadsi fisheries resources. Finally, as a potential cultured fish, the genome of D. maruadsi will help to the breeding program for selecting excellent growth-related traits.
Decapterus maruadsi is one of the representative offshore fish in the Western Pacific. Since the last century, it has become a commercially valuable marine fishery species in the Western Pacific region. Despite its high economic value, there is still a lack of high-quality reference genome of D. maruadsi in germplasm resource evaluation research. Here we report a chromosome-level reference genome of D. maruadsi based on Nanopore sequencing and Hi-C technologies. The whole genome was assembled through 169 contigs with a total length of 723.69 Mb and a contig N50 length of 24.67 Mb. By chromosome scaffolding, 23 chromosomes with a total length of 713.58 Mb were constructed. In addition, a total of 199.49 Mb repetitive elements, 33,515 protein-coding genes, and 6,431 ncRNAs were annotated in the reference genome. This reference genome of D. maruadsi will provide a solid theoretical basis not only for the subsequent development of genomic resources of D. maruadsi but also for the formulation of policies related to the protection of D. maruadsi . Subject terms
Data Records The raw sequencing reads of all libraries have been deposited into NCBI SRA database via the accession number of SRP408505 43 . The assembled genome has been deposited at Genbank under the accession number GCA_030347415.2 44 . Moreover, data of the assembled genome and sequence annotations are available at Figshare 45 . Technical Validation Genome assembly and annotation completeness evaluation To ensure the accuracy and integrity of the assembly, we assessed the completeness of the final genome assembly using Benchmarking Universal Single-Copy Orthologues (BUSCO) 46 with the Actinopterygii_odb10 lineage database. Out of 3,640 single-copy orthologues, approximately 97.8% were completely identified in the D. maruadsi genome (Supplementary Table 3 ). Besides, the Illumina short reads were aligned to the genome using the BWA MEM algorithm. Subsequently, employing samtools on the generated BAM files, we calculated the sequencing depth across the genome. The non-zero sequencing depth positions were tallied and summed, then compared to the total base positions for the final coverage percentage. This yielded a mapping ratio of 99.76% and a genome coverage of 98.80% (Supplementary Table 2 ). Moreover, a total of 33,515 protein-coding genes were successfully obtained by combining ab-initio strategies, homologous searching and transcriptome-assisted approaches. A total of 25,933 genes were successfully functionally annotated in at least one of these databases (Fig. 2B & Table 5 ). The high integration efficiency, mapping ratio, recognition rate of single-copy orthologues and gene number collectively suggest that the assembled D. maruadsi genome was of superior quality. Genome assembly accuracy evaluation To validate the precise arrangement of the D. maruadsi genome, we aligned the assembly to the O. latipes genome using minimap2 47 with a unit of 1 Kbp (Fig. 1C ). Additionally, we performed the same alignment method with Trachurus trachurus , a closely related species in the Carangidae family. The 23 chromosomes identified in the D. maruadsi genome showed a significant level of collinearity with the other two species, indicating the high genomic continuity of our assembly (Supplementary Fig. 2 & Supplementary Fig. 3 ). Notably, Chromosome 2 of D. maruadsi aligns with both chromosome 2 and chromosome 4 of O. latipes and T. trachurus . To confirm the accuracy of the chromosome number, we performed a nucmer 48 alignment of chromosome 2 of D. maruadsi with chromosome 2 and chromosome 4 of O. latipes and T. trachurus . The results revealed that chromosomes 4 and 2 of O. latipes aligned to the regions 0.34 M - 31.19 M and 31.82 M - 44.96 M, respectively, on chromosome 2 of D. maruadsi . Similarly, chromosomes 4 and 2 of T. trachurus aligned to the regions 0.09 M - 31.62 M and 31.66 M - 45.06 M, respectively, on chromosome 2 of D. maruadsi (Supplementary Fig. 4B ). These comparative analyses collectively indicate the presence of a distinct alignment gap within the 31 M - 32 M region of the reported chromosome 2 of D. maruadsi . This alignment gap suggests the possibility of a structural alteration and connection region between the two chromosomes in this location. Moreover, based on the Hi-C assisted assembly data, we have identified that this connection region is completely covered by the precisely assembled contig (Supplementary Fig. 4 ). Additionally, we selected the genomic range from 31 M to 32 M and utilized the minimap2 tool to align Illumina and Nanopore reads to this region. Subsequently, we implemented a sliding window approach with a window size of 50 kb to calculate the average depth at each genomic position. In this important region, both Illumina and Nanopore data consistently exhibited stable depth profiles, with no significant decrease in depth observed (Supplementary Fig. 5 ). These observations further emphasize the integrity and continuity of our assembly results. Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s41597-024-02912-1. Acknowledgements We acknowledge financial support from the Fundamental Research Funds for the Central Universities (No.20720200119). Author contributions P.X. conceived and supervised the study. Z.X.Z., Y.C.D. and P.X. colledcted the sample. L.Y.C., Z.X.Z. and Z.Y.Z. performed bioinformatics analysis. L.Y.C., Z.X.Z. and J.Y.Y. drafted the manuscript. F.P. helped on manuscript preparation. P.X., T.Z. and Y.L.B. provided review and modification of the manuscript. All authors read and approved the final manuscript. Code availability Genome annotation: (1) RepeatMasker: parameters: -e ncbi -a -nolow -no_is -norna. (2) TE-class: parameters: all parameters were set as default. (3) Braker2: parameters: all parameters were set as default. (4) PASA: --ALIGNERS blat. (5) EvidenceModeler: parameters: all parameters were set as default. Genome assembly: (1) NextDenovo: parameters: all parameters were set as default. Gene family identifcation and phylogenetic analysis: (1) RAxML: parameters: -f a -m PROTGAMMAAUTO. (2) MCMCTREE: parameters: all parameters were set as default. Other analysis modules that were not mentioned parameters were used default parameters. The other custom codes used in this analysis were mentioned in methods sections. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Data. 2024 Jan 13; 11:69
oa_package/4c/cb/PMC10787795.tar.gz
PMC10787796
38218961
Introduction All-carbon quaternary stereocenters are an important synthetic motif found in natural products and bioactive molecules (Fig. 1 ) that are especially difficult to synthesize enantioselectively 1 – 4 . Successful strategies have recently been developed for cyclic systems (Fig. 2A ) 2 , 5 – 10 ; however, constructing quaternary centers in acyclic molecules remains a significant synthetic challenge due to the combination of high levels of steric congestion and greater conformational freedom 11 – 21 . Despite its known benefits, organocatalysis 22 – 24 has been used for asymmetric quaternary center construction only a few classes of acyclic systems 25 – 28 , though asymmetric 1,4-conjugate addition for tertiary carbon synthesis is well documented 29 – 33 . This report details the successful organocatalytic synthesis of valuable acyclic 1,4-dicarbonyl products with vinylated and arylated quaternary centers 34 . Organocatalyzed Michael additions to acyclic proquaternary substrates have been reported for nitromethane or cyanide but have otherwise been rare (Fig. 2B ) 35 – 42 . A newer class of easily synthesized BINOL–derived enantioselective 1,4–addition organocatalysts have proven to be useful, recyclable, and functional group tolerant in many transformations 43 – 53 ; however, these reactions have only produced chiral tertiary carbon centers to date. In fact, β,β–disubstituted enones were investigated for quaternary carbon formation but completely lacked reactivity 54 .
Results and discussion To overcome steric deactivation via increased electrophilic activation, we looked to enones bearing additional electron-withdrawing groups. In particular, the use of 2–ene–1,4–diones (1) could allow an approach to often difficult-to-access chiral 1,4-diketones with beta quaternary carbons (Fig. 2C ) and functionality for the total synthesis of natural products 55 – 57 . However, few precedents related to such a β-vinylation or arylation of ketones to construct quaternary centers exist 58 – 63 . To test the diketone activation hypothesis, enediketone 1a was synthesized as a mixture of cis and trans isomers, which was then purified, and potassium styrenyl trifluoroborate was chosen as an exploratory nucleophile (Fig. 3 ). Unsurprisingly, the E –isomer was almost completely unreactive (<2% yield, ~77:23 er), so further experimentation was conducted with pure ( Z )–enediketone. While we fully expected competing regioselectivity with addition at both alkene positions, substrate ( Z )–1a reacted with superb regioselectivity with the use of typical conditions for BINOL 4–catalyzed conjugate additions 9 , 10 to give a single product. In fact, ( Z )−1a defied all initial expectations of addition at the less hindered carbon and provided quaternary carbon-bearing 1,4-diketone 3a in 99% yield and 96:4 er. A control experiment without any catalyst gave 3a from ( Z )–1a in 69% yield after 18 h, indicating that a significant racemic background reaction was operative. These results suggested that two cis-disposed ketone carbonyls must be present for reactivity and may provide Z -dependent cooperative activation. We were then able to demonstrate a one-pot reaction using ( E )−1a , which was first converted to the Z -form via photo-isomerization 64 and then could undergo conjugate addition to provide the quaternary carbon product in good yield (79%) and the same er obtained from pure ( Z )-enediketone. Such an isomerization may even have been the source of activity seen in Fig. 3A . The presence or absence of light had no effect on the reaction using pure Z -enediketone. Various BINOL derivatives were tested for increased reactivity and stereoselectivity, and the effects of solvent and temperature were also investigated, but with no improvement (See “Synthesis of BINOL-based catalysts” in the Supplementary Information). Various enediketones showed productive reactivity with the catalytic conditions identified above. Both electron–donating and electron–withdrawing groups on the ketone’s aromatic ring provided effective reaction ( 3a – 3e , Fig. 4 ). A heteroaromatic enone substituent was likewise accommodated (see 3f ). However, changing the phenylketone to a methylketone resulted in the formation of product 3 g with slightly reduced stereoselectivity and competing regioselectivity to form a tertiary stereocenter in 11% yield (see 5g ). Despite both carbonyls being equally Lewis basic, quaternary carbon formation was still favored in a 4:1 ratio. Interestingly, moving the branching vinyl methyl from the alkyl ketone side of the alkene to the phenyl ketone side reversed the regioselectivity so that the major diastereomer of dione 5h was formed with low enantioselectivity in 63% yield. The minor diastereomer of 5h was produced along with the quaternary product in 17% yield as an inseparable 2:1 mixture. An initial regioselectivity hypothesis was that the relative locations of the phenyl and aliphatic ketones have a directing effect on the addition, where C–C bond formation is more strongly favored at the β-carbon of an aryl ketone than that of an alkyl ketone 43 – 53 . Based on work by Goodman and Pellegrinet 44 , we believed that the phenyl ketone directed an intramolecular 5-exo-trig 1,4-addition that reinforced the favored quaternary carbon formation. Replacing the methyl groups of 1a with ethyl groups gave improved stereoselectivity ( 3i ) but slightly lowered the yield, likely due to an increase in steric repulsion. Changing the methyl ketone of 1a to a phenyl ketone afforded the quaternary center in 3j with moderate er; however, replacing both methyl groups with phenyls precluded reaction so triphenyl product 3k was not observed. The series 3a, 3j , and 3k shows the negative impact of increasing the size of substituents on yield. It is noteworthy that a cyclic diketone system gave the alpha quaternary center in 3l in high yield but with little enantioselectivity. Apparently, without rotation of the carbonyl-olefin bond where C–C bond formation occurs reactivity is retained, but the stereoselectivity is almost completely lost. Various vinyl, alkynyl, and heteroaromatic nucleophiles were also examined. A few substrates gave a lower er, but use of trifluorotoluene and/or an increased loading of the catalyst improved the enantioselectivity (Fig. 5 ). For example, the electron rich styrenyl nucleophiles that afforded 3m and 3n originally showed a significant racemic background reaction, but increasing the catalyst loading improved the er to 80:20 and 92:8, respectively. An electron-withdrawing group on the styrene system in 3o produced a high yield and er without adjustment. Nucleophiles with alkyl chains gave the dienyl adducts 3p, 3q , and trans alkenyl 3r – 3x in high yield and enantioselectivity. Having two vinyl substituents resulted in diminished enantioselectivity (see 3y and 3z ), but a synthetically useful bromo vinyl borate synthesized 3aa in moderate yield and improved er. Alternatives to the vinyl nucleophilic system were also tested. Alkynyl reagents provided useful reactivity, but decreased stereoselectivity (see 3ab and 3ac ). It is worth noting that the isomerization from ( Z )−1a to ( E )−1a occurred competitively during the formation of 3z and 3ab , which may have reduced both the yield and stereoselectivity for those reactions. The use of other strong nucleophiles, like furanyl borate, similarly formed quaternary carbons in high yield but with low enantioselectivity due to the competitiveness of the background reaction (see 3ad ) 65 . On the other hand, a thienyl borate produced 3ae in good yield and high er. To explain (A) why only the Z -isomer was reactive and (B) why quaternary regioselectivity was favored over tertiary carbon formation, we pursued a computational investigation using substrates ( Z )−1a and ( E )−1a with styrenyl boronate. Based on prior mechanistic investigations relevant to tertiary carbon formation via similar catalysis 47 , it is likely that the potassium trifluoroborate salt dissociates fluoride and condenses with the BINOL 4 to form an activated chiral boronate ester that then coordinates to the enone carbonyl. To simplify the calculations, BINOL 4 was modelled as 3,3 ́-difluorobisphenol. We initially modeled the formation of the Lewis acid/Lewis base complexed boronate-ketone adduct 6 , and our calculations supported Goodman’s finding 44 that this complex formed as a discrete mechanistic intermediate prior to the transition state (Fig. 6 ). Note that this stable intermediate was taken as the zero-point reference for all other calculated geometries. Conjugate addition transition states derived from ketone-coordinated boronates with both endo and exo modes of addition 66 , 67 were next examined (Figs. 6 and 7 ). Where Goodman’s work showed 6-endo cyclization (see 8c and 8e ), we found that 5-exo modes 68 were lower in energy for both quaternary and tertiary carbon formation (compare 8a to 8c and 8d to 8e ). This new mode of reactivity is enabled by the additional ketone. Close examination of the exo transition states revealed a fascinating stabilizing effect; the ketone distal to the Lewis acid coordination not only enabled the 5-exo addition but also participated in an n→π* donation to the bound carbonyl ( 8a ) 69 . This ouroboros-like activation 70 , 71 is evidenced by the short C = O→C = O bond (1.53 Å in 8a ) and the tetrahedral geometry of the carbon of the bound C = O (C16). Such interactions have been described for static protein structure 72 and utilized for the synthesis of Lewis acid/base heteroaromatics 69 , but to our knowledge has not been proposed as a stabilizing factor in reaction catalysis 73 . The LUMO of the C16 carbonyl thus acts as a Lewis acid activating the planar enone for 5-exo-trig conjugate addition, lowering the LUMO energy, and the electron donation of the planar ketone to the C16 carbonyl simultaneously increases the electron density in the nucleophilic system (C28), raising the HOMO energy. Additionally, this stabilizing interaction was not accessible in the 6-endo geometries (see longer O to C distances in 8c and 8e ), and we believe this to be a reason for their relatively higher energy pathways. In investigating the generation of the ouroboros stabilization, we could identify that the formation of iso-furan 7 occurred prior to C–C bond formation. Some pathways, such as that shown in Fig. 6 , have 7 formed as a meta stable intermediate as a local minimum. In others, it is a shoulder or part of a continuous slope to the transition state. These calculations also showed that the ( R )-biaryl introduces torsion in the coordinated system that favors 8a over 8b , giving the major observed enantiomer. In considering why reactivity is unfavorable for ( E )-enediketones, several potential transitions states derived from methyl or phenyl ketone-coordinated isomers of 6 were examined, but only the transition states 8f and 8g converged reliably. Notably, ouroboros activation was not observable for any reasonable geometries corresponding to transition states derived from ( E )-substrates, the 6-endo-trig transition state was thus lower in energy, and the resulting higher overall barrier explains the lack of reactivity of β-disubstituted enones in all previous studies 9 . Given other recent efforts that also observe such a dependence on ( Z )-enone geometry 25 – 28 , ouroboros stabilization may be operative in many catalytic reactions. The poor er arising from the lack of rotation about the alkene-ketone bond in forming 3l also aligns with this hypothesis, as the carbonyl could not fully rotate out of plane to provide 5-exo reactivity, forcing it through a 6-endo transition state like 8e , which also lacks ouroboros activation and therefore has reduced stereocontrol. The decreased enantioselectivity and altered regioselectivity seen for 5g and 5h could also be due to competitive 6-endo reactivity for those substrates rather than due to Lewis basicity. An examination of the calculated LUMOs in intermediate 6 and its phenylketone-coordinated isomer showed localization on the planar enone (HOMO/LUMO illustrated in the Source Data file), but the carbons undergoing nucleophilic attack (C13 or C14) bear different proportions of the LUMO 74 . For both isomers, C14 has significant LUMO character, leading to better HOMO/LUMO overlap, but less of the LUMO is located on C13. The relative localization of the LUMO in these structures and the relative energies of the subsequent transition states correspond to previously characterized experimental rate dependencies on the stabilization of developing cationic charge at the β-carbon of the enone in this class of conjugate additions 75 , 76 . An additional insight into the regioselectivity was obtained by examining two possible n→π* interactions in the Z-enediketone. That arising from the Ph-ketone donating into a twisted Me-ketone resulted in a 1.7 kcal/mol more stable conformation than that arising from the Me-ketone donating into a twisted Ph-ketone. The stereoelectronic and steric interactions in these conformations are also likely to be present in 8a and 8d , and the relative stability of their geometries contributing to the difference in regioisomeric transition state energies. To demonstrate the utility of these quaternary 1,4-diketone products 58 – 63 , 77 , examples were transformed into key synthetic building blocks (Fig. 8 ). Chiral cyclopentenones (see 8), which exist widely in bioactive compounds 78 – 80 , could be formed in high yield and er via aldol condensation. The absolute stereochemistry of 9 was confirmed by X-ray crystallography 81 . Chemoselective hydrogenation reduced the benzoyl carbonyl and the alkene of 3a or 3q to form the aliphatic quaternary carbon centers in 10a and 10q , respectively, with the latter containing an otherwise difficult to access alkylated quaternary center with high er. The oxidative cleavage of the styrenyl olefin gave ketoaldehyde 11 , which would be useful in recently reported pyrrolidine syntheses 77 , 82 . Non-planar heterocycle dihydropyridazine 12 , a bioactive pharmacophore 83 , 84 , could be generated in good yield. Using the bromo substrate 3e to incorporate an intramolecular Heck coupling reaction gave the quinone-like derivative 12 85 . In conclusion, we successfully synthesized challenging quaternary centers enantioselectively from ( Z )-1,4-enediketones via organocatalyzed conjugate addition. Control experiments showed that the cis-relationship of the ketones was vital to reactivity, and keto-ene bond rotation at the location of C–C bond formation was important for enantioselectivity. DFT calculations showed that the additional ketone provided 5-exo-trig reactivity and a stabilizing interaction through an n→π*cyclic ouroboros activation. The regioselectivity for quaternary carbon formation appeared to be based primarily on the greater HOMO/LUMO overlap in the formation of the quaternary carbon relative to the tertiary carbon. A broad substrate scope of chiral α-quaternary 1,4-diketones were synthesized. Further transformations to quaternary carbon-containing enantio-enriched cyclopentenones, linear hydrocarbons, dihydropyridazines, and quinone methides were demonstrated in good yield and er. These building blocks will enable synthetic endeavors in many areas.
Results and discussion To overcome steric deactivation via increased electrophilic activation, we looked to enones bearing additional electron-withdrawing groups. In particular, the use of 2–ene–1,4–diones (1) could allow an approach to often difficult-to-access chiral 1,4-diketones with beta quaternary carbons (Fig. 2C ) and functionality for the total synthesis of natural products 55 – 57 . However, few precedents related to such a β-vinylation or arylation of ketones to construct quaternary centers exist 58 – 63 . To test the diketone activation hypothesis, enediketone 1a was synthesized as a mixture of cis and trans isomers, which was then purified, and potassium styrenyl trifluoroborate was chosen as an exploratory nucleophile (Fig. 3 ). Unsurprisingly, the E –isomer was almost completely unreactive (<2% yield, ~77:23 er), so further experimentation was conducted with pure ( Z )–enediketone. While we fully expected competing regioselectivity with addition at both alkene positions, substrate ( Z )–1a reacted with superb regioselectivity with the use of typical conditions for BINOL 4–catalyzed conjugate additions 9 , 10 to give a single product. In fact, ( Z )−1a defied all initial expectations of addition at the less hindered carbon and provided quaternary carbon-bearing 1,4-diketone 3a in 99% yield and 96:4 er. A control experiment without any catalyst gave 3a from ( Z )–1a in 69% yield after 18 h, indicating that a significant racemic background reaction was operative. These results suggested that two cis-disposed ketone carbonyls must be present for reactivity and may provide Z -dependent cooperative activation. We were then able to demonstrate a one-pot reaction using ( E )−1a , which was first converted to the Z -form via photo-isomerization 64 and then could undergo conjugate addition to provide the quaternary carbon product in good yield (79%) and the same er obtained from pure ( Z )-enediketone. Such an isomerization may even have been the source of activity seen in Fig. 3A . The presence or absence of light had no effect on the reaction using pure Z -enediketone. Various BINOL derivatives were tested for increased reactivity and stereoselectivity, and the effects of solvent and temperature were also investigated, but with no improvement (See “Synthesis of BINOL-based catalysts” in the Supplementary Information). Various enediketones showed productive reactivity with the catalytic conditions identified above. Both electron–donating and electron–withdrawing groups on the ketone’s aromatic ring provided effective reaction ( 3a – 3e , Fig. 4 ). A heteroaromatic enone substituent was likewise accommodated (see 3f ). However, changing the phenylketone to a methylketone resulted in the formation of product 3 g with slightly reduced stereoselectivity and competing regioselectivity to form a tertiary stereocenter in 11% yield (see 5g ). Despite both carbonyls being equally Lewis basic, quaternary carbon formation was still favored in a 4:1 ratio. Interestingly, moving the branching vinyl methyl from the alkyl ketone side of the alkene to the phenyl ketone side reversed the regioselectivity so that the major diastereomer of dione 5h was formed with low enantioselectivity in 63% yield. The minor diastereomer of 5h was produced along with the quaternary product in 17% yield as an inseparable 2:1 mixture. An initial regioselectivity hypothesis was that the relative locations of the phenyl and aliphatic ketones have a directing effect on the addition, where C–C bond formation is more strongly favored at the β-carbon of an aryl ketone than that of an alkyl ketone 43 – 53 . Based on work by Goodman and Pellegrinet 44 , we believed that the phenyl ketone directed an intramolecular 5-exo-trig 1,4-addition that reinforced the favored quaternary carbon formation. Replacing the methyl groups of 1a with ethyl groups gave improved stereoselectivity ( 3i ) but slightly lowered the yield, likely due to an increase in steric repulsion. Changing the methyl ketone of 1a to a phenyl ketone afforded the quaternary center in 3j with moderate er; however, replacing both methyl groups with phenyls precluded reaction so triphenyl product 3k was not observed. The series 3a, 3j , and 3k shows the negative impact of increasing the size of substituents on yield. It is noteworthy that a cyclic diketone system gave the alpha quaternary center in 3l in high yield but with little enantioselectivity. Apparently, without rotation of the carbonyl-olefin bond where C–C bond formation occurs reactivity is retained, but the stereoselectivity is almost completely lost. Various vinyl, alkynyl, and heteroaromatic nucleophiles were also examined. A few substrates gave a lower er, but use of trifluorotoluene and/or an increased loading of the catalyst improved the enantioselectivity (Fig. 5 ). For example, the electron rich styrenyl nucleophiles that afforded 3m and 3n originally showed a significant racemic background reaction, but increasing the catalyst loading improved the er to 80:20 and 92:8, respectively. An electron-withdrawing group on the styrene system in 3o produced a high yield and er without adjustment. Nucleophiles with alkyl chains gave the dienyl adducts 3p, 3q , and trans alkenyl 3r – 3x in high yield and enantioselectivity. Having two vinyl substituents resulted in diminished enantioselectivity (see 3y and 3z ), but a synthetically useful bromo vinyl borate synthesized 3aa in moderate yield and improved er. Alternatives to the vinyl nucleophilic system were also tested. Alkynyl reagents provided useful reactivity, but decreased stereoselectivity (see 3ab and 3ac ). It is worth noting that the isomerization from ( Z )−1a to ( E )−1a occurred competitively during the formation of 3z and 3ab , which may have reduced both the yield and stereoselectivity for those reactions. The use of other strong nucleophiles, like furanyl borate, similarly formed quaternary carbons in high yield but with low enantioselectivity due to the competitiveness of the background reaction (see 3ad ) 65 . On the other hand, a thienyl borate produced 3ae in good yield and high er. To explain (A) why only the Z -isomer was reactive and (B) why quaternary regioselectivity was favored over tertiary carbon formation, we pursued a computational investigation using substrates ( Z )−1a and ( E )−1a with styrenyl boronate. Based on prior mechanistic investigations relevant to tertiary carbon formation via similar catalysis 47 , it is likely that the potassium trifluoroborate salt dissociates fluoride and condenses with the BINOL 4 to form an activated chiral boronate ester that then coordinates to the enone carbonyl. To simplify the calculations, BINOL 4 was modelled as 3,3 ́-difluorobisphenol. We initially modeled the formation of the Lewis acid/Lewis base complexed boronate-ketone adduct 6 , and our calculations supported Goodman’s finding 44 that this complex formed as a discrete mechanistic intermediate prior to the transition state (Fig. 6 ). Note that this stable intermediate was taken as the zero-point reference for all other calculated geometries. Conjugate addition transition states derived from ketone-coordinated boronates with both endo and exo modes of addition 66 , 67 were next examined (Figs. 6 and 7 ). Where Goodman’s work showed 6-endo cyclization (see 8c and 8e ), we found that 5-exo modes 68 were lower in energy for both quaternary and tertiary carbon formation (compare 8a to 8c and 8d to 8e ). This new mode of reactivity is enabled by the additional ketone. Close examination of the exo transition states revealed a fascinating stabilizing effect; the ketone distal to the Lewis acid coordination not only enabled the 5-exo addition but also participated in an n→π* donation to the bound carbonyl ( 8a ) 69 . This ouroboros-like activation 70 , 71 is evidenced by the short C = O→C = O bond (1.53 Å in 8a ) and the tetrahedral geometry of the carbon of the bound C = O (C16). Such interactions have been described for static protein structure 72 and utilized for the synthesis of Lewis acid/base heteroaromatics 69 , but to our knowledge has not been proposed as a stabilizing factor in reaction catalysis 73 . The LUMO of the C16 carbonyl thus acts as a Lewis acid activating the planar enone for 5-exo-trig conjugate addition, lowering the LUMO energy, and the electron donation of the planar ketone to the C16 carbonyl simultaneously increases the electron density in the nucleophilic system (C28), raising the HOMO energy. Additionally, this stabilizing interaction was not accessible in the 6-endo geometries (see longer O to C distances in 8c and 8e ), and we believe this to be a reason for their relatively higher energy pathways. In investigating the generation of the ouroboros stabilization, we could identify that the formation of iso-furan 7 occurred prior to C–C bond formation. Some pathways, such as that shown in Fig. 6 , have 7 formed as a meta stable intermediate as a local minimum. In others, it is a shoulder or part of a continuous slope to the transition state. These calculations also showed that the ( R )-biaryl introduces torsion in the coordinated system that favors 8a over 8b , giving the major observed enantiomer. In considering why reactivity is unfavorable for ( E )-enediketones, several potential transitions states derived from methyl or phenyl ketone-coordinated isomers of 6 were examined, but only the transition states 8f and 8g converged reliably. Notably, ouroboros activation was not observable for any reasonable geometries corresponding to transition states derived from ( E )-substrates, the 6-endo-trig transition state was thus lower in energy, and the resulting higher overall barrier explains the lack of reactivity of β-disubstituted enones in all previous studies 9 . Given other recent efforts that also observe such a dependence on ( Z )-enone geometry 25 – 28 , ouroboros stabilization may be operative in many catalytic reactions. The poor er arising from the lack of rotation about the alkene-ketone bond in forming 3l also aligns with this hypothesis, as the carbonyl could not fully rotate out of plane to provide 5-exo reactivity, forcing it through a 6-endo transition state like 8e , which also lacks ouroboros activation and therefore has reduced stereocontrol. The decreased enantioselectivity and altered regioselectivity seen for 5g and 5h could also be due to competitive 6-endo reactivity for those substrates rather than due to Lewis basicity. An examination of the calculated LUMOs in intermediate 6 and its phenylketone-coordinated isomer showed localization on the planar enone (HOMO/LUMO illustrated in the Source Data file), but the carbons undergoing nucleophilic attack (C13 or C14) bear different proportions of the LUMO 74 . For both isomers, C14 has significant LUMO character, leading to better HOMO/LUMO overlap, but less of the LUMO is located on C13. The relative localization of the LUMO in these structures and the relative energies of the subsequent transition states correspond to previously characterized experimental rate dependencies on the stabilization of developing cationic charge at the β-carbon of the enone in this class of conjugate additions 75 , 76 . An additional insight into the regioselectivity was obtained by examining two possible n→π* interactions in the Z-enediketone. That arising from the Ph-ketone donating into a twisted Me-ketone resulted in a 1.7 kcal/mol more stable conformation than that arising from the Me-ketone donating into a twisted Ph-ketone. The stereoelectronic and steric interactions in these conformations are also likely to be present in 8a and 8d , and the relative stability of their geometries contributing to the difference in regioisomeric transition state energies. To demonstrate the utility of these quaternary 1,4-diketone products 58 – 63 , 77 , examples were transformed into key synthetic building blocks (Fig. 8 ). Chiral cyclopentenones (see 8), which exist widely in bioactive compounds 78 – 80 , could be formed in high yield and er via aldol condensation. The absolute stereochemistry of 9 was confirmed by X-ray crystallography 81 . Chemoselective hydrogenation reduced the benzoyl carbonyl and the alkene of 3a or 3q to form the aliphatic quaternary carbon centers in 10a and 10q , respectively, with the latter containing an otherwise difficult to access alkylated quaternary center with high er. The oxidative cleavage of the styrenyl olefin gave ketoaldehyde 11 , which would be useful in recently reported pyrrolidine syntheses 77 , 82 . Non-planar heterocycle dihydropyridazine 12 , a bioactive pharmacophore 83 , 84 , could be generated in good yield. Using the bromo substrate 3e to incorporate an intramolecular Heck coupling reaction gave the quinone-like derivative 12 85 . In conclusion, we successfully synthesized challenging quaternary centers enantioselectively from ( Z )-1,4-enediketones via organocatalyzed conjugate addition. Control experiments showed that the cis-relationship of the ketones was vital to reactivity, and keto-ene bond rotation at the location of C–C bond formation was important for enantioselectivity. DFT calculations showed that the additional ketone provided 5-exo-trig reactivity and a stabilizing interaction through an n→π*cyclic ouroboros activation. The regioselectivity for quaternary carbon formation appeared to be based primarily on the greater HOMO/LUMO overlap in the formation of the quaternary carbon relative to the tertiary carbon. A broad substrate scope of chiral α-quaternary 1,4-diketones were synthesized. Further transformations to quaternary carbon-containing enantio-enriched cyclopentenones, linear hydrocarbons, dihydropyridazines, and quinone methides were demonstrated in good yield and er. These building blocks will enable synthetic endeavors in many areas.
The chemical synthesis of molecules with closely packed atoms having their bond coordination saturated is a challenge to synthetic chemists, especially when three-dimensional control is required. The organocatalyzed asymmetric synthesis of acyclic alkenylated, alkynylated and heteroarylated quaternary carbon stereocenters via 1,4-conjugate addition is here catalyzed by 3,3 ́-bisperfluorotoluyl-BINOL. The highly useful products (31 examples) are produced in up to 99% yield and 97:3 er using enediketone substrates and potassium trifluoroorganoborate nucleophiles. In addition, mechanistic experiments show that the ( Z )–isomer is the reactive form, ketone rotation at the site of bond formation is needed for enantioselectivity, and quaternary carbon formation is favored over tertiary. Density functional theory-based calculations show that reactivity and selectivity depend on a key n→π* donation by the unbound ketone’s oxygen lone pair to the boronate-coordinated ketone in a 5-exo-trig cyclic ouroboros transition state. Transformations of the conjugate addition products to key quaternary carbon-bearing synthetic building blocks proceed in good yield. All-carbon quaternary stereocenters are an important synthetic motif but are especially difficult to synthesize enantioselectively. Here, the authors demonstrate the organocatalytic regio- and enantioselective synthesis of valuable acyclic 1,4-dicarbonyl products with vinylated and arylated quaternary centers. Subject terms
Supplementary information Source data
Supplementary information The online version contains supplementary material available at 10.1038/s41467-024-44744-y. Acknowledgements The authors are grateful for generous financial support from the NSF (grant CHE-2102282, JAM) and the Welch Foundation (grant E−1744, JAM). Professor Judy Wu at the University of Houston is thanked for aid in access to computational facilities. The computational work was completed in part with resources provided by the Research Computing Data Core at the University of Houston. We are grateful to Dr. Sasha Sundstrom and AbbVie North Chicago for aiding in the acquisition of specific rotation data. Author contributions P.-K.P.: Conceptualization, experimental methodology development and investigation, validation of structures, writing of first draft. A.I.: computational model development and investigation, validation of transitiona states and intermediates, and electronic data curation. J.A.M.: conceptualization of project, writing and editing, project supervision, and acquisition of funding. Peer review Peer review information Nature Communications thanks Giovanni Ghigo, Silvina Pellegrinet and the other, anonymous, reviewer for their contribution to the peer review of this work. A peer review file is available. Data availability The experimental data generated in this study and computational procedures with optimized structures are provided in the Supplementary Information file. The molecular coordinate data generated in this study are provided in the Source Data file. All primary data files, such as.fid files for NMR spectra or coordinate files for molecular structures, are available from the corresponding author for free upon request. The X-ray crystallographic coordinates for structures reported in this study have been deposited at the Cambridge Crystallographic Data Centre (CCDC), under deposition number 2121715 ( 9 ). These data can be obtained free of charge from The Cambridge Crystallographic Data Centre via www.ccdc.cam.ac.uk/data_request/cif . Source data are provided with this paper. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Nat Commun. 2024 Jan 13; 15:504
oa_package/11/e7/PMC10787796.tar.gz
PMC10787799
38218996
Introduction The axillary nerve is a branch of the posterior cord of the brachial plexus and contains nerve components from the fifth and sixth cervical spinal segments 1 . It passes through the posterior wall of the axilla via the quadrangular space with the posterior circumflex humeral vessels 2 . The axillary nerve can be injured by blunt trauma, traction injury, penetrating trauma, and nerve compression in the quadrangular space 3 . The most common cause of axillary nerve injury is trauma during orthopedic surgery such as shoulder arthroscopy, thermal shrinkage of the shoulder capsule, and plate fixation on the proximal humerus 4 . Axillary nerve injury can result in partial or total inactivation of the deltoid and teres minor muscles 4 , 5 . Because damage to the axillary nerve can result in shoulder instability or dysfunction, it is important to protect the axillary nerve and its branches and to restore the damaged nerves to maintain the functions of shoulder. Nerve transfer is one of the therapeutic options for functional restoration of denervated muscles 5 , 6 . To restore the deltoid muscle, nerve transfer using triceps motor branches of the radial nerve to the axillary nerve are widely used 7 – 10 . The most appropriate choice of branch of the radial nerve to use as the donor and which branch of the axillary nerve to use as the recipient can improve the results of nerve transfer. The purpose of this study was to understand the anatomy of axillary and radial nerves to prevent nerve damage and restore muscle action. Additionally, this study aimed to identify potential donor branches of the radial nerve that are good matches for potential recipient branches of the axillary nerve in nerve transfer.
Materials and methods Fifty upper extremities, 24 right and 26 left, from 29 formalin-embalmed cadavers (mean age, 75.8 years; range 52–95 years) with no pathologies or surgical history in the upper extremity were used in this study. All study procedures approved by the Surgical Anatomy Education Centre, Yonsei University College of Medicine (approval number: YSAEC: 23-006). The participants have provided informed consent to donate their bodies for research purposes. The authors state that every effort was made to follow all local and international ethical guidelines and law that pertain to the use of human cadaveric donors in anatomical research 27 . After removal of the skin and subcutaneous tissue of the shoulder, the deltoid and teres minor muscles were identified. The deltoid was detached from its origin on the clavicle, acromion, and scapular spine and divided into three parts according to origin. The axillary nerve was traced from the level of the quadrangular space to the points of entry into the perimysium of the deltoid and teres minor muscles. Number and length of axillary nerve branches from the posterior and anterior divisions were compared. We measured the location where the axillary nerve entered the deltoid muscle, with the acromial end as 0% and the deltoid tuberosity as 100%. To expose the medial branch of the radial nerve, the lateral head of the triceps brachii was divided. Other branches of the radial nerve were traced proximally to their origins from the radial nerve. The number, diameter, and length of branches were recorded. From ten sides of fresh cadavers, 5-mm-long segments were harvested from the main trunk of the axillary nerve, its anterior and posterior divisions, and radial branches, to estimate the number of axonal fibers in these nerve branches. Branches were fixed overnight in 4% paraformaldehyde, dehydrated, and then cleared in toluene. They were then embedded in paraffin and cut into transverse sections of 2 μm thickness. Sections were stained with toluidine blue and photographed with a microscope-mounted camera. Morphometric measurements were performed at 100-fold magnification. The number of axons in randomly selected 0.04 mm 2 cross sectional areas was measured by ImageJ software version 1.53t (NIH, Bethesda, MD, USA). Independent t-tests and chi-square tests were used to compare the significance of differences between body sides or genders. Statistical analyses were performed using the software SPSS version 21 (IBM SPSS Software, Armonk, NY, USA). One-way ANOVA was used to verify the significance of differences in measurements between the types of nerves.
Results Branching patterns of the axillary nerve The clavicular and acromial parts of the deltoid muscle were constantly innervated by the anterior division of the axillary nerve. The teres minor muscle was constantly innervated by the posterior division of the axillary nerve. Variations were observed in the innervation of the spinous part of the deltoid muscle, which was innervated by the posterior division, anterior division, or both. According to the origin of the nerve branches to the spinous part of the deltoid muscle, variations in the distribution of the axillary nerve to the deltoid and teres minor muscles were grouped into three types (Fig. 1 A–F). In 48.0% of upper limbs, the axillary nerve was “mixed type,” as the anterior division of the axillary nerve innervated all parts of deltoid muscle and the posterior division innervated the spinous part of the deltoid muscle as well as the teres minor muscle (Fig. 1 A, D). In these upper limbs, the spinous part of the deltoid muscle was innervated by both the anterior and posterior divisions. In 38.0%, the axillary nerve was “anterior dominant type,” as the anterior division innervated all three parts of the deltoid muscle and the posterior division innervated the teres minor muscle only (Fig. 1 B, E). In these upper limbs, the spinous part of the deltoid muscle was innervated by only the anterior division. In 14.0%, the axillary nerve was “posterior dominant type” with the anterior division innervating the clavicular and acromial parts of the deltoid muscle, while the posterior division innervated the teres minor and spinous part of the deltoid muscle (Fig. 1 C, F ). Size of the axillary nerve The axillary nerve bifurcated into anterior and posterior divisions before passing through the quadrangular space in every upper extremity (Fig. 2 ). The length of the axillary nerve from its origin from the posterior cord to its bifurcation was 44.0 ± 10.5 mm on average. This length was 6.5 mm longer in males than in females, which was a statistically significant difference (p < 0.05). Ramification of the muscular branches from the anterior division was distal to that of the muscular branches from the posterior division. The location of the first ramification of a muscular branch from the anterior division was an average of 23.6 ± 10.0 mm from the bifurcation. The first ramification of a muscular branch from the posterior division was an average of 16.9 ± 9.4 mm from the bifurcation. The first ramification of a muscular branch from the anterior division was distal to the quadrangular space in 78.0% of upper limbs. In contrast, the first ramification of a muscular branch from the posterior division was proximal to the quadrangular space in 66.0% of upper limbs. In 82.0% of upper limbs, the diameter of the anterior division was larger than the diameter of the posterior division. The average diameters of the axillary nerve, the anterior division, and the posterior division were 3.0 ± 0.5 mm, 2.4 ± 0.5 mm, and 2.0 ± 0.5 mm, respectively. Divisions and muscular branches of the axillary nerve The average number of deltoid branches from the axillary nerve was 11.1 ± 2.4. The spinous part of the deltoid muscle was innervated by the smallest number of nerve branches among the parts (Table 1 ). Mixed type axillary nerves provided more branches to the spinous part than the other axillary nerve types (p < 0.05). The number of branches to the spinous part was 2.5 for mixed type axillary nerves, and 57.7% of these were from the anterior division. The average lengths of the terminal branches from the main division were 9.4 ± 3.7 mm to the clavicular part, 22.4 ± 11.9 mm to the acromial part, and 46.3 ± 14.7 mm to the spinous part. Branch lengths to the spinous part were greater on the left than the right with statistical significance (p < 0.05). Site of junction of the nerve branch to the deltoid muscle Nerve terminals on the deltoid muscle were concentrated in the second upper quarter of the deltoid muscle. The junction of nerve branches to the deltoid muscle was located between the level of 28.7 ± 7.6% and the level of 47.0 ± 6.7% from the acromial tip to the deltoid tuberosity (Fig. 3 ). In 80.0% of shoulders, the highest entry point of nerve branches into the deltoid muscle was the acromial part. The lowest entry point was also in the acromial part in 86.0% of shoulders, with a diamond-shaped area of entry. Teres minor branch of the axillary nerve In all upper limbs, the teres minor muscle was innervated only by the posterior division of the axillary nerve. The mean number of branches innervating the teres minor was 1.5 ± 0.6, and the mean length of the longest branch to the teres minor was 36.8 ± 11.2 mm. Entry points of nerve branches were on the deeper side of the teres minor muscle. Anatomy of the radial nerve to the triceps brachii muscle All muscular branches to the triceps brachii muscle branched from the radial nerve before passing through a triangular inlet. Each head of the triceps brachii muscle was innervated by branches that arose separately from the radial nerve or by common branches for multiple triceps heads. In 74.0% of upper limbs, there was a common branch to the medial and lateral heads. In contrast, in 26.0% of upper limbs, all triceps heads were innervated by muscular nervous branches that arose from the radial nerve separately. There were more branches to the long head than to the other heads. The average numbers of branches to the long head, lateral head, medial head, and common branch to the lateral and medial heads were 2.1 ± 0.8, 1.6 ± 0.9, 1.5 ± 1.0, and 0.7 ± 0.6, respectively. The number of branches was greater for the long and lateral heads than for the medial head, with statistical significance (p < 0.05). The nerve branch to the medial head was longer than the branches to the other heads. Lengths of the medial head branch, lateral head branch, long head branch, and the common branch to the lateral and medial heads were 63.5 ± 36.4 mm, 38.1 ± 26.7 mm, 23.2 ± 19.5 mm, and 48.5 ± 25.8 mm, respectively. Branches to the long head had a larger diameter than branches to the other heads. Diameters of branches to the long head, lateral head, and medial head were 1.4 ± 0.3 mm, 1.3 ± 0.3 mm, and 1.1 ± 0.4 mm, respectively. The average diameter of common branches to the lateral and medial heads was 1.7 ± 0.4 mm, which was significantly larger than that of branches to each head. Cross-section between the axillary nerve and radial nerve To identify the most suitable branch for nerve transfer, the diameter and axon numbers per nit area of branches between the axillary and radial nerves were compared. The diameter of the muscular branch was larger in the axillary nerve and its divisions than in the radial nerve and the branches to the triceps brachii (Table 2 ). The mean diameter of the axillary nerve before division was 3.0 ± 0.5 mm. In the anterior and posterior divisions, this diameter was 2.5 ± 0.6 mm and 2.3 ± 0.6 mm, respectively. In the radial nerve, the diameters of branches to the triceps brachii long head, lateral head, medial head, and common branch of the medial and lateral head were 1.4 ± 0.3 mm, 1.3 ± 0.3 mm, 1.1 ± 0.4 mm, and 1.7 ± 0.4 mm, respectively. In cross-section, the mean number of axons was 214.8 ± 43 and the number of axons in a 0.04 mm 2 area was similar among all branches (Fig. 4 ).
Discussion The axillary nerve ran a mean distance of 44.0 mm in the axilla and divided into anterior and posterior divisions before passing through the quadrangular space in every specimen we examined. Early bifurcation of the axillary nerve before passing through the quadrangular space is an expected finding because of innervation of the teres minor muscle on the muscle’s deep surface, which is proximal to the quadrangular space. If the division was distal to the space, branches to the teres minor muscle would have to take a recurrent pathway to re-enter the axilla from the posterior side of the arm. The axillary nerve is one of the most commonly injured nerves during surgical procedures of the shoulder 4 , 11 – 13 and axillary nerve injuries account for 6–10% of all brachial plexus injuries 11 , 12 .The entire axillary nerve or its divisions can be damaged by an injury near the quadrangular space. The axillary nerve is damaged most commonly at the posterior opening of the quadrangular space 3 . Owing to its early bifurcation, when the axillary nerve is injured near the quadrangular space, innervation of the teres minor muscle is protected by the muscle itself, as there is no superficial exposure to the quadrangular space. In contrast, denervation of the deltoid muscle can be caused by injury of the anterior division or injury of single muscular branches 13 – 15 . Each division of the axillary nerve is related to different movements of the glenohumeral joint. Therefore, the function of individual branches in terms of their effect on the deltoid muscle should be considered when performing nerve replacement. In shoulders with a mixed type axillary nerve where the spinous part of the deltoid is innervated by both divisions, the anterior division is responsible for medial rotation, flexion, abduction, extension, and lateral rotation of the shoulder; while the posterior division is responsible for extension and lateral rotation of the shoulder by the deltoid and teres minor muscles. In the anterior dominant type, the anterior division contributes to medial rotation, flexion, abduction, extension, and lateral rotation of the shoulder; while the posterior division is responsible only for extension and lateral rotation of the shoulder by the teres minor muscle. In the posterior dominant type, the anterior division helps the shoulder rotate medially, flex, and abduct but is not involved in lateral rotation or extension of the glenohumeral joint, which is performed by only the posterior division. This type corresponds with the nerve distribution of the deltoid muscle as described by Frohse and Frenkel 16 . In these types, the anterior division is important for abduction of the shoulder, for which other muscles such as the supraspinatus cannot compensate properly. In contrast, dysfunction of the teres minor and spinous part of the deltoid can be compensated by other muscles such as the infraspinatus, teres major, and latissimus dorsi 17 – 19 . Therefore, the posterior division may not be as important as the anterior division. Damage of the anterior division is therefore more serious than damage of the posterior division because of the possibility of compensation by other muscles in the latter case. The axillary nerve is vulnerable to injury from intramuscular injection of the deltoid muscle for vaccination 20 . In the present study, junctions of the axillary nerve branches to the deltoid muscle were confined to the second upper quarter of the muscle (Fig. 3 ). In most cases, the thickness of the deltoid muscle will protect against needle penetration 21 . However, especially in thin patients with small deltoid muscles, avoidance of this area during needle insertion is recommended. The distribution of the junctions is also important to consider when performing intraosseous infusion into the humeral head. During intraosseous access, the humerus is medially rotated, and the needle is inserted into the greater tubercle through its superolateral surface at a 45° angle to the horizontal plane 22 . According to our findings, insertion of the intraosseous needle is superior to the area of nerve-muscle junctions and favorable. There are two important principles to maximize the outcomes of nerve transfers. The first is to reinnervate the recipient nerve as close to the target muscle as possible 23 . Although some studies recommend nerve transfer using the branch to the long head of triceps brachii using a posterior approach 10 , 24 , the length of this branch was relatively short in our study. Because of the possibility of damage to the posterior branch of the axillary nerve and difficulty in dissecting the teres minor branch through a posterior approach 25 , 26 , a deltopectoral approach (anterior approach) is preferable for axillary nerve transfer 7 , 26 . The radial nerve branch to the lateral head is not easily accessible because it is covered by the triceps brachii muscle as it usually arises from the radial groove. Instead, use of the branch to the medial head of triceps brachii is recommended for restoration of the deltoid muscle. The medial branch of the radial nerve is easy to find because it runs across the medial head superficially. The second principle is to use combinations of similarly behaving neuromuscular units, maximized when agonistic donor and recipients are chosen, as cortical readaptation is the physiological basis for functional recovery 23 . In our study, the number of axons in a given cross-sectional area was similar between nerve branches, which means that the number of axons is proportional to the diameter of the nerve branch. It is reasonable to transfer a donor nerve with a similar diameter to the recipient nerve. This study has several limitations. The most significant limitation is that the measurement of lengths was conducted on formalin-fixed cadavers, which may result in deviations due to tissue shrinkage and damage from microdissection, potentially causing differences from values measured in living subjects. Additionally, the limited number of samples used for tissue staining means the findings may not fully represent the variation seen in a larger population, indicating the need for further studies with increased sample sizes to validate and extend our conclusions.
Conclusions The significant role of the axillary nerve in maintaining deltoid muscle functionality and in preventing palsy or malfunction highlights its vital importance in clinical and surgical procedures. Functional loss of the anterior division may be more critical because it can lead to difficulty in abduction by the acromial part of the deltoid 5 . Consequently, for axillary nerve transfer, the medial and long branches of the radial nerve, with their suitable length and axon density, are recommended as optimal donor choices for effective functional restoration.
This study investigated the anatomical details of the axillary and radial nerves in 50 upper limbs from 29 adult formalin-embalmed cadavers, and ten fresh upper limbs. The focus was on understanding the course, division, and ramifications of these nerves to improve treatment of shoulder dysfunction caused by axillary nerve damage. The axillary nerve divided anteriorly and posteriorly before passing the quadrangular space in all specimens, with specific distances to the first ramifications. It was found that the deltoid muscle's clavicular and acromial parts were always innervated by the anterior division of the axillary nerve, whereas the spinous part was variably innervated. The longest and thickest branches of the radial nerve to the triceps muscles were identified, with no statistically significant differences in fiber numbers among triceps branches. The study concludes that nerve transfer to the anterior division of the axillary nerve can restore the deltoid muscle in about 86% of shoulders, and the teres minor muscle can be restored by nerve transfer to the posterior division. The medial head branch and long head branch of radial nerve were identified as the best donor options. Subject terms
Acknowledgements The authors sincerely thank those who donated their bodies to science so that anatomical research could be performed. Results from such research can potentially increase mankind's overall knowledge that can then improve patient care. Therefore, these donors and their families deserve our highest gratitude 28 . Author contributions S.-J.K., J.-H.B., H.-J.Y., S.-H.M., Y.-R.C., and H.-Y.L. contributed to the study’s conception and design. S.-J.K., and J.-H.B. proceeded with gross dissection, histological experiments, and photographic work. H.-J.Y. provided assistance in histologic analysis and gave technical advice. S.-J.K., and J.-H.B. performed formal analysis, and writing of the original manuscript. H.-J.Y., S.-H.M., and Y.-R.C. contributed to reviewing and editing. H.-Y.L. proceeded with project administration, manuscript editing, and critical revision of the manuscript for intellectual content. All authors read and provided final approval of the version to be published. Data availability The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1262
oa_package/d0/4a/PMC10787799.tar.gz
PMC10787801
38218873
Introduction Reducing fossil fuel use and global climate change requires a fast energy transition, and nations across the globe have successively set out their own targets and pathways to carbon neutrality 1 . Since 2009, as the fastest-growing renewable power source, the generating capacity of solar photovoltaic (PV) energy has grown globally by 41% per year 2 . It has put forward higher requirements for the conversion efficiency and capital cost reduction of PV energy generation 3 , which is always impacted by cloud cover, aerosol and panel soiling 4 – 9 . Yet, in a stark contrast to aerosol and panel soiling, cloud cover or advection can dramatically and intermittently affect incident solar radiation, resulting in unbalance between the load demand and PV energy generation, which poses a considerable risk to the stability of power grids 10 – 12 . Therefore, reliable and powerful PV energy generation or global tilted irradiance (GTI, the radiation captured by solar photovoltaic panels) forecast technique, particularly short-term forecasts of the intra-day GTI or PV power generation (at the leading time of 0–4 h), is also highly beneficial to power smoothing processes and other load-following applications 9 , 13 . In addition, currently, in most European countries, short-term prices for the physical delivery of electricity are formed by spot markets, such as the European Power Exchange SPOT ( https://www.europex.org/members/epex-spot/ ). Although ~80% of trade volume is controlled by the day-ahead trading market, the intra-day auctions from hourly to 15-min intervals determine real-time electricity prices 14 . Thus, sophisticated solar PV power generation nowcasting technique not only can improve the stability of power generation, but also facilitates the developments of more commercially viable PV systems, the current electricity market and price transactions, and increases the competitiveness of the solar PV energy source 15 , 16 . In recent years, rapid advances in artificial intelligence have promoted the application of data-driven machine learning-based approaches in Earth system science 17 , 18 . Particularly, some recent studies 12 , 13 , 19 – 21 on the prediction of solar radiation also explicitly indicate that advanced prediction approaches based on machine learning perform better compared with empirical models, time series, and hybrid algorithms, such as artificial neural networks and support vector machines. Nevertheless, it is still a great challenge to predict cloud motion, formation, deformation and dissipation under complex atmospheric dynamics, geography, and climatic conditions 9 , 22 , 23 . Thus, there is still no solar radiation forecast model that can work well in every region and at every time 21 . Cloud cover nowcasting remains a field of interest for forecasting the electricity production of PV plants 24 . We are committed to developing a daytime hourly intra-day cloud fraction (CF) prediction algorithm for small areas over PV plants. Based on the recurrent-neural-networks-based (RNNs) long short-term memory (LSTM) algorithm framework, the newly developed PredRNN and PredRNN++ (an extended and latest version of PredRNN) 25 , 26 can well learn to predict long-term future imageries in various spatio-temporal tasks by modeling their spatial and temporal dependencies, including video frame prediction, human motion prediction, etc. Therefore, our primary objective is to develop an innovative and easy-to-promote algorithm or system based on the key framework of the PredRNN++ model. Through this algorithm, the 0–4 h CF at solar PV plants under all weather conditions can be predicted by using sequential Himawari-8/9 geostationary satellite images with high spatio-temporal resolutions 27 . Compared with the previous study 28 , it only used a single visible channel of geostationary satellite and a constant model to predict cloudiness. Some former studies directly used surface solar global horizontal irradiance (GHI) as model input to predict GHI values in the next few hours 29 , achieving the purpose of estimating the power generation of PV plants. Nevertheless, the presence of clouds is still identified as the primary uncertainty in current surface solar GHI forecasts 30 . In contrast, our investigation only predicts geostationary satellite Level 1B (L1B) radiance data. With the prediction results of satellite L1B radiance data and accurate cloud detection algorithm, this approach is expected to provide reliable and variable CF information for further improving the predictability of current GTI or PV power generation.
Methods The newly developed EPM and fast cloud mask algorithm in the NCP_CF system are applied to predict CF (or CSR) at the leading time of 0–4 h at two test PV plants. In this system, the −4–0 h geostationary satellite radiance data is used as input to dynamically provide 0–4 h satellite cloud images and fractions. To verify the reliability of the EPM, we first use the CF observations from twelve widely distributed ground-based manual stations and three all-sky imager stations for the period from 2019 to 2022 as true values to compare with the predictions in the corresponding period. Moreover, correlations of the predicted CSRs with the actual PV power generation and surface solar radiation at five test PV plants from October 2022 to March 2023 are analyzed. The benefits of the geostationary satellite data with high spatio-temporal resolutions and the advanced EPM to improve the PV power generation efficiency are investigated, as well as the wide applicability and generalizable value of the EPM system. Ground-based observation data The total cloud cover or CF (~20 km × 20 km square area) used in this study is obtained through manual observation at twelve ground-based meteorological stations (Fig. 1 ) in January, April, July, and October of 2019. Note that due to the relatively large errors in low-visibility conditions, the CF data with the matched and automatically measured visibility <2 km is removed. Besides, the view zenith angles of ground-based stations from H8/AHI field of view used here, as stated in Supplementary Table 2 , are smaller than 60°. Therefore, the parallax effect is negligible (error < 1 km) in the collocation between satellite pixels and ground-based stations for this study (For more explanations and details, please refer to Supplementary Note and Supplementary Figs. 7 and 8 ). Firstly, considering the daytime nowcasting applications and sunshine conditions at PV plants, only the manual observations at 11:00, 14:00 and 17:00 are collected for validation in this research. Secondly, three ground-based all-sky imager stations (equipped with a Japan EKO ASI-16 all-sky imager, https://www.eko-instruments.com/us/categories/products/all-sky-imagers/asi-16-all-sky-imager ) can provide the high temporal resolution (5 min) and valuable CF and cloud cover data during the daytime, which are retrieved by the standard EKO ASI-16 cloud detection algorithm ( https://www.eko-instruments.com/media/z2aalysq/asi-16-software-manual-find-clouds.pdf ) 36 . ASI, equipped with a digital camera coupled with an upward looking fisheye lens, could provide field of view (FOV) of ~180°, but pixels at a FOV > 140 ° are excluded due to distortion. Digital images of the sky obtained by ASI are classified pixel by pixel into clear sky, optically thin and optically thick clouds, respectively. The cloud detection and opacity classification (CDOC) algorithm developed by Ghonima et al., 2012 36 could provide 96%, 60%, and 96.3% accuracy in the validation for clear, thin, and thick cloud, respectively. Finer CF data allow more accurate validation of the NCP_CF system, thereby demonstrating the specific prediction effect at the forecast leading time of 0–4 h. The study periods of the local CF data from three all-sky imagers located in Zhuhai, Nanjing and Beijing (Fig. 1 ) are from September 2022 to February 2023, April 2021 to September 2021, and September 2022 to December 2022, respectively. In addition, the actual power generation (MW, temporal resolution of 15 min) and GTI (W·m −2 ) measured at Sangge and Leling PV plants from November 2022 to March 2023 and at Xiaochengzi, Lijiamen and Shiziyan PV plants in November 2022 (Fig. 1 ) are also used to analyze the agreement with the predicted CF. These real power generation data are obtained from the SCADA (Supervisory Control and Data Acquisition) system of China General Nuclear Power Group Wind Energy Co. Ltd. Geostationary satellite data and calculating the cloud fraction The NRT 16-band full-disk AHI level-1B radiance data from the Himawari-8/9 satellite (the new-generation Japanese geostationary meteorological satellite) with spatio-temporal resolutions of 1–4 km and 10 min are obtained from the Japan Meteorological Agency Himawari-Cast in China 32 . Additionally, the offline Himawari-8/9 data at the original resolution (0.5–2 km) are also available for free download from the JAXA (Japan Aerospace Exploration Agency) Himawari satellite data FTP (File Transfer Protocol) site (ftp.ptree.jaxa.jp) from July 7, 2015 ( http://www.jma-net.go.jp/msc/en/ ). The nadir point of the Himawari-8/9 satellite is located at 140.7°E, and the coverage of this satellite includes the Japan island and the eastern parts of China. Based on the real-time Himawari-8/9 AHI full-disk observation data, each site would be centered around its precise location and matched with a 32 × 32 pixels box as the experimental area. The special fast cloud mask algorithm 32 combines five inherited and improved cloudy/clear pixel tests in visible and infrared bands to determine the final confidence value ( c ) of every pixel of the satellite imager, i.e., c > 0.99 = clear, 0.95 < c ≤ 0.99 = probably clear, 0.66 < c ≤ 0.95 = probably cloudy, and c ≤ 0.66 = cloudy. As the real viewing field at a ground-based station approximates a 20 km × 20 km square area, a 5 × 5 pixels box of cloud mask centered around a targeted PV plant is used in this study to calculate the CF predictions, which is expressed as Eq. ( 1 ). where and indicate the total numbers of the cloudy and probably cloudy pixels in the 5 × 5 pixel box 37 , respectively. The complementary CSR is equal to 1-CF. Model The PredRNN++ model 24 (an improved prediction RNN), dedicated to short-term prediction and nowcasting, is adopted as a key model in this investigation for 0–4 h CF nowcasting. This advanced neural network successfully overcomes the spatio-temporal predictive learning dilemma between deep-in-time structure and vanishing gradient. Previous research has demonstrated that the PredRNN++ consistently outperforms the ConvLSTM, TrajGRU, Discrete Fracture Network, MCnet and PredRNN at every future time step for both peak signal-to-noise ratio and structural similarity index measure 24 . To achieve greater spatio-temporal modeling capability, in this investigation, we re-design and develop the EPM with five convolutional layers, whose elaborated structure is shown in Fig. 5 . The details of the casual LSTM and the Gradient Highway Unit (GHU) 24 in the EPM structure are also illustrated in Supplementary Fig. 2 . The causal LSTM, an upgraded version of the LSTM, increases the recurrence depth from one time step to the next and derives a more powerful modeling capability for stronger spatial correlations and short-term dynamics. As shown in Supplementary Fig. 2a , a causal LSTM unit contains two memories, namely a temporal memory and a spatial memory , where the superscripts and t denote the k th hidden layer in the stacked causal LSTM network and the t th time step, respectively. The temporal memory depends on its preceding state and is controlled by an input gate , a forget gate and an input modulation gate . The spatial memory relies on which is in the deep transition route. Notably, the topmost spatial memory is assigned to the bottom spatial memory . For the k th layer, the updated equations of the causal LSTM can be expressed as Eqs. ( 2 – 7 ). where “⊚" denotes the convolution, "⨂" the element-wise multiplication, tan h the element-wise hyperbolic tangent function, σ the element-wise sigmoid function, and “[]” a concatenation of tensors. the convolutional filters. All the equations in the causal LSTM can be briefly expressed as Eq. ( 8 ). where is replaced by and between the first and second layers. The gradient highway, a shorter route from future outputs back to distant inputs, can alleviate the vanishing gradient problem. As shown in Supplementary Fig. 2b , in a GHU, represents the convolutional filters, is a switch gate and enables adaptive learning between the transformed input and the hidden state . The equation of the GHU can be written as Eqs. ( 9 – 11 ). The equations of the GHU can be briefly expressed as: As presented in Fig. 5a , combined with Eqs. ( 8 ) and ( 12 ), the key equations of the entire EPM framework can be written as Eqs. ( 13 – 18 ). In the EPM framework, the GHU is injected between the first and second causal LSTMs. The causal LSTM and GUH respectively capture short-term and long-term data or image dependencies. The gradient highway (blue line in Fig. 5a ) supplies a quick path from the first to the last time step by quickly updating hidden state . It is worth noting that, unlike temporal skip connections, the GHU controls the proportions of and the deep transition feature through , which allows the EPM to adaptively learn both short-term and long-term frame relations. In the causal LSTM of the EPM, the spatial memory is a function of the temporal memory through another set of gate structures. As the recurrence depth along the spatio-temporal transition paths grows considerably, each pixel in the final generated frame has a bigger receptive field of the input sequence at each time step, which is why the EPM has a better ability to model short-term video dynamics and sudden changes. Figure 5b displays the data flow process in the spatial memory of the EPM. The dimension of the input sequential satellite data (4 h and a time interval of 10 min) is 24 × 32 × 32. At the first convolutional layer, the input terms and are concatenated to form a larger tensor, and then is generated by the convolution calculation. The EPM performs a total of five convolution calculations, with dimensions of hidden state from 10 × 32 × 32 to 1 × 32 × 32. The EPM training process involves the use of the Adam optimizer with a learning rate of 0.001 and the mean square error as the loss metric. The input is a four-dimensional tensor of size [ c, t, h, w ] (6, 24, 32, 32), where c represents the number of input channels, t represents time (spanning 4 h), h represents height, and w represents width. Our research area focuses on PV plants, which provides us with images measuring 32 × 32 pixels (~128 km × 128 km). To train our model, we created a dataset consisting of sequences of 48 images for each channel, spanning 4 h. During the initial training process of the optimal network structure, data were obtained from resized AHI images at six bands over twelve manual observation stations and three all-sky imager stations. The training set utilized data from January to October in 2018, while the remaining months of 2018 were used as the validation set. In the application scenario of PV plants, the training method is described in the section of NRT and cyclically updated prediction system. Validation and assessment . The primary metrics used to evaluate the accuracy of the NCP_CF system for forecasting the CF are the RMSE, MBE and R, defined as shown in Eqs. ( 19 )-( 21 ). where P i denotes the predicted CF, T i denotes the actual CF obtained from ground-based observations mentioned above, and N is the total number of the matched samples. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Results Cloud fraction nowcasting and validations In order to better simulate real application scenarios, a quasi-operational and near real-time (NRT) and cyclically updated prediction system is newly developed for 0–4 h CF nowcasting at solar PV plants (hereafter referred to as the NCP_CF). The predicted CF from this NCP_CF nowcasting system is also compared with the real PV power generation and the GTI to verify its feasibility, reliability and adaptability. The results from the NCP_CF system are examined and validated by using the observed CF values from twelve manual meteorological observation stations of the China Meteorological Administration (only the observations at 14:00 and 17:00 are used) and three all-sky imager stations (Fig. 1 ). Figure 2 shows the root mean square errors (RMSEs) and mean bias errors (MBEs) of the CF predictions at targeted stations with the forecast horizon, local time (diurnal cycle) and time series. In terms of forecast horizon (Figs. 2 a– 2f , Supplementary Table 1 ), the RMSE increases from 0.18 to 0.35 at all stations for 0–4 h forecast periods, and the MBE fluctuates around −0.1. Notably, the RMSE is less than 0.25 within the 2-h leading period, but the forecast accuracy decreases faster when the forecast leading period exceeds 2 h, indicating that the forecast performance threshold of this system is ~2-h leading time. Considering the continuity and coverage of observation time, Fig. 2g only shows the diurnal cycle of CF forecast accuracy at three all-sky imager stations. Within a one-day forecast window, the most and least accurate predictions occur during 12:00–17:00 and 08:00–09:30, respectively, whereas the relatively moderate decrease in the forecast accuracy before 09:30 is mainly attributed to the invalid satellite visible images before 08:00. Regarding the predicted performance for different months, the monthly mean RMSE and MBE values in time series at three all-sky imager stations do not vary considerably, indicating the weak monthly dependence and stability of this cyclically updated CF nowcasting algorithm and system. Furthermore, the correlation coefficients (R) between the 1–4 h predicted clear sky ratios (CSRs; CSR = 1−CF) from the NCP_CF system and the actual power generation (GTI) are calculated. Figure 3 displays the comparisons among the 1–4 h predicted CSRs, the actual power generation and the GTI at Sangge, Leling, Xiaochengzi, Lijiamen, and Shiziyan PV plants from 09:00 to 17:00 in November 2022. The mean R values between the 1–4 h predicted CSRs and the actual PV power generation (the GTI) in November 2022 at five PV plants are 0.81, 0.73, 0.65, and 0.55 (0.81, 0.72, 0.64, and 0.55). This result highlights the good consistency of the predicted CSRs with the actual PV power generation and the GTI, especially for the first 2-h leading time. Overall, the EPM-model-based NCP_CF system developed in this research is applicable to provide high-quality CF estimations at PV plants in advance, which can be used to predict the GTI and power generation at the forecast leading time of 0–4 h. The results at Sangge and Leling PV plants from December 2022 to March 2023 are displayed in Supplementary Figs. 3 – 6 . Near real-time and cyclically updated prediction system The satellite-based NRT and cyclically updated prediction system (Fig. 4 ) for 0–4 h CF nowcasting, operating at five real PV plants (Sangge, Leling, Xiaochengzi, Lijiamen and Shiziyan PV plants) belonged to China General Nuclear Power Group Wind Energy Co. Ltd., mainly consists of three subsystems, i.e., preprocessing, prediction and retrieval modules. Specifically, the preprocessing module regularly adjusts the real-time down-sampling Himawari-8/9 Advanced Himawari Imager (AHI) data received from the direct broadcast receiving system. The AHI is an advanced imager with 16 spectral bands ranging from 0.47 μm to 13.3 μm, which has spatio-temporal resolutions of 4 km and 10 min 31 . The full-disk Level-1B radiance data at 0.65 μm, 0.86 μm, 3.9 μm, 7.0 μm, 11.2 μm and 12.3 μm with high-quality geolocation and radiometric calibration is grided into a 32 × 32 pixel box (~128 km × 128 km) centered around the targeted PV plant. Then, these sequential and resized AHI images at six bands are converted into a tensor [tile size = 6 (band number) × 24 (4-h time sequence) × 32 × 32] of the prediction model for forecasting the following 0–4 h satellite images (6 × 24 × 32 × 32). As a key function of the prediction module, an enhanced PredRNN + + model (EPM; more details in “Model” section) is developed in this study to predict 0–4 h sequential geostationary satellite images. In order to better track the fast and stochastic changes in cloud images, the neural network of the EPM is always cyclically generated by the scheduled training process of the model, which has a 1-h update frequency. The cumbersome cyclic process should take 40–50 min when using the latest sequential satellite images (6 × 24 × 32 × 32) between −5 h and −1 h as training samples and one graphics processing unit processor (NVIDIA Tesla-V100). For instance, a cyclic training process of the model starts at the scheduled local time of 12:08 and ends at ~12:53, the imported training samples are from 08:00 to 12:00, and the latest updated EPM is timely activated for CF nowcasting at ~13:08 (nowcasting from 13:00 to 17:00) and 13:38 (nowcasting from 13:30 to 17:30), respectively, with an update frequency of a frequency of half an hour. During the same period, the network of the next EPM is also trained simultaneously by using the sequential satellite images from 09:00 to 13:00. This cyclical procedure will continually update and replace the existing EPM every hour, ensuring that we always have the most up-to-date nowcasting model. Considering the use of satellite visible images, the NCP_CF system only operates from 07:20 to 17:20, which still meets the requirement of CF nowcasting at PV plants. In the retrieval module, the 0–4 h predicted and resized cloud images mentioned above are used by a fast cloud mask algorithm 32 , which is able to calculate the number of cloudy pixels and the CF or cloud cover in the observation field of the targeted PV plant. Note that compared with the operational cloud mask algorithm, the fast cloud mask algorithm can fast retrieve the CF without any ancillary data, which is crucial for CF nowcasting. Supplementary Figure 1 presents the comparisons between the predicted and actual satellite images and cloud mask results at Zhuhai station on 17 November 2022, illustrating the good agreement between them.
Discussion Our study demonstrates that the NCP_CF system can provide high-efficiency, high-quality and adaptable 0–4 h CF nowcasting data for PV plants. As shown in Figs. 2 a– f , the mean RMSE (MBE) values are 0.21 (−0.09), 0.25 (−0.08), 0.3 (−0.07), and 0.35 (−0.03) for the forecast leading time of 1 h, 2 h, 3 h, and 4 h, respectively. Particularly, the CF nowcasting results from the NCP_CF system remain highly reliable within the forecast leading time of 2 h, with RMSE values staying almost at 0.2 and not increasing within the 1-h leading time. Conversely, the prediction performance of the NCP_CF system gradually deteriorates as the forecast leading time increases to more than 2 h, which may be due to the vanishing gradient problem 22 . By the limited spatial domain, the rapid movement of clouds may cause a small bias between the predicted CF and the actual CF. Further analyses on the daily and seasonal scales are also conducted, as shown in Fig. 2 g, a–f, respectively. On the daily scale, the NCP_CF system performs particularly well from 09:30 to 18:00, whereas it shows poor performance before dawn, mainly due to the poor quality of satellite data at the visible band during that time. Fortunately, this issue is mitigated due to the low power generation of PV plants before dawn, and thus cloud cover has a low impact on power generation in this period. For the seasonal scale, except for the forecast results at Nanjing station in May 2021, the NCP_CF system shows stable forecast performance and seasonal biases at different stations and in different seasons. Its salient adaptability thus is the largest advantage compared with other solar radiation nowcasting methods summarized in the previous review 21 . Overall, the CF nowcasting results of the NCP_CF system have good stability, strong generalizability and non-sensitivity to geographical locations and climatic characteristics. Given that the present electricity spot markets in Europe work within different time horizons, specific and professional forecast techniques are required for each leading time 13 . The NCP_CF system with the 0–4 h forecast leading time within a 10-min interval shows more advantages than other existing nowcasting methods, such as the manners based on all-sky imager observation (the forecast leading time only ranging from 0 to 20 mins) 33 and numerical weather prediction (from 6 h to day-ahead time frames) 34 for solar PV power generation. Principally, this system is applicable to fast varying small-scale weather and environmental conditions and can accurately capture cloud motion over PV plants without relying on long-term historical in-situ meteorological data. Besides, the predictions of the NCP_CF system are not markedly affected by seasonal climate changes on long-term scales, which underscores the stable operation of this system. The system shows excellent forecast performance within the first 2-h leading time, with an average R value between the predicted CF and the actual power generation or GTI at PV plants close to or more than 0.80. Since one of the greatest challenges facing solar PV renewable energy production is its instability and intermittency, accurate CF nowcasting is still vital for the efficient operation of PV plants and their power systems. Improving the stability of PV power production can directly facilitate policy-making of feed-in tariffs and attract more investment in solar PV power generation 14 , 35 . However, most importantly, the nowcasting technique developed in this research deserves attention in terms of promoting the overall penetration of solar power on the electric grid and having a non-negligible impact on electricity price trading in the intra-day spot market 14 . Furthermore, it is evident that increasing the share of renewable energy in the global energy system can contribute to the reduction of global carbon emissions 3 . Therefore, our future mission is to further promote applications and improve the accuracy of this cloud cover nowcasting technique, especially for the forecast leading time of more than 2 h, by using higher spatial-resolution satellite data (i.e., 1–2 km) and combining the short-term forecast data from a rapidly updated regional high-resolution numerical weather prediction.
Accurate nowcasting for cloud fraction is still intractable challenge for stable solar photovoltaic electricity generation. By combining continuous radiance images measured by geostationary satellite and an advanced recurrent neural network, we develop a nowcasting algorithm for predicting cloud fraction at the leading time of 0–4 h at photovoltaic plants. Based on this algorithm, a cyclically updated prediction system is also established and tested at five photovoltaic plants and several stations with cloud fraction observations in China. The results demonstrate that the cloud fraction nowcasting is efficient, high quality and adaptable. Particularly, it shows an excellent forecast performance within the first 2-hour leading time, with an average correlation coefficient close to 0.8 between the predicted clear sky ratio and actual power generation at photovoltaic plants. Our findings highlight the benefits and potential of this technique to improve the competitiveness of solar photovoltaic energy in electricity market. Accurate nowcasting of cloud cover or fraction and its movement remains a significant challenge for stable solar photovoltaic electricity generation. Here, the authors combine continuous radiance images with high spatio-temporal resolutions to develop a nowcasting algorithm for predicting cloud cover at a leading time of 0–4 h. Subject terms
Supplementary information Source data
Supplementary information The online version contains supplementary material available at 10.1038/s41467-023-44666-1. Acknowledgements We appreciate the Facebook Inc. for freely providing the Pytorch toolkit ( https://pytorch.org ). This study was supported partly by the Innovation Group Project of Southern Marine Science and Engineering Guangdong Laboratory (Zhuhai) (No.SML2023SP208), the National Natural Science Foundation of China (Grants 42175086 and U2142201), the FengYun Meteorological Satellite Innovation Foundation under Grant FY-APP-ZX-2022.0207, and the Science and Technology Planning Project of Guangdong Province (2023B1212060019). Author contributions P.X. and M.M. conceived and designed the study. P.X., M.M., L.Z. and J.L. collaborated in discussing the results and writing the paper. Y.W. provided the actual power generation and surface solar radiation data. Y.Y. and S.J. provided the ground-based observation data. All authors contributed edits. Peer review Peer review information Nature Communications thanks Sakshi Mishra and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Data availability The Himawari-8/9 data are available for free download from website [ http://www.jma-net.go.jp/msc/en/ ]. The photos above PV plants are available for free exported from Google Earth Pro. Source data are provided with this paper. Code availability Data processing, drawing and FCMA were conducted using PYTHON. Those codes can be accessed at [ https://zenodo.org/doi/10.5281/zenodo.10148796 ]. The EPM code generated during the current study is available from the corresponding author upon request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Nat Commun. 2024 Jan 13; 15:510
oa_package/ba/ab/PMC10787801.tar.gz
PMC10787822
38218883
Introduction Long fusion surgery for adult spinal deformity (ASD), performed only in a limited number of centers more than a decade ago, has rapidly spread and is now a standard and widely performed procedure 1 . ASD surgery was primarily performed for de novo scoliosis in the early days. Later, ASD became a broad disease concept that included sagittal imbalance as a surgical target. Thus, although ASD has complex conditions, patients with symptoms that warrant surgical treatment should have specific common problems. The Scoliosis Research Society-22 Patient Questionnaire (SRS-22) is a standard questionnaire used to evaluate the treatment of scoliosis 2 . The SRS-22 is sometimes used to assess ASD as well, because no ASD-specific scale currently exists. However, the questions in the SRS-22 were designed primarily for adolescent idiopathic scoliosis (AIS). AIS and ASD have different ages of onset, various pathologies, and main complaints. In addition, in AIS, the lowest end of fixation is usually more proximal than L3, whereas, in ASD, the level of fixation often includes the pelvis, which is often accompanied by postoperative mobility restrictions 3 , 4 (Fig. 1 ). Recently, Hart et al. developed the lumbar stiffness disability index to evaluate the limitation of motion of the spine due to long fusion surgery 5 . They called the restriction for activities of daily living (ADL) due to long fusion the collateral outcome. There is a trade-off relationship, so to speak, between improving pain due to fusion and restriction of range of motion. This trade-off is considered to be well established if the patient’s needs are met 6 . Thus, ASD presents a unique condition among spinal disorders that has elements of scoliosis but also kyphosis, as well as pain and limited postoperative range of motion. Although surgery for ASD is becoming more widespread, some researchers are concerned about the cost of the procedure and the high complication rate 7 . Conversely, conservative treatment of ASD includes medication, orthotics, Nordic walking canes, and walkers. These conservative treatments have the advantage of being less risky and less expensive than surgery and do not cause a postoperative range of motion limitations. However, conservative treatment could be less effective with respect to improving posture and pain. Furthermore, the use of a cane may be inconvenient for household activities because both hands are occupied when walking 8 . Currently, there is no HR-PRO that evaluates these life inconveniences from the perspective of ASD patients. Therefore, we thought that a specific scale was needed to evaluate ASD. This study aimed to create a disease-specific patient-reported outcome measure (PROM) for ASD.
Methods Patients This study was a multicenter, self-report questionnaire survey conducted at two spine centers. In total, 106 patients were included: 97 patients who underwent long fusion surgery between 2007 and 2020 and nine patients who were undergoing conservative treatment and considering surgery for spinal deformity. The conservative patients had spinal deformities but preferred conservative treatment because their clinical symptoms were milder than those of the operative patients. A questionnaire consisting of 29 questions was mailed to these patients, and they were asked to complete and return it. Patients who had undergone surgery were asked to answer both preoperative and postoperative conditions. Conservatively treated patients were asked to answer questions about their current condition. A five-point satisfaction rating scale for surgery and Short-Form-8 (the physical component summary; PCS, and the mental component summary; MCS) were enclosed for criterion-related validation. Of the 106 patients, eight did not receive the mailing due to a change of address. The 98 patients (89 surgical patients) who responded were included in the study (Fig. 2 ). Long fusion was defined as the fusion of five or more vertebrae, including the lumbar spine. Fixation across the sacroiliac joint to the pelvis was counted as one vertebral segment. On imaging evaluation, all patients had a coronal plane Cobb angle > 30°, SVA > 40 mm, or pelvic tilt > 20° 9 . Selection of 29 questions COnsensus-based Standards for the selection of health Measurement Instruments (COSMIN) aimed at improving the selection of PROM in research and clinical practice and some guidelines exist. We conducted this study in accordance with the COSMIN guidelines 10 . Content validity is the most important measurement property of PROM. It is the degree to which the content of an instrument is an adequate reflection of the construct to be measured. The criteria of content validity include the relevance, comprehensiveness, and comprehensibility of the PROM for the target population. We conducted a literature search to select questions relevant to ASD. We assumed that ADL, appearance, pain, mental health, and satisfaction would be the assessment items necessary to capture the disease concept of ASD 1 , 3 , 6 , 11 – 13 . To develop the comprehensive questions, we reviewed a wide variety of existing questionnaires (Table 1 ), including Short-Form-36 14 , patient-reported outcomes measurement information system (PROMIS) 15 , Oswestry disability index (ODI) 16 , Roland–Morris questionnaire 17 , SRS-22 2 , Japanese Orthopedic association back pain evaluation questionnaire (JOABPEQ) 18 , Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) 19 , Knee Society Score 20 , Bath Ankylosing Spondylitis Functional Index (BASFI) 21 , Health Assessment Questionnaire (HAQ) 22 , pain disability assessment scale 23 , Zurich claudication questionnaire (ZCQ) 24 , EuroQol 5-dimensions 5-levels (EQ5D) 25 , lumbar stiffness disability index (LSDI) 5 , 25-question geriatric locomotive function scale (Locomo-25) 26 , gastroesophageal reflux disease questionnaire (GerdQ) 27 , and the Frequency Scale for the symptoms of gastroesophageal reflux disease (FSSG) 28 . In total, 390 items were placed into the following categories by content: (1) pain, (2) appearance, (3) sleeping, getting up from bed or floor and bedtime-related activities (4) sitting, standing up, and other sitting-related activities, (5) standing, walking, and stairs, (6) toilet and bathing-related activities, (7) dressing-related activities, (8) transportation, (10) housework, (11) sports, (12) social activities, (13) meals, and (14) mental health. From these categories, we extracted 114 items that were considered useful for assessing ASD (Fig. 3 ). Sexual life, although an important item, was not included because of the expected large number of non-responses 28 . To ensure the relevance of questions to ASD in content validity, eight surgeons with extensive experience in operating on patients with ASD gave these 114 items a score from 3 to 0 according to their level of importance. We used the total score as a reference and selected 29 question items after discussion among the senior surgeons (Table 2 ). We modified detailed wording partially modified as appropriate. To examine the results comprehensibility, the developed questionnaire was given to three patients and one nurse, who reviewed the items in terms of text, meaning, and ambiguity and who provided feedback. Responses were on a five-point scale 29 , with an additional free-text field. Ethics statement The study was conducted in accordance with the ethical standards of the Declaration of Helsinki. The study was approved by the local ethical review board (Osaka University Hospital Ethics Review Committee. No.11360). Written informed consent was obtained from each patient. Statistical analysis The COSMIN guidelines introduce classical test theory and Rasch analysis for construct validation. We used classical test theory and factor analysis. Factor analysis was used to reduce and group the questions in order to create a valid, simple, and easy-to-use questionnaire. An exploratory factor analysis was performed using the maximum likelihood method on data from a total of 98 patients, including 89 postoperative responses and nine conservative cases. The number of factors was determined using the scree method. Because correlations between factors can be assumed, oblique rotation was performed using the Promax method. Finally, reliability was evaluated for content consistency using Cronbach’s coefficient alpha. Score calculation formula Factor score coefficients obtained from factor analysis were used as a reference to correct the coefficients so that the scale’s total score ranged from 0 to 100. Specifically, individual items were weighted so that the difference between the minimum and maximum factor scores was approximately 100 depending on the choice of response 14 . However, we provided greater weight to those questions that clinicians deemed important. For example, 0 represented a limited health status and 100 represented an excellent health status. Comparison of scores and responsiveness We compared the scores of the created scale, the PCS, and the MCS before and after surgery (paired t-test). Similarly, we compared the scale scores between the operated and conservative groups (unpaired t-test). We calculated Cohen’s d effect size by taking the difference between two means and dividing it by the standard deviation of the data. Cohen’s d effect size was used to evaluate the internal responsiveness of the scales. Next, we calculated Spearman’s correlation coefficients between the five satisfaction levels and the amount of score change on each scale. The external responsiveness of the scales was evaluated using Spearman’s correlation coefficients. An effect size of 0.2–0.49 was considered small, an effect size of 0.5–0.79 was considered moderate, and an effect size of 0.80 or greater was considered large 30 . A correlation coefficient of 0.2–0.39 was considered weak, a correlation coefficient of 0.4–0.69 was considered moderate, and a correlation coefficient of 0.70 or greater was considered strong. A p-value < 0.05 was considered statistically significant for two-tailed tests. SPSS Statistics (version 20; IBM, Armonk, NY, USA) was used for statistical analysis. External validation We collected new patients with ASD from another institution for external validation. We applied our ASD disease-specific scale for these patients and compared the results with the internal validation data.
Results Demographics of the patients Of a total of 98 patients, 88 were women. The mean age of the 89 operative patients was 68 ± 7 years, and the mean time since the last surgery was 56 ± 35 months (Table 3 ). The mean number of fixed vertebral segments was 10 ± 3, including the sacrum or pelvis, in 76 patients (85%). The preoperative PCS was 31 ± 7 and improved to 41 ± 8 postoperatively (p < 0.0001). Postoperative satisfaction was 23 (26%) very satisfied, 42 (47%) satisfied, 18 (20%) neither satisfied nor dissatisfied, and 6 (7%) dissatisfied. Response of the patients The results of the responses to each question are shown in Table 4 , and the correlation coefficients are shown in Table 5 . Seven patients had a free-text response of not performing Q23 heavy housework. Therefore, Q23 heavy housework was deemed inappropriate and excluded from the factor analysis. Regarding Q16 walking distance, four patients answered that they did not know the distance. Because there was a strong correlation between Q16 walking distance and Q17 walking time, we considered that Q17 walking time could be substituted for Q16 walking distance and excluded Q16. Factor analysis was conducted on the remaining 27 questions. Factor analysis The two-factor solution was adopted based on the decay status of the eigenvalues (scree criteria). The proportion of the total variance of the 27 items explained by the two factors before rotation was 47%. Each item was ordered by factor loadings (Table 6 ). The first factor was named the main symptom because many of the symptoms were related to the patient’s primary complaints, such as the ability to do housework and walk, including Q25 dishwashing, Q21 laundry, Q20 shelving, and Q17 walking. The loadings for Q1 appearance, Q2 back pain, and Q29 anxiety were relatively low but were included because we considered these questions essential. We selected Q19 ride, Q24 garbage disposal, and Q15 standing as the remaining questions, according to factor loadings. Because Q22 light housework was strongly correlated with Q25 dishwashing (r = 0.82) and Q21 laundry (r = 0.80) and was considered to refer to the same thing, we excluded Q22. A total of 10 question items (Q1 appearance, Q2 back pain, Q15 standing, Q17 walking, Q19 ride, Q20 shelving, Q21 laundry, Q24 garbage disposal, Q25 dishwashing, Q29 anxiety) were used for the main symptom factor. The second factor was named the collateral symptom because many items were related to postoperative limitation of movement, such as Q12 socks wearing and Q9 picking up. Because wearing Q11 pants and Q12 socks were highly correlated (r = 0.76), we excluded Q11 because Q12 socks could be substituted for Q11 pants. According to factor loadings, we selected five question items (Q7 standing up floors, Q8 toilet, Q9 picking up, Q10 washing, Q12 socks) as collateral symptom factors. Reliability Internal consistency The Cronbach’s alpha coefficient was 0.90 for the main symptom and 0.84 for the collateral symptom. Calculation of scores The factor score coefficients were used as weighting coefficients for each question, rounding the factor score coefficients to whole numbers to distribute the total scale score was distributed from 0 to 100. Because Q1 appearance and Q2 back pain are particularly important items, we gave them the same coefficients as Q25 dishwashing, which had a higher factor score coefficient. The better symptoms were set to 100 and the worse symptoms were set to 0. The calculation formulas are shown below (Supplement File 1 ). Score and responsiveness Score change The scores calculated based on the above formula are shown in Table 7 . Comparing the operative and conservative groups, the main symptom of the operative group was 47 ± 21 preoperatively, while the conservative group was 63 ± 15. The operative group had significantly worse preoperative main symptoms than the conservative group (p = 0.029). However, the main symptom of the surgical group significantly improved to 70 ± 22 after surgery (p < 0.0001), exceeding those of the conservative group. As a result of the surgical improvement, there was no significant difference between the postoperative main symptom of the operative group and the main symptom of the conservative group (p = 0.3). The mean collateral symptom score in the operative group worsened from 76 ± 25 preoperatively to 60 ± 25 postoperatively (p < 0.0001). The preoperative collateral symptom score in the operative group was significantly worse than that in the conservative group, 92 ± 12 (p = 0.005). Effect size The effect size measured by Cohen’s d was 1.09, indicating a large effect size, for the main symptom for comparison of the preoperative and the postoperative score (Table 7 ). In the same comparison, the effect size of the collateral symptom was 0.65 (moderate), and that of the PCS was 1.26 (large). In a comparison of operative and conservative groups, the effect size was 0.77 for the main symptom and 0.67 for the collateral symptom, indicating a moderate effect size. Correlation coefficient The Spearman’s correlation coefficient between satisfaction and the amount of score change was 0.48 (p < 0.001) for the main symptom and 0.38 for the PCS, both showing a moderate correlation (Table 8 ). The correlation coefficient between the main symptom and the PCS was 0.43, indicating a moderate correlation (p = 0.002). Ceiling and floor effects The main symptoms had no floor or ceiling effect either preoperatively or postoperatively (Figs. 4 , 5 ). Conversely, the collateral symptom had a ceiling effect preoperatively, but no floor effect postoperatively (Figs. 6 , 7 ). External validation We added a new sample of 30 surgical patients with ASD in another facility for a disease-specific scale for ASD that we had created. This scale consisted of 10 main symptom and 5 collateral symptom questions, as described above. Total scores were calculated using the above formulas (Supplementary File 1 ). The SF-8 and satisfaction scale were enclosed, as well as the date when the scale was created. Twenty-five people responded (Table 9 ). There was a significant difference in the age and fixation range between 25 patients for external validation and 89 patients for internal validation. However, no other background information was significantly different. The main symptom improved from 56 ± 19 preoperatively to 76 ± 19 postoperatively with an effect size of 1.05. The collateral symptom worsened from 75 ± 23 preoperatively to 64 ± 24 postoperatively with an effect size of 0.48. In both domains, the effect size was not different from the effect size at the time of scale creation, indicating the robustness of the scale.
Discussion This study is the first to use factor analysis to create a disease-specific scale for ASD. The most important point in a scale is to be able to measure the construct it is trying to measure 10 , 31 . Factor analysis is a technique used to explore and validate constructs, and is often used to create scales. “Intelligence” and “health” are examples of constructs that cannot be observed or measured directly. However, it is considered that they can be measured indirectly through multiple behaviors and events related to the construct 31 . In the present study, factor analysis allowed us to detect two factors that constitute the construct of ASD. The first factor was named the main symptom because it reflected the patient’s main problems, such as appearance, pain, and housework activities. The second factor was named the collateral symptom, and was related to postoperative movement limitations such as putting on socks, picking up, and using the toilet. We considered that these two factors could measure the construct of ASD. The Cronbach’s alpha coefficients for each were 0.90 and 0.84, respectively, and had reliabilities that were acceptable for a clinically used measure. In this study, the scale scores of the main symptom and the collateral symptom were calculated by weighting them according to the factor score coefficients. Both the main symptom and the collateral symptom showed significant differences in preoperative and postoperative comparisons of the surgery groups, and the effect size was large. Comparing preoperative scores of the surgery group and the conservative group also showed significant differences, and the effect size was moderate. In addition, the main symptom was significantly correlated with satisfaction and the PCS. These results indicated that the created scale had adequate responsiveness and criterion-related validity. The items included in the factor analysis in this study were selected from various representative scales by physicians with extensive experience in ASD surgery and had content validity. The SRS-22 is a commonly used outcome for assessing ASD, but several problems were noted 32 . Faraj et al. reviewed the strengths, weaknesses, and gaps of current outcomes in measuring ASD outcomes 33 . According to their study, the most frequently used outcome was the ODI, with the SRS-22s. However, they stated that both the ODI and the SRS-22 had weaknesses in their use to assess ASD. The ODI is a low back pain-specific questionnaire and does not necessarily include the concept of deformity. Conversely, the SRS-22 was developed for AIS, which is less functionally impaired and, therefore, is less relevant for ASD, which seeks to restore pain and quality of life. Faraj et al. stated that there was an overlap between the two outcomes and the need to develop a core outcome set that is more specific to the assessment of ASD. Mannion et al. performed a factor analysis of the SRS-22 on ASD patients 34 . They found a poor fit for four questions on the SRS-22: Q3 (nervous person), Q14 (personal relationship), Q15 (financial difficulties), and Q17 (sick days). They recommended the deletion of these four questions. Zaina et al. compared the newly developed Italian spine youth quality of life (ISYQOL) with the SRS-22 using Rasch analysis 35 . According to this group, Q15 (financial difficulties) in the SRS-22 was a poor fit, and they recommended 21 items except for that one. By excluding this item, the revised SRS-22 showed construct validity comparable with the ISYQOL. Scheer et al. devised a patient generated index, a questionnaire that patients were asked to fill out freely 36 . The top 10 concerns of patients with ASD were walking, activities, posture, pain, sports, housework, relationships, gardening, sleeping, and traveling. The 29 items we selected almost covered these items. Of these items, about sports, some patients in this study indicated in their free-text sections that they did not engage in these activities. The term “sports” covers an extensive range, from light gymnastics and walking to running and swimming. We did not select Q26 sports because the factor loading was small and also because different people perceived this item differently. Housework activities, conversely, are important for patients with ASD. In particular, as ASD is more common in older women, it is essential to include kitchen activities in the assessment. A kitchen elbow sign, for example, is a skin abnormality that develops on the elbow when working in the kitchen, as the patient must rest her elbow on a table to maintain a standing position 36 . In the current study, the factor loadings for washing dishes and laundry were large. Kitchen elbow sign is especially likely to occur when washing dishes because both hands are used, and the patient cannot hold a cane or walker during the task. Large factor loading of these two items suggests that patients with ASD have kyphosis, making it difficult for them to maintain an intermediate or dorsiflexed position. Restriction of lumbar spine mobility after long fusion is a concern for both surgeons and patients 6 . Ishikawa et al. conducted a study about ADL for 36 long fusion patients 13 . They found that patients after long fusion performed better than preoperatively in activities such as sleeping supine, standing upright, vacuuming, doing laundry, and reaching for objects placed at heights. Conversely, strenuous activities such as shoveling snow worsened postoperatively. Overall surgical satisfaction was 70%. Their report suggests that long fusion surgery for ASD requires evaluating both positive and negative aspects. Hart et al. investigated functional limitations due to lumbar stiffness in 62 patients 5 . They reported that 91% of the patients were satisfied with the trade-off between postoperative improvement in back pain and associated restriction of motion. In the present study, 73% of the patients were satisfied with their surgery. Their study included 24 cases (40%) of one vertebral fusion and only 19 cases (31%) of five or more vertebral fusion. Our patients had five or more intervertebral fusions, with an average of 10 fused vertebrae. This difference in fixation levels may have influenced the difference in satisfaction. One of the advantages of our scoring system was that factor analysis divided the questions into two domains. The effect of surgery on ASD resulted in improved ADLs associated with improved pain and posture, but also movement limitations. Simply adding up these improvements and any worsening could result in a total score of plus or minus 0. By dividing this score into two domains, we could assess each symptom with each domain having the appropriate responsiveness. This represents two aspects of surgery for ASD and is a necessary component for improving treatment efficacy and explaining surgery to patients. Another strength of this study was that the subject patients had an average of 10 long fixed vertebral intervertebral spaces, and 78% underwent fusion from the pelvic to the thoracic spine. Previous studies have focused on short lumbar intervertebral fusion procedures. Our patients are a more suitable population to assess ADLs for long fusion, especially as including L5/S in the fusion range would result in greater limitation. There were some limitations in this study. The number of the patients was limited. Factor analysis was performed on 98 patients, slightly less than 100 patients. However, considering the two factors that were found, this could be considered sufficient. Because this study was conducted in one country, the results may not be generalizable to other countries. The burden of housework activities may differ between developed and developing countries. Reliability was assessed by content consistency, and a test–retest was not conducted in this study. The preoperative score was based on memory and there may have been recall bias. Patients with a longer follow-up period become more accustomed to their current symptoms and may underestimate the difference between their preoperative and current conditions. These issues should be addressed in future studies.
Conclusion We developed a disease-specific outcome for ASD using factor analysis. This analysis is the first scientifically validated measure that could simultaneously assess the benefits and limitations of ASD surgery. This tool can complement existing outcomes and will be useful for explaining surgery to patients and for future clinical trials.
Adult spinal deformity (ASD) is a complex condition that combines scoliosis, kyphosis, pain, and postoperative range of motion limitation. The lack of a scale that can successfully capture this complex condition is a clinical challenge. We aimed to develop a disease-specific scale for ASD. The study included 106 patients (mean age; 68 years, 89 women) with ASD. We selected 29 questions that could be useful in assessing ASD and asked the patients to answer them. The factor analysis found two factors: the main symptom and the collateral symptom. The main symptom consisted of 10 questions and assessed activity of daily living (ADL), pain, and appearance. The collateral symptom consisted of five questions to assess ADL due to range of motion limitation. Cronbach’s alpha was 0.90 and 0.84, respectively. The Spearman’s correlation coefficient between the change of main symptom and satisfaction was 0.48 (p < 0.001). The effect size of Cohen’s d for comparison between preoperative and postoperative scores was 1.09 in the main symptom and 0.65 in the collateral symptom. In conclusion, we have developed a validated disease-specific scale for ASD that can simultaneously evaluate the benefits and limitations of ASD surgery with enough responsiveness in clinical practice. Subject terms
Supplementary Information
Supplementary Information The online version contains supplementary material available at 10.1038/s41598-024-51783-4. Acknowledgements This work was supported by a Grant from JSPS KAKENHI No. JP21K20966. Author contributions T.F. prepared the manuscript. T.F. and Y.N. collected the data. S.T. provided a critical comment. T.K., Y.K., Y.U., M.F., T.M., Sh.O. provided research advices. M.I. and Se.O. supervised the entire project. All authors reviewed the paper. Data availability The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 14; 14:1286
oa_package/7e/00/PMC10787822.tar.gz
PMC10787823
38218984
Introduction Sugarcane juice is a nutritious and energetic green brownish beverage with low acidity (pH 4.8-5.5) and high water activity (Aw~0.99). The juice’s composition depends on the variety, cultivar, maturation stage, soil, climatic and agricultural conditions. Cane juice has a limited shelf life due to its rapid microbiological and enzymic deterioration 1 – 3 , and when processed in industrial plants is subjected to heat treatment in order to inactivate enzymes, spoilage and potentially pathogenic microorganisms. Thermal processing, depending on its intensity, may however damage the sensory, functional and nutritional juice’s quality 4 . During extraction, the juice is exposed to oxygen, a reactant for enzymic browning, catalyzed by polyphenol oxidase (PPO) and peroxidase (POD), so the enzymic inactivation guarantees a higher quality product 1 . Additionally, it is necessary to eliminate spoilage microorganisms, such as Leuconostoc mesenteroides , molds and yeasts. L. mesenteroids produces lactic acid and changes the juice’s viscosity during storage 1 . The industrialization of cane juice is rapidly growing in Brazil, and over the last few years more than 10 brands of processed cane juice have been launched. The preservation technologies frequently combine acidification, chemical preservatives, heat treatment and refrigeration, and vary among brands. Nevertheless, the sensory quality of the products available on the market is questionable. As an alternative to heat treatment, the technology that employs supercritical carbon dioxide (SC-CO 2 ) is a promising intervention to inactivate pathogens and spoilage enzymes and microorganisms. This non-thermal technique consists of exposing food or beverages to high pressure (beyond 74 bar), and the main advantage is the preservation of the food sensory attributes 1 . In contrast, conventional heat treatments may trigger the onset of nutritional and sensory losses. A supercritical fluid is defined as any substance maintained above its critical temperature and pressure. The critical temperature is the highest temperature at which a gas can be converted into a liquid by increasing pressure. Critical pressure is the highest pressure at which a liquid can be converted into a gas by increasing the temperature of the liquid 5 . In SC-CO 2 treatment, the food is exposed to pressurized CO 2 for a certain period of time. The supercritical fluid diffuses through the food, showing a microbicidal effect, whose intensity depends on the pressure and holding time 6 – 8 . The temperature and pressure that characterize the critical state/point of carbon dioxide are 31.1 °C and 73.8 bar 4 . Gómez-López et al. 9 reported the specific energy required by the pressure change technology (PCT) application. The energy consumption per unit mass of treated product has been estimated and compared to that required by conventional indirect thermal technology. The estimated specific energy consumption was respectively 162.6 kJ/kg for indirect thermal and 26.3 kJ/kg for PCT. On the other hand, completely non-thermal processes such as PCT do not involve any energy costs for product and equipment cooling. Therefore, the energy consumption involved in this technology is only due to product and inert gas pumping and compression. Vignali et al. 10 stated a comparison of typical specific working energy costs for thermal and non-thermal treatments in terms of KJ per kilograms of processed product. Non-thermal approaches seem to offer the most effective alternative in terms of nutrients and fresh-like characteristics preservation as well as working energy costs saving. Nevertheless, the SC-CO 2 treatment has not been considered in that study because the literature is scarce and the data available for microbial inactivation are very low in comparison to the other technologies. According to Bocker and Silva 11 , CO 2 stands out as the best cost-benefit among supercritical fluids since it has a low cost and is non-toxic. The critical conditions of CO 2 (73,8 Bar and 31 °C) are moderate compared to those of other fluids used in the supercritical state. These moderate conditions reduce the process energy expenditure and promote less damage to the nutritional properties of the food matrices. Many studies address the direct injection application of SC-CO 2 in the inactivation of enzymes and microorganisms in fruit juices 1 , 12 – 14 . The SC-CO 2 proved to be efficient in the inactivation of microorganisms and enzymes. Additionally, the low toxicity and cost are important advantages of CO 2 since it is naturally found in the atmosphere. The use of SC-CO 2 under mild conditions is a technique that, when used in juices, allows greater preservation of thermally unstable constituents, such as phenolic compounds, flavonoids and anthocyanins 15 . Nevertheless, no work targeting the stabilization of SC-CO 2- treated cane juice has been found. This study was primarily conducted to evaluate the combined effect of mild temperatures and SC-CO 2 on microorganisms and enzymes in cane juice.
Methods The cane juice was procured from a local vendor, in the city of Pirassununga/SP-Brazil. The freshly extracted juice was collected in a plastic bottle by the vendor, kept on ice in an isothermal container, and rapidly transported to the laboratory in the Food Engineering Department at the University of Sao Paulo. The juice’s sample was divided into two parts and collected in previously sterilized glass bottles with screw caps. One fraction was used as a control (unprocessed juice) and the other one was treated with SC-CO 2 . Figure 4 illustrates the juice processing. The treatment of cane juice with direct injection of SC-CO 2 was carried out in a 100 mL-reactor of a supercritical fluid system (Thar Technologies SFE-500, Pittsburgh/USA) available at the Laboratory of High Pressure Technology and Natural Products. The juice sample (100 mL) was transferred to the reactor and kept under the pre-set conditions (subsequently described). At the end of the treatment the sample was moved out through rapid depressurization to a sterilized glass flask. Table 6 points out the independent variables and their actual and coded levels tested according to Rodrigues and Iemma 29 . Pressures (P) in the range of 74 to 351 bar, temperatures (T) between 33 and 67 °C, and holding times (t) varying from 20 to 70 min were tested in a central composite rotatable design. Seventeen trials were performed. This study aimed at exploring a wide range of CO 2 pressure (above the critical one – 73.8 bar). Also the operational limits of the equipment available to conduct this study were considered in the range of the investigated parameters set. Because no study carried out with cane juice was found, no reference is herein mentioned. As for the temperature, mild values were targeted to preserve the original quality of the juice. To get an approximate statistical inference, three trials were conducted at the central point of the experimental space; they can provide valuable information on the behavior of the responses between the levels attributed to the factors, and demonstrate the repeatability of the process (Rodrigues and Iemma) 29 . The physicochemical, enzymic, microbiological analysis and instrumental determination of color parameters were carried out on raw and processed samples to evaluate the performance of multiple combinations of the processing’s parameters (pressure, temperature and holding time) and are as follows. All assays were performed in triplicate as described in Petrus and Simões 30 . The physicochemical tests were performed according to the Association of Official Analytical Chemists (AOAC, 2010) 31 . An Analyzer model 300 M was used to determine the pH. The soluble solids content (expressed in °Brix) was determined in a Reichert model AR 200 portable digital refractometer. Counts of mesophiles, molds and yeasts, lactic bacteria, and coliforms (at 45 °C) were conducted following the protocol described in the Compendium of Methods for the Microbiological Examination of Foods (Salfinger and Tortorello) 32 . The protocols adapted from ref. 33 were used to determine the polyphenol oxidase (PPO) and peroxidase (POD) activities. Five and half milliliters of 0.2 M phosphate buffer solution (pH 6.0) and 1.5 mL of 0.2 M catechol were added into a test tube and maintained at 25 °C for 10 min. Then 1.0 mL of the diluted sample in deionized water (1:10) was added. The tube was stirred for 15 s and returned to the water bath at 25 °C for 30 min. The absorbance was read in a spectrophotometer at 425 nm. The blank was prepared by diluting the sample in deionized water. Seven milliliters of 0.2 M phosphate buffer solution (pH 5.5) and 1.0 mL of the diluted sample (juice) in deionized water (1:10) were added to a test tube and maintained in a heat bath at 35 °C for 10 min. Then 1.5 mL of 0.05% guaiacol and 0.5 mL of 0.1% hydrogen peroxide were added. The tube was magnetically stirred for 15 s and returned to the bath at 35 °C for 15 min. Finally, the absorbance was read in a spectrophotometer at 470 nm. One (1) unit of enzyme activity (U) was defined as the amount of enzymic extract capable of increasing absorbance at 425 and 470 nm for PPO and POD, respectively, at rates of 0.001 units per minute. The color parameters (L*, a* and b*) of unprocessed and treated juice’s samples were measured in a Hunterlab Ultra-Scan colorimeter (Hunter Associates Laboratory, Model SN7877 Reston, VA/USA). The iluminant D65 and observation angle at 10° were set up. The parameters a* and b* were used to determine chroma (C) and hue angle (°hue) (Eq. 3 and 4 ). To compare the raw juice to the processed one, total color difference (TCD) was calculated by Eq. 5 . Also L*, a* and b* were inserted in the EasyRGB to convert them into color image (EasyRGB) 34 . L*: lightness (0 a 100). a*: coordinate red (+60) / green (−60). B: coordinate yellow (+60) / blue (−60). ΔL*: lightness variation Δa*: red/green variation Δb*: yellow/blue variation Data from the central composite rotatable design were first subjected to the analysis of effects to identify the variable(s) (P, T and t) that had significant effect on the responses (PPO, POD, coliforms, mesophiles, molds and yeasts, and lactic bacteria reduction, total color difference, pH and soluble solids variation), at 10% of significance. Due to the high variability of processes involving microorganisms and enzymes, p -values below 10 percent ( p ⩽ 0.1) are considered significant parameters, as stated by Rodrigues and Iemma (2015). The analysis of regression was performed for both 1st and 2nd (responses including axial points) orders. Then the mathematical model was re-parameterized considering only the statistically significant coefficients. The analysis of variance (ANOVA) was undertaken to evaluate if the model was statistically significant. If so, the response surface was generated. Statistical tests were performed using the software Protimiza Experimental Design ( http://experimental-design.protimiza.com.br ). Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Results and discussion Physicochemical tests Table 1 exhibits the pH and soluble solids values determined in raw and processed cane juice. The pH values ranged from 4.6 to 6.0 in the raw juice and between 4.4 and 6.3 for the processed one. The treatments reduced up to 0.4 units in the pH; however, the pH from trials 3 (295 bar/40 °C/30 min), 6 (213 bar/50 °C/45 min) and 16 (213 bar/50 °C/20 min) remained unchanged. The variety of cane, type of soil, fertilization, climatic conditions, degree of maturity, harvesting and extraction methods are important factors to be considered in the variation of juice’s pH. Bomdespacho et al. 16 evaluated different cultivars of raw cane juice and reported an average pH equivalent to 5.05. This data is close to the values found in most treatments performed in the present study, with the exception of trials 3, 9 and 12. Regarding the soluble solids content, variations between 18.5 and 25.3 °Brix were determined in the raw juice, and between 18.2 and 25.0 for the processed beverage. The variations (Δ) in this parameter caused by the treatment ranged between 0.0 and 0.4. With the exception of trials 15 and 16, there was a reduction in this parameter. These phenomena may be related to the variation of the treatments submitted, as well as to the batch used on the day of the respective trials. Bomdespacho et al. 16 reported an average of 21.2 °Brix in fresh juice extracted from different cultivars. This result is in the range obtained in this study. In all 17 trials, no meaningful variations (Δ ≤ 0.4) were observed between processed and raw juice. These findings are positive as they lead to the hypothesis that there was no significant difference between the pH values and soluble solids after the treatments applied. Microbiological assays Table 2 reports the microbial counts in raw and SC-CO 2 -treated cane juice as well as the log reduction achieved in each trial. The results exhibited in Table 2 show the potential of SC-CO 2 in the reduction of contaminants in raw cane juice. The reductions achieved by the different trinomials were 2.5 log for coliforms, 3.9 log for aerobic mesophiles, 2.1 log for lactic acid bacteria, and 4.1 log for molds and yeasts. The lactic bacteria counts in raw juice varied between 1.0 and 4.0 logCFU/mL. For the processed sample, counts ranged from <1.0 est to 3.5 logCFU/mL; comparison with data from other studies was not possible, once counts were carried out after cane fermentation, as reported by Silva et al. 17 , who performed counts after 3, 11 and 24 h of fermentation. The lactic bacteria contamination, such as Leuconostoc mesenteroides and some species of the Lactobacillus , can trigger the synthesis of dextrans (polysaccharides formed by glucose units) (Koblitz) 18 , forming gums in the juice, leading to its rejection. The discrepancy among counts within the same group of microorganisms in raw juice may be attributed to failures in the hygiene procedures of the raw material, utensils and/or equipment used in the extraction. This event is usual when it comes to street vending. According to Prati, Moretti and Cardello 19 , mesophilic counts above 6.0 logCFU/mL may be related to hygienic-sanitary deficiencies in the extraction and/or storage of cane juice. In this study both the raw and processed juice exhibited counts within the range 1.6–6.0 logCFU/mL. For molds and yeasts, counts were between 2.4 and 5.4 logCFU/mL; Jay 20 holds that values above 3.0 logCFU/mL can cause undesirable changes. The efficiency of SC-CO 2 treatment on microbial inactivation is associated with the modification of intracellular and extracellular pH, and also the length of time CO 2 diffuses into the cells. Therefore, the holding time of treatment greatly impacts the microbial inactivation rate 4 . Dhansu et al. 21 pasteurized cane juice at 65 °C/25 min, and stored it under refrigeration, achieving a shelf life of 60 days. Oliveira et al. 1 pasteurized the juice at 70 °C/25 min; the lactic acid bacteria counts in raw and processed cane juice were (5.9 and 1.3) logCFU/mL respectively, reaching 4.6 log reduction. The molds and yeasts’ counts in raw and processed juice were (6.1 and 1.7) logCFU/mL respectively. Gomes et al. 22 optimized the time x temperature binomial used in the pasteurization of whole cane juice; temperatures and holding times ranging between 78 and 92 °C, and from 16 to 44 s, were tested. Regarding the reduction of microorganisms, the treatment at 90 °C/40 s was the most efficient, achieving 4.6 log reductions for mesophiles. For molds and yeasts, 3.2 log reductions were reached. Hart et al. 23 reported the application of SC-CO 2 in the inactivation of spores in foods, highlighting how this technique can be more efficient in preserving nutritional and sensory characteristics as compared to high hydrostatic pressure techniques and thermal methods at high temperatures. The action of SC-CO 2 occurs through disruption of the cell wall, coating, cortex and membranes, and degradation of proteins. More in-depth studies on larger scales are needed to disseminate this technology in the processing of fruits and juices. Eggleston 24 reported the microbiological, enzymic and chemical deterioration (acid degradation) of sucrose in cane juice. The findings indicated that the growth of microorganisms is relevant for sucrose degradation. After 14 h, the largest contribution was microbiological, accounting for 93% of losses, while enzymatic degradation contributed with 5.7% of losses and chemical degradation with 1.3%. Enzymic tests The endogenous enzymes activity (polyphenol oxidase and peroxidase) as well as the percentages of reduction achieved by different treatments are shown in Table 3 . The results exhibited in Table 3 suggest the potential of SC-CO 2 combined with mild temperatures to inactivate the endogenous enzymes that are responsible for the degradation of the color, flavor and the nutritional values of cane juice. The percentages of PPO (3.3–64.5%) and POD (0.0–40.9%) reduction varied widely. The trinomial applied in trial 5 (295 bar/60 °C/30 min) reached the greatest PPO inactivation (64.5%), suggesting that temperature had a more significant effect on the percentage of reduction; however, this hypothesis will be confirmed in the light of the statistical analysis of the effects of the variables studied. POD exhibited greater resistance to the treatments in most trials. The trinomial applied in trial 15 (213 bar/67 °C/45 min) reached the highest percentage of inactivation (40.9%). Marszałek et al. 12 studied the effect of supercritical carbon dioxide on PPO and POD in mushroom and radish; surprisingly, PPO was more resistant to temperature and pressure than POD. In this study, similar result was observed in trials 7, 9, 10 and 15, i.e., PPO was more resistant to SC-CO 2 . In most trials, however, the percentage of POD reduction in the juice was lower than PPO. This finding suggests that the type of food matrix also influences the impact of the technology that is applied. The food matrix can interfere with the intermolecular bonds of the two enzymes depending on the amount of water in the medium, since the impact of pressure on intra and intermolecular interactions can also be correlated with the ability of the functional groups of the enzymes to interact with water (Marszałek et al.) 25 . The deactivation of enzymes exposed to high temperatures and prolonged times is explained by changes in the tertiary and secondary structures of the protein. The thermal stability of enzymes depends on a number of factors such as source, species, nature of the food matrix (Iqbal et al.) 26 . Color parameters The color parameters instrumentally measured in cane juice are presented in Table 4 . The parameter L*, which represents lightness, varied widely for raw (32.3–72.9) and processed (32.5–73.1) juice. Most treatments positively influenced the lightness of the juice samples, which is most likely related to enzymic inactivation (Table 3 ). This result could favor the juice’s sensory acceptance, assuming the consumers prefer a lighter drink. Similarly, there was a great variation in the a* parameter for fresh (1–16.1) and processed (4.0–14.7) juice. The b* parameter also varied considerably for raw (26.0–46.2) and processed (24.8–50.7) samples. Regarding the chroma parameter (C*), significant variations were also observed for raw (28.1–46.4) and processed (25.4–52.8) juice. Chroma correlates to saturation, characterizing the sample’s color as “vivid” or opaque (dull). This attribute is independent of lightness and °hue. Saturation ranges from purple-red to green, and increases from the center (0) to the edge of the color wheel. Oliveira et al. 1 stated that low C* values represent gray, and values close to 60 represent vivid colors; they found C* values close to 9 (more neutral color) for raw cane juice, in contrast to the present study. Meerod 27 studied different cultivars of raw material, which showed divergent colors at various levels. Following the same behavior as the previous parameters, the hue angle also showed great variation for raw (68.1–88.5) and processed (71.1–84.5°) juices. Hue, measured in degrees, classifies color (green, yellow, blue, etc.). The ranges determined in this study are positioned in the first quadrant of the color circle, and can be classified between yellow-red and yellow. The wide variation ranges in the color parameters can be explained by the variability inherent to the raw material; juice samples extracted on different days, from different stalks, were used during the course of this research. Figure 1 illustrates the total color difference (TCD) between raw and processed cane juice. The highest (12.3) and lowest (2.0) values of total color difference (∆E*) were determined for juice treated at 295 bar/ 60 °C/ 60 min and (130 bar/ 40 °C/ 60 min and 213 bar/ 33 °C/45 min), respectively. Bernard et al. 28 states that ∆E* values less than 3 cannot be easily detected by the human eye, and values greater than 12 represent different color “spaces”. Therefore, of the 17 tests carried out, only six preserved the original color of the juice, in terms of its sensory perception. Statistical analysis Because the log reduction in coliforms and lactic bacteria could not be calculated in some trials (Table 3 ), these responses were not subjected to the statistical analysis. Figure 2 demonstrates the Pareto diagrams, built to investigate which parameters/variables (pressure/P/x 1 , temperature/T/x 2 , holding time/t/x 3 ) were significant ( p ≤ 0.1) in the studied responses. The terms that were not statistically significant were incorporated into the lack-of-fit to calculate the coefficient of determination (R 2 ). As for mesophiles, molds and yeasts reduction, and soluble solids variation, none of variables or their interactions were significant. In terms of polyphenol oxidase (PPO) reduction, only t (x 3 ) was significant; however, the parameters T (x 2 ), t (x 3 ), and the interaction between them (x 2 .x 3 ) played a significant effect on the peroxidase (POD) reduction. In regards to pH variation, P (x 1 ) and the interaction between T and t (x 2 .x 3 ) were significant. Finally, P, T, t, and the interaction between T and t were significant in the total color difference. Only the significant variables were encompassed into the mathematical model, whose statistical significance was evaluated through analysis of variance (ANOVA). Table 5 exhibits the ANOVA carried out for the models (1st and 2nd orders) generated for the responses POD reduction and total color difference (TDC); the models for other responses were not statistically significant ( p > 0.1). The coded predicted models obtained for the aforementioned responses are represented by Eqs. 1 and 2 . Y 1 – POD reduction (%) x 2 – Temperature (T) x 3 – holding time (t) Y 2 – Total color difference x 1 – Pressure (P) x 2 – Temperature (T) x 3 – holding time (t) For practical purposes, it is desirable that the fitted model be as simple as possible and contain the smallest possible number of parameters without giving up the quality assured in the careful selection of the experimental design. The models herein presented were re-parameterized/reduced because the parameters with little or no influence on the outcome of the final fit were excluded. Regarding the response POD reduction , Table 5 shows that the 1 st order model (R 2 = 0.86) better fitted to experimental data than the 2 nd order model (R 2 = 0.47). For both orders, F calc was greater than F tab . Similarly, as for the total color difference (TCD), the 1st order mathematical model (R 2 = 0.90) best fitted to experimental data. The coded Eqs. 1 and 2 can be used to predict the percentage of POD reduction and the TCD that can be achieved in cane juice processed under the same conditions of this study. The coded model is that whose regression coefficients are obtained from the matrix of coded variables (−α, −1, 0, +1, +α). Given this, to obtain a predicted value from the model one must replace the values in the coded equation. In contrast, if using real values for the variables in the model, the predicted value may be incorrect and even absurd. Of particular relevance is the claim that the first order mathematical models hereby presented (Eqs. 1 and 2 ) are only valid in a range of pressure from 130 to 295 bar, temperature from 40 to 60 min, and holding time between 30 and 60 min (Table 6 ). Figure 3 depicts the response surfaces and contour curves that represent Eqs. 1 and 2 . By analyzing the surface for POD reduction, one can identify the existence of an optimal range for the temperature (57–60 °C) and holding time (56–60 min), regardless the pressure (in the range 130–295 bar). As for TDC, the ranges 130–150 bar, 40–43 °C and 30–35 min, within which the color difference between raw and processed juice is minimal, represent the optimal conditions in this experiment. This is of much greater interest than a simple point value, because it provides information about the “robustness” of the process, and most notably, it is the variation in pressure, temperature and holding time that may be permitted around optimal values which still maintains the process under optimized conditions. This finding is fundamental for the control engineer to define and maintain the pressure, temperature and time sensors and controller levels. This directly affects viability and process implementation (Rodrigues and Iemma) 29 . A fact worth highlighting is that studies addressing the use of SC-CO 2 in cane juice processing have not been found. In this way, data comparison could not be made. The combination of supercritical carbon dioxide and mild temperatures exhibited a meaningful effect on microorganism’s reduction in sugarcane juice, under the conditions of this study. Endogenous enzymes that deteriorate the juice’s quality were partially inactivated. None of variables (pressure/P, temperature/T, holding time/t) or their interactions were significant in mesophiles, molds and yeasts reduction, or soluble solids variation. In terms of polyphenol oxidase (PPO) reduction, only t was significant; however, T, t and the interaction between them played a significant effect on the peroxidase (POD) reduction. In regards to pH variation, P and the interaction between T and t were significant. Finally, P, T, t, and the interaction between T and t were significant in the total color difference. The optimal parameters (P, T and t) determined in this study varied for different responses. The combination of mild temperatures and SC-CO 2 can be potentially used for cane juice preservation.
Results and discussion Physicochemical tests Table 1 exhibits the pH and soluble solids values determined in raw and processed cane juice. The pH values ranged from 4.6 to 6.0 in the raw juice and between 4.4 and 6.3 for the processed one. The treatments reduced up to 0.4 units in the pH; however, the pH from trials 3 (295 bar/40 °C/30 min), 6 (213 bar/50 °C/45 min) and 16 (213 bar/50 °C/20 min) remained unchanged. The variety of cane, type of soil, fertilization, climatic conditions, degree of maturity, harvesting and extraction methods are important factors to be considered in the variation of juice’s pH. Bomdespacho et al. 16 evaluated different cultivars of raw cane juice and reported an average pH equivalent to 5.05. This data is close to the values found in most treatments performed in the present study, with the exception of trials 3, 9 and 12. Regarding the soluble solids content, variations between 18.5 and 25.3 °Brix were determined in the raw juice, and between 18.2 and 25.0 for the processed beverage. The variations (Δ) in this parameter caused by the treatment ranged between 0.0 and 0.4. With the exception of trials 15 and 16, there was a reduction in this parameter. These phenomena may be related to the variation of the treatments submitted, as well as to the batch used on the day of the respective trials. Bomdespacho et al. 16 reported an average of 21.2 °Brix in fresh juice extracted from different cultivars. This result is in the range obtained in this study. In all 17 trials, no meaningful variations (Δ ≤ 0.4) were observed between processed and raw juice. These findings are positive as they lead to the hypothesis that there was no significant difference between the pH values and soluble solids after the treatments applied. Microbiological assays Table 2 reports the microbial counts in raw and SC-CO 2 -treated cane juice as well as the log reduction achieved in each trial. The results exhibited in Table 2 show the potential of SC-CO 2 in the reduction of contaminants in raw cane juice. The reductions achieved by the different trinomials were 2.5 log for coliforms, 3.9 log for aerobic mesophiles, 2.1 log for lactic acid bacteria, and 4.1 log for molds and yeasts. The lactic bacteria counts in raw juice varied between 1.0 and 4.0 logCFU/mL. For the processed sample, counts ranged from <1.0 est to 3.5 logCFU/mL; comparison with data from other studies was not possible, once counts were carried out after cane fermentation, as reported by Silva et al. 17 , who performed counts after 3, 11 and 24 h of fermentation. The lactic bacteria contamination, such as Leuconostoc mesenteroides and some species of the Lactobacillus , can trigger the synthesis of dextrans (polysaccharides formed by glucose units) (Koblitz) 18 , forming gums in the juice, leading to its rejection. The discrepancy among counts within the same group of microorganisms in raw juice may be attributed to failures in the hygiene procedures of the raw material, utensils and/or equipment used in the extraction. This event is usual when it comes to street vending. According to Prati, Moretti and Cardello 19 , mesophilic counts above 6.0 logCFU/mL may be related to hygienic-sanitary deficiencies in the extraction and/or storage of cane juice. In this study both the raw and processed juice exhibited counts within the range 1.6–6.0 logCFU/mL. For molds and yeasts, counts were between 2.4 and 5.4 logCFU/mL; Jay 20 holds that values above 3.0 logCFU/mL can cause undesirable changes. The efficiency of SC-CO 2 treatment on microbial inactivation is associated with the modification of intracellular and extracellular pH, and also the length of time CO 2 diffuses into the cells. Therefore, the holding time of treatment greatly impacts the microbial inactivation rate 4 . Dhansu et al. 21 pasteurized cane juice at 65 °C/25 min, and stored it under refrigeration, achieving a shelf life of 60 days. Oliveira et al. 1 pasteurized the juice at 70 °C/25 min; the lactic acid bacteria counts in raw and processed cane juice were (5.9 and 1.3) logCFU/mL respectively, reaching 4.6 log reduction. The molds and yeasts’ counts in raw and processed juice were (6.1 and 1.7) logCFU/mL respectively. Gomes et al. 22 optimized the time x temperature binomial used in the pasteurization of whole cane juice; temperatures and holding times ranging between 78 and 92 °C, and from 16 to 44 s, were tested. Regarding the reduction of microorganisms, the treatment at 90 °C/40 s was the most efficient, achieving 4.6 log reductions for mesophiles. For molds and yeasts, 3.2 log reductions were reached. Hart et al. 23 reported the application of SC-CO 2 in the inactivation of spores in foods, highlighting how this technique can be more efficient in preserving nutritional and sensory characteristics as compared to high hydrostatic pressure techniques and thermal methods at high temperatures. The action of SC-CO 2 occurs through disruption of the cell wall, coating, cortex and membranes, and degradation of proteins. More in-depth studies on larger scales are needed to disseminate this technology in the processing of fruits and juices. Eggleston 24 reported the microbiological, enzymic and chemical deterioration (acid degradation) of sucrose in cane juice. The findings indicated that the growth of microorganisms is relevant for sucrose degradation. After 14 h, the largest contribution was microbiological, accounting for 93% of losses, while enzymatic degradation contributed with 5.7% of losses and chemical degradation with 1.3%. Enzymic tests The endogenous enzymes activity (polyphenol oxidase and peroxidase) as well as the percentages of reduction achieved by different treatments are shown in Table 3 . The results exhibited in Table 3 suggest the potential of SC-CO 2 combined with mild temperatures to inactivate the endogenous enzymes that are responsible for the degradation of the color, flavor and the nutritional values of cane juice. The percentages of PPO (3.3–64.5%) and POD (0.0–40.9%) reduction varied widely. The trinomial applied in trial 5 (295 bar/60 °C/30 min) reached the greatest PPO inactivation (64.5%), suggesting that temperature had a more significant effect on the percentage of reduction; however, this hypothesis will be confirmed in the light of the statistical analysis of the effects of the variables studied. POD exhibited greater resistance to the treatments in most trials. The trinomial applied in trial 15 (213 bar/67 °C/45 min) reached the highest percentage of inactivation (40.9%). Marszałek et al. 12 studied the effect of supercritical carbon dioxide on PPO and POD in mushroom and radish; surprisingly, PPO was more resistant to temperature and pressure than POD. In this study, similar result was observed in trials 7, 9, 10 and 15, i.e., PPO was more resistant to SC-CO 2 . In most trials, however, the percentage of POD reduction in the juice was lower than PPO. This finding suggests that the type of food matrix also influences the impact of the technology that is applied. The food matrix can interfere with the intermolecular bonds of the two enzymes depending on the amount of water in the medium, since the impact of pressure on intra and intermolecular interactions can also be correlated with the ability of the functional groups of the enzymes to interact with water (Marszałek et al.) 25 . The deactivation of enzymes exposed to high temperatures and prolonged times is explained by changes in the tertiary and secondary structures of the protein. The thermal stability of enzymes depends on a number of factors such as source, species, nature of the food matrix (Iqbal et al.) 26 . Color parameters The color parameters instrumentally measured in cane juice are presented in Table 4 . The parameter L*, which represents lightness, varied widely for raw (32.3–72.9) and processed (32.5–73.1) juice. Most treatments positively influenced the lightness of the juice samples, which is most likely related to enzymic inactivation (Table 3 ). This result could favor the juice’s sensory acceptance, assuming the consumers prefer a lighter drink. Similarly, there was a great variation in the a* parameter for fresh (1–16.1) and processed (4.0–14.7) juice. The b* parameter also varied considerably for raw (26.0–46.2) and processed (24.8–50.7) samples. Regarding the chroma parameter (C*), significant variations were also observed for raw (28.1–46.4) and processed (25.4–52.8) juice. Chroma correlates to saturation, characterizing the sample’s color as “vivid” or opaque (dull). This attribute is independent of lightness and °hue. Saturation ranges from purple-red to green, and increases from the center (0) to the edge of the color wheel. Oliveira et al. 1 stated that low C* values represent gray, and values close to 60 represent vivid colors; they found C* values close to 9 (more neutral color) for raw cane juice, in contrast to the present study. Meerod 27 studied different cultivars of raw material, which showed divergent colors at various levels. Following the same behavior as the previous parameters, the hue angle also showed great variation for raw (68.1–88.5) and processed (71.1–84.5°) juices. Hue, measured in degrees, classifies color (green, yellow, blue, etc.). The ranges determined in this study are positioned in the first quadrant of the color circle, and can be classified between yellow-red and yellow. The wide variation ranges in the color parameters can be explained by the variability inherent to the raw material; juice samples extracted on different days, from different stalks, were used during the course of this research. Figure 1 illustrates the total color difference (TCD) between raw and processed cane juice. The highest (12.3) and lowest (2.0) values of total color difference (∆E*) were determined for juice treated at 295 bar/ 60 °C/ 60 min and (130 bar/ 40 °C/ 60 min and 213 bar/ 33 °C/45 min), respectively. Bernard et al. 28 states that ∆E* values less than 3 cannot be easily detected by the human eye, and values greater than 12 represent different color “spaces”. Therefore, of the 17 tests carried out, only six preserved the original color of the juice, in terms of its sensory perception. Statistical analysis Because the log reduction in coliforms and lactic bacteria could not be calculated in some trials (Table 3 ), these responses were not subjected to the statistical analysis. Figure 2 demonstrates the Pareto diagrams, built to investigate which parameters/variables (pressure/P/x 1 , temperature/T/x 2 , holding time/t/x 3 ) were significant ( p ≤ 0.1) in the studied responses. The terms that were not statistically significant were incorporated into the lack-of-fit to calculate the coefficient of determination (R 2 ). As for mesophiles, molds and yeasts reduction, and soluble solids variation, none of variables or their interactions were significant. In terms of polyphenol oxidase (PPO) reduction, only t (x 3 ) was significant; however, the parameters T (x 2 ), t (x 3 ), and the interaction between them (x 2 .x 3 ) played a significant effect on the peroxidase (POD) reduction. In regards to pH variation, P (x 1 ) and the interaction between T and t (x 2 .x 3 ) were significant. Finally, P, T, t, and the interaction between T and t were significant in the total color difference. Only the significant variables were encompassed into the mathematical model, whose statistical significance was evaluated through analysis of variance (ANOVA). Table 5 exhibits the ANOVA carried out for the models (1st and 2nd orders) generated for the responses POD reduction and total color difference (TDC); the models for other responses were not statistically significant ( p > 0.1). The coded predicted models obtained for the aforementioned responses are represented by Eqs. 1 and 2 . Y 1 – POD reduction (%) x 2 – Temperature (T) x 3 – holding time (t) Y 2 – Total color difference x 1 – Pressure (P) x 2 – Temperature (T) x 3 – holding time (t) For practical purposes, it is desirable that the fitted model be as simple as possible and contain the smallest possible number of parameters without giving up the quality assured in the careful selection of the experimental design. The models herein presented were re-parameterized/reduced because the parameters with little or no influence on the outcome of the final fit were excluded. Regarding the response POD reduction , Table 5 shows that the 1 st order model (R 2 = 0.86) better fitted to experimental data than the 2 nd order model (R 2 = 0.47). For both orders, F calc was greater than F tab . Similarly, as for the total color difference (TCD), the 1st order mathematical model (R 2 = 0.90) best fitted to experimental data. The coded Eqs. 1 and 2 can be used to predict the percentage of POD reduction and the TCD that can be achieved in cane juice processed under the same conditions of this study. The coded model is that whose regression coefficients are obtained from the matrix of coded variables (−α, −1, 0, +1, +α). Given this, to obtain a predicted value from the model one must replace the values in the coded equation. In contrast, if using real values for the variables in the model, the predicted value may be incorrect and even absurd. Of particular relevance is the claim that the first order mathematical models hereby presented (Eqs. 1 and 2 ) are only valid in a range of pressure from 130 to 295 bar, temperature from 40 to 60 min, and holding time between 30 and 60 min (Table 6 ). Figure 3 depicts the response surfaces and contour curves that represent Eqs. 1 and 2 . By analyzing the surface for POD reduction, one can identify the existence of an optimal range for the temperature (57–60 °C) and holding time (56–60 min), regardless the pressure (in the range 130–295 bar). As for TDC, the ranges 130–150 bar, 40–43 °C and 30–35 min, within which the color difference between raw and processed juice is minimal, represent the optimal conditions in this experiment. This is of much greater interest than a simple point value, because it provides information about the “robustness” of the process, and most notably, it is the variation in pressure, temperature and holding time that may be permitted around optimal values which still maintains the process under optimized conditions. This finding is fundamental for the control engineer to define and maintain the pressure, temperature and time sensors and controller levels. This directly affects viability and process implementation (Rodrigues and Iemma) 29 . A fact worth highlighting is that studies addressing the use of SC-CO 2 in cane juice processing have not been found. In this way, data comparison could not be made. The combination of supercritical carbon dioxide and mild temperatures exhibited a meaningful effect on microorganism’s reduction in sugarcane juice, under the conditions of this study. Endogenous enzymes that deteriorate the juice’s quality were partially inactivated. None of variables (pressure/P, temperature/T, holding time/t) or their interactions were significant in mesophiles, molds and yeasts reduction, or soluble solids variation. In terms of polyphenol oxidase (PPO) reduction, only t was significant; however, T, t and the interaction between them played a significant effect on the peroxidase (POD) reduction. In regards to pH variation, P and the interaction between T and t were significant. Finally, P, T, t, and the interaction between T and t were significant in the total color difference. The optimal parameters (P, T and t) determined in this study varied for different responses. The combination of mild temperatures and SC-CO 2 can be potentially used for cane juice preservation.
Sugarcane juice is a nutritious and energetic drink. For its processing, the use of supercritical carbon dioxide (SC-CO 2 ) technology as an intervention potentially capable of rendering a high quality product can be considered. This study evaluated the combined effect of SC-CO 2 and mild temperatures, primarily aiming for the reduction of endogenous microorganisms and enzymes in sugarcane juice (pH~5.5). Pressures (P) ranging from 74 to 351 bar, temperatures (T) between 33 and 67 °C, and holding times (t) between 20 and 70 min were tested in a central composite rotational design. Seventeen trials were performed, comprising three replicates at the central points. Counts of aerobic mesophiles, molds and yeasts, lactic acid bacteria and coliforms at 45 °C, determination of polyphenol oxidase (PPO) and peroxidase (POD) activities, and measurement of color parameters in freshly extracted and processed juice’s samples were carried out. The pH of fresh and processed juice varied between 4.6 and 6.0, and between 4.6 and 6.3, respectively. The number of decimal reductions achieved in mesophiles, molds and yeasts, lactic acid bacteria and coliforms varied between 0.1 and 3.9, 2.1 and 4.1, 0.0 and 2.1, and 0.3 to 2.5, respectively. The percentages of PPO reduction ranged from 3.51% to 64.18%. Regarding the POD, reductions between 0.27% and 41.42% were obtained. Color variations between fresh and processed samples varied between 2.0 and 12.3. As for mesophiles, molds and yeasts reduction, and soluble solids variation, none of the variables or their interactions were significant. In terms of polyphenol oxidase (PPO) reduction, only t was significant; however, T, t, and the interaction between them significantly affected the peroxidase (POD) reduction. In regards to pH variation, P, and the interaction between T and t were significant. P, T, t, and the interaction between T and t played a significant effect on color. The combination of mild temperatures and SC-CO 2 can be potentially used for cane juice preservation. Subject terms
Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s41538-023-00242-x. Acknowledgements Grateful acknowledgment is made for the grant (2020/08011-5) provided by the Sao Paulo Research Foundation/Brazil (Fundação de Amparo à Pesquisa do Estado de São Paulo), and support offered by the Laboratory of High Pressure Technology and Natural Products. Author contributions F.C.P. performed the assays and reviewed the manuscript. T.C.K.M. performed the assays and reviewed the manuscript. G.C.D. reviewed the manuscript. A.L.dO. helped to design the study and reviewed the manuscript. R.P. designed the study and wrote the manuscript. Data availability The datasets generated during and/or analyzed during the current study are available from the corresponding author upon reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
NPJ Sci Food. 2024 Jan 13; 8:6
oa_package/9f/0e/PMC10787823.tar.gz
PMC10787824
38218946
Introduction Proton exchange membrane fuel cells (PEMFCs) have shown great potential as power sources for electric vehicle applications owing to their high energy conversion efficiency and environmental benign characteristics 1 – 4 . However, the widespread uptake of such PEMFCs in automotive vehicles is largely impeded at present because a high loading amount of precious Pt on the cathode (~0.4 mg cm −2 ) is needed to promote the sluggish kinetics of oxygen reduction reaction (ORR) and compensate the insufficient catalyst durability 5 – 8 . Alloying Pt with a transition metal, such as Fe, Co, Ni, and Cu, has been identified as an effective way to boost the catalytic activity toward ORR through the advantageous ligand and strain effects 9 – 14 , which has thus been regarded as the most promising solution to reduce the excessive Pt usage. As a convincing example, the spherical PtCo alloy nanocatalysts are adopted in the Toyota Mirai fuel cell vehicle at present 4 . Unfortunately, the long-term durability of these active PtM (M = transition metals) alloy catalysts still remains a serious problem, which can be generally ascribed to the rapid leaching of the reactive transition metal under detrimentally corrosive ORR conditions and “start-up/shutdown” process (the applied potential is generally over 1.0 V) 15 – 17 . Under this context, how to stabilize the active PtM alloy catalyst for long-term fuel cell operation has become the most concerned research topic for the industrial application of PEMFCs. Doping a specific element, with notable examples of Mo, Au, and Rh, into active PtM alloy catalysts and metal oxide support has been discovered as a straightforward and potential strategy to enhance the catalytic durability of PtM alloy catalysts for ORR 18 – 25 . Mechanistically, the prior studies suggested that these dopants may stabilize the alloy catalysts via suppressing the surface Pt migration and/or mitigating the outward diffusion of reactive transition metal 18 . However, it needs to be pointed out that these two main reasons (i.e., surface Pt migration and outward diffusion of reactive transition metal) are essentially entangled 18 , 26 – 28 , making great difficulty to gain a clear mechanistic understanding. Moreover, those arguments on the stabilization mechanism of dopants were mostly established via theoretical simulations, probably because of the challenges in constructing the model catalysts, more or less leading to the discrepancy with a realistic situation. As a result, the atomistic understanding of the role of dopants that is concluded from a comprehensive combining of convincing experiments and theoretical computations is still lacking presently, which indeed greatly limits the rational design of active and stable PtM alloy catalysts for practical fuel cells. Because Mo and Au elements are resistant to corrosion under acidic ORR, these elements offer ideal cases for the stability study. As a result, we systematically investigate the roles of Mo and Au dopants in stabilizing PtNi alloy catalysts based on the well-defined PtNi-based nanowires (NWs) model catalysts. Combining ex-situ experimental characterizations and density of functional theory (DFT) calculations, the distinct roles of Mo and Au in stabilizing PtNi NWs catalysts are identified. We find that the Mo dopant plays a major role in suppressing the outward diffusion of Ni atoms and the Au dopant mainly stabilizes the surface Pt overlayer. Based on this atomistic understanding, PtNiMoAu NWs are rationally designed and constructed by integrating the different functions of Mo and Au into PtNi NWs. As expected, the as-synthesized PtNiMoAu NWs catalysts present an unprecedented activity and stability toward ORR, with a 16.2% loss in its mass activity (MA) after 80,000 (80 K) cycles of durability test. When assembling the PtNiMoAu NWs catalysts into the fuel cell cathode, a high MA retention of 77.4% (H 2 -O 2 , 0.9 V iR-free ) and a low voltage loss of 25 mV (H 2 -Air, 0.8 A cm −2 ) after 30 K cycles of durability test are output, proving the highly durable fuel cell performance.
Methods Chemicals Platinum (II) acetylacetonate (Pt(acac) 2 , 97%), nickel (II) acetylacetonate (Ni(acac) 2 , 97%), molybdenum (III) acetylacetonate (Mo(acac) 3 , 97%), gold (III) chloride trihydrate (HAuCl 4 ·3H 2 O, 99%), cetyltrimethylammonium bromide (CTAB, 99%), tungsten hexacarbonyl (W(CO) 6 , 97%), oleylamine (OAm, 70%) and Nafion (5 wt%) were purchased from Sigma-Aldrich. Ethanol (CH 3 CH 2 OH, 99%) and perchloric acid (HClO 4 , 70–72%) were purchased from Sinopharm Chemical Reagent Co. Ltd. (Shanghai, China). The water (18.2 MΩ/cm) was freshly prepared through an ultra-pure purification system (Master-515Q, HHitech). All the chemicals were used without further purification. Synthesis of ultrathin Pt-based nanowires In a typical synthesis of ultrathin PtNiMoAu NWs, Pt(acac) 2 (20 mg), Ni(acac) 2 (10 mg), Mo(acac) 3 (8 mg), HAuCl 4 ·4H 2 O (2.7 mg), CTAB (75 mg), and 4 mL OAm were added into a 30 mL glass vial. After sonicated for 30 min, W(CO) 6 (20 mg) was added into the pre-dispersed solution and capped. The resulting homogeneous mixture was then heated and kept at 175 °C for 2 h in an oil bath. Finally, the products were collected by centrifugation at 13000 × g and cleaned four times with a hexane/ethanol mixture (v/v = 2/1), then dried under vacuum. The preparation of other ultrathin Pt-based nanostructures was similar to that of PtNiMoAu NWs except that Mo(acac) 3 or HAuCl 4 ·4H 2 O were taken out and the reaction temperature was changed. The detailed synthetic parameters for other Pt-based NWs have been listed in Supplementary Table 1 . Characterization techniques XRD patterns were collected to analyze the crystal structures of NWs by X-ray diffractometer (Rigaku Miniflex-600) with Cu K α radiation ( λ = 0.15406 nm, 40 kV). TEM images were carried out on a JEOL 2100-Plus operating at 120 kV with the samples deposited on carbon-coated copper TEM grids. HAADF-STEM images and EDS line-scan file were taken on a JEOL ARM-200F microscope with a spherical aberration corrector operating at 200 kV. Elemental analysis of ultrathin NWs was quantitatively determined by ICP-MS with a SPECTRO BLUE SOP. The XPS spectra were collected using an Escalab 250Xi equipped with an Al Ka (1486.6 eV) excitation source. All the spectra collected were corrected using a Shirley background. All XAFS data were collected at BL14B station of Shanghai Synchrotron Radiation Facility (SSRF), China. The storage ring of SSRF operates at 3.5 GeV with a maximum current of 210 mA. The electrochemical in situ XAFS measurements were carried out in a custom-fabricated three-electrode system with a 1.4 × 0.7 cm 2 carbon cloth loaded with PtNi and PtNiMoAu NWs catalysts as the working electrode and 0.1 M continuously O 2 -saturated HClO 4 solution as the electrolyte. For the in situ XAFS spectral data acquisition of Pt L 3 -edge (11564 eV), we calibrated the positions of absorption edges (E 0 ) by using Pt foil and Fe foil standard samples, respectively, and all spectra were collected in the same beam time by fluorescence mode to ensure comparability. Preparation of working electrode For different NWs/C catalysts, 4 mg of the prepared NWs was added to a chloroform solution (8 mL) and sonicated for 1 h. The above-dispersed solution was dropwise added to an ethanol solution containing 16 mg of carbon support (Vulcan XC-72) under vigorous stirring for 30 min. After centrifugation and washing twice with hexanes by centrifugation, the NWs/C catalysts were re-dispersed in acetic acid and then heated at 70 °C for 12 h to remove the surfactants on the surface of NWs. A certain amount of as-obtained catalyst was mixed with 0.5 mL isopropanol, 0.495 mL ethanol, and 0.005 mL Nafion (5 wt%) and sonicated for 1 h to form the homogenously mixed catalyst ink solution. For the commercial Pt/C catalysts (20 wt% loading on Vulcan XC-72 carbon support, Johnson Matthey), the ink solution (2 mg/mL) was prepared and sonicated for 1 h. Finally, prepared catalyst ink was dropped onto the glassy carbon rotating disk electrode (GC, RDE with geometric area of 0.196 cm 2 ) with the loading amount of Pt at 2.0 μg, 2.1 μg, 2.2 μg, 2.0 μg, 2.3 μg, and 4.0 μg for PtNiMoAu NWs/C, PtNiAu NWs/C, PtNiMo NWs/C, PtNi NWs/C, Pt NWs/C, and commercial Pt/C catalysts, respectively (based on ICP-MS). Electrochemical testing Electrochemical tests were conducted using a three-electrode cell on a CHI760e electrochemical workstation (Chenhua Instrument, China). A glassy carbon Rotating Disk Electrode (RDE, diameter: 5 mm) was used as the working electrode, the Ag/AgCl (3 M KCl) electrode was used as a reference electrode, and a platinum wire was used as a counter electrode. All potentials were converted to the reversible hydrogen electrode reference. The CVs were tested in a N 2 -saturated 0.1 M HClO 4 electrolyte with a sweep rate of 50 mV s −1 . The ECSAs were calculated by integrating the hydrogen adsorption/desorption charge area between 0.05 and 0.38 V RHE from the CVs. The equation for the calculation of ECSAs is shown as follows: The charge density for the adsorption of one monolayer hydrogen on Pt (Q H ) was assumed to be 210 μC cm −2 . L Pt (mg Pt cm −2 ) was the working electrode Pt loading. A g (cm 2 ) was the geometric surface area of the glassy carbon electrode (0.196 cm 2 ). The ORR polarization curves were measured in an O 2 -saturated 0.1 M HClO 4 solution between 0.05 and 1.05 V RHE using a sweep rate of 10 mV s −1 at a rotation rate of 1600 rpm. The ADTs were performed via cyclic sweeps between 0.6 and 1.0 V RHE in an O2-saturated 0.1 M HClO 4 electrolyte at a sweep rate of 100 mV s −1 for different cycles. MEA fabrication and test The catalytic activity was also evaluated under PEMFC operating conditions. Specifically, each catalyst was mixed with isopropanol, deionized water, and Nafion solution by ultrasonicating for 1 h to form homogeneous ink with the ratio of ionomer to carbon (I/C) of 0.8. The ink was then sprayed the ink onto the Nafion® 211 membrane (DuPont). The catalyst-coated-membrane (CCM) with an active geometric area of 12.25 cm 2 was applied to the gas diffusion layer (GDL, Toray TGP-H-060). The compression ratio of GDL was calculated to be 31.5%. PtNiMoAu NWs/C, PtNiAu NWs/C, PtNiMo NWs/C, PtNi NWs/C, and commercial Pt/C were employed as the cathode catalysts and Pt/C (20 wt% loading, Johnson Matthey) was used for the anode. The Pt loading at the cathode was 0.10 mg Pt cm −2 for PtNi-based NWs/C and 0.12 mg Pt cm −2 for commercial Pt/C, respectively. As for the anode, Pt/C was used with a loading of 0.05 mg Pt cm −2 . Fuel cell testing was performed in a single cell using a commercial fuel cell test system (Scribner 850e, Hephas Energy Corporation). The MEA was sandwiched between two graphite plates with single serpentine flow channels. The corresponding pressure drop across the cathode and anode was measured using two pressure sensors connected to the inlet and outlet of the cell, respectively, while the cell was operated at 80 o C, 150 kPa (absolute, abs) back pressure, relative humidity (RH) 100%, H 2 flow rate of 200 mL min −1 and O 2 /Air flow rate of 500 mL min −1 . Fuel cell polarization curves were recorded using potential step mode with 50 mV/point (holding 2 min for each point). The MA was calculated by normalizing the measuring currents with Pt loading amount in 150 kPa abs H 2 /O 2 (80 o C, 100% RH, 200 anode /500 cathode sccm) at 0.9 V iR-free . The ADTs were conducted by the standard 30 K-cycle square-wave protocol to evaluate the durability of catalysts. Specifically, the cathode was held at 0.60 V for 3 s and 0.95 V for 3 s in each cycle. DFT calculations All of the DFT calculations were performed using the Vienna Ab-initio simulation package program 40 – 42 , which uses a plane-wave basis set and a projector augmented wave method (PAW) for the treatment of core electrons 41 . The Perdew, Burke, and Ernzerhof exchange-correlation functional within a generalized gradient approximation (GGA-PBE) 43 was used in our calculations, and the van der Waals (vdW) correction proposed by Grimme (DFT-D3) 44 was employed due to its good description of long-range vdW interactions. For the expansion of wavefunctions over the plane-wave basis set, a converged cutoff was set to 450 eV. In order to simulate the Pt NWs and PtNi bimetallic NWs, seven-layer 2×2 Pt (111) (a = b = 5.456 Å) and 2 × 2 Pt 3 Ni (111) (a = b = 5.383 Å) slabs with periodical boundary conditions were used, respectively. Besides, to model the PtNiMo trimetallic NWs and PtNiMoAu tetrametallic NWs, a Pt atom in the subsurface layers of the Pt 3 Ni (111) slab was substituted by a Mo atom and two Pt atoms in the subsurface layers of the Pt 3 Ni (111) slab were substituted by a Mo and an Au atom. Moreover, considering the surface Ni and Mo atoms will be leached into an acidic solution, the outmost layers of the Pt 3 Ni (111), Mo-doped Pt 3 Ni (111), and Mo- and Au-doped Pt 3 Ni (111) slabs were all replaced as shown in Supplementary Fig. 10 . The vacuum space was set to 17 Å in the z direction to avoid interactions between periodic images. In geometry optimizations, all the atomic coordinates were fully relaxed up to the residual atomic forces smaller than 1 × 10 −4 eV/Å, and the total energy was converged to 10 −5 eV. The Brillouin zone integration was performed on the (9 × 9× 1) Monkhorst–Pack k-point mesh 45 . The ORR pathways on various NWs were calculated in detail according to the electrochemical framework developed by Nørskov and his co-workers 46 , 47 . For ORR, the four-electron reaction mechanism follows several elementary steps: where the represents the active site on the electrocatalyst surface, and refer to liquid and gas phases, respectively, and , and are adsorbed intermediates. The binding energies of , and were obtained by DFT calculations as follows: 46 , 47 in which, , , , and are the ground state energies of a clean surface and surfaces adsorbed with , and , respectively. and are the calculated DFT energies of H 2 O and H 2 molecules in the gas phase. If we consider the zero point energy (ZPE) and entropy correction, the free energies of adsorption, ΔG ads , can be transformed from DFT binding energies, ΔE ads , as follows: where ΔE ads is the binding energy of adsorption species OOH * , O * and OH * . ΔZPE, ΔS, U and e are the ZPE changes, entropy changes, applied potential at the electrode, and charge transferred. Using the adsorption free energies obtained from (10) and (7)-(9), the reaction free energies of ORR reactions (2)-(6) can be calculated as: Thus, for the ORR reactions, the onset potential, , and the overpotential, η ORR , can be expressed as: 46 , 47 The vacancy formation energy is calculated as: where E slab-VPt represents the energy of a slab with one Pt vacancy, μ Pt represents the chemical energy of a Pt atom, E slab represents the energy of a perfect slab. We also calculated the vacancy formation energy of Ni (E slab - VNi ) in PtNi slab, PtNiMo slab, PtNiAu slab and PtNiMoAu slab by formula (17), where E slab-VNi represents the energy of a slab with one Ni vacancy on the subsurface layer, μ Ni represents the chemical energy of a Ni atom, E slab represents the energy of a perfect slab. In addition, we constructed 2 × 2 Pt 3 Ni conventional cells with a Ni vacancy model to calculate the Ni diffusion energy. The bulk Ni diffusion was proposed to diffuse to the nearest Ni vacancy site and the corresponding transition state searches were conducted with the climbing image nudged elastic band method (CI-NEB) method. For the Ni diffusion energy in the PtNiMo slab and PtNiAu slab, a Pt atom near the Ni diffusion pathway of the Pt 3 Ni (111) slab was substituted by the Mo atom and Au atom, respectively. For the Ni diffusion energy in the PtNiMoAu slab, two Pt atoms near the Ni diffusion pathway of the Pt 3 Ni (111) slab were substituted by the Mo atom and Au atom.
Results Synthesis and characterizations of PtNi-based model catalysts To explore the effect of Mo and Au dopants on the ORR performance of the PtNi catalyst, we synthesized a series of PtNi-based NWs, containing PtNi NWs, PtNiMo NWs, and PtNiAu NWs with the same diameter, as model catalysts by slightly adjusting the synthetic method (Supplementary Table 1 ). The detailed structure and composition of the three NWs were carefully analyzed. The low-magnification transmission electron microscopy (TEM) images in Supplementary Fig. 1a–c shows the successful preparation of PtNi NWs, PtNiMo NWs, and PtNiAu NWs in high purity and uniformity. The average diameters of these three NWs are all about 1.2 nm, which were determined by counting 100 NWs (Supplementary Fig. 1d–f ). To acquire the atomic-level structural information of three NWs, we captured the atomically resolved high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image of a single NW (Fig. 1a–c ). Specifically, for all these three NWs, a typical NW is made up of about six atomic layers, agreeing well with the average diameter. And the interplanar spacings of {111} planes for three NWs are almost the same, suggesting that the negligible strain is introduced with the doping of Mo or Au atoms into PtNi NWs. As shown in Fig. 1d , the X-ray diffraction (XRD) pattern of three NWs presented the same features in terms of peak number and peak position, being consistent with the data from its interplanar spacing. Figure. 1 a–c and Supplementary Fig. 2 show the energy-dispersive X-ray spectroscopy (EDS) elemental mapping and EDS line-scanning profile, clearly demonstrating the homogeneous distribution of Pt/Ni in PtNi NWs, Pt/Ni/Mo in PtNiMo NWs, and Pt/Ni/Au in PtNiAu NWs. According to the inductively coupled plasma mass spectrometry (ICP-MS), the atomic ratio of Pt/Ni in PtNi NWs, Pt/Ni/Mo in PtNiMo NWs, and Pt/Ni/Au in PtNiAu NWs were determined to be 3.03:1.00, 2.99:1.00:0.10, and 3.01:1.00:0.11, indicting the similar atomic ratio of Pt:Ni in three NWs. Taken together, the as-prepared PtNi-based NWs possess almost identical morphological structure, diameter, ratio of Pt/Ni, and interplanar spacing, thus offering the ideal platform to study the effects of Mo and Au dopants on the catalytic stability of PtNi NWs toward ORR. Atomistic Insights into the Roles of Mo and Au Dopants for Catalytic Stability Prior to ORR measurements, three PtNi-based NWs, as well as the Pt NWs using as the reference catalyst (Supplementary Fig. 3 ), were first loaded on Vulcan XC-7 carbon to prepare the carbon-supported NWs (NWs/C) catalysts (Supplementary Fig. 4 ). The cyclic voltammograms (CVs) and positive-going polarization curves of all NWs/C catalysts were further recorded to obtain the initial electrochemically active surface areas (ECSAs) and MA, respectively (Supplementary Fig. 5 and Fig. 1e, f ). Clearly, the MA of PtNi NWs/C (1.96 A mg −1 Pt ) catalyst displayed an obvious increase with respect to the Pt NWs/C (0.84 A mg −1 Pt ), in line with the prior studies 29 . When the Mo and Au dopants are introduced, the MA could be further increased to 2.54 A mg −1 Pt (PtNiMo NWs/C) and 2.24 A mg −1 Pt (PtNiAu NWs/C), respectively. The long-term catalytic stability for these three PtNi-based NWs catalysts was then assessed via an accelerated durability test (ADT) between 0.6 and 1.0 V versus reversible hydrogen electrode (V RHE ) in O 2 -saturated 0.1 M HClO 4 at room temperature. The CVs curves and polarization curves of NWs catalysts after certain cycles of ADT were then recorded, where the corresponding specific activity (SA), ECSAs, and MA were derived (Supplementary Figs. 6 and 7 ). The results demonstrate that both the Mo and Au dopants could significantly enhance the catalytic stability of PtNi NWs/C catalyst for ORR. Specifically, the PtNiMo NWs/C and PtNiAu NWs/C only show a drop of 22.8% and 18.3% in MA after 20 K cycles of ADT, respectively, contrasting with the big decrease of 53.5% for PtNi NWs/C catalyst (Fig. 1e ). The changes in structure for three PtNi-based catalysts after 20 K cycles of ADT were further examined to support the observations on durability trend. Self-consistently, the PtNiMo NWs/C and PtNiAu NWs/C catalysts both present smaller changes in ECSAs and NWs diameters after ADT measurements relative to PtNi NWs/C catalyst (Fig. 1f , Supplementary Figs. 8 and 9 ). These results together manifest that both the Mo and Au dopants could improve the ORR performance of the PtNi NWs/C catalyst (Supplementary Table 2 ). The in-depth insights into how Mo and Au dopants stabilize the PtNi catalyst for ORR operation were further explored. In principle, the Pt-based ORR catalysts degradation can be traced back to the following four possible reasons: (i) degradation of carbon support; (ii) dissolution of reactive metal components, e.g. the Pt and Ni for PtNi catalyst; (iii) detachment of catalyst nanoparticles; (iV) loss of active surface area induced by electrochemical Ostwald ripening and nanoparticle coalesce 30 , 31 . Because our PtNi-based model catalysts have almost identical structures and the same carbon support, we can safely exclude the carbon degradation and catalyst detachment as the reasons for the different stability performance of PtNi-based model catalysts. Besides, it should be noted that the dissolution of reactive metal components can trigger the electrochemical Ostwald ripening and nanoparticle coalesce 32 . In this case, we could reasonably assume that the dissolution of Pt and Ni elements should be the main cause for the PtNi-based catalysts degradation under long-term operation 33 , 34 . Consequently, we systematically detected the dissolution process of Pt and Ni elements for these catalysts during ADT measurements by a stationary rotating disk electrode coupled with an ex-situ ICP-MS. Fig. 1g, h shows the leaching amounts of Pt and Ni in the electrolyte after cyclic testing at intervals, showing the continuous leaching of Pt and Ni for each catalyst. Intriguingly, it is found that the Pt dissolution is maximally suppressed when the Au dopant is introduced, as indicated by the lowest dissolution rate of Pt in PtNiAu after 20 K cycles (Fig. 1i ). The difference is that the Ni dissolution is suppressed most when the Mo dopant is adopted. These experimental evidences definitely indicate that the Mo and Au dopants play distinct roles in improving the stability of PtNi NWs/C catalyst. Specifically, the Mo dopant mainly contributes to the Ni stabilization while the Au dopant dominantly restrains the Pt dissolution. To further verify the distinct roles of Mo and Au dopants, the DFT calculations were performed. In light of the corresponding structural and compositional parameters, we first constructed the PtNiAu (111) slab, PtNiMo (111) slab, PtNi (111) slab (Supplementary Fig. 10 ) as the models to represent the PtNiAu NWs/C, PtNiMo NWs/C, and PtNi NWs/C catalysts, respectively. Since the vacancy formation energies (E vac ) of Pt and Ni atoms could reflect the tendency of Pt atoms dissolution and Ni atoms leaching 35 , 36 , the E vac of Pt and Ni atoms for PtNi-based (111) slabs were calculated (Supplementary Fig. 11 ). In comparison with PtNi (111) slab, both PtNiMo (111) slab and PtNiAu (111) slab present increased E vac of Pt and Ni atoms, self-supporting the mitigated dissolution of Pt and Ni for PtNiMo NWs/C and PtNiAu NWs/C catalysts. We further found that the E vac of Pt atoms for the PtNiAu (111) slab and E vac of Ni atoms for PtNiMo (111) slab present the largest increment with respect to the PtNi-based slab, respectively (Fig. 2a ). The results thus confirm that the Mo and Au dopants could separately mitigate the leaching of Ni and Pt atoms, in accord with experimental observations from the ICP-MS. Besides the E vac , the diffusion energy barrier of Ni is also crucial for Ni leaching. We thus calculated the Ni diffusion energy diagrams for PtNi-based slabs (Fig. 2b , Supplementary Fig. 12 ). It could be seen that the diffusion energy barrier of Ni for PtNiMo (111) slab (0.68 eV) was the largest, followed by PtNiAu (111) slab (0.57 eV) and PtNi (111) slab (0.40 eV), indicating that Mo dopant could obviously inhibit the outward diffusion of bulk Ni atoms (Fig. 2b ). All these preceding results together confirm that, in improving the durability of PtNi NWs, the Mo dopant mainly increases E vac and diffusion energy of Ni atoms, while the Au dopant mainly increases E vac of Pt atoms. We further attempted to understand how the Mo and Au dopants play distinct roles in stabilizing Ni and Pt, which should be, in principle, originated from the different interaction strengths between dopants and Pt/Ni 18 , 26 . To explore the interaction strength for the atomic pair of Mo-Pt, Au-Pt, Mo-Ni, and Au-Ni, the partial density of states (PDOS) of PtNiMo (111) and PtNiAu (111) slabs were analyzed (Fig. 2c, d ). In general, the degree of overlaps between d -orbitals could reflect bonding strength between different metals. The results show that the degree of overlap between Pt d -orbital and Au d -orbital is larger than that of Pt d -orbital and Mo d -orbital, indicating the reinforced interaction between Pt and Au. Meanwhile, the stronger interaction between Mo d -orbital and Ni d -orbital is found relative to that between Au d -orbital and Ni d -orbital. These insights into the interaction between dopants and Pt/Ni atoms reasonably underpin the distinct roles of Mo and Au dopants in stabilizing the PtNi catalyst. Besides, the improved activities of PtNiMo NWs/C and PtNiAu NWs/C catalysts were also rationalized by DFT calculations (Supplementary Fig. 13 ). Supplementary Fig. 14a, b shows the Gibbs free energies of reaction intermediates for the ORR on different slabs. The results testify that the overpotential on Pt (111), PtNi (111), PtNiMo (111), and PtNiAu (111) slabs follows the trend of PtNiMo (111) < PtNiAu (111) < PtNi (111) < Pt (111), well-consistent with our experimental observations on the activity trend. Relative to PtNi NWs/C catalyst, we could conclude that the weakened adsorption of oxygenated species on PtNiMo NWs/C and PtNiAu NWs/C catalysts arising from the ligand effect is the intrinsic cause for the improved activity, which is further supported by the downshifted d -band center on PtNiMo (111) and PtNiAu (111) slabs (Supplementary Fig. 14c ) 35 , 37 , 38 . Rational design and synthesis of PtNiMoAu NWs Since Au and Mo dopants could respectively stabilize the surface Pt overlayer and suppress the leaching of Ni atoms, it is reasonably hypothesized that integrating these two dopants into PtNi would further improve the ORR stability. The hypothesis is further verified by the DFT calculations, as indicated by the largest E vac of Pt and Ni atoms, as well as the highest diffusion energy barrier (0.76 eV) of Ni atoms, on PtNiMoAu (111) slab (Supplementary Figs. 10 – 12 and 15a ). The strong coupling degrees between Mo, Pt, and Au d orbitals, as well as Mo, Ni, and Au d orbitals may account for the increased E vac and diffusion energy barrier (Supplementary Fig. 15b, c ). Moreover, it is predicted that the PtNiMoAu would present the enhanced ORR activity, as indicated by the lowest overpotential and downshifted d -band center of the PtNiMoAu (111) slab (Supplementary Fig. 14 ). Inspired by this conceptual design, we subsequently synthesized PtNiMoAu NWs catalysts by using a modified synthetic procedure (see Methods for detail). Figure 3a,b shows low-magnification TEM and STEM images of the as-synthesized PtNiMoAu NWs, respectively. The NWs with a diameter of 1.1 ± 0.3 nm and length of 62 ± 21 nm are identified by statistic counting (Fig. 3c ). The atomic-level structural information was analyzed based on the HAADF-STEM image of a single NW (Fig. 3d ). Specifically, the well-defined lattice spacing can be assigned to the {200} and {111} planes, matching with the fast Fourier transform (FFT) pattern in the inset. From the distinguished lattice planes, the grown direction of <110> orientation could be deduced. The EDS elemental mapping (Fig. 3e ) and line-scanning profile (Fig. 3f ) both indicate the homogeneous distribution of Pt, Ni, Mo, and Au elements throughout a single NW. The Pt:Ni:Mo:Au atomic ratio was estimated to be 3.02:1.00:0.11:0.13 by ICP-MS, very close to the value (3.10:1.00:0.15:0.14) determined by the XPS spectrum (Supplementary Fig. 16 ). These structural characterizations prove the successful formation of ultrathin PtNiMoAu tetrametallic NWs. The chemical states for the constituent elements of PtNi NWs and PtNiMoAu NWs were then analyzed by XPS. Compared with the Pt 4 f of PtNi NWs, the decreased binding energy for Pt 4 f of PtNiMoAu NWs implies more electrons accumulate around Pt in PtNiMoAu NWs (Fig. 3g ). To examine the local coordination and electronic structures of the PtNi NWs and PtNiMoAu NWs, we further employed X-ray absorption spectroscopy (XAS), in comparison with a bulk Pt foil and PtO 2 . Figure 3h shows the X-ray absorption near-edge structure (XANES) spectra of Pt L 3 -edge. Agreeing well with XPS results and Bader analysis (Supplementary Table 3 ), the Pt white line of PtNiMoAu NWs is found to possess lower intensity and be close to the metallic state (Pt foil) when compared to the PtNi NWs. These results manifest that, after introducing the Mo and Au atoms, Pt atoms in PtNiMoAu NWs could gain more electrons. The local structure was further derived from the Fourier transform of the phase-corrected extended X-ray absorption fine structure (EXAFS). Fig. 3i and Supplementary Fig. 17 show the fitted R-space and K-space data of the Pt edge for PtNiMoAu NWs and PtNi NWs. Clearly, the first-shell Pt-Pt length of PtNi NWs and PtNiMoAu NWs was obviously shorter than that of Pt/C, attributing to the smaller radius of Ni/Mo atoms. Besides, the PtNiMoAu NWs present a similar Pt-Pt coordination number (5.5 ± 0.4) with that of PtNi NWs (6.3 ± 0.4), due to the similar NWs diameter (Supplementary Table 4 ). These results together demonstrate that the Mo and Au dopants could redistribute the electrons and induce the compressive strain, which are believed as the structural foundations for the improved ORR performance as predicted. ORR Performance of PtNiMoAu NWs To verify the validity of our rational design for a practical catalyst, we further assessed the ORR performance of PtNiMoAu NWs/C catalyst by recording the CVs and polarization curves in 0.1 M HClO 4 solution (Supplementary Fig. 18 and 19 ). Specifically, the PtNiMoAu NWs/C catalyst exhibits specific activity (SA) and MA of 3.13 mA cm −2 and 2.89 A mg −1 Pt at 0.9 V RHE , much higher than those of Pt/C (0.27 mA cm −2 and 0.18 A mg −1 Pt ), PtNi NWs/C (2.34 mA cm −2 and 1.96 A mg −1 Pt ), PtNiMo NWs/C (2.93 mA cm −2 and 2.54 A mg −1 Pt ) and PtNiAu NWs/C catalysts (2.61 mA cm −2 and 2.24 A mg −1 Pt ) (Fig. 4a and Supplementary Fig. 19 ). These results prove that the co-doping of Mo and Au elements into PtNi NWs evidently boost the ORR activity, which is in accordance with DFT prediction. The long-term stability of PtNiMoAu NWs/C catalyst was further evaluated by ADT. Figure 4b shows the polarization curves of PtNiMoAu NWs/C catalyst after different cycles of ADT. Remarkably, the ECSAs (86.1 m 2 g −1 Pt ) and MA (2.42 A mg −1 Pt ) of PtNiMoAu/C catalyst only displayed a drop of 6.7% and 16.2% after 80 K cycles (Fig. 4c and Supplementary Fig. 20 ), respectively. Of note, such excellent long-term stability of PtNiMoAu NWs surpasses the recently-reported Pt-based ORR electrocatalysts at different cycles of ADT (Fig. 4d and Supplementary Table 5 ). By contrast, after only 20 K cycles, the ECSAs and MA for commercial Pt/C, Pt NWs/C, PtNi NWs/C, PtNiMo NWs/C, and PtNiAu NWs/C catalysts severally decrease 68.4% and 70.2%, 30.8% and 34.5%, 49.4% and 53.5%, 19.2% and 22.8%, 14.4% and 18.3% (Fig. 1e and Supplementary Fig. 21 ). The structure and composition of PtNiMoAu NWs/C catalyst after 80 K of ADT were further probed. The STEM image demonstrates the well-maintained morphology of PtNiMoAu NWs/C after 80 K cycles (Fig. 5a ), whereas the commercial Pt/C and other Pt-based NWs/C catalysts exhibit varying degrees of sintering after only 20 K cycles (Supplementary Fig. 8 , 22a and 23 ). Its diameter only increases from 1.1 nm (initial) to 1.4 nm (after 80 K cycles) (Fig. 5a and Supplementary Fig. 22b ), consistent with the ECSAs losses. And the EDS mapping and line-scan profiles further confirm its well-reserved elemental distribution and composition (Fig. 5b, c ). The atomic ratio of Pt:Ni:Mo:Au (3.65:1.00:0.16:0.20) in PtNiMoAu NWs/C catalyst after 80 K cycles is relatively close to its initial composition (3.02:1.00:0.11:0.13), contrasting with the big change for PtNi NWs/C catalyst after only 20 K cycles from 3.03:1.00 to 11.01:1.00. The dissolution of Pt and Ni for PtNiMoAu NWs/C catalyst at different ADT cycles was further examined based on ex-situ ICP-MS measurements. Self-consistent with our expectation from the distinct roles of Mo and Au dopants, the PtNiMoAu NWs/C presents the lowest leaching mass of Pt and Ni metal among all PtNi-based NWs (Fig. 5d ). Besides, in-situ XANES was carried out to experimentally support the remarkable stability of the PtNiMoAu NWs/C catalyst for ORR. From XANES spectra of the Pt L 3 edge, we can find that the normalized white line intensity (μ norm ) of PtNiMoAu NWs slightly increases with the potential (Fig. 5e ), while the PtNi NWs display a much larger increase at the same operational condition (Supplementary Fig. 24 ). The difference is more clear in the relative change of the white line intensity [(Δμ E -Δμ 0.42VRHE )/Δμ 0.42VRHE )] of the Pt L 3 edge spectra for PtNiMoAu NWs/C and PtNi NWs/C catalysts as a function of applied potential 20 , 39 . As shown in Fig. 5f , the increase in white line intensity for PtNiMoAu NWs/C catalyst occurs at a higher applied potential compared with that for PtNi NWs/C catalyst, corroborating the elevated Pt oxidation potential for PtNiMoAu NWs/C catalyst. In addition, anodic shift of Pt/Pt 2+ redox potential for PtNiMoAu NWs/C catalyst also supports this result (Supplementary Fig. 25 ). This finding is indeed consistent with the electron-rich Pt atoms for PtNiMoAu NWs as demonstrated by the XANES spectrum (Fig. 3h ) and Bader analysis (Supplementary Table 3 ), which could curb the Pt dissolution and in turn stabilize the catalyst. All these preceding evidences strongly support our conceptual design that integrating the different functions of Mo and Au into PtNi NWs would maximize the catalyst stability. Fuel cell performance We further fabricated the membrane electrode assembly (MEA) to evaluate the fuel cell performance of PtNi NWs/C, PtNiMo NWs/C, PtNiAu NWs/C, and PtNiMoAu NWs/C. The polarization curves and power density curves were recorded in H 2 -O 2 (Fig. 6a ) and H 2 -Air environment (Fig. 6b ) (80 °C and 150 kPa absolute pressure), in which the Pt loadings for Pt/C as anode catalyst and PtNi-based NWs/C as cathode catalyst were 0.05 mg Pt cm −2 and 0.10 mg Pt cm −2 , respectively (Pt/C catalyst with Pt loading of 0.12 mg Pt cm −2 on the cathode was also employed as the reference). Before evaluating MEA performance of catalysts, we first tested the pressure drop between the inlet and outlet at different flow rates (Supplementary Fig. 26 ), in which the gap pressure drop of anode (H 2 , 200 mL min −1 ) and cathode (O 2 , 500 mL min −1 ) is severally 3 kPa and 6 kPa. Given the relatively low pressure drop, we reasonably believe that it has a negligible impact on MEA performance. The H 2 –O 2 /H 2 -Air fuel cell assembled with PtNiMoAu NWs/C cathode shows a peak power density of 1.71/0.93 W cm −2 , exceeding the corresponding values of 1.24/0.82 W cm −2 for commercial Pt/C. Besides, the PtNiMoAu NWs/C presents a 6.6-, 2.0-, 1.3-, and 1.7-fold enhancement in the beginning-of-life (BOL) MA (0.93 A mg −1 Pt ) relative to the Pt/C (0.14 A mg −1 Pt ), PtNi NWs/C (0.46 A mg −1 Pt ), PtNiMo NWs/C (0.70 A mg −1 Pt ), and PtNiAu NWs/C (0.55 A mg −1 Pt ) at 0.9 V iR-free (Fig. 6c and Supplementary Fig. 27 ). Even comparing with the value that was set in the 2025 targets by U.S. Department of Energy (DOE) (MA of 0.44 A mg −1 Pt ), the remarkable mass activity of PtNiMoAu NWs/C catalyst presents 2.1-fold increasement, showing the great potential for PEMFCs. Moreover, the PtNiMoAu NWs/C under the H 2 -Air condition delivers the improved current density of 0.35 A cm −2 at the kinetic dominant region of 0.8 V when compared to the Pt/C (0.16 A cm −2 at 0.8 V) and the DOE 2025 target (0.30 A cm −2 ) (Supplementary Fig. 28 ). For ADT, the standard square-wave protocol by holding the cathode at 0.60 V for 3 s and 0.95 V for 3 s was employed to evaluate the durability of catalysts (see Methods). The end-of-life (EOL) MA at 0.9 V iR-free (H 2 -O 2 ) and a voltage loss at 0.8 A cm −2 (H 2 -Air) for PtNiMoAu NWs/C are 0.72 A mg −1 Pt (22.6% MA loss) and 25 mV after 30 K cycles of ADT, respectively, which comprehensively exceed the DOE’s 2025 durability goal (MA loss <40% and voltage loss <30 mV). As a contrast, the Pt/C, PtNi NWs/C, PtNiMo NWs/C, and PtNiAu NWs/C show substantial MA loss of 45.7%, 58.7%, 41.4%, and 32.7%, respectively. Besides, the variation trend of ECSAs, voltages loss (H 2 -O 2 ), Ni content, and morphology of PtNi NWs/C, PtNiAu NWs/C, PtNiMo NWs/C, and PtNiMoAu NWs/C catalysts during the MEA durability tests also was in accordance with the results obtained under the rotating disk electrode measurements (Fig. 6d, e , Supplementary Figs. 29 – 31 , and Supplementary Table 6 ). Even comparing with the Pt-based catalysts reported recently, the excellent BOL MA and durability of PtNiMoAu NWs/C indeed render it as the top catalyst for durable fuel cells (Fig. 6f and Supplementary Table 7 ). Beyond this, we can envision our delicate design would also enhance the catalyst durability a more challenging start-up and shutdown process of the fuel cells (generally under the applied potential over 1.0 V) through mitigation of the sintering process. All these preceding results together strongly evidence the successful design of active and durable PtNiMoAu NWs/C catalyst for practical fuel cells.
Discussion To sum up, we have identified the distinct roles of Mo and Au dopants in inhibiting the dissolution of Ni and Pt metal, respectively, in terms of PtNi-based NWs/C as model catalysts, by combining the experimental evidences and DFT calculations. Insightful studies indicated that the distinct roles of Mo and Au dopants are essentially derived from the stronger interaction of atomic orbital for Pt-Au ( d - d ) and Ni-Mo ( d - d ). Based on this atomistic understanding, we delicately designed a remarkable PtNiMoAu NWs/C catalyst that possesses a concurrently high MA of 2.89 A mg −1 Pt and preeminent stability, with only 16.2% loss of MA after 80 K cycles of ADT. By contrast, PtNi NWs/C catalyst showed a 53.5% loss of MA after only 20 K cycles of ADT, persuasively confirming the advantage of Mo/Au co-doping in improving its ORR stability. Moreover, when assembling the PtNiMoAu NWs/C catalyst into the fuel cell cathode, a high MA retention of 77.4% (H 2 -O 2 , 0.9 V iR-free ) and a low voltage loss of 25 mV (H 2 -Air, 0.8 A cm −2 ) after ADT were output, proving the highly durable fuel cell performance. This work advances the design of robust PtM catalysts to a more precise stage by providing insights into the functions of dopants, which is thus of general importance for the fields of catalysis, sensing, and even beyond.
Stabilizing active PtNi alloy catalyst toward oxygen reduction reaction is essential for fuel cell. Doping of specific metals is an empirical strategy, however, the atomistic insight into how dopant boosts the stability of PtNi catalyst still remains elusive. Here, with typical examples of Mo and Au dopants, we identify the distinct roles of Mo and Au in stabilizing PtNi nanowires catalysts. Specifically, due to the stronger interaction between atomic orbital for Ni-Mo and Pt-Au, the Mo dopant mainly suppresses the outward diffusion of Ni atoms while the Au dopant contributes to the stabilization of surface Pt overlayer. Inspired by this atomistic understanding, we rationally construct the PtNiMoAu nanowires by integrating the different functions of Mo and Au into one entity. Such catalyst assembled in fuel cell cathode thus presents both remarkable activity and durability, even surpassing the United States Department of Energy technical targets for 2025. Doping strategies have been shown to stabilize the active platinum-nickel (PtNi) catalyst in fuel cells, however, the atomistic mechanism is less known. Here, the authors identify the roles of Mo and Au dopants in improving the durability of a PtNi nanowire catalyst for fuel cells. Subject terms
Supplementary information Source data
Supplementary information The online version contains supplementary material available at 10.1038/s41467-024-44788-0. Acknowledgements This work was supported by the National Key Research and Development Program of China (No. 2021YFA1502000 to H.H.), NSFC (Nos. 22322902, U22A20396, 22211540385, and 22309050 to H.H. and L.G.), the Science and Technology Innovation Program of Hunan Province (No. 2021RC3065, 2021RC2053, and 2023JJ40117 to H.H. and L.G.), the Jiebang Guashuai Project of Changsha City (Grant No. kq2301009 to H.H.), the China Postdoctoral Science Foundation (2023T160205 and 2023M741118 to L.G.), Shenzhen Science and Technology Program (Nos. JCYJ20210324120800002 and JCYJ20220818100012025 to H.H.). The X-ray ab-sorption structure (XAS) spectra were performed at the BL11B beam line of the Shanghai Synchrotron Radiation Facility (SSRF). Author contributions L.G. and H.H. conceived the idea and designed experiments. L.G. performed the preparation of samples, carried out the electrochemical measurements, and analyzed the experimental data. Z.Y. and W.Z. conducted the DFT simulation and theoretical analyses. T.S. and M.L. conducted the TEM and EDX characterizations. X.C., W.L. and Q.Y. helped with the analysis and discussion of experimental data. H.H. wrote the manuscript. All the authors were involved in the discussion and analysis of the manuscript. Peer review Peer review information Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available. Data availability The data that support the plots within this paper and other finding of this study are available from the corresponding authors upon request. Source data are provided with this paper. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Nat Commun. 2024 Jan 13; 15:508
oa_package/49/d6/PMC10787824.tar.gz
PMC10787825
38218897
Introduction Abiotic factors play a pivotal role in the decline of agricultural and allied sector production. Climate change, pollution, and degraded water quality stand out as major abiotic influencers shaping the life patterns of aquatic organisms, including fish 1 , 2 . Ammonia holds a crucial position in the nitrogen cycle, undergoing conversion into nitrite (NO 2 ) by Nitrosospira and Nitrosomonas bacteria in aquatic systems through the nitrification process. Additionally, it originates from fish waste, high-protein diets, and metabolic processes, contributing to the presence of ammonia in aquatic systems 3 , 4 . The breakdown of amino acids, pyrimidines, and purines also generates ammonia 5 , existing in two forms: unionized ammonia (NH 3 ) and ionized ammonium (NH 4 + ) 6 . Ammonia toxicity can lead to noticeable reductions in growth performance 2 , immunity, tissue erosion, neurotoxicity, oxidative stress, and ultimately result in high mortality 7 . Naturally occurring arsenic, typically harmless in its natural state, can undergo transformation into inorganic arsenic, thereby contaminating groundwater sources used for drinking or irrigating crops. The accumulation of arsenic from one trophic level to another level, depends not only on the total arsenic content but also significantly on its bioavailability 8 . The chemical forms of arsenic, such as inorganic and organic forms present in crops, vegetables, and fish, play a crucial role in determining bioavailability, which is essential for estimating its toxicity. Humans can also uptake arsenic from contaminated sources such as rice, vegetables, milk, and meat. Consequently, 'plant–human' and 'plant–animal–human' represent potential food chain pathways for arsenic accumulation 9 , 10 . It is also considered to be consumption of even low dose of arsenic can cause deadly diseases, including cancer 11 , 12 . In all around the world, such as in India, Bangladesh, Argentina, China, Ghana, USA, and Vietnam, more than 200 million peoples are at high risk 13 , 14 . Further, in Bangladesh, 43,000 peoples die annually due to arsenic pollution 14 . Fish are classified as poikilothermic animals; however, even with slight temperature variations, their physiology undergoes abrupt changes. These changes include alterations in growth, metabolism, food consumption, thermal tolerance, and an inability to maintain internal homeostasis in response to the fluctuating external environment 15 , 16 . Moreover, elevated temperatures diminish the availability of oxygen to aquatic animals, creating challenges in meeting metabolic demands, especially as the water flow rate increases across the gills 1 . Interestingly, manganese (Mn) plays a vital role as an essential micronutrient in the growth and development of the vertebral column, serving as an antioxidant and acting as a cofactor for numerous enzymes 17 . Typically, the requirement for Mn is met through waterborne sources, but additional supplementation is necessary to fulfil the physiological needs of the fish 18 , 19 . Therefore, Mn supplements are provided to meet the physiological requirements and metabolic scope of the fish. A deficiency in Mn for fish can lead to retarded growth performance, skeletal deformities (dwarfism), eye lens cataracts, decreased activities of copper-zinc superoxide dismutase (Cu–Zn-SOD), manganese superoxide dismutase complex (Mn-SOD), and reduced reproductive performance 19 – 21 . Mn is primarily located in the mitochondria and plays a crucial role in activating several enzymes, including decarboxylases, kinases, hydrolases, and transferases. Key manganese metallo-enzymes, such as pyruvate carboxylase, catalyze the conversion of pyruvate to oxaloacetate 22 . Apoptosis is a programmed cell death crucial for regular cell repair, cellular function, immune and hormone-related gene development, and chemical cell death in all organisms, including fish 23 . Cytokines, serving as essential signaling molecules, are released during both physiological and pathological conditions. They play a role in stress responses and modulate the host's inflammatory response and immunobiological mechanisms 24 , 25 . Furthermore, NF-kB regulates and controls the transcription of genes related to immune cells, inflammation, proliferation, the cell cycle, and cell death 26 . Pangasianodon hypophthalmus exhibits great potential as a fish species suitable for cultivation in challenging conditions, displaying tolerance to high abiotic and biotic stress 16 , 27 , 28 . Moreover, it is an ideal species for studying gene regulation involved in both abiotic and biotic stress 2 . Consequently, the present investigation aims to study the role of manganese in mitigating arsenic and ammonia toxicity, as well as high-temperature stress. This study also explores gene regulation associated with abiotic and biotic stress in response to dietary manganese in P. hypophthalmus .
Materials and methods Ethics statement The Research Advisory Committee (RAC) of the Institute (ICAR-National Institute of Abiotic Stress Management, Baramati, Pune) has approved the experimental procedures, and this study adheres to the Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines. All methods were conducted in strict accordance with the relevant guidelines and regulations. Experimental animal and design Pangasianodon hypophthalmus specimens with an average weight of 6.71 ± 0.52 g and a length of 5.12 ± 0.17 cm was utilized in the current investigation. These fish were sourced from the NIASM farm pond and were in a healthy condition. At Prior to the commencement of the experiment, a two-week acclimatization period was provided to the fish in a Fiberglass Reinforced Plastic (FRP) tank. During this acclimatization period, the fish received regular feeding and other necessary maintenance. Subsequently, the eighteen fish were evenly distributed in a plastic rectangular tank with a capacity of 150 L. The experiment was designed with 12 treatments, each replicated in triplicate, employing a Completely Randomized Design (CRD). The treatments followed as 1. Control 2. As exposed group 3. Ammonia exposed group 4. Concurrent exposure to arsenic and ammonia group 5. Concurrent exposure to ammonia and high temperature group 6. Concurrent exposure to arsenic, ammonia and high temperature group 7. Group fed with Mn at 4 mg kg −1 diet 8. Group fed with Mn at 8 mg kg −1 diet 9. Group fed with Mn at 12 mg kg −1 diet 10. Group fed with Mn at 4 mg kg −1 diet and concurrently exposed to arsenic, ammonia and high temperature 11. Group fed with Mn at 8 mg kg −1 diet and concurrently exposed to arsenic, ammonia and high temperature 12. Group fed with Mn at 12 mg kg −1 diet and concurrently exposed to arsenic, ammonia and high temperature. The details of the treatment is shown in Table 1 . The experimental diets were administered twice daily to the fish at 9:00 AM and 5:00 PM. Continuous aeration was maintained throughout the experiment using an aerator. Daily removal of uneaten feed and faecal matter was carried out through siphoning. Periodic analysis of water quality parameters was conducted using the APHA method 29 , and the results consistently fell within acceptable ranges throughout the experiment. Every alternate day, 2/3rd of the water in the tank was manually replaced. Additionally, (NH 4 ) 2 SO 4 was added as a source of ammonia toxicity (NH 3 ), and sodium arsenite, NaAsO 2 , was introduced as a source of arsenic. The concentrations used were Ammonium sulfate (1/10th of LC 50 , 2.0 mg L −1 of (NH 4 ) 2 SO 4 ) 2 and As (1/10th of LC 50 , 2.68 mg L −1 of arsenic) 1 . The water temperature was maintained at a high level (34 °C) throughout the experiment to induce stress. Four iso-caloric (365 kcal/100 g) and iso-nitrogenous (35% crude protein) pelleted diets containing manganese were prepared. The feed ingredients included wheat flour, groundnut meal, soybean meal, and fish meal. Cod liver oil, lecithin, vitamin C, and other labile nutrients were added after heating the feed ingredients. A manganese-free mineral mixture was manually prepared. Proximate analysis was conducted using the AOAC method 30 , while ether extract (EE) was determined through solvent extraction, crude protein by nitrogen content, and ash content by using a muffle furnace at 550 °C. Total carbohydrate content was calculated using the formula 100—(CP% + EE% + Ash %+moisture). Gross energy was determined using the Halver method 31 (Table 2 ). Tissue homogenate preparation and blood collection The gill, muscle, brain, liver, and kidney were dissected from anesthetized fish (clove oil, 50 μl L −1 ) under aseptic conditions. The chilled sucrose (5% w/v, 0.25 M) and EDTA solution (1 mM) were used as homogenates for tissue homogenization using a homogenizer (Omni Tissue Master Homogenize, Kennesaw, GA) for enzyme analysis. The gene expression and quantification, the liver and muscle tissues samples were processed with liquid nitrogen. For the enzymes analysis, the tissues were homogenized and centrifuged at 5,000 × g for 15 min at 4 °C to homogenated samples. The supernatants were collected and stored at -20 °C until further analysis. During dissection, the blood (3 fish) was also collected from the same fish of each tank and serum (5 fish) was processed from the same collected blood. Lowry protein assay 32 was used for tissue protein analysis. RNA isolation and quantification and cDNA synthesis and quantitative PCR The TRIzol method was employed for total RNA isolation from the liver tissue of P. hypophthalmus . Liquid nitrogen was utilized for the homogenization of the liver tissue. Subsequently, chloroform was added to the homogenized samples, and the mixture was incubated for 5 min to allow for phase separation. The resulting solution was then centrifuged to separate the RNA, followed by the addition of 75% ethanol and air drying. The RNA pellet was dissolved in free water and stored at -80 °C for future use. To assess RNA integrity, 1% agarose gel electrophoresis was conducted, and the RNA bands were visualized using a Gel documentation system (ChemiDocTM MP imaging system, Bio-Rad). RNA quantification was performed using a NanoDrop spectrophotometer (Thermo Scientific). Revert Aid First Strand cDNA synthesis kit (Thermo Scientific) was utilized. DNase I was employed to remove trace amounts of DNA. The reaction mixture, consisting of oligo dT primers (15 pmol) and RNA template (100 ng) in 12 μl, was heated for 5 min at 65 °C and then chilled on ice. Subsequently, 1.0 μl of reverse transcriptase enzyme, 2 μl dNTP Mix (10 mM), and 1 μl Ribo Lock RNase Inhibitor (20 U/μL) were added to the chilled mixture, followed by a brief centrifugation. The reaction mixture was incubated for 42 min at 60 °C, then at 70 °C for 5 min, and the synthesized cDNA was stored at -20 °C. β-actin was used as a reference for confirming the synthesized cDNA. Real-time PCR was conducted using SYBR green and gene-specific primers (Bio-Rad). The quantification protocol included an initial denaturation for 10 min at 95 °C, followed by 39 cycles of cDNA amplification, denaturation at 95 °C for 15 s, and annealing at 60 °C for 1 minute 33 . Details of the primers are recorded in Table 3 . Genes The genes were investigated in liver tissues in this study viz. catalase (CAT), glutathione-s-transferase (GST), superoxide dismutase (SOD), nitric oxide synthase (iNOS), heat shock protein (HSP 70), Caspase 3a (CAS 3a and 3b), cytochrome P450 (CYP 450), tumor necrosis factor (TNFα), toll like receptor (TLR), metallothionine (MT), growth hormone receptor (Ghr1 and Ghrb), interleukin (IL), immunoglobulin (Ig), insulin like growth factor 1 and 2 (IGF1X1 and IGF1X2), somatostatin (SMT), myostatin (MYST), and growth hormone (GH), studied for real-time quantification. Antioxidant enzyme activities Superoxide dismutase (SOD) (EC 1.15.1.1) activities in different fish tissues were determined by Misra and Fridovich 34 . Catalase (EC 1.11.1.6) was determined as followed as a procedure of Takahara et al 35 . The glutathione S-transferase (GST) (EC 2.5.1.18) was determined as per the procedure of Habing et al 36 . Glutathione peroxidase (GPx) (EC 1.11.1.9) activity was accomplished following the method of Paglia and Valentine 37 . Neurotransmitter enzyme activities Hestrin modified by Augustinsson 38 method was applied to determine the acetylcholine esterase activities (AChE) (EC. 3.1.1.7) in brain tissue. Lipid peroxidation (LPO) and Vitamin C Uchiyama and Mihara 39 method was followed to determine the LPO in liver and kidney tissues. Similarly, Roe and Keuther 40 used to determine the Vitamin C in brain and muscle tissues. Hematological parameters Blood was drawn from the caudal peduncle region of the fish using heparinised syringe. Indices measured included erythrocyte count (RBC), hemoglobin concentration (Hb), WBC (total leucocyte count) and the procedures were based on unified methods for hematological examination of fish. Immunological attributes Total serum protein, albumin, globulin, and A:G ratio was determined using the protein estimation kit. Secombes 41 , with some modification by Stasiack and Baumann 42 used for the estimation of respiratory burst activity. The blood glucose was determined using Nelson 43 and Somoyogi 44 . Moreover, Quade and Roth 45 , with some modifications 46 and Anderson and Siwicki 47 were applied for the determination of myeloperoxidase and total immunoglobulin. Cortisol Serum cortisol was determined using ELISA kit (Commercially available Cortisol EIA kit, catalogue no. 500360, Cayman Chemicals, USA). The assay was performed as per instruction provided with the kit using ELISA plate reader (Biotek India Pvt. Ltd.). Arsenic and manganese analysis from fish tissues and experimental water Liver, muscle, gill, brain, and kidney were collected to determine in arsenic concentration. Whereas, Mn concentration was determined in the feed and fish muscle. The tissues and diets were processed in a microwave digestion system (Microwave Reaction System, Multiwave PRO, Anton Paar GmbH, Austria, Europe) using Inductively Coupled Plasma Mass Spectrometry (ICP-MS) (Agilent 7700 series, Agilent Technologies, USA) as followed the method of Kumar et al. 48 , 49 . Alkaline single-cell gel electrophoresis (SCGE)/Comet assay Alkaline single cell gel electrophoresis/comet assay was applied for determination of DNA damage in kidney tissue using Ali et al 50 . with slight modification 51 . The slides coating and other procedure were followed the above method. Then prepared slides for genotoxicity were analysed in fluorescent microscope (Leica Microsystems Ltd, DM 2000, Heerbrugg, Switzerland). The position of DNA damage was captured using the microscope and analyzed using Open comet. The parameter selected for quantification of DNA damage was percent tail DNA (i.e., % tail DNA = 100% head DNA) as determined by the software. Aspartate aminotransferase (AST) and alanine aminotransferase (ALT), Lactate dehydrogenase (LDH), and malate dehydrogenase (MDH) AST (E.C.2.6.1.1) and ALT (E.C.2.6.1.2) were determined using Wooten 52 method. Similarly, LDH activities were determined using Wroblewski and Ladue 53 . Similarly, MDH was determined using Ochoa 54 . A similar reaction mixture was used except for substrate oxaloacetate instead of sodium pyruvate. Growth performance The growth performance was determined by evaluating the following method. The sampling/weighing of the fish was observed by every 15 days up to 105 days. Challenge study with Aeromonas hydrophila After 105 days of the feeding trial, 8 fishes per replicates in each treatment were challenged with virulent A. hydrophilla (Lot no. 637–51-5 and Ref 0637P, HiMedia, Mumbai). A. hydrophilla was grown on a nutrient broth for 24 h at 37 °C in a BOD incubator and harvested by centrifuging the culture broth at 10,000 × g for 10 min at 4 °C. The cells were then washed thrice in sterile PBS (pH 7.2), and the final concentration was maintained at 10 8 CFU ml −1 . The fish were intraperitoneally injected with 0.15 ml of bacterial suspension in each treatment group. The fish mortality in each treatment group was recorded up to 7 days of challenge study. The tissues were dissected out from morbid fish for confirmation of A. hydrophilla as a causative agent for death. Statistics The data were analysed using Statistical Package for the Social Sciences (SPSS) version 16 software. The data were tested for normality and homogeneity of variance using Shapiro–Wilk’s and Levene's test and Shapiro–Wilk's test, respectively. One way ANOVA (analysis of variance) using Duncan’s multiple range tests were applied in the present study. The data were analysed and significant at p < 0.05. Ethics approval The Institute (ICAR-National Institute of Abiotic Stress Management, Baramati, Pune) Research Advisory Committee (RAC) has approved the experimental procedures and this study compliance with Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines. Consent to participate All authors are aware and agree with this submission for publication.
Results 3.1. Effect of Mn on cortisol levels. In the present investigation, cortisol levels were assessed in response to dietary manganese (Mn) at 4, 8, and 12 mg kg −1 diet fed to P. hypophthalmus . The fish were reared under normal conditions as well as under the arsenic and ammonia pollution, coupled with high-temperature stress, over a period of 105 days. The corresponding data are illustrated in Fig. 1 A. Cortisol levels exhibited a noticeable increase ( p = 0.0025) in the group subjected to concurrent exposure to arsenic, ammonia toxicity, and high-temperature stress, followed by the group exposed to arsenic and ammonia, when compared to the control and other groups. Furthermore, dietary manganese at 8 and 4 mg kg −1 diet, with or without stressors (As + NH 3 + T), significantly reduced cortisol levels ( p = 0.0025) compared to the control and other groups. However, manganese at 12 mg kg −1 diet did not exhibit an inhibitory effect on cortisol levels in fish reared under both control and stressor conditions. Effect of Mn on Heat shock protein (HSP 70) and cytochrome P450 (CYP P450) The expression of the HSP70 gene in liver tissue exhibited a significant increase ( p = 0.0017) under concurrent exposure to ammonia and arsenic toxicity, along with high-temperature stress. This was followed by the group exposed to ammonia and high temperature, then arsenic and ammonia, ammonia alone, and arsenic alone groups, as compared to the control and Mn-supplemented groups. This observation held true for fish reared both in control conditions and under multiple stressors (As + NH 3 + T). Interestingly, the group fed with a Mn-containing diet at 8 mg kg −1 with stressors, followed by the same group without stressors, and then the Mn at 4 mg kg −1 diet, exhibited a significant difference compared to the control and other groups (see Fig. 1 B). Intriguingly, the expression of the CYP 450 gene in liver tissue was significantly upregulated ( p = 0.0013) in response to a combination of different stressors (As + NH 3 + T, NH 3 , As, As + NH 3 , and NH 3 + T) compared to the control and Mn-supplemented groups (Mn at 4, 8, and 12 mg kg −1 in the diet). Conversely, the supplementation of dietary Mn at 8 and 4 mg kg −1 , with or without stressors, resulted in a substantial downregulation of CYP 450 gene expression compared to the control and other groups (Fig. 1 C). Effect of Mn on DNA damage-inducible protein (DDIP), DNA damage and metallothionine (MT) DNA damage inducible protein (DDIP) exhibited a noticeable upregulation ( p = 0.0011) with concurrent exposure to ammonia, arsenic, and high-temperature stress. This was followed by the group exposed to ammonia and high temperature, and then the arsenic and ammonia, ammonia alone, and arsenic alone groups, in comparison to the control and other groups. Surprisingly, dietary Mn at 8 mg kg −1 in the diet, with or without stressors, demonstrated the ability to substantially downregulated DDIP gene expression. This effect was followed by Mn at 4 mg kg −1 in the diet, as compared to the control and other groups. However, Mn at 12 mg kg −1 in the diet was not as effective in modulating DDIP gene expression against multiple stresses (Fig. 1 C). Similarly, this study determined DNA damage in terms of tail DNA %, head DNA %, comet length, comet DNA, and head area, with the data recorded in Table 4 . The tail DNA % was significantly highest in the group exposed to As + NH 3 + T, NH 3 + T, As + NH 3 , NH 3 , followed by the As alone group, compared to the control and Mn-supplemented groups. Further, the noticeably least tail DNA % was determined in the group fed with Mn at 4, 8, and 12 mg kg −1 in the diet without stressors, followed by the same feeding group but with stressors. Similarly, the results of head DNA % were inverse to tail DNA %. Interestingly, the expression of the metallothionein (MT) gene was substantially downregulated ( p = 0.0002) with dietary Mn at 8 mg kg −1 in the diet, followed by Mn at 4 mg kg −1 in the diet, with or without stressors, in comparison to the control and other groups. However, concurrent exposure to As, NH 3 , and high temperature noticeably upregulated MT gene expression compared to arsenic and ammonia alone groups, in comparison to the control and Mn-supplemented groups (Fig. 1 C). Effect of Mn on caspase 3a and 3b (Cas 3a and 3b) The gene regulation of caspase 3a and 3b (Cas 3a and 3b) in liver tissue was significantly upregulated by concurrent exposure to ammonia, arsenic, and high-temperature stress. This upregulation was followed by the arsenic and ammonia alone group compared to the control and other treatments. Furthermore, the supplementation of Mn at 8 mg kg −1 diet, followed by Mn at 4 mg kg −1 diet, noticeably downregulated the Cas 3a ( p = 0.0052) and 3b ( p = 0.0003) gene regulations compared to the control and stressors group. However, Mn at 12 mg kg −1 in the diet was not effective for the gene regulation of Cas 3a and 3b (Fig. 2 A). Effect of Mn on tumor necrosis factor (TNFα) and immunoglobulin (Ig), (TLR), and interleukin (IL) gene regulation In the present investigation, the gene regulation of tumor necrosis factor (TNFα) and immunoglobulin (Ig) is presented in Fig. 2 B. The gene regulation of TNFα was significantly downregulated ( p = 0.0006) with the supplementation of dietary Mn at 8 mg kg −1 diet, with or without stressors. In contrast, TNFα was significantly upregulated, compared to the control and other groups, by concurrent exposure to As + NH 3 + T, As, NH 3 , As + NH 3 , and As + T groups. Moreover, the Ig gene expression was noticeably upregulated ( p = 0.0012) by the supplementation of Mn at 8 mg kg −1 diet, compared to other Mn-supplemented, control, and stressors groups. Further, the Ig gene expression was significantly downregulated with concurrent exposure to As + NH 3 + T and NH 3 + T, followed by As + NH 3 , NH 3 , and As alone groups, compared to the control and other groups. The toll-like receptors (TLR) ( p = 0.0046) and interleukin (IL) (0.0029) were substantially upregulated by concurrent exposure to ammonia, arsenic, and high temperature, followed by NH 3 + T, As + NH 3 , NH 3 , and As alone, in comparison to the control and Mn-supplemented groups. Furthermore, the gene regulation of TLR and IL was noticeably downregulated by Mn at 8 mg kg −1 diet, compared to the control and other groups (Fig. 2 C). 3.6. Effect of Mn on catalase (CAT), superoxide dismutase (SOD), glutathione-s-transferase (GST) and glutathione peroxidase of biochemical activities and gene expressions. This is subtitle caption, Please make the subtitled like others. In the present investigation, the activities of anti-oxidative enzymes such as catalase (CAT), superoxide dismutase (SOD), glutathione-s-transferase (GST), and glutathione peroxidase were determined in the liver and gill tissues of P. hypophthalmus reared under arsenic and ammonia toxicity, along with high-temperature stress. The corresponding data are recorded in Table 5 . CAT, GST, and GPx activities in the liver and gill were notably higher ( p < 0.01) in the group concurrently exposed to As + NH 3 + T, followed by NH 3 + T, compared to the control and supplemented groups. Similarly, CAT, GPx, and GST activities in the liver and gill were also higher in the group exposed to arsenic and ammonia, compared to the control and Mn-supplemented groups (4 and 8 mg kg −1 diet). Interestingly, the supplementation of dietary Mn at 8 mg kg −1 diet, with or without stressors (As + NH 3 + T), noticeably reduced CAT, GPx, and GST activities compared to the control and other treatment groups. The supplemented group with Mn at 4 mg kg −1 in the diet also effectively controlled CAT, GPx, and GST activities. Regarding SOD activities in the liver ( p = 0.018) and gill ( p = 0.037), they were significantly higher in the group treated with all stressors (As + NH 3 + T, NH 3 + T, As + NH 3 , NH 3 , and As) compared to the control and Mn-supplemented groups. The supplemented groups of Mn at 4, 8, and 12 mg kg −1 diet exhibited SOD activities similar to the control group in both liver and gill tissues. Interestingly, the gene expression of CAT, SOD, and GPx was also quantified in the present investigation, and the related data are noted in Fig. 3 A,B. The CAT ( p = 0.002), SOD ( p = 0.0061), and GPx ( p = 0.014) gene expressions were substantially upregulated with concurrent exposure to arsenic, ammonia, and high-temperature stress, followed by other stressor groups, in comparison to the control and Mn-supplemented groups. Furthermore, the gene expressions of CAT, SOD, and GPx were noticeably downregulated by Mn at 8 mg kg −1 diet with stressors (As + NH 3 + T), followed by the same diet group without stressors, Mn at 4 mg kg −1 diet, compared to the control and other treatment groups. Effect of Mn on inducible nitric oxide synthase (iNOS) and Na + K + ATPase gene expression In the present investigation, the inducible nitric oxide synthase (iNOS) in liver tissue was quantified, and the results are presented in Fig. 3 B. The iNOS gene expression was remarkably upregulated ( p = 0.0036) by the As + NH 3 + T group, followed by arsenic alone, compared to the control and other groups. Exposure to NH 3 , As + NH 3 , and NH 3 + T groups showed similar iNOS gene expression levels. Dietary supplementation of Mn at 8 mg kg −1 group noticeably downregulated iNOS gene expression, with or without stressors, followed by Mn at 4 mg kg −1 diet, compared to the control and other treatments. Additionally, the Na + K + ATPase gene expression was quantified in the present investigation, and the data are presented in Fig. 3 C. The gene expression of Na + K + ATPase was noticeably upregulated ( p = 0.0022) by As + NH 3 and As + NH 3 + T, followed by As alone and NH 3 + T, compared to the control and other groups. Moreover, Na + K + ATPase was significantly downregulated with the supplementation of Mn at 8 mg kg −1 diet, with or without stressors, compared to the control and other treatment groups. Effect of Mn on growth performance attributes and gene regulation In the present investigation, genes related to growth performance, such as growth hormone (GH), myostatin (MYST), somatostatin (SMT), growth hormone regulatory (GHR1 and GHRβ), and insulin-like growth factors (IGF1X1 and IGF1X2), were quantified, and the data are presented in Figs. 3 C and 4 A-C. GH was significantly upregulated ( p = 0.0016) by the supplementation of Mn at 8 mg kg −1 diet, with or without stressors, compared to the control and other treatment groups, including other Mn-supplemented diets. Moreover, GH gene regulation was noticeably downregulated by As + NH 3 + T, As + NH 3 , and NH 3 + T, in comparison to the control and supplemented groups. Furthermore, Mn at 4 and 12 mg kg −1 in the diet did not effectively regulate GH gene expression (Fig. 3 C). On the other hand, MYST ( p = 0.0023) and SMT ( p = 0.0042) gene regulations were remarkably upregulated by concurrent exposure to the As + NH 3 + T stress group, in comparison to the control and other treatment groups. Moreover, dietary supplementation of Mn at 8 mg kg −1 diet significantly downregulated MYST and SMT in the liver tissue of P. hypophthalmus , compared to the control and other treatment groups (Fig. 4 A). Surprisingly, GHR1 ( p = 0.0027), GHRβ ( p = 0.0033), IGF1X1 ( p = 0.015), and IGF1X2 ( p = 0.0072) genes were substantially upregulated with the supplementation of Mn at 8 mg kg −1 in the diet, with or without stressors (As + NH 3 + T), followed by Mn at 4 mg kg −1 in the diet, in comparison to the control and other treatment groups. In contrast, GHR1, GHRβ, IGF1X1, and IGF1X2 genes were significantly downregulated by stressors (As + NH 3 + T, As + NH 3 , NH 3 + T, As, and NH 3 ), compared to the control and other treatment groups (Fig. 4 B,C). In the present investigation, the growth performance indicators of P. hypophthalmus viz. final weight gain %, FCR, SGR, PER, DGI, TGC and RFI were presented in Table 6 . Results of final weight gain and SGR, PER, DGI and RFI were significantly reduced with concurrent exposure to arsenic, ammonia, and high-temperature stress, followed by NH 3 + T, respectively As + NH 3 , NH 3 , and As groups compared to control and Mn-supplemented groups. Further, the supplementation of Mn at 8 mg kg −1 diet with or without stressors was noticeably enhanced, followed by Mn at 4 mg kg −1 diet compared to control and other treatments. Whereas, the results of FCR were significantly inverse to SGR and PER. The Mn diet at 8 mg kg −1 diet was observed significantly lowest FCR followed by Mn at 4 mg kg −1 diet with or without stressors. However, the highest FCR was observed in the concurrent exposure to As + NH 3 + T and control fed group. Effect of Mn on LPO, Vit C and haematological parameters The results of lipid peroxidation (LPO) in the liver and kidney, and Vitamin C levels in muscle and brain, as well as the counts of RBC, WBC, and Hb in P. hypophthalmus reared under control conditions and multiple stressors (As, NH 3 , As + NH 3 , NH 3 + T, As + NH 3 + T), and fed with control and Mn-supplemented diets, are presented in Table 7 . LPO levels in the liver ( p = 0.0022) and kidney ( p = 0.0053) were significantly higher with concurrent exposure to ammonia, arsenic, and high temperature, followed by other stressors, compared to control and Mn-supplemented diets. Interestingly, Mn at 8 mg kg −1 diet, with or without stressors, significantly reduced LPO levels in the liver and kidney, followed by Mn at 4 mg kg −1 diet, compared to the control and other treatment groups. Vitamin C levels in muscle ( p = 0.015) and brain ( p = 0.021) were noticeably elevated with dietary Mn at 8 mg kg −1 diet, with or without stressors, compared to control and other treatment groups. Further, Vitamin C levels in muscle and brain were significantly lowered with concurrent exposure to arsenic, ammonia, and high temperature, followed by other stressor groups, compared to control and Mn-supplemented groups. Surprisingly, total blood counts such as RBC and Hb were significantly elevated with stressors (As + NH 3 + T, As + T, As + NH 3 , NH 3 , and As) compared to control and Mn-supplemented groups. Moreover, RBC ( p = 0.0066) and Hb ( p = 0.017) counts were noticeably reduced with dietary Mn at 8 mg kg −1 diet, followed by Mn at 4 mg kg −1 diet, with or without stressors, in comparison to control and other treatment groups. In contrast, WBC counts ( p = 0.019) were inversely related to RBC and Hb, as they were significantly reduced with different stressors such as arsenic, ammonia, and high temperature. Moreover, Mn supplementation considerably enhanced WBC counts (Mn at 8 and 4 mg kg −1 diet). Effect of Mn on immunological attributes The data on immunological attributes, such as nitroblue tetrazolium (NBT), blood glucose (BG), total protein, albumin, globulin, A:G ratio, and myeloperoxidase (MPO) in P. hypophthalmus reared under arsenic, ammonia, and high temperature, and fed with different levels of Mn, are presented in Table 8 . The levels of NBT ( p = 0.018), total protein ( p = 0.023), and globulin ( p = 0.011) were noticeably inhibited with concurrent exposure to ammonia, arsenic, and high-temperature groups, in comparison to all other groups. However, NBT was significantly reduced in the ammonia and high-temperature groups. Further, Mn at 8 mg kg −1 diet was noticeably elevated, with supplementation of Mn at 8 mg kg −1 diet, followed by other Mn-supplemented groups, in comparison to the control and other treatment groups. Whereas BG ( p = 0.0046) and A:G ratio ( p = 0.029) were significantly reduced with supplementation of Mn at 8 mg kg −1 diet, compared to the control, Mn at 4 and 12 mg kg −1 diet, and stressor groups. Similarly, levels of MPO were significantly elevated with Mn at 8 mg kg −1 diet, with or without stressors, followed by Mn at 4 mg kg −1 diet, compared to the control and other treatment groups. Effect of Mn on protein and carbohydrate metabolic enzymes In the present investigation, the data on ALT, AST, LDH, and MDH activities in the liver and gill, as well as acetylcholine esterase (AChE) in the brain of P. hypophthalmus , are recorded in Table 9 . ALT and AST activities in the liver and gill were noticeably elevated ( p < 0.01) with concurrent exposure to As, NH 3 , and high temperature, followed by As + NH 3 , NH 3 + T, and other stressor groups, compared to the control and Mn-supplemented groups. Moreover, ALT, AST, LDH, and MDH activities in the liver and gill were remarkably reduced with Mn at 8 mg kg −1 diet, with or without stressors, followed by Mn at 4 mg kg −1 diet, compared to the control and other treatment groups. Furthermore, Mn at 12 mg kg −1 diet was not effective in modulating the activities of ALT, AST, LDH, and MDH against multiple stressors. Effect of Mn on neurotransmitter Interestingly, the AChE activities in the brain were remarkably inhibited ( p = 0.0039) with concurrent exposure to arsenic, ammonia, and high temperature, followed by NH 3 + T, As + NH 3 , NH 3 , and As groups, in comparison to the control and Mn-supplemented groups. Conversely, AChE activities were noticeably elevated with Mn at 8 mg kg −1 in the diet, compared to the control, Mn at 4 and 12 mg kg −1 in the diet, and stressor groups (Table 9 ). Effect of Mn on bioaccumulation of arsenic The results of bioaccumulation and concentration of arsenic in different fish tissues and experimental water are presented in Table 10 . The arsenic concentration in water was determined to be the highest in the group treated under arsenic, ammonia, and high temperature and fed with a control diet (1776 μg L −1 ), followed by Mn at 12 mg kg −1 diet with stressors (As + NH 3 + T) (1337 μg L −1 ), the arsenic alone group (1186 μg L −1 ), and Mn at 4 mg kg −1 diet with stressors (1040 μg L −1 ) groups. Meanwhile, the bioaccumulation of arsenic was found to be highest in the liver and kidney tissues treated under As + NH 3 + T. Arsenic was below the detection limit in the groups treated with Mn at 4 and 8 mg kg −1 diet, as well as in the control groups, in muscle and brain tissues. Moreover, in the same diets but exposed to As + NH 3 + T, the arsenic concentration was the least in muscle and brain tissues. Furthermore, Mn bioaccumulation was highest in the Mn-12 mg kg −1 diet, followed by Mn at 4 and 8 mg kg −1 in the diet. Effect of Mn on bacterial infection After the experiment, the fish were infected with the Aeromonas hydrophila . Cumulative and relative % survival was determined up to seven days after the fish were infected. Cumulative mortality was observed to be higher in the group treated with concurrent exposure to As, NH 3 , and T, followed by As + NH 3 (63), NH 3 + T (61), and Mn at 12 mg kg −1 in the diet with stressors (61). In contrast, the least mortality was observed in Mn at 8 mg kg −1 in the diet (27) with stressors (38). Similarly, the relative % survival was observed as -37.5, -37.5, -48.3, -37.5, -56, 0, 27, -25, -12.5, 12.5, -37% for As, NH 3 , As + NH 3 , NH 3 + T, As + NH 3 + T, Mn at 4, 8, and 12 mg kg −1 in the diet, and Mn at 4, 8, and 12 mg kg −1 diet with stressors, respectively (Fig. 5 ).
Discussion Cortisol is secreted from the inter-renal cell of head kidney released directly into the blood 55 . It is an important stress hormone plays a role in growth, reproduction, and osmoregulation 56 . The multiple stressors (As + NH 3 + T) induces stress in the fish which needed energy to combat the stress which compensated through glucogenic pathway. The stress indues by As + NH 3 + T, that elevated cortisol levels could be due to the stressors targeting the multiple sites in the hypothalamus-pituitary-interrenal axis and altered the adrenocorticotropic hormone (ACTH) secretion 57 – 59 . Moreover, the dietary Mn at 8 mg kg −1 diet followed by 4 mg kg −1 diet noticeably reduced the cortisol levels which could be due to Mn support the energy provided to glucogenic pathway 60 . During this stage the physiological change also occurred and adapted by fish and homeostasis return which was supported by Mn diet. HSP 70 are chaperones protein, maintain the normal structure and function of cell protein and help in folding protein into unfolded protein 61 and indicating physiological conditions of fish during stress. HSPs protein expression is generally upregulated during temperature and metal stress 62 , 63 . Indeed, the Mn-containing diet downregulated the HSP 70 expression might be due to its role in ApoA-1 gene regulation. It is also proven that if brain could not control oxidative challenges, heat shock protein upregulates the HSPs during stress. The present study showed that Mn at 8 mg kg −1 diet helps in maintenance of the cellular homeostasis through correct folding of nascent and stress-accumulated misfolded proteins in the cell 64 . This might be due to activating the transcription factor of HSP by Mn as it loses the binding activity of heat shock elements, and thus Mn downregulates the HSP70 expression 65 , 66 . The cytochrome P450 is the heme-thiolate protein, is a major component of the membrane-bound microsomal monooxygenase system (MMO), which helps in catalyzes the oxygenation of exogenous and endogenous compounds (Xenobiotics, drugs, and carcinogens) 67 . Moreover, the toxicity of arsenic, ammonia, and high-temperature stress upregulated the CYP 450 gene expression as it generates ROS, which potentially causes lipid peroxidation, cell toxicity, and death. Interestingly, it is inferred that CYP 450 involved in the arsenic, ammonia toxicity and high temperature stress which take part for apoptosis and upregulated the transcription of bcl2-associated X (Bax) 68 . Bax is the important cell death promoting gene in fish which induce release of cytochrome c, leading to caspase activation 69 . Surprisingly, Mn at 8 mg kg −1 diet was remarkably downregulated the CYP 450 gene expression might be due to it regulating and controlling the generated reactive oxygen systems (ROS), cytokines regulation, and lipid peroxidation. In the present study, results of ROS, LPO, and cytokines regulation supported the role of Mn in the control CYP 450 regulation. DNA damage and DNA damage inducible protein (DDIP) gene expression was upregulated by ammonia, arsenic and high temperature stress could be due to extensive generation of reactive oxygen species, dysregulation of cell proliferation, apoptosis, diminished DNA repair, aberrant in histone post-translational modification and DNA methylation 70 . However, the dietary Mn at 8 mg kg −1 diet protect against DNA damage and downregulated DDIP might be due to at this lower dose of Mn enhances the viability of SH-SY5Y cells, reduced the ROS production and LPO levels as well as enhances GSH levels 71 . Interestingly, the MT gene expression was highly upregulated by arsenic, ammonia, and high temperature stress and downregulated by Mn diet. Moreover, the higher dietary Mn induces overexpression of MT gene 72 . Apoptosis indicates cell programming death in which Cas 3a and 3b belong to the apoptosis gene. The Cas 3a and 3b were upregulated due to arsenic, ammonia, and high-temperature stress could incur apoptosis using p53 and regulated the apoptosis in the liver tissue 73 and upregulated gene related to oxidative stress and inflammatory response as shown in the present study. Indeed, the supplementation of Mn at 8 and 4 mg kg −1 diet help in controlling the regulation of Cas 3a and 3b, which might be due to it has a role in the activation of the caspase cascade and DNA fragmentation in the liver cell 74 . The present study revealed that stressors (As + NH 3 + T) reduced the immunity of the fish through cytokines gene upregulation such as TNFα, TLR and IL and downregulated the Ig gene. Zhang et al 75 . reported that the ammonia toxicity altered the immunity of the fish. The stressors induced the stress and showed higher inflammation rate in liver tissue in fish and hence the higher upregulation of TNFα, IL and TLR was determined in the present study 2 . The TNFα, IL and TLR acts as an essential pro-inflammatory cytokine that enhances the immunity in aquatic animals including fish 76 . Notably, the Mn at 8 mg kg −1 diet was improved the immunity of the fish using the strengthening/downregulating the TNFα, IL and TLR. This might be due to Mn have an important role in immunostimulants and activating the NF-κB signaling pathways to enhance immunity of the fish against multiple stresses. In contrast to the results of TNFα, IL, TLR and Ig was downregulated with stressors (As + NH 3 + T, NH 3 + T, As + NH 3 , NH 3 and As) and upregulated by Mn diet. This could be due to the role of Mn in enhancing humoral and cell-mediated immunity and improving antibody affinity, early β cell development, complement system, cell mediated immunity, phagocytose activity, and antibody reaction. The present study pointed out the remarkable reduction of oxidative stress enzymes (SOD, CAT, GST, and GPx activities) and gene expression of SOD, CAT, and GPx through Mn diet at 8 mg kg −1 . This could be due to the role of manganese in substituting as a cofactor for iron in certain enzymes which is responsible for oxidative stress elevation 77 . Mn is also a cofactor for many enzymes, including pyruvate carboxylase and manganese superoxide dismutase (Mn-SOD). It also protects the cell against reactive oxygen species (ROS) as Mn is part of metalloenzyme by catalyzing the one-electron reduction of peroxide anion to hydrogen peroxide 78 . Mn is found in the Mn-SOD complex, which is useful in maintaining the structure of antioxidant enzymes 79 , 80 , affecting the Fenton reaction. Mn also enhances the organism's anti-oxidative status through synthesizing and activating certain enzymes such as oxidoreductase transferase, hydrolases, and ligase, as well as vitamins C and B. It also involved metalloenzymes such as arginase, glutamine synthetase, phosphoenol pyruvate, and decarboxylase 81 . Moreover, Mn is the essential mineral nutrient for managing aquatic animals' oxidative stress. Interestingly, iNOS gene expression was notably highly upregulated by stressors (As + NH 3 + T, NH 3 + T, As + NH 3 , NH 3 and As) could be due to higher accumulation of NH 3 in fish tissues. Moreover, the NH 3 is converted into urea via ornithine-urea cycle (OUC) and then converted into glutamine via the glutamine synthetase including non-essential amino acids 82 . Similarly, the blood carrying the high ammonia concentration and affecting the liver tissue 83 . Moreover, nitric oxide provided protection to cellular system against oxidative stress 84 . Moreover, the dietary Mn at 8 mg kg −1 diet remarkably downregulated the iNOS gene expression in liver tissue. Further, during stress condition, the organism needs more energy in the form of ATPase, therefore the gene expression of Na + K + ATPase was highly upregulated. Notably, the Mn diet help in formation of more ATPase and supplied to the fish reread under multiple stress condition. The growth performance related gene expression viz. GH, GHR1, GHRβ, IGF1X1 and IGF1X2 were remarkably downregulated by stressors (As + NH 3 + T, NH 3 + T, As + NH 3 , NH 3 and As) could be due to disruption of endocrine receptor which control the growth related gene expression. The GH gene bind with GHR and controlled by hypothalamic regulation viz. GH-releasing hormone, ghrelin, dopamine and somatostatin 85 , 86 . The growth-related genes mainly regulated by genetically, endocrinologically and environmentally. It is also related with better nutrition, optimum temperature, good husbandry condition and better functioning of endocrine regulation 87 . It is also observed that the Mn diet notably downregulated MYST and SMT at 8 mg kg −1 diet. It might be due to the role of MYST in decreasing the myoblast, which results in terminal differentiation and division of fiber enlargement 88 . Further, the IGF1X1 and IGF1X2 gene expressions have important role in biomolecular regulation such as carbohydrates, lipid, protein, and mineral metabolism, differentiation and proliferation of the cell and ultimately growth 89 . As the GH bind to the receptor in the liver cell to stimulate, release and synthesize IGF gene expression and dietary Mn help in this process, the Mn diet is responsible for growth enhancement and biomolecular function in the cell of aquatic organism. The stressors (As + NH 3 + T, NH 3 + T, As + NH 3 , NH 3 , and As) drastically inhibited the growth performance (final weight gain %, FCR, SGR, PER, DGI, TGC, and RFI) of the fish might be due to arsenic and ammonia toxicity, and high-temperature stress reduces the feed intake and metabolic rate, which was reported by our previous study 2 . Interestingly, the Mn diet improved growth performance could be due to the role of Mn in improving feed efficiency, feed utilization, growth rate, and immunity of the fish. It also improved the specific growth rate, daily growth index %, relative feed intake and protein efficiency in the fish 90 . Moreover, the deficiency and inadequate supply of Mn result in reduced growth rate, reduced feed intake, skeletal abnormalities such as dwarfism and cataract in fish 91 . Mn has also provided the uptake of glucose, insulin receptors and triglyceride synthesis 92 . However, the dietary Mn at optimum levels is beneficial for growth enhancement of fish reread under control and stressed environment. The present study revealed that ammonia and arsenic toxicity and high-temperature stress elevated the LPO level in the liver and kidney tissues might be due to the formation of ROS by stressors. A free radical producing system mainly generates it. Excessive ROS generation may cause oxidative stress and damage critical biomolecules, resulting in deleterious biological effects 93 . Moreover, the dietary Mn at 4 and 8 mg kg −1 diet remarkably reduced the ROS and LPO levels. Similarly, the muscle and brain Vit C were noticeably elevated by dietary Mn at 4 and 8 mg kg −1 diet. It is crucial for collagen synthesis and in metabolism of biomolecules, including steroids and detoxification of xenobiotics 94 . Therefore, Mn is important in maintaining the Vit C in fish tissues. The blood profiling viz. Hb, WBC and RBC were important component which were altered by arsenic, ammonia and high temperature, whereas, the dietary Mn at 8 mg kg −1 diet was corrected the count of Hb, WBC and RBC. Hb helps in aerobic metabolism, distribution of the gases and maintenance of the physiological attributes in the fish via fish growth and health 95 . The RBC helped absorb oxygen through gill and circulated in the different tissues in the body. Stress changed not only the RBC count but also the shape of the RBC. Further, the WBC is an important component for acquired and innate immune response. It constituted eosinophils, neutrophils, lymphocytes, monocytes, and basophils. Hence, the dietary Mn enhances the WBC count in fish reared in control or stress conditions. NBT, blood glucose, total protein, albumin, globulin, A:G ratio, and MPO are important attributes of immunity. In the present study, stressors altered the immunity, whereas the dietary Mn approved the immunity in the fish. NBT indicates the health of the fish as elevated levels mention higher immunity. It involved the phagocytes for intercellular superoxide radicals produced by leucocytes 96 . Moreover, the globulin are also major component and four types such as α 1 , α 2 , β and γ 97 , which the gamma globulin is essential for blood immunological protein 98 . Further, albumin helps in transportation of hormones, metal, bilirubin, drug and vitamin. It also regulates the free available hormones 99 and fat metabolism. Interestingly, the supplementation of Mn diet enhances the production of B-lymphocytes and it maintained the higher immunity of the fish. However, MPO is the haemoprotein and important during respiratory burst using H 2 O 2 to produce hypochlorous acid 100 . Hypochlorous acid is a potent oxidant that elicit the cytotoxic effect on bacterial cells 101 . Moreover, the Mn containing diets helps in released of neutrophils and O 2 derived species (H 2 O 2 ) and H 2 O 2 to oxidize Cl - ions to form HOCl. Moreover, the blood glucose are indicators for good health and Mn diet improved the BG level after exposure to stressors. The role of Mn in regulating blood glucose might be due to, it enhances the gluconeogenesis viz. synthesis of glucose from non-carbohydrate source mainly protein and amino acid, and the enhancement of secretion of catecholamine 102 . The carbohydrate and protein metabolic enzymes viz. LDH, MDH, ALT and AST activities were notably elevated whereas the dietary Mn at 4 and 8 mg kg −1 diet reduced the activities. This might be due to Mn fulfilled the energy demand during stress conditions, and hence, it reduces the activities of LDH and MDH. Notably, the LDH is the glycolytic enzyme that catalyzes the interconversion of pyruvate and lactate using the nicotinamide adenine dinucleotide (NAD) as a coenzyme. Moreover, the MDH is the limiting enzyme for the oxidative catabolism of carbohydrates 103 . The ALT and AST activities were also reduced by the Mn diet, possibly because Mn is required for many cofactors for biomolecular enzymes. The stressors (As + NH 3 + T, As + T, As + NH 3 , NH 3 , and As) significantly inhibited AChE activities might be due to arsenic, ammonia, and high temperature preventing the hydrolysis of acetylcholine 58 . Moreover, acetylcholine helps dominate cholinergic synapses and neuromuscular junctions in the fish's central nervous system (CNS). This results in the hydrolysis of acetylcholine and choline after the activation of acetylcholine receptors at the postsynaptic membrane 104 . Surprisingly, the AChE activities were improved by dietary Mn. It also showed that a higher Mn diet at 12 mg kg −1 diet significantly inhibited AChE activities, which might be due to its nature to induced the toxicity to the postsynaptic membrane. The stressors group, such as arsenic, ammonia toxicity, and high-temperature stress, enhances arsenic bioaccumulation in the fish, whereas the Mn diet at 8 mg kg −1 diet reduced the arsenic bioaccumulation. This might be due to the ability of Mn to enhance the detoxification of arsenic in all tissues. Moreover, the kidney and liver tissues had higher arsenic bioaccumulation reported in the present investigation. These results revealed that Mn could detoxify arsenic efficiently in all the tissues. The present study also showed that dietary Mn at 8 mg kg −1 diet enhanced fish survival after an infection of a bacterial pathogen. The results of the present study showed the Mn diet improved the antioxidant and immunity of the fish. However, this might be the reason for the higher survival of the fish against pathogenic infection after the dietary application of Mn. It is also reported that Mn helps generate neutrophils, which enhances the effector cells function for defencing the fish against pathogenic bacteria 105 .
Conclusion The present study is the first report on the role of manganese (Mn) on gene regulations and biochemical regulators in response to arsenic and ammonia toxicity and high-temperature stress in P. hypophthalmus . The immunity, anti-oxidative status, growth performance, genotoxicity, and other stress-responsive genes were controlled and regulated by dietary Mn at 8 mg kg −1 diet. Mn at 8 mg kg −1 diet efficiently regulates cortisol, HSP 70, and apoptosis and protects against genotoxicity. Mn at 8 mg kg −1 diet is also efficient in enhancing the detoxification of arsenic in different fish tissues. Moreover, the results revealed that Mn at 8 mg kg −1 efficiently controls the gene regulation involved in the multiple stressors (As + NH 3 + T). Indeed, dietary Mn at 8 mg kg −1 diet improved gene regulation, maintained fish hemostasis, and noticeably reduced the bioaccumulation of arsenic in fish tissues. Overall results of the present investigation concluded that Mn at 8 mg kg −1 diet should be included in the fish diet to maintain gene regulation of the NFkB signaling pathway and mitigate the multiple stresses in fish.
The ongoing challenges of climate change and pollution are major factors disturbing ecosystems, including aquatic systems. They also have an impact on gene regulation and biochemical changes in aquatic animals, including fish. Understanding the mechanisms of gene regulation and biochemical changes due to climate change and pollution in aquatic animals is a challenging task. However, with this backdrop, the present investigation was conducted to explore the effects of arsenic (As) and ammonia (NH 3 ) toxicity and high-temperature (T) stress on gene regulation and biochemical profiles, mitigated by dietary manganese (Mn) in Pangasianodon hypophthalmus . The fish were exposed to different combinations of As, NH 3 , and T, and fed with dietary Mn at 4, 8, and 12 mg kg −1 to evaluate the gene expression of immunity, antioxidative status, cytokine, and NfKB signaling pathway genes. HSP 70, cytochrome P450 (CYP 450), metallothionein (MT), DNA damage-inducible protein (DDIP), caspase (CAS), tumor necrosis factor (TNFα), toll-like receptor (TLR), interleukin (IL), inducible nitric oxide synthase (iNOS), catalase (CAT), superoxide dismutase (SOD), and glutathione peroxidase (GPx) were noticeably highly upregulated by As + NH 3 + T stress, whereas Mn diet at 8 mg kg −1 downregulated these genes. Further, total immunoglobulin (Ig), myostatin (MYST), somatostatin (SMT), growth hormone (GH), growth hormone regulator 1 and β, insulin-like growth factors (IGF1X1 and IGF1X2) were significantly upregulated by Mn diets. The biochemical profiles were highly affected by stressors (As + NH 3 + T). The bioaccumulation of arsenic in different tissues was also notably reduced by Mn diets. Furthermore, the infectivity of the fish was reduced, and survival against pathogenic bacteria was enhanced by Mn diet at 8 mg kg −1 . The results of the present investigation revealed that dietary Mn at 8 mg kg −1 controls gene regulation against multiple stressors (As, NH 3 , As + NH 3 , NH 3 + T, As + NH 3 + T) in fish. Subject terms
Acknowledgements The present work was supported by Indian Council of Agricultural Research (ICAR), New Delhi, India under Project “Lal bahadur Shastri Young Scientist Award (Project code: OXX5181). Authors also sincerely acknowledged to Director ICAR-NIASM for providing research facilities for this work. Author contributions N.K., Conceived and designed the experiments; performed the experiments; analysed the data; contributed reagents/materials/analysis tools; wrote the paper S.T.T. Perform gene analysis S.A.K. Data Validation K.S.R. Supervision and editing. Funding The present work was supported by Indian Council of Agricultural Research (ICAR), New Delhi, India under Project “Lal bahadur Shastri Young Scientist Award (Project code: OXX5181). Data availability The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1273
oa_package/a7/4f/PMC10787825.tar.gz
PMC10787826
38218973
Introduction Ki-67 immunohistochemistry (IHC) serves as a reliable marker of cell proliferation and is widely used to evaluate the aggressiveness and prognosis of human tumors. Notably, Ki-67 has been adopted for prognostication in breast cancer, with elevated Ki-67 expression correlating with poorer prognosis 1 , 2 . The Ki-67 proliferation index (PI) in breast cancer is a measure of the percentage of tumor cells with nuclear immunoreactivity relative to the total number of malignant cells assessed 3 . A meta-analysis of 64,196 patients revealed that higher Ki-67 PI values are associated with worse overall survival in breast cancer, with 25% being a cutoff of strong outcome prognostication 4 . The monarchE committee reported that among patients with early-stage HR+, HER2− breast cancer, and nodal involvement, the addition of abemaciclib to hormone therapy significantly improves cancer-specific free survival and decreases the risk of disease recurrence 5 – 7 . For tumor stage 1 to 2, nodal stage 0 to 1, ER+/HER2− breast cancer, the International Ki-67 in Breast Cancer Working Group’s (IKWG) consensus in 2021 recommended using Ki-67 to aid in the decision-making of adjuvant chemotherapy only for cases with a very low (< 5%) or very high (> 30%) PI due to substantial inter-rater variability within this range 8 , 9 . The panelists of the St. Gallen International Consensus Conference in 2021 generally support this recommendation 10 . The monarchE phase III clinical trial studied the impact of a high Ki-67 PI on disease recurrence in a cohort of patients with HR+/HER2− node-positive breast cancer with high-risk clinicopathological features (at least 4 positive lymph nodes, or 1 to 3 positive lymph nodes with either tumor size ≥ 5 cm or histological grade 3 disease). Their analyses demonstrated that a Ki-67 PI ≥ 20% in patients treated with endocrine therapy alone was associated with a significantly increased risk of recurrence within three years compared to patients with lower Ki-67 expression 6 , 11 . Following this, the American Food and Drug Administration and Health Canada approved the use of abemaciclib (CDK4/6 inhibitor) for patients with HR+/HER2− high-risk early breast cancer and a Ki-67 PI of ≥ 20% 12 . In a recently published landmark study 13 based on 500 patients, it was demonstrated that a PI score threshold of < 13.25% derived from Ki-67 slides effectively identified women with luminal A breast cancer who could be safely treated without local breast radiation therapy. This underscores the clinical significance of Ki-67 as a marker with significant promise in guiding management decisions for breast cancer patients. The current gold standard for quantifying Ki-67 PI is to manually evaluate at least 500 malignant cells based on IKWG recommendations 8 , 9 . However, this method is labor-intensive, time-consuming, and prone to poor inter-rater reproducibility and errors 14 , 15 . As a result, it is hard to standardize and use Ki-67 for clinical assessments. As shown in the recommendations from the IKWG 8 , 9 , the assessment by the pathologist is most reliable for PI values below 5% and above 30% (the 5 to 30% range is subject to the most interpretation variability). The Canadian Association of Pathologists recommends that a second pathologist assess PIs in this range, or use a computer assessment tool to improve robustness 12 . Considering this range is critical for treatment decisions, its reliability must be improved. The recent emergence of digital pathology and high-performance AI algorithms offers the possibility that automated PI scoring can overcome these challenges by accurately and efficiently measuring cell count. There have been several AI-based Ki-67 assessment tools developed 16 – 21 , and the advantages are becoming increasingly evident. Several comparative studies have reported the role of AI-assisted assessments of Ki-67 PI in breast cancer 19 , 22 , 32 . These studies demonstrated that AI-aided assessment of Ki-67 could achieve a lower mean error 23 and a lower standard error deviation 19 , however, the impact on inter-rater agreement is less clear. Additionally, while these studies have encompassed broad PI ranges from 0 to 100%, the effect of AI assistance in the clinically crucial 5 to 30% PI interval has not yet been studied. Herein, we conducted a large-scale, international study that analyzed the effects of AI assistance on key aspects of pathologists' work, including accuracy, inter-rater agreement, and turnaround time (TAT) in the context of Ki-67 scoring for breast cancer. Our focus was on assessing these metrics within the 5 to 30% PI range to better understand the implications and usability of AI-assisted Ki-67 evaluations. Additionally, we gathered insights into pathologists' perspectives, trust levels, and readiness to adopt AI technologies, highlighting the importance of user acceptance. This study provides a strong foundation for understanding the future impact and potential of AI tools for Ki-67 scoring in the daily routine of pathologists.
Materials and methods Ethics approval for the study was obtained from Toronto Metropolitan University (REB: 2022-154). All experiments were performed in accordance with the Tri-Council policy statement 2 for the ethical conduct of research involving humans. Case selection, TMA preparation, and image acquisition A subset of ten TMAs from the Toronto-British Columbia trial was used for this study 24 , which was composed of node-negative patients above the age of 50 years with invasive breast cancer 25 . Tissue microarrays (TMAs) were constructed using a 0.6 mm tumor core procured from formalin-fixed, paraffin-embedded specimens. TMA sections, with a thickness of 0.5 μm, were stained using a 1:500 dilution of SP6 (ThermoFisher Scientific, Waltham, MA, USA)—a Ki-67 antibody—and counterstained with hematoxylin. The study incorporated ten TMAs with high tumor cellularity, averaging 2093 neoplastic cells per TMA and a PI range of 7 to 28% 16 . This range, which poses a challenge for pathologists, encompasses the clinically relevant PI cutoffs identified in prior studies 6 , 11 , 13 . AI tool A deep learning-based AI tool for IHC quantification, UV-Net, developed by Toronto Metropolitan University, was used in the study 18 . This tool detects neoplastic cells in IHC-stained tissue and differentiates Ki-67 positive from Ki-67 negative tumor cells. Its underlying architecture, a modified U-Net, includes additional connections for densely packed nuclei and replaces the standard 3 × 3 convolutional layers with 'V-Blocks'. These V-Block connections maintain high-resolution nuclear features for precise differentiation between nuclei; each V-Block inputs n channels and outputs 2n channels, forming a 'V' shape across four successive stages. The AI tool was trained using 256 × 256 RGB patches of WSIs from St. Michael's Hospital, and an open-source dataset "Deepslides" 26 from × 20 Aperio AT Turbo and × 40 Aperio ScanScope scanners respectively. Images were annotated with single-pixel centroid markers distinguishing Ki-67 positive and Ki-67 negative tumor nuclei cells 20 , defining positive nuclei as any brown color above the background, following the IKWG’s recommendations 8 , 9 . Single-pixel markers were extended into circular areas using a Gaussian function, this allocated the highest value to the center of the nuclei, incorporated more contextual information, and improved the efficiency of the training process. A Huber loss function was used to regress and predict the centroid of nuclei. For a given image, the AI tool generates an automated Ki-67 positive and Ki-67 negative overlay (Fig. 1 ), providing an accessible visual interpretation along with the automated PI calculation. The generalizability of UV-Net was previously validated on multi-institutional datasets from 5 institutions 18 , including WSIs and TMAs from breast cancer images. UV-Net consistently outperformed other architectures across all image variations, registering an average F1-score of 0.83 on expertly annotated data. In comparison, alternative architectures achieved scores between 0.74 and 0.79. The images on which UV-Net was trained differed from those used in this study, originating from datasets with different scanners and institutions. None of the pathologists involved in this study participated in annotating the training or validation datasets. Study design A cross-sectional study was performed using an anonymous, self-administered, and structured online survey developed using QualtricsTM, which included hyperlinks for viewing digitized TMAs on the cloud through PathcoreFlowTM, a browser-based commercial image management solution and viewer for digital pathology 27 . The AI tool for Ki-67 scoring was integrated into PathcoreFlowTM using an Application Programming Interface. The tool provided an overlay of the Ki-67 positive and negative nuclei and calculated PI scores (Fig. 1 ). Participants were presented with a digital invasive breast cancer TMA stained for Ki-67 for each question and were asked to assign a Ki-67 score by entering a percentage value into QualtricsTM. Examples of questions with and without AI assistance are shown in Supplementary Figs. 1 and 2 . Each Ki-67 TMA was reviewed by respondents twice—once without AI assistance and once with AI assistance—resulting in a total of 20 assessment questions. Participants were not explicitly told to use the AI, but rather to observe the AI results and estimate their PI score. They were instructed to compute the Ki-67 PI by counting individual cells with a denomination of 500 cells and to regard any brown staining beyond the background as positive, in line with current guidelines 8 , 9 . They were also guided to spend approximately the same time they would during standard procedures with no limit on the time for assessments. Each pathologist used a distinct viewer from a separate workstation. To minimize bias, the order of cases was randomized, ensuring that TMAs with AI assistance were not shown immediately before or after the same TMA without assistance. Additionally, the AI-assisted images were altered in orientation to look different from the unaided images. At the end of the study, participants were requested to provide their demographic information and respond to inquiries regarding their perspectives on AI. Study population Participants were recruited through the professional networks of the authors between September and November 2022. Contact channels included pathology associations, local pathology residency programs, pathologist colleagues, and social media platforms (LinkedIn, Twitter). Eligible participants were trained pathologists with experience in Ki-67 PI scoring. The study included all participants who provided consent and identified themselves as pathology specialists. There were no limitations based on gender, age, or employment status, and only those who finished the study were considered, in total there were 116 completed responses. Spurious responses defined as outliers with large PI errors (more than 20% on a single response) were excluded from the analysis (N = 26 participants). Consequently, the main analysis included 90 respondents, all experienced in using digital pathology. Demographic characteristics are described in Supplemental Table 1 . The participants' median age ranged from 40 to 49 years; however, the most common age group was 30 to 39 years, accounting for 34.4% of the respondents. While the median work experience falls within the 10 to 19 years range, the most prevalent work experience category is 0 to 9 years, representing 26.7% of the total. The majority of respondents are male, with many being retired clinical pathologists from North America. Among those currently working, most practice in academic health sciences centers. Ground truth scores The ground truth Ki-67 PI scores for the 10 TMAs were determined using the gold standard manual counting method, where any brown staining above the background level was deemed positive, following current guidelines 8 , 9 . Each TMA was divided into five rows and five columns, creating 400 × 400 pixel tiles, and annotations were made in each region. Nuclei were annotated at the center of each cell, with tumor cells marked as Ki-67 positive if any discernible brown staining above the background was observed and the cell border was visible; otherwise, they were marked as Ki-67 negative. In cases of overlapping tumor cells, each cell was marked individually if its borders were discernible. An anatomical pathology resident (N.N.J.N.) performed the manual annotations, which were verified by a breast pathologist (S.D.). Ground truth PI scores were calculated from these manual annotations. The ground truth PI scores of the ten cases ranged from 7 to 28%. Statistical analysis Statistical analyses were performed to assess the PI scoring error, inter-rater agreement, and TAT among pathologists when using the AI tool, compared to a standard clinical workflow (i.e., without AI). The experiment involved two groups: a control group where pathologists evaluated Ki-67 PI using standard clinical methods, and an experimental group where the same pathologists used the AI tool to assist with Ki-67 PI assessment on the same TMAs. For each participant, two PI estimations and TATs were obtained per TMA, resulting in 900 paired assessments (90 pathologists × 10 cases). For every assessment, several metrics were recorded, including the clinician-estimated raw PI score, the PI error (the absolute difference between the estimated and ground truth PI), and TAT, which denotes the time taken to score the TMA. The paired Wilcoxon signed-rank test 28 was used to compare the differences between the two groups, with significance determined based on the median values of the paired differences. This test was chosen due to the non-normal distribution of the data, as indicated by the Shapiro–Wilk test. All statistical analyses were two-sided, with significance set at p < 0.05. PI scores and PI errors were assessed with and without AI assistance, using continuous and binary values. PI scores and PI errors were first treated as continuous values and summarized by the mean and standard deviation. Box and bar plots were used to visually depict case-based and sub-demographic PI errors, respectively. PI scores and errors were additionally binarized and assessed using low-risk Ki-67 PI < 20%, and high-risk ≥ 20% stratification 12 . The consistency of scoring among pathologists, with and without AI assistance, was examined using both continuous and binary metrics. For the continuous analysis, the Two-Way Random-Effects Model for single-rater consistency agreement was chosen to assess the inter-rater agreement using the Intraclass Correlation Coefficient (ICC) 29 , 30 . This model was selected since all cases were evaluated by all raters. The choice of the single-rater model stemmed from the clinical reliance on a singular clinician's decision for Ki-67 scores, rather than averaging scores from multiple clinicians 29 . The ICC between the pathologists’ PI scores and the ground truth PI was assessed twice: once with and once without AI assistance. Complementary to ICC, Krippendorff's α was calculated to measure inter-rater agreement and chosen for its adaptability in handling continuous data 31 . Bland–Altman and linear regression plots of the PI scores were incorporated to supplement the measure of inter-rater agreement, with parameters such as Pearson’s correlation coefficient, slope, offset, mean, and limits of agreement being considered. Using binarized PI scores (with scores ≥ 20% assigned a 1, and scores < 20% a 0), the percent agreement and Fleiss’ Kappa 32 were calculated for both groups. The TAT among pathologists was considered the time in seconds to perform the PI score estimation, starting from the moment they began examining the case to the point when the PI score was saved. TATs were summarized by the mean and standard deviation. Box and bar plots were used to visually depict case-based and sub-demographic TATs, respectively. Additionally, the percentage of time reduction computed by the time savings was determined by the formula: (total time saved/total time spent on conventional assessment) X 100%. Statistical analyses were performed using SPSS Version 28 (Armonk, NY, USA). Ethics approval This research study has been reviewed by the Toronto Metropolitan University Research Ethics Board (REB 2022-154). Participants voluntarily consented to participate and to share contact information if they wanted to.
Results Scoring accuracy The respondents' PI scores and PI errors per case and within ranges are shown in Table 1 . Responses including outliers are shown in Supplementary Table 2 . The overall mean PI error was found to be 2.1 (2.2) using the AI tool, and 5.9 (4.0) without the AI (difference of − 3.8%, 95% CI: −4.10% to −3.51%, p < 0.001). The AI tool significantly improved the accuracy of PI scoring. The PI error was plotted per case (Fig. 2 A) and for each PI interval (Supplementary Fig. 3 ). Cases 2 through 10 had significantly less error ( p < 0.001), and both the < 20% and ≥ 20% PI ranges had statistically significant decreases in error with AI ( p < 0.001). Furthermore, Fig. 2 B, C, which display the PI error across various demographics, revealed that AI-aided scoring was superior across all pathologist age ranges and experience levels—indicating that despite variable background and training, AI improved PI accuracy for all groups of pathologists. Supplementary Fig. 4 shows that AI-aided scoring was statistically superior ( p < 0.001) across all pathologist subdisciplines. The AI tool demonstrated high accuracy in the study, with a mean PI error rate of 0.6%, which ranged from 0.0 to 6.1%, as shown in Table 1 . To quantify the increase of PI estimation accuracy when pathologists used the AI tool, Supplementary Fig. 5 shows the difference in PI error for each case. This difference is calculated as the PI error for the estimated PI score with AI assistance minus the error without AI assistance, highlighting the extent to which the AI tool reduces error rates. Most pathologists experienced increased accuracy with the AI tool, as indicated by positive differences seen in Supplementary Fig. 5 . Inter-rater agreement AI assistance led to a significant improvement in inter-observer reproducibility (with AI assistance: ICC = 0.92 [95% CI 0.85–0.98], Krippendorff’s α = 0.89 [95% CI 0.71–0.92], without AI assistance: ICC = 0.70 [95% CI 0.52–0.89], Krippendorff’s α = 0.65 [95% CI 0.41–0.72]). These statistics are visually depicted in Supplementary Fig. 6 . Bland–Altman analyses (Fig. 3 B, D) revealed that pathologists with AI assistance exhibited less bias (mean of 0.7 vs. − 2.7) and tighter limits of agreement (6.5 to − 5.1 vs. 10.2 to − 15.6) compared to the ground truth scores. Linear regression models (Fig. 3 A, C) further support the notion that AI assistance improves inter-rater agreement (with AI assistance: y = 1.06x − 0.46, r = 0.92, SSE = 7792; without AI assistance: y = 0.64x + 3.62, r = 0.58, SSE = 33,992). After binarizing the pathologists' responses, with scores ≥ 20% assigned as 1 and scores < 20% as 0, the Fleiss’ Kappa values showed better agreement with AI assistance (with AI assistance: 0.86 [95% CI 0.85–0.86]; without AI assistance: 0.40 [95% CI 0.40–0.41]). Table 2 shows that agreement levels are increased for every case when using AI, with some cases achieving 100% agreement. Turnaround time A visual depiction of TATs for each case is provided in Fig. 4 A. Table 3 displays the mean response time, standard deviation, and time saved for each TMA case for PI scoring with and without the AI aid. Without AI assistance, pathologists required an average of 23.3 s to assess each TMA, with a median time of 7.5 s and an interquartile range (IQR) of 5.5 to 16.2 s. AI assistance led to a statistically significant increase ( p < 0.001) in efficiency where the average TAT per TMA reduced to 18.6 s, a median time of 6.4 s and a narrower IQR from 4.6 to 12.1 s. Figure 4 B illustrates the TAT for each question, showing the progression of TAT across cases as they were presented to the pathologists. Due to initially high response times, likely caused by participants acclimating to the software and study setup, question 1 (Case 2 without aid and Case 7 with aid) was excluded from further analyses. For evaluations without AI, pathologists averaged 18.3 s per TMA, with a median time of 7.2 s and an IQR of 5.5 to 14.0 s. With AI support, the average TAT per TMA decreased to 16.8 s, the median time was 6.4 s, and the IQR narrowed to 4.7 to 11.6 s. The reduction in TAT was statistically significant among pathologists with experience ranging from 10 to 39 years (Fig. 4 C) ( p < 0.001), and for pathology fellows, practicing and retired pathologists (Fig. 4 D) ( p < 0.001). Supplemental Fig. 7 shows the mean TAT with and without aid for various disciplines, where roles such as clinical and forensic pathologists were statistically faster ( p < 0.001). AI assistance resulted in an average reduction of 1.5 s per TMA [95% CI, −2.4 to −0.6 s, p < 0.001]. Supplementary Fig. 8 displays a histogram of the distribution of the total percentage of time saved, calculated using the formula: (total time saved/total time spent on conventional assessment) × 100%. The mean percentage saving was 9.4%, with a median of 11.9%. Pathologists’ opinions Pathologists’ opinions on the use of AI for Ki-67 assessment in breast cancer are summarized in Fig. 5 . The majority of respondents considered the AI tool's suggestion, found it to be appropriate and agreed that this AI tool could improve accuracy, inter-rater agreement and TAT for Ki-67 assessments (Fig. 5 A). Many respondents also agreed that they would personally implement and agree with the routine implementation of AI aid for Ki-67 assessments within the next decade (Fig. 5 B, C).
Discussion Ki-67 serves as a crucial indicator for predicting cancer recurrence and survival among early-stage high-risk breast cancer patients 1 , 2 . It informs decisions regarding adjuvant chemotherapy 12 and radiation therapy opt-out for Luminal A breast cancer patients 13 . These clinical decisions often rely on PI scores between 5 and 30%; however, this range exhibits significant scoring variability among experts, making standardization and clinical application challenging 8 , 9 , 12 . This inconsistency, combined with long assessment times using the current Ki-67 scoring system, has limited the broader clinical application of Ki-67 and resultantly, has not yet been integrated into all clinical workflows 16 . AI technologies are being proposed to improve Ki-67 scoring accuracy, inter-rater agreement, and TAT. This study explores the influence of AI in these three areas by recruiting 90 pathologists to examine ten breast cancer TMAs with PIs in the range of 7 to 28%. Two previous studies aimed to quantify PI accuracy with and without AI 19 , 23 . One study demonstrated that AI-enhanced microscopes improved invasive breast cancer assessment accuracy 23 . They had 30 pathologists use an AI microscope to evaluate 100 invasive ductal carcinoma IHC-stained whole slide images (WSIs), which provided tumor delineations, and cell annotations. AI use resulted in a mean PI error reduction from 9.60 to 4.53. A similar study was conducted 19 , where eight pathologists assessed 200 regions of interest using an AI tool. Pathologists identified hotspots on WSIs, after which the AI tool provided cell annotations for the clinician's review. The study found that this method significantly improved the accuracy of Ki-67 PI compared to traditional scoring (14.9 error without AI vs. 6.9 error with AI). Similarly, this study found that using AI assistance for PI scoring significantly ( p < 0.001) improved pathologists’ accuracy, reducing both the PI error and its standard deviation across various demographics, including years of experience and specialties. This indicates that AI assistance leads to higher PI accuracy across all levels of pathologists' training, enabling professionals at every career stage to deliver more precise PI scores in the range critical for clinical decision-making. This improvement may help bridge experience gaps and is critical for PI scoring standardization. An underestimation trend, previously reported by 33 , was also noted in this study, as shown by the PI correlation and Bland–Altman analysis (Fig. 3 ). However, scoring with the support of AI improved PI accuracy for all cases and corrected this underestimation bias. This is exemplified by the scoring near the 20% cutoff, which simulates a clinical decision threshold. In conventional assessments, many pathologists select the incorrect range (≥ 20% or < 20%), particularly for TMAs 7, 8, and 9, with ground truths of 19.8, 23.7, and 28.2, respectively. For instance, TMA 8 had 76.7% of respondents incorrectly estimated the score as < 20%. Errors like these would result in incorrect therapy decisions and poor patient outcomes. Fortunately, with AI assistance, the percentage of pathologists agreeing with the ground truth greatly improved, providing a strong incentive for the clinical use of AI tools in Ki-67 scoring. All cases showed a statistically significant PI error decrease with AI assistance, except for Case 1, with a ground truth PI score of 7.3% (p = 0.133). This exception could be attributed to fewer Ki-67 positive cells requiring counting, which likely simplified the scoring process. In addition to accuracy, PI scoring agreement is critical to ensure that patients with similar disease phenotypes are delivered the proper therapeutic regimes. However, significant variability in Ki-67 scoring is widely recognized, even in established laboratories. A study led by 34 , found reproducibility among eight labs was only moderately reliable with contributing factors such as subjective judgements related to PI scoring and tumor region selection. Standardizing scoring methods becomes imperative, as transferring Ki-67 PIs and cutoffs between laboratories would compromise analytical validity. In another study by 35 , the variability in breast cancer biomarker assessments, including Ki-67, among pathology departments in Sweden was investigated. While positivity rates for HR and HER2 had low variability, there was substantial variation in Ki-67 scoring, where 66% of labs showed significant intra-laboratory variability. This variability could potentially affect the distribution of endocrine and HER2-targeted treatments, emphasizing the need for improved scoring methods to ensure consistent and dependable clinical decision-making. The study by 23 , aimed to improve Ki67 scoring concordance with their AI-empowered microscope. They found a higher ICC of 0.930 (95% CI: 0.91–0.95) with AI, compared to 0.827 (95% CI: 0.79–0.87) without AI. Similarly 22 , aimed to quantify the inter-rater agreement for WSIs with AI assistance across various clinical settings. The AI tool evaluated 72 Ki-67 breast cancer slides by annotating Ki-67 cells and providing PI scores. Ten pathologists from eight institutes reviewed the tool and input their potentially differing PI scores. When the scores were categorized using a PI cutoff of 20%, there was an 87.6% agreement between traditional and AI-assisted methods. Results also revealed a Krippendorff's α of 0.69 in conventional eyeballing quantification and 0.72 with AI assistance indicative of increased inter-rater agreement, however, these findings were not significant. In this study, we evaluated the scoring agreement with and without AI across 90 pathologists, representing one of the largest cohorts analyzed for this task. It was found that over the critical PI range of 7 to 28%, AI improved the inter-rater agreement, with superior ICC, Krippendorff’s α and Fleiss’ Kappa values compared to conventional assessments and higher correlation of PI estimates with the ground truth PI score. Additionally, there was a decrease in offset and variability, as shown in Fig. 3 . These agreement metrics align with findings from earlier studies 22 , 23 and signify that AI tools can standardize Ki-67 scoring, enhance reproducibility and reduce the subjective differences seen with conventional assessments. Therefore, using an AI tool for Ki-67 scoring could lead to more robust assessments and consistent therapeutic decisions. AI applications have predominantly focused on automating the laborious tasks for pathologists, thereby freeing up time for high-level, critical decision-making, especially those related to more complex disease presentations 16 – 20 , 36 . Some research into AI support tools in this field has demonstrated a notable decrease in TAT for pathologists. For instance, a study led by 37 , which involved 20 pathologists analyzing 240 prostate biopsies, reported that an AI-based assistive tool significantly reduced TAT, with 13.5% less time spent on assisted reviews than on unassisted ones. Similarly, the study by 38 , demonstrated a statistical improvement ( p < 0.05) in TATs when 24 raters counted breast mitotic figures in 140 high-power fields, with and without AI support, ultimately achieving a time saving of 27.8%. However, the study by 23 , reported a longer TAT using an AI-empowered microscope in their study, which involved 100 invasive ductal carcinoma WSIs and 30 pathologists (11.6 s without AI vs. 23.8 s with AI). Our study found that AI support resulted in faster TATs (18.3 s without AI vs. 16.8 s with AI, p < 0.001), equating to a median time saving of 11.9%. Currently, our team only performs Ki-67 testing upon oncologists' requests, as routine Ki-67 assessment is not yet standard practice. This is partly due to the difficulties in standardizing Ki-67, compounded by pathologists' increasing workloads and concerns over burnout 39 , 40 . Pathologists' caseloads have grown in the past decade, from 109 to 116 annually in Canada and 92 to 132 in the U.S. 41 . With the Canadian Cancer Society expecting 29,400 breast cancer cases in 2023 42 , routine Ki-67 assessments would significantly increase workloads. Therefore, the implementation of AI tools in this context could alleviate workload pressures by offering substantial time savings and supporting the clinical application of this important biomarker. The gold standard for assessing Ki-67 PI is manual counting 8 , 9 ; however, due to the labor-intensive nature of this method, many pathologists often resort to rough visual estimations 43 , 44 . As indicated in Table 3 and Fig. 4 , the shorter TATs suggest that respondents may have relied on visual estimations for Ki-67 scoring. Despite this, the TATs significantly improved ( p < 0.001) when using AI. This improvement was evident among experienced pathologists; however, some encountered longer TATs after integrating AI, possibly due to unfamiliarity with the AI tool or digital pathology viewing software. Although participants received a brief orientation and two initial examples, the novelty of the tool might have posed a learning curve. Addressing this challenge involves integrating the tool into regular practice and providing comprehensive training before its use. The perspectives of pathologists highlight a growing enthusiasm towards AI integration for Ki-67 evaluations for breast cancer. A significant 84% of participants agreed the AI’s recommendations were suitable for the task at hand. They recognized AI's ability to improve pathologists' accuracy (76%), enhance inter-rater consistency (82%), and reduce the TAT for Ki-67 evaluations (83%). Additionally, 49% expressed their intent to incorporate AI into their workflow, and 47% anticipated the routine implementation of AI within the next decade. An important observation is that many respondents who were hesitant about personally or routinely implementing AI in clinical practice were retired pathologists. In total, 83% of retired pathologists reported they would not currently implement AI personally or routinely, which is a stark contrast to only 15% of practicing pathologists who expressed the same reluctance. This positive outlook in the pathologist community supports the insights of this study and signals an increasing momentum for the widespread adoption of AI into digital pathology. The strength of this research is highlighted by the extensive and diverse participation of 90 pathologists, which contributes to the study's generalizability in real-world clinical contexts. Adding to the study's credibility is the focus on Ki-67 values around the critical 20% threshold, which is used for adjuvant therapy decisions. Moreover, the AI nuclei overlay addresses the transparency concerns often associated with AI-generated scores, thus improving clarity and comprehensibility for users. The ongoing discussion around 'explainable AI' highlights the importance of transparency in AI tools' outputs, a crucial factor for their acceptance and adoption 45 . The outcomes of the study emphasize the positive outlook and readiness of pathologists to embrace AI in their workflow and serve to reinforce the growing need for the integration of AI into regular medical practice. The study has its limitations, one of which includes the potential unintentional inclusion of non-pathologists. The survey required respondents to confirm their status as pathologists through agreement before beginning; however, due to confidentiality limitations, no further verification was possible. In some instances, pathologists' scores deviated from the ground truth by more than 20%, with PI errors reaching up to 50%. Such large errors would render any PI score diagnostically irrelevant, as the variance exceeds the clinical threshold of 20%. These errors might be attributed to input errors or a lack of experience in Ki-67 assessments. Consequently, we used this threshold to filter out potentially erroneous responses. In total, 26 participants who logged responses exceeding the 20% error threshold were subsequently excluded from the study. For completeness, Supplementary Table 2 discloses the PI scores and PI errors of all respondents, including outliers, where the data trends appear similar. The demographics of the study's participants reveal there was limited participation from currently practicing pathologists, representing 14.4% of respondents. This may be attributed to the time constraints faced by practicing pathologists. In future research, efforts will be made to include more practicing pathologists and to evaluate intra-observer variability. Additionally, while the survey provided specific guidelines for calculating the PI and applying Ki-67 positivity criteria, the accuracy and thoroughness of each pathologist's evaluations could not be verified. Lastly, the study deviated from standard practice by using TMAs instead of WSIs for Ki-67 clinical assessments. The rationale behind this choice was the expectation of more precise scoring with TMAs, as this eliminates the need to select high-power fields (a subjective process) and involves a lower number of cells to evaluate, leading to better consistency in visual estimations. Future research should focus on evaluating the accuracy achieved with AI assistance in identifying regions of interest and analyzing WSIs. This should also incorporate a broader range of cases and a wider PI range. Prospective studies involving solely practicing breast pathologists could also yield valuable insights into the real-world application of the AI tool and its impact on clinical decision-making. In conclusion, this study provides early insights into the potential of an AI tool in improving the accuracy, inter-rater agreement, and workflow efficiency of Ki-67 assessment in breast cancer. As AI tools become more widely adopted, ongoing evaluation and refinement will be essential to fully realize its potential and optimize patient care. Such tools are critical for robustly analyzing large datasets and effectively determining PI thresholds for treatment decisions.
The Ki-67 proliferation index (PI) guides treatment decisions in breast cancer but suffers from poor inter-rater reproducibility. Although AI tools have been designed for Ki-67 assessment, their impact on pathologists' work remains understudied. 90 international pathologists were recruited to assess the Ki-67 PI of ten breast cancer tissue microarrays with and without AI. Accuracy, agreement, and turnaround time with and without AI were compared. Pathologists’ perspectives on AI were collected. Using AI led to a significant decrease in PI error (2.1% with AI vs. 5.9% without AI, p < 0.001), better inter-rater agreement (ICC: 0.70 vs. 0.92; Krippendorff’s α: 0.63 vs. 0.89; Fleiss’ Kappa: 0.40 vs. 0.86), and an 11.9% overall median reduction in turnaround time. Most pathologists (84%) found the AI reliable. For Ki-67 assessments, 76% of respondents believed AI enhances accuracy, 82% said it improves consistency, and 83% trust it will improve efficiency. This study highlights AI's potential to standardize Ki-67 scoring, especially between 5 and 30% PI—a range with low PI agreement. This could pave the way for a universally accepted PI score to guide treatment decisions, emphasizing the promising role of AI integration into pathologist workflows. Subject terms
Supplementary Information
Supplementary Information The online version contains supplementary material available at 10.1038/s41598-024-51723-2. Acknowledgements This research acknowledges the Canadian Cancer Society and Mitacs. Author contributions A.D. designed the study, did the statistical analyses, gathered, and interpreted the data, conceived the tables and figures, and drafted the manuscript. N.N.J.N. did the pathologic annotations of the study samples for the ground truth references, contributed to the interpretation of the data, and helped draft the manuscript. J.M. helped design the study. S.D. helped design the study and validated the pathologic annotations of the study samples for the ground truth references. A.K. (principal investigator) designed the study, contributed to the interpretation of the data, helped draft the manuscript, and supervised the project. All authors helped with the recruitment of participants, revised the manuscript, and agreed with the final version of the manuscript. Funding This publication is funded by the Canadian Cancer Society. Data availability The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1283
oa_package/90/05/PMC10787826.tar.gz
PMC10787827
38218952
Introduction In schizophrenia, only ~70% of affected individuals respond to antipsychotic drug treatment, and even fewer go into remission [ 1 , 2 ]. Pre-treatment prediction of the subsequent response could reduce the time spent on ineffective treatments, shorten patient suffering, and reduce possible mortality [ 3 , 4 ]. The ability to identify brain biomarkers of antipsychotic nonresponders using magnetic resonance imaging may lead to improved prognosis and the detection of malleable central nervous system targets for the development of new treatment strategies. Second-generation antipsychotics (SGAs) are widely believed to work by decreasing striatal dopamine via dopamine receptor blockage within the mesolimbic pathway to alleviate positive symptoms and increasing cortical dopamine via 5-HT(2A) antagonism in presynaptic neurons within the mesocortial pathway to improve negative symptoms [ 5 , 6 ]. These neurotransmitters modulate synapses at glutamate (N-methyl-d-aspartate [NMDA]) receptors that are involved in synaptic plasticity and may cause delayed corollary discharges [ 7 ]. Animal studies noted that antipsychotic treatment following administration of the copper chelator cuprizone promoted oligodendrocyte development and remyelination [ 8 , 9 ]. Thus, the integrity of WM may be a sensitive index of the brain pharmacological mechanism of action of SGA. Diffusion tensor imaging (DTI) is widely used to evaluate the structural integrity of WM, and voxelwise metrics such as fractional anisotropy (FA) and mean diffusivity (MD) are generally considered sensitive measures for axonal/myelin damage [ 10 – 12 ]. A handful of DTI studies have documented the association between FA/MD of WM and response to treatment [ 13 – 15 ]. Further, inconsistencies prior work may be related to differences in clinical variables such as the illness course/chronicity, substance misuse, and previous exposure to antipsychotics. Therefore, investigating patients with drug-naive first-episode psychosis may address to disentangle which brain white matter changes may predict response to treatment. To date, only three studies have investigated the effects of antipsychotic use on WM tracts in drug-naive first-episode schizophrenia. These studies show that antipsychotic medications appear to alter or improve FA or MD of WM, such as in the bilateral anterior cingulate gyrus (ACG), corticospinal tract (CT), anterior thalamic radiation (ATR), longitudinal fasciculi (ILF), inferior fronto-occipital fasciculi (IFOF), and uncinated fasciculus (UF) structural abnormalities, especially at remission [ 16 – 18 ]. However, while these studies typically report group-level WH tract differences between psychosis before and after antipsychotic administration, whether these findings have predictive value for individualized drug responses is unclear. WH alterations may occur before the onset of psychosis, and FA/MD of WM changes have been reported in subjects with a high risk of psychosis [ 19 , 20 ]. The severity of WM alterations such as fronto-temporal, fronto-occipital, and fronto-striatal WH at onset were associated with the severity and persistence of the signs and symptoms of schizophrenia, suggesting that baseline WM alterations may serve as an early marker for differentially characterizing patients with poor or good outcomes [ 21 – 23 ]. Furthermore, the cortex, especially the frontal lobe, is the core component of the dopamine projection system [ 24 , 25 ]. However, it is still unknown whether psychotic symptom-related alterations in FA and MD of WM at the early stage of the disorder may provide aid to individualized prediction of drug response. In this study, we investigated the above questions using DTI data acquired from first-episode schizophrenia patients with no prior medication. Patients underwent baseline structural MRI scans and were subsequently randomized to receive a single atypical antipsychotic throughout the first 12 weeks. It was hypothesized that the altered FA/MD of WM was related to the severity of psychotic symptoms at baseline, and those changes would show potential as individualized predictive biomarkers of response to SGA.
Methods Participants A total of 68 drug-naive patients between 18 and 45 years old were diagnosed with schizophrenia based on the Structured Clinical Interview for the DSM-IV Axis I disorder and had no previous psychiatric treatment. Patients underwent MRI scans and symptom ratings before assigned to a randomized open-label treatment with risperidone, olanzapine or aripiprazole for up to 1 year (Clinical trials.gov ID: NCT01057849). Clinical symptoms were evaluated using the eight “core symptoms” selected from the Positive and Negative Syndrome Scale (PANSS-8), which has more acceptable internal consistency and comparable sensitivity to early improvement in psychotic symptoms than the PANSS-30 [ 26 ]. This analysis included data only from the first 12 weeks of treatment. During this period, patients received a single antipsychotic that started with low dosage and gradually increased to a standard therapeutic range (3–6 mg risperidone, 15–30 mg aripiprazole, or 10–25 mg olanzapine per day) in 2 weeks. Follow-up assessments were conducted at the 4th, 8th, and 12th weeks by trained psychiatrists. To ensure the consistency and reliability of ratings across the study, three psychiatrists with more than 5 years of experience in clinical psychiatry attended a 1-week training workshop on the use of the rating instruments prior to the study. After training, they achieved an interrater reliability of 0.80 for the PANSS-8 score. Evaluation of treatment response Treatment response was operationalized as a reduction in symptom severity to the levels required by the remission criteria of the Schizophrenia Working Group Consensus [ 27 ]. According to these criteria, clinical improvement is reached when a simultaneous rating of mild or less (equivalent to 1, 2, or 3) is given in all the following items of the PANSS-8: delusions (P1), conceptual disorganization (P2), hallucinatory behavior (P3), mannerisms and posturing (G5), unusual thought content (G9), blunted affect (N1), social withdrawal (N4), and lack of spontaneity and flow of conversation (N6). The clinical recommendation is that antipsychotic treatment with a specific drug should be continued for 6–8 weeks before switching to a different medication owing to lack of efficacy or adverse effects. Hence, in this study, we defined treatment response as meeting the remission criteria at the 8th or 12th week. Only Fifty patients completed both clinical follow-up assessments at the 8th and 12th weeks and were therefore included in this study as the final sample. The drop-out participants did not differ from the rest of the sample in characteristics and symptom severity (Supplementary Table S1 ). MRI data acquisition The MRI scans were performed before medication using a GE Signa EXCITE 3.0-T scanner (GE Healthcare, Milwaukee, Wisconsin) equipped with an 8-channel phase array head coil. The DTI data were acquired using a bipolar diffusion-weighted spin‒echo planar imaging (EPI) sequence (TR = 10000 ms, TE = 70 ms) with a 128 × 128 matrix over a field of view of 240 × 240 mm and 42 axial slices of 3 mm thickness to cover the whole brain without gap. Each DTI dataset included 20 images of unique diffusion directions (B = 1000) and a nondiffusion image (B = 0). High-resolution T1 data were acquired using a 3D spoiled gradient (3D-SPGR) sequence: TR = 8.5 ms, TE = 3.5 ms, TI = 400 ms, flip angle = 12, 240*240 matrix over a field of view of 240*240 mm, and 156 axial slices of 1 mm thickness. All scans were reviewed by an experienced neuroradiologist to exclude gross brain abnormalities. Imaging processing The routine DTI preprocessing included head motion and eddy current correction, brain extraction, and tensor model fitting was performed using FSL (FMRIB Software Library, http://www.fmrib.ox.ac.uk/fsl ). We used automated fiber quantification software (AFQ) to identify 20 white matter traces in individual subjects. The identification procedure included three primary steps: whole-brain deterministic fiber tractography, waypoint ROI-based tract segmentation, and probability map-based fiber refinement using the 20-tract Johns Hopkins University white matter template. The 20 identified tracts were the left and right ATR, cingulum–cingulate (CC), cingulum–hippocampus pathway, inferior fronto-occipital fasciculus (IFOF), inferior longitudinal fasciculus (ILF), superior longitudinal fasciculus (SLF), uncinated fasciculus and arcuate fasciculus, and the forceps major of the splenium and the forceps minor of the genu of the corpus callosum. After tract identification, we smoothed each tract using a 10-point moving average filter to reduce local variation caused by imaging noise. The diffusion measurements along the tract core, defined as the tract profile, were extracted from each fiber tract, including the FA and MD values. Hence, each tract had two features, and each subject had 40 features to depict their global white matter status. Diffusion properties and clinical associations To confirm the correlation between diffusion properties and symptoms at baseline, partial Pearson correlation was performed to examine relations between 40 white matter features and PANSS-8 scores at baseline, with age, gender, and duration of untreated psychosis as covariates. To adjust the significant values for multiple comparisons, we used the Benjamini–Hochberg false discovery rate (FDR q value selected to maintain the false positive error rate <0.05). Prediction of treatment outcome with diffusion properties We next sought to investigate whether baseline FA and MD of WM would be capable of distinguishing antipsychotic responders from nonresponders at the individual level. To this end, we trained a cross-validated generalized LASSO regression model with treatment outcome as the dependent variable and baseline diffusion properties as predictors. To constrain the number of features in the model and meanwhile include all features relevant to the disorder, we preselected the FA and MD measures for model training. Here, only measures significantly associated with baseline symptoms at uncorrected P < 0.05 were included in the model as input predictors. We also trained the model with all baseline diffusion properties as predictors as a supplementary analysis (see Supplementary Materials). The LASSO regression is an L1-norm regularization method that incorporates a shrinkage penalty term λ to avoid model overfitting, which coerces the coefficients of some less important predictors to be shrunken to zero. Specifically, the predictors included in the model were adjusted for age, sex, antipsychotic drug dosage, duration of untreated psychosis, and PANSS-8 scores at baseline. Similar to our prior work [ 28 – 30 ], a repeated nested cross-validation (CV) method (10 outer folds, each with 10 inter folds) was used in which the tuning parameter λ was optimized within the inner cycles and subsequently utilized to predict remaining subjects in the outer cycles. This procedure eventually yielded predicted probabilities of nonresponders for each individual in the main dataset, based on which the classification accuracy was calculated. To ensure the robustness of the results, we repeated the CV 100 times, each time by randomly parcellating the sample. The final classification performance was determined as the average area under curve (AUC) of the receiver operating characteristic (ROC) curves from the 100 runs, and the significance of the performance was determined by 1000 permutations. We also investigated whether the top features selected by the model (at least 8 out of 10 cycles) would be capable of predicting individualized symptom changes in first-episode schizophrenia as a supplementary analysis (see Supplementary Materials).
Results 12-Week treatment outcome By the end of the 12th week, 30 patients met the remission criteria as responders, and 20 patients were classified as nonresponders. Demographic and clinical variables at baseline were not significantly different between responders and nonresponders (Table 1 ). Diffusion properties and clinical associations at baseline In partial correlation analysis between PANSS-8 scores and FA/MD of each fiber tract, we found positive correlations between PANSS-8 scores and average MD of the following fiber tracts after FDR correction: left and right IFOF ( r = 0.564, q = 0.002; r = 0.456, q = 0.013), and left and right ILF ( r = 0.493, q = 0.009; r = 0.425, q = 0.03) (Fig. 1 ). At a more liberal threshold without FDR correction, PANSS-8 score was significantly correlated with the FAs of the left ILF ( r = -0.296, P = 0.044) and right ILF ( r = −0.374, P = 0.01), as well as MDs of the left ATR ( r = 0.358, P = 0.013), left corticospinal ( r = 0.362, P = 0.012), right corticospinal ( r = 0.389, P = 0.007), genu of corpus callosum ( r = 0.346, P = 0.017), and right SLF ( r = 0.376, P = 0.009). Therefore, these eleven measures (two FA measures and nine MD measures) were subsequently used as predictors in the LASSO regression model. Classification performance of treatment response The average AUC from the 100 repeats of the LASSO regression model was 0.828 (range: 0.81–0.86) ( P < 0.001, average sensitivity = 0.867 and average specificity = 0.636). Here, the FA of the right ILF and MDs of the left IFOF and the right SLF had nonzero coefficients at least 8 out of 10 cycles during all 100 repeats, and were therefore selected as final features. The post-hoc t test revealed significantly higher FA of the right ILF ( t = 5.69, P < 0.001) but lower MDs of the left IFOF and the right SLF in responders compared with nonresponders ( t = −2.25, P = 0.029; t = −2.62, P = 0.012) (Table 2 and Fig. 2 ).
Discussion Here, we provide evidence for a baseline psychotic symptom-related white matter tract biomarker that potentially predicts response to antipsychotic treatment in first-episode schizophrenia. Importantly, the study sample was treatment-naive to ensure that the findings were not confounded by the drug. Two main findings emerged from this study. First, the altered diffusion properties (FA or MD) of fiber tracts in the ATR, corticospinal tract, callosum forceps minor, IFOF, ILF and SLF were related to the severity of symptoms, demonstrating that early microstructural WH changes contribute to the pathophysiology of psychosis. Second, these abnormal fiber tracts, especially the ILF, IFOF, and SLF, significantly predicted the response to antipsychotic treatment at the individual level. This suggests that these symptom-related WH changes could be an outcome marker after the onset of psychosis or even a target for intervention and preventive strategies. Our findings decreased FA and increased MD of several fiber bundles throughout the brain correlated with core positive and negative symptom severities consistent with a “disconnection” hypothesis of symptoms in schizophrenia [ 31 , 32 ]. Several meta-analytic studies have investigated the role of WH irregularities in schizophrenia spectrum disorders [ 33 – 36 ]. A meta-analysis of WM alteration in patients with FES indicated widespread abnormalities across white matter tracts, with evidence for reductions in FA in the corpus callosum, the left ILF and IFOF [ 34 ]. Even in chronic schizophrenia, the meta-analysis of 15 DTI studies also observed significant FA reductions in the genu and splenium of corpus callosum, the left anterior thalamic radiation, the left IFOF and ILF [ 36 ]. Moreover, the relationship between aberrant FA of these WM tracts and psychotic symptoms of schizophrenia were reported among previous studies [ 22 , 37 ]. Consistent with this study, several studies found the inverse relationship between FA of SLF, ILF, and IFOF and negative symptoms and auditory verbal hallucinations in schizophrenia [ 37 – 39 ]. Taken together, these implied that white matter dysintegrity may represent a “trait” marker, related to the underlying pathophysiology in schizophrenia. Beyond the group level for white matter correlated with symptoms at baseline, our longitudinal follow-up study also provided evidence that these baseline psychotic symptom-related FA and MD of WM may serve as an individualized predictor for antipsychotic treatment response in patients with schizophrenia. Similar to our findings, previous group-level analysis studies found more widespread FA decreases at baseline in FEP patients with a subsequent poorer response, and that baseline global white matter network organization showed greater alterations in FEP patients who subsequently showed a poorer treatment response [ 40 – 42 ]. Taken together, all these studies highlight the usefulness of baseline WM integrity in predicting response to treatment. Furthermore, the LASSO regression model in this study correctly classified 82.8% of patients as responsive, which may represent an important preliminary step to provide clinicians with decision support in selecting the ideal antipsychotic treatment for schizophrenia in a personalized manner. Further studies using larger and independent samples are required to replicate these findings. We focused on predicting biomarkers of symptom-related FA and MD in WM. Longitudinal studies have observed that longitudinal increases in FA values, especially in the IFOF, ILF, SLF, and anterior thalamic radiation, are significantly correlated with improved symptoms at follow-up [ 17 , 18 , 43 ]. Consistent with these findings, the top treatment response predicting features in our study were located in the ILF, IFOF, and SLF. These WM tracts connect frontal, temporal, parietal and occipital areas, which have been implicated in several cognitive functions, such as visuospatial processing, emotional regulation, memory and language, in schizophrenia [ 44 , 45 ]. In addition, altered FAs in the SLF and IFOF have been found to be a biomarker for auditory hallucinations, and FA in the ILF has been linked to positive symptoms [ 46 , 47 ]. Furthermore, these regions are major target of dopamine signaling. Evidence from animal models suggested that the upregulation of D2 receptors in the frontal, parietal, temporal and occipital lobes and the downregulation of D1 receptors in the prefrontal and temporal cortices may be an important component of the therapeutic response to neuroleptic drugs [ 48 ]. These receptor blockades have been shown to promote oligodendrocyte repopulation and remyelination of experimentally demyelinated cells in mice [ 49 ]. We also found that nonresponders had more severe WM damage as lower FA and higher MD in these fiber bundles at baseline than responders. These findings together suggest that the alterations in the IFOF, ILF and SLF may represent neural markers of the severity and persistence of the signs and symptoms of schizophrenia, and may compromise the potential effects of antipsychotics. This study has some limitations. First, participants in our study were randomized to receive a single standardized treatment with one of three antipsychotics, but the sample was too small to rule out changes in response patterns based on different medications. Future investigations are encouraged to provide a comparison of the effectiveness of different drugs. Second, the study did not include a placebo control group, so a potential placebo or time effect cannot be excluded. Due to ethical issues, these effects are normally nested in clinical studies and cannot be completely removed. Third, as a machine learning study focusing on individual prediction, the sample size in this study was relatively small, and it lack an independent validation sample. Therefore, these findings should be externally verified with larger samples in the future. In conclusion, altered FA and MD of fiber tracts in the ATR, corticospinal tract, callosum forceps minor, IFOF, ILF, and SLF were related to the severity of symptoms in first-episode schizophrenia. These effects on WM tracts are not influenced by pharmacotherapy and therefore appear to be disease-related. These baseline psychotic symptom-related WM tracts, especially ILF, IFOF, and SLF, may serve as meaningful individualized predictors of response to SGA. These results may represent an important first step of the translational value of baseline brain structural measures in precision psychiatry.
There is significant heterogeneity in individual responses to antipsychotic drugs, but there is no reliable predictor of antipsychotics response in first-episode psychosis. This study aimed to investigate whether psychotic symptom-related alterations in fractional anisotropy (FA) and mean diffusivity (MD) of white matter (WM) at the early stage of the disorder may aid in the individualized prediction of drug response. Sixty-eight first-episode patients underwent baseline structural MRI scans and were subsequently randomized to receive a single atypical antipsychotic throughout the first 12 weeks. Clinical symptoms were evaluated using the eight “core symptoms” selected from the Positive and Negative Syndrome Scale (PANSS-8). Follow-up assessments were conducted at the 4th, 8th, and 12th weeks by trained psychiatrists. LASSO regression model and cross-validation were conducted to examine the performance of baseline symptom-related alterations FA and MD of WM in the prediction of individualized treatment outcome. Fifty patients completed both clinical follow-up assessments by the 8th and 12th weeks. 30 patients were classified as responders, and 20 patients were classified as nonresponders. At baseline, the altered diffusion properties of fiber tracts in the anterior thalamic radiation, corticospinal tract, callosum forceps minor, longitudinal fasciculi (ILF), inferior frontal-occipital fasciculi (IFOF) and superior longitudinal fasciculus (SLF) were related to the severity of symptoms. These abnormal fiber tracts, especially the ILF, IFOF, and SLF, significantly predicted the response to antipsychotic treatment at the individual level (AUC = 0.828, P < 0.001). These findings demonstrate that early microstructural WM changes contribute to the pathophysiology of psychosis and may serve as meaningful individualized predictors of response to antipsychotics. Subject terms
Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s41398-023-02714-w. Acknowledgements This study was supported by the National Major Project of Scientific and Technical Supporting Programs of China during the 11th Five-year Plan Period (Grant No. 2017BAI17B04), the Key research and development project of Science and Technology, department of Sichuan Province (22ZDFY2064, 2022YFS0179). Author contributions YC, HC, HD, and XY conceptualized the study. HD, SL, BZ, GZ, ZZ, SL, and HL collected the data. YC and HC performed statistical analysis. YC drafted the manuscript, HC critically reviewed the manuscript. All authors reviewed the manuscript and approved the final version for submission. Data availability The data that support the findings of this study are available from the corresponding author upon reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Transl Psychiatry. 2024 Jan 13; 14:23
oa_package/51/c8/PMC10787827.tar.gz
PMC10787828
38218965
Introduction Modern science has become increasingly collaborative over the past decades 1 . Large teams have become almost necessary to tackle complex problems in various disciplines, requiring a large pool of knowledge and skills. On the other hand, small teams may introduce novel paradigms 2 . A powerful representation of the collaborative nature of science is given by a collaboration network, in which nodes are authors, and two nodes are connected if they have coauthored at least one paper. With the growing availability of bibliometric data, collaboration networks have been extensively studied, and their structural properties are now well known 3 – 6 . Collaboration networks are concrete manifestations of homophily between scholars, i.e . , of the tendency of individuals to interact with people similar to themselves. People working on the same topic or problem may decide to team up and leverage their respective skills to increase their chances of discovering new results. This is an example of selection , where homophily results from the choice of people to engage with similar individuals. On the other hand, collaboration could also induce social influence , in that scholars might affect the future behavior of their coauthors. For a thorough discussion on homophily, selection, and social influence, we refer the reader to chapter 4 of the book by Easley and Kleinberg 7 . Coauthors often expose us to new tools, methods, and theories, even when the latter is not being used for the specific project carried out by the team. The link between diffusion of knowledge and collaboration has been highlighted and explored for some time. For instance, it is known that knowledge flow occurs with a greater probability between scholars who have collaborated in the past 8 and those who are in close proximity in the network 9 . In particular, once scholars discover new research topics, they may decide to work on them in the future. Switches between research interests have become increasingly frequent over time 10 and have recently been subjected to investigation 11 , 12 . The decision to switch may actually be induced by the coauthors in a social contagion process 13 – 17 where scholar a , who spreads the new topic, influences scholar b to adopt it. For this reason, epidemic models have been applied to describe the diffusion of ideas 18 – 20 . In these models, an infected individual a exposes a susceptible individual b to a disease with a certain probability of getting infected and continuing the spread. In the case of an idea or a topic, the infection spreads if b adopts the new idea or starts working on the new topic. On a macro level, dynamics within collaboration networks like topic switches guide the evolution of disciplines 21 , 22 . Here we present an extensive empirical analysis of the relationship between topic switches of scientists and their collaboration patterns. We distinguish active authors, i.e . , those who have papers on the new topic, from inactive authors who have never published in that area. For simplicity, we focus only on the first-order neighborhoods in the collaboration network. We find that the probability that the inactive coauthors of an active scholar switch topic grows with the productivity and impact of the latter. The larger the average number of inactive coauthors of active scientists, the smaller the effect. Also, the topic-switch probability for an inactive scholar grows with the number of their active coauthors, with a profile suggesting that the contributions of each coauthor are not independent.
Methods Data We analyze papers from the February 2023 snapshot of the bibliometric dataset OpenAlex: the successor to Microsoft Academic Graph (MAG). We found incomplete citation coverage for papers published before 1990. So, we only consider papers published between 1990 and 2022 and having at most thirty authors. Papers are tagged with concepts (topics) by a classifier trained on the MAG. We use concept tags to construct snapshots for three fields: Physics, Computer Science (CS), and Biology and Medicine (BioMed). Physics contains 19.7M papers, while CS and BioMed each have 27.6M and 43.52M papers, respectively. Within each domain, we select seven, six, and seven topics, respectively. Within each topic, we consider reference years between 1995 and 2018, where the respective interaction and activation windows contain at least 3000 papers. This threshold ensures a critical mass of papers and authors to conduct the analyses. Each topic we selected has at least ten reference years satisfying the constraint. The statistical tests in the manuscript are aggregated over the different reference years. More information is available in Supplementary Tables S1–3 online. Overlap coefficient We use the overlap coefficient to measure the degree of overlap between the different sets of authors picked based on productivity and impact. In our case, the two sets are the same size, so a score of 10% implies that both sets share 10% of the elements. Author ranking metrics Let P be the set of papers published on topic t authored by the set of active authors A during the interaction window IW. Let a be an active author who wrote papers during the IW. We define the following metrics to rank active authors and select the top and bottom 10%. Productivity: the count of papers a has authored on topic t during the IW. More formally, it is the cardinality of the set . Impact: the average citation count of from the papers in P . We argue that restricting incoming citations from P is a good proxy for the impact that a has made on that topic. The average number of citations is a better indicator of excellence than the total citation count 31 . Also, considering the average instead of the sum lowers its correlation with productivity, here measured by the overlap coefficient , as often the most productive authors are also the most cited ones 12 . A low correlation lets us safely disregard the confounding effects of the two metrics and allows us to treat them as fairly independent variables. Correlation statistics are reported in Supplementary Tables S4–6 online. Although citation-based measures are frequently used to quantify research impact, we are aware of the influence of social structures and other hidden biases on scholarly citation behavior 32 . Using more sophisticated measures, however, is beyond the scope of this present work. Statistical test for difference of samples To test whether two independent samples and are different concerning their means and , we assume the null hypothesis that their means are the same, i.e . , . Next, we compute the mean and 95% confidence interval of the distribution of the difference of their means, i.e . , , using bootstrapping 33 . We reject the null hypothesis at if the confidence interval of does not contain 0 34 . In other words, and are considered statistically different at if the 95% confidence interval of the difference of their respective means does not contain 0. Furthermore, a positive mean of the difference indicates that , while a negative mean indicates . In our experiments, we aggregate the differences across the reference years for a given topic, and then carry out the procedure described above. Target activation probability Let n ( k ) be the number of inactive authors with exactly k contacts during the exposure window, of whom m ( k ) become active in the observation window. The target activation probability P ( k ) is the probability of becoming active after having exactly k contacts, defined as The cumulative target activation probability C ( k ) with k or more contacts is given by Simple baseline for membership closure Let p represent the probability of activation from a single contact. The probability of activation having k contacts, acting independently of each other, is . We compute p from the observed data using Eq. ( 1 ) as . This is the fraction of inactive authors with exactly one contact who became active as . Like before, we calculate the cumulative target activation probability for the baseline with k or more contacts as The denominator is the same as in Eq. ( 1 ) and comes from the observed data. The numerator represents the expected number of active authors if the contacts affect the activation independently. Source activation probability Let be the number of exclusive inactive coauthors of an active author a in the IW. Let be the number of those exclusive inactive coauthors who become active in the AW. The source activation probability of scholar a is thus We stress that, for the probability to be well-defined, must be greater than zero. Therefore, in our calculations, we focused on active authors with at least one exclusive inactive coauthor. For any , we compute the fraction of all active authors whose source activation probability is greater than or equal to f . is the complementary cumulative probability distribution of the source activation probability . As expected, quickly decreases to 0 with increasing f . Because the curves corresponding to two sets of active authors are effectively indistinguishable at the tail, we compare a pair of points at some threshold . We call the cumulative source activation . The choice of the threshold is important. Setting it to 0 or 1 would return the same probability for both sets of authors. It should not also be too small for numerical reasons. For example, if there are only five inactive coauthors, the smallest non-zero fraction cannot be smaller than . Choosing too high a value instead would lead to weaker statistics. So, we fix the value at 0.10 for the results in the main text (Figs. 4 and 5 ), and at 0.20 in the Supplementary Figs. S3 and S4 online. Chaperoning propensity Let be the number of exclusive inactive coauthors of an active author a who become active in the AW, which is the same as the numerator of Eq. ( 4 ). Let be the number of those authors who write their first paper on topic t with a in the AW. The chaperoning probability of a is defined as We define the chaperoning propensity corresponding to a specific threshold as the fraction of all active authors with . We use the aforementioned values of 0.10 (Figs. 4 and 5 ) and 0.20 (Supplementary Figs. S3 and S4 online) for the threshold f .
Results We use the scientific publication dataset OpenAlex 23 . We present the results for twenty topics belonging to three disciplines: Physics, Computer Science, and Biology & Medicine. See “ Methods ” for details. Our approach is inspired by the pioneering work by Kossinets and Watts on social network evolution 24 . In it, the authors estimated triadic closure of two individuals a and b , i.e., the probability that a and b become acquainted as a function of the number of common friends. They took two snapshots of the network at consecutive time ranges: in the earlier snapshot, one keeps track of all pairs of disconnected people, and in the latter, one counts how many of those pairs become connected. A similar approach has been adopted to compute membership closure , i.e . , the probability that an individual starts participating in an activity having been connected to k others who participate in it 25 . We now describe how we adapt this framework to measure how collaborations induce topic switches. Given a scientific topic t , reference year , and window size T , we construct two consecutive non-overlapping time ranges spanning years and respectively. We call the first range the interaction window (IW), where we track author interactions in the collaboration network, and the latter range, the activation window (AW), where we count topic switches. We then identify the set of active authors A who published papers P on topic t during the IW. For example, in Fig. 1 a, . We construct the collaboration network G by considering all papers written by authors after a becomes active. Note that includes papers outside of P , like the ones drawn in gray in Fig. 1 a. We classify the non-active authors in G as inactive authors who are the candidates for topic switches in the AW. They turn active when they publish their first paper on topic t . In Fig. 1 b, authors , and are inactive, with and becoming active in the AW. Furthermore, we rank each active author based on two metrics of scientific prominence: productivity and impact , described in Methods, and calculated at the end of the IW to capture the current perception of a ’s scholarly output. Finally, for each metric, we identify and mark the authors who rank in the top and the bottom 10%. Given this general setup, we conduct two complementary experiments that we describe in depth in the following sections. In Experiment I, we measure membership closure among inactive authors to quantitatively assess how past collaborations with active authors manifest in topic switches. In Experiment II, we instead focus on active authors, quantifying the propensity of their inactive coauthors to start working on their topic of expertise. All the measures used in these sections are formally defined in Methods . Experiment I Here we investigate membership closure among inactive authors. Specifically, we will answer the following questions: How is the probability of topic switches related to k , the number of contacts with active authors? Does this probability depend on the relative prominence of the active authors? To compute the measure, we first must define what construes as contact with an active author in the IW. We consider two definitions as described below. The number of active coauthors, with the same coauthor counted as many times as the number of collaborations. In the collaboration network, this corresponds to the weighted degree when considering only active coauthors. The number of papers written with active coauthors. For example, in Fig. 1 c, author has five contacts based on the first definition (two each from and and one from ), and three if we use the second (the second, the fifth, and the seventh papers in the IW). We report the findings based on the first definition in the main text. The results from the second definition do not alter the main conclusions and can be found in Supplementary Figs. S1 and S2 online. To address the first question, we compute the cumulative target activation probability C ( k ), i.e . , the fraction of inactive authors who become active in the AW as a function of the number of contacts k . In Fig. 2 , we plot C ( k ) (in purple) for each of the twenty topics under investigation. Error bars derive from averaging over different time windows for each field. As expected, we see an increasing trend. In particular, the jump from k = 0 to k = 1 is remarkable, showing that the probability of spontaneous activation in the absence of previous contacts ( k = 0) is much lower than that of activation through collaboration ( k 1). We observe that the higher the number of contacts, the larger the probability. Most of the growth occurs for low values of k . To put these numbers in context, we consider a simple baseline where we assume each contact has a constant, independent probability of producing a topic switch. Within each topic, we compute the difference between the curves for each value of k (see “ Methods ”) over all reference years and plot them below the x -axis. Except for the topics of Cluster Analysis, Parallel Computing, and Peptide Sequence, the observed curves deviate from the baseline. This provides some empirical evidence to ascertain that the baseline cannot capture the nuances in the observed data. A positive deviation for the majority of the topics indicates a compounding effect. Fluid Dynamics and Statistical Physics are exceptions, as they undershoot the baseline. This may be because they are broad interdisciplinary fields unlike the others, and having collaborators in different fields may lessen their effect. Next, we explore the second research question, checking if the contact source’s prominence affects activation chances. Recall that in every IW for a topic, we select active authors in the top 10% and the bottom 10% based on productivity and impact. This separates the most prominent active authors from the least prominent. To mitigate confounding effects, we only consider the subset of inactive authors who are neighbors with strictly one of the two sets of active authors. In Fig. 3 , we assess the significance of the difference between the cumulative target activation probabilities for inactive authors in contact with active authors in the two bins. Each heatmap row corresponds to a topic, and the color of each cell indicates whether the difference is positive (red), negative (blue), or non-significant (gray). The two panels correspond to prominent authors selected based on productivity (panel a ) and impact (panel b ). For productivity, all differences are significant and positive, meaning that contacts with highly productive active authors lead to higher target activation probabilities. For impact, there are a handful of exceptions. Overall, having prominent contacts increases the target activation probability. Experiment II Here we focus on the active authors and their collaborators. For every active author a , we consider the subset of their inactive coauthors who have exclusively collaborated with a in the IW. We call this set the exclusive inactive coauthors of a . For example, in Fig. 1 d, active author has four coauthors , of whom only and exclusively collaborate with in the IW. We do this because effects due to active authors different from a would be difficult to disentangle and could confound the analysis and the conclusions. The relevant measure here is the source activation probability , i.e . , the fraction of exclusive inactive coauthors who become active in the AW. The fraction controls for the collaboration neighborhood sizes which could vary widely for different scholars. In Fig. 1 d, for is = 50%, as only becomes active in the AW. For a given set of active authors, we obtain , the complementary cumulative probability distribution of their source activation probabilities. We select the pools of the most and the least prominent authors as described in Experiment I. The relative effects of the two groups are estimated by comparing the cumulative source activations , i.e . , points on the respective cumulative distributions at a specific threshold . Results are reported in Fig. 4 a for a threshold . Our conclusions also hold when considering a threshold , which can be found in Supplementary Fig. S3 online. In Fig. 4 a, each row corresponds to a topic. The different ranges represent the confidence intervals of the mean difference between the cumulative source activations for the two pools of authors for productivity (green) and impact (pink), respectively. For productivity, the difference is significant for all topics but one (Gravitational Wave). The differences are somewhat less pronounced for impact, but are still significant in most cases. To further corroborate this finding, we specialize the analysis by checking how many exclusive coauthors of a also published their first paper on topic t in the AW with a . This is a way to assess the chaperoning propensity of active authors 26 , and we define the measure in Methods . In Fig. 4 b, we report the confidence intervals of the average difference between the chaperoning propensities for the most prominent and the least prominent active authors for threshold . Similar to Fig. 4 a, we find that the more productive/impactful an active author is, the more likely their coauthors will start working with them on a new topic. Results for , which confirm this trend, can be found in Supplementary Fig. S4 online. While our analysis clearly shows that prominence is a factor, one may wonder if the number of coauthors also plays a role. We posit that, on average, the more collaborators one has, the more tenuous the contact with any of them will be, resulting in lower source activation probabilities. From each group of most prominent authors, we, therefore, pick the top and the bottom 20% based on the average number of coauthors on papers published with exclusive inactive coauthors. By construction, this excludes any paper written on the focal topic. In Fig. 5 , we perform the same analysis as in Fig. 4 for the two pools of authors described above. We observe that the confidence intervals of the differences lie to the left of zero, i.e . , are negative. For productivity, all values are significant. For impact, there are only two topics (Chemotherapy and Radiation Therapy) that are not significant. Overall, inactive coauthors of prominent authors with more collaborators have a lower probability of switching topics. This is consistent with the intuition that the interactions with each coauthor are less frequent/strong in that case and, consequently, less effective at inducing topic switches.
Discussion Collaboration allows scholars to deepen existing knowledge and be exposed to new ideas. In this paper, we assessed if and how collaboration patterns affect the probability of switching research topics. We determined that the probability for a scholar to start working on a new topic depends on earlier contacts with people already active in that topic. This effect is proportional to the number of contacts, with more contacts resulting in higher probabilities. In most topics, this behavior is distinct from a simple baseline assuming independent effects from the contacts, which likely indicates effects of non-dyadic interactions that prompt further investigation. Similarly, we measured the probability that inactive coauthors of an active author end up publishing on the new topic, which singles out the effect of the association with that author in the activation process. Specifically, we checked whether the activation probability depends on some features of the active authors. We found that the more prolific and impactful authors have higher chances of inducing coauthors to switch topics and become coauthors in their first paper on the topic. We stress that, by design, previous interactions between inactive and active authors are limited to works dealing with topics different from the focal topic. Therefore, our analysis suggests that an active author may expose an inactive one to a new topic, even when their interactions do not directly concern that topic. This underlines the social character of scientific interactions, where discussions may deviate from the context that mainly motivates them. Furthermore, we showed that the larger the number of coauthors of an active author, the lower the chance of a topic switch. This is consistent with a dilution of the effect, resulting from the inability to interact strongly with collaborators when their number is large. To the best of our knowledge, we are disclosing this effect for the first time. A possible explanation of our findings is that topic switches result from a social contagion process, much like the adoption of new products 15 , 27 , or the spreading of political propaganda 17 . However, we cannot discount selection effects in observational studies like ours 28 . Having large numbers of active coauthors on a topic may be associated with strong latent homophily between the authors, which may facilitate the future adoption of the topic even without interventions from the active authors. Therefore, the effects we observed may be due to a combination of social contagion and selection. Our work uses OpenAlex, a valuable open-access bibliometric database. We rely on their author disambiguation and topic classification algorithms to conduct the analyses. These processes are inherently noisy and can introduce implicit biases. In addition, there appears to be incomplete citation coverage which might partly explain why the results for impact are not so robust as those for productivity. Future releases of OpenAlex might mitigate these problems. To counter these issues, we repeated our analysis on multiple topics from three distinct scientific disciplines. While the size of the effects varies with the topic, the paper’s main conclusions hold across topics, with very few exceptions. In conclusion, our work offers a platform for further investigations on the mechanisms driving topic switches in science. A thorough understanding of these mechanisms requires effective integration of all factors that may play a role. Besides productivity and impact, topic switches may be affected by the institutional affiliations of those involved. On the one hand, it is plausible that people in the same institution have more chances to interact and affect each other’s behavior. On the other hand, collaborations with people from renowned institutions are expected to weigh more in the process. Another discriminating factor could be the number of citations to the collaborator’s papers. The higher the number of citations, the closer the association between collaborators. We could also include the scientific affinity between coauthors through the similarity of their papers. Modern neural language models 29 , 30 allow to embed papers and, consequently, authors in high-dimensional vector spaces, where the distance between two authors is a good proxy of the similarity of their outputs. The analysis we have conducted here can be extended to other sectors of human activity where collaboration plays a key role, like software development and patent design.
Collaboration is a key driver of science and innovation. Mainly motivated by the need to leverage different capacities and expertise to solve a scientific problem, collaboration is also an excellent source of information about the future behavior of scholars. In particular, it allows us to infer the likelihood that scientists choose future research directions via the intertwined mechanisms of selection and social influence. Here we thoroughly investigate the interplay between collaboration and topic switches. We find that the probability for a scholar to start working on a new topic increases with the number of previous collaborators, with a pattern showing that the effects of individual collaborators are not independent. The higher the productivity and the impact of authors, the more likely their coworkers will start working on new topics. The average number of coauthors per paper is also inversely related to the topic switch probability, suggesting a dilution of this effect as the number of collaborators increases. Subject terms
Supplementary Information
Supplementary Information The online version contains supplementary material available at 10.1038/s41598-024-51606-6. Acknowledgements We acknowledge the support of the AccelNet-MultiNet program, a project of the National Science Foundation (Award #1927425 and #1927418). This work is also supported by the Air Force Office of Scientific Research under award #FA9550-19-1-0354.This research was also supported in part by Lilly Endowment, Inc., through its support for the Indiana University Pervasive Technology Institute. Author contributions S.F. designed the research; S.V. and S.S. performed the experiments and data analysis. S.V., S.S., F.R., F.T., and S.F. wrote the manuscript. All authors reviewed the manuscript. Data availability The datasets generated during and/or analyzed during the current study are available in the Collaboration-Topic-Switches repository on GitHub . Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1258
oa_package/58/e7/PMC10787828.tar.gz
PMC10787829
38218954
Introduction Coronary artery disease (CAD) is the leading cause of mortality around the globe. In developed and developing countries, CAD is the foremost cause of death. According to a previous study, people in the Middle East demonstrate heart disease at younger ages 1 . In Iran, 50% of annual deaths are the consequence of CAD 2 . CAD usually affects people aged above 50 years. Nonetheless, in some cases, women under 55 and men under 45 years develop CAD, termed “premature CAD” 3 . No universally accepted age threshold exists for premature CAD. A prior investigation assigned the age of 49 as the cutoff for premature CAD in males based on several autopsy reports 4 . Here, we determined the age of 49 as a cutoff for assigning premature CAD in males. Patients with premature CAD tend to develop consequent ischemic events at a higher rate than other age groups 5 .CAD is the consequence of atherosclerosis, an inflammatory disease contributing to the accumulation of fatty deposits within the arterial wall which leads to atherosclerotic plaque formation 6 . The results of a recent genome-wide association study (GWAS) ascribed many loci as risk loci for premature CAD. Among these, the 9p21.3 locus is the most striking genetic risk factor. The 58-kb region of the 9p21 locus is assigned as the CAD risk interval region. No protein-coding gene resides in this segment. The risk haplotype in this region includes interlinked noncoding single-nucleotide polymorphisms (SNPs) 7 . CDKN2B-AS1 , also known as “ ANRIL (antisense noncoding RNA in the INK4A locus)”, is a long noncoding RNA (lncRNA) overlapping the CDKN2B gene. Further, rs10757274 (NC_000009.12:g.22096056A>G), rs2383206 (g.22115027A>G), rs2383207 (g.22115960A>G), rs496892 (g.22024352C>T), rs10757278 (g.22124478A>G), and rs10738605 (g.22049131C>G) reside in the 9p21 locus. While rs10757274, rs2383206, rs2383207, and rs496892 are in the intronic regions of the ANRIL sequence, rs10738605 resides in exon six of ANRIL (isoform. 1). On the other hand, rs10757278 is in the intergenic region downstream of the ANRIL gene. Extensive GWAS analyses have been conducted on the associations between the abovementioned SNPs and CAD 8 , 9 . In our study, however, we narrowed down the CAD-associated GWAS variants to SNPs that appeared in publications most frequently as CAD-associated variants in different populations. Of the 14 splicing isoforms of ANRIL annotated on Ref Seq, five isoforms are short and nine are long. The expression pattern of the two groups differs noticeably. The expression of short isoforms is higher than that of long ones. Specific genetic variants are associated with increased levels of short variants and a reduced expression of long isoforms suggesting the contribution of genetic variants to ANRIL expression 9 , 10 . A prior study discovered that the relatively lower ANRIL expression could be increased following treatment 11 . The interplay between the genotypes of 9p21 risk SNPs and ANRIL expression has been extensively explained, with research having validated differential exon expression levels of ANRIL , along with various circular and linear isoforms 10 , 12 – 14 . In vascular smooth muscle cells and mononuclear cells, ANRIL promotes proliferation and atherosclerosis progression 12 . It has been demonstrated that ANRIL is upregulated in serum samples from patients with multiple malignancies, including glioma and breast cancer 15 , 16 . The higher expression of circulating ANRIL in the peripheral blood and serum of patients with diabetes and ischemic stroke has been reported 17 , 18 . As demonstrated by Holdt and Teupser 12 , atherosclerosis progression is associated with decreased expression of circular ANRIL and increased expression of linear ANRIL . Circulating transcripts are enriched in serum compared to whole blood sharing 80% of RNAs with other tissues 19 . Serum is more sensitive than plasma in biomarker discoveries 20 and being free of EDTA, which induces platelet activation, provides a more realistic picture of circulating RNA's physiological pathways 21 . As proposed by Lawford, from the therapeutic intervention perspective, the earlier the diagnosis is established, the better the outcome will be 6 . Early diagnosis of CAD is crucial in Iran due to the increasing incidence of premature CAD in recent years. Sedentary lifestyles, familial CAD, type 2 diabetes (T2D), hyperlipidemia, and smoking are the most frequently correlated risk factors, although the precise involvement of genetic susceptibility factors should not be underestimated in population-based studies 22 , 23 . In the present study, we aimed to investigate the associations between the genotypes of six SNPs within (rs10757274, rs2383206, rs2383207, rs496892, and rs10738605) and proximal to (rs10757278) the ANRIL gene sequence, in addition to ANRIL expression, and PCAD solely and in combination. According to similar studies in other populations, we hypothesized that circulating ANRIL expression differs between premature CAD and non-CAD groups. We also assumed that in premature CAD patients of the Iranian population, a correlation exists between the genetic profile of ANRIL SNPs, rs10757278 and ANRIL expression, since it has been proposed that genetic factors are more important than environmental factors in the development of CAD among young adults. The growing number of premature CAD occurrences in Iran prompted us to investigate the reasons.
Methods Sample collection The present study recruited 93 premature CAD and 87 non-CAD age-matched subjects hospitalized at Rajaie Cardiovascular Medical and Research Center, Tehran, Iran, between 2017 and 2019. All patients were diagnosed by angiography. A questionnaire and a formal consent form were completed and signed by all the subjects. The study protocol was approved by the Ethics Committee of Rajaie Cardiovascular Medical and Research Center (RHC.AC.IR.REC.1396.62), and was conducted in accordance with the Helsinki Declaration. The informed consent was obtained from all subjects and/or their legal guardian(s). Patients (women < 55 y/o and men < 49 y/o) diagnosed with premature CAD by coronary angiography were included, while patients with malignant tumors, severe liver disease, severe kidney dysfunction, infectious diseases, immune diseases, communication dysfunction, or cognitive dysfunction were excluded. The non-CAD group included women < 55 y/o and men < 49 y/o with no evidence of coronary artery disease. The exclusion criteria was as well as the one applied for CAD subjects. T2D patients with and without PCAD were also included in the study. All patients in Rajaie Cardiovascular Medical and Research Center were evaluated by the medical team for their diabetes status before being included in the current study. Clinical considerations regarding diabetes have been taken into account by clinicians prior to any treatment or intervention. DNA extraction DNA was extracted from peripheral blood using the DNeasy Blood & Tissue Kit (QIAGEN, Germany). The quantity of the extracted DNAs was assessed using NanoDrop 2000/2000c Spectrophotometers (Thermo Fisher Scientific, USA). Serum isolation, RNA extraction, and cDNA synthesis Following blood collection, serum was isolated by incubation at room temperature for 30 min. Afterward, clots were removed by centrifuge at 3500× g for 20 min. Subsequently, the serum was isolated in a ribonuclease (RNase)-free tube and kept at − 80 °C for longer preservation. Total RNAs were isolated from serum using a plasma/ serum RNA purification kit (NORGEN BIOTEK, Canada). The RNAs were then concentrated by the RNA Clean-up and Concentration Kit (NORGEN BIOTEK, Canada). Subsequently, reverse transcription was performed with a PrimeScript First-Strand cDNA Synthesis Kit (Takara, Japan). Real-time polymerase chain reaction (PCR) Specific PCR primers were designed using Gene Runner (version 6.5.51) and Primer3, synthesized by Macrogen (Korea) ( S1 Table). The primers were designed for different exons of ANRIL (RefSeq NR_003529.3): one pair on exon one, one on exon five-six, and one on exon 19 (Fig. 1 ). The real-time PCR analysis was performed using the BioFACT 2X Real-Time PCR Master Mix (High ROX), including SYBR Green (BioFACT, South Korea), in an Applied Biosystems StepOne Plus instrument (Applied Biosystems, USA). Next, cDNA was added to 10 μL of SYBR Green and 0.5 μM of each primer in a 20 μL reaction. All the reactions were repeated in duplicates, and mean threshold cycles were used for further analyses. The real-time thermal program was as follows: 95 °C for 30 s and 62 °C for 35 s, for which 5srRNA was utilized as an internal control. The expression level was calculated by 2 −ΔCt . PCR and sequencing For SNP genotyping, primers were designed to detect a 300–700 base pair (bp) product. The amplicon was then sequenced. The sequence and characteristics of the primers used in this study are listed in S1 Table. Afterward, 100 ng DNA was added to 10 μL of Taq DNA Polymerase Master Mix RED (Ampliqon, Denmark) for PCR reactions, along with 0.5 μM of each primer in each 20 μL reaction. PCR was performed for 30 cycles in the following thermal program: 95 °C for 5 min, 30 cycles at 95 °C for 30 s, 61 °C for 30 s, 72 °C for 30 s, and a final extension at 72 °C for 5 min. The PCR products were electrophoresed on a 1–2% agarose gel, stained with FluoroDye (SMOBIO, Taiwan), and visualized under ultraviolet light. Subsequently, Sanger sequencing was performed, and the sequencing files were analyzed using CodonCode Aligner 10.0.2 for Windows ( www.codoncode.com ). Linkage disequilibrium (LD) analysis Haploview (Haploview Software, USA, www.broadinstitute.org/haploview ) was employed to analyze LD 42 . Haploview generated D’, logarithm of the odds (LOD), and r 2 factors. The results are presented in S2 Table. In silico analysis Finding transcription-factor binding sites in the SNPs’ regions Transcription-factor (TF) binding sites in ANRIL introns were discovered through the application of related databases in the UCSC Genome Browser track data hubs ( https://genome.ucsc.edu/ ) 43 . The list of the databases is as follows: TF ChIP-seq Clusters from ENCODE3, eCLIP (by biosample) (ENCODE), eCLIP (by target) (ENCODE), ENCODE TFs, ENC RNA Binding, ENC TF Binding, TFBS Conserved, JASPAR CORE 2022-Predicted TF Binding Site, TOBIAS footprint prediction, human P53 Binding And Expression Resource (BAER) hub, ReMap 2022 Regulatory Atlas, and UniBind 2021 Hub. Statistical analysis The mean and the standard deviation were calculated in GraphPad Prism 9 for Windows (GraphPad Software, San Diego, California USA, www.graphpad.com ) for each parameter between the premature CAD and non-CAD subjects. Risk-factor probabilities were analyzed with the aid of the χ 2 test for qualitative variables and the t test for quantitative variables. The D'Agostino–Pearson omnibus normality test was also performed. Additionally, a nonparametric t test (Mann–Whitney) was applied for values not fitting in a normal distribution. One-way ANOVA was utilized to analyze the values of more than two groups, and the Spearman correlation was applied to generate a correlation matrix for all variables. The Receiver Operating Characteristic (ROC) analysis was applied to assess the diagnostic potential of ANRIL expression. ROC curves were drawn using GraphPad Prism 9. Ethics approval The study protocol was approved by the Ethics Committee of Rajaie Cardiovascular Medical and Research Center (RHC.AC.IR.REC.1396.62), and the study was conducted in accordance with the Helsinki Declaration. The informed consent was obtained from all subjects and/or their legal guardian(s).
Results Clinical and demographic features represented an association with premature CAD The results of clinical and demographic characteristic analysis are included in Table 1 . The levels of triglyceride, total cholesterol, low-density lipoprotein, and fasting blood sugar were significantly associated with premature CAD. Furthermore, T2D, stroke history, and familial CAD were significantly more frequent in the premature CAD group than in the non-CAD group. The distribution of age in CAD and non-CAD groups respective to and irrespective of sex is depicted in S1 Figure. ANRIL was expressed differentially in the premature CAD group compared with the non-CAD group ANRIL featured 14 splicing variants in the Ref Seq database and encompassed many regulatory elements along its sequence (Fig. 1 ). ANRIL was significantly downregulated in the serum samples of the premature CAD group compared with the non-CAD control group ( P < 0.05) (Fig. 2 ). The detection of ANRIL yielded diverse results through different primer sets. The expression analysis of ANRIL was performed using three different sets of primers amplifying exons 1, 5–6, and 19. The detection of middle exons (exons five-six) showed a nonsignificant decrease in ANRIL expression in the premature CAD group compared with the non-CAD group (Fig. 2 D), while the detection of terminal exons manifested a significant downregulation (Fig. 2 A,G). Diabetes correlated with higher ANRIL expression levels Patients with both premature CAD and T2D showed lower expression of ANRIL than those categorized as non-CAD. Variations in ANRIL expression using different primers were adopted here as well (Fig. 2 B,E,H). Diabetic patients with no symptoms of premature CAD indicated higher levels of ANRIL than those who did not have T2D or had premature CAD. Nevertheless, in the patients with premature CAD and T2D, the expression of ANRIL remained low altogether. The history of familial CAD was in line with lower ANRIL expression levels A history of familial CAD per se lacked contribution to significantly lower ANRIL expression. However, along with premature CAD, it preserved the low expression of ANRIL altogether. Genetic variants within and proximal to the ANRIL sequence displayed different genotype and allele frequencies but not a significant association with premature CAD Associations between premature CAD status and the genotypes of six SNPs were investigated. No significant associations were found between premature CAD and rs10757274, rs2383206, rs2383207, rs496892, rs10757278, and rs10738605. An overview of the studied variants is summarized in S3 Table. The odds ratios and P values for the different genetic models are listed in Table 2 . The distributions of the genotypes for all 6 SNPs were consistent with the Hardy Weinberg law at a significance level of 0.05 (Table 2 ). The frequencies of risk genotypes for all the SNPs were higher in the premature CAD group than in the non-CAD group but the differences lacked statistical significance. An association existed between ANRIL expression and the genotypes of rs10757274, rs2383206, rs2383207, rs496892, rs10757278, and rs10738605 While rs10738605 resided in exon six of ANRIL , the other SNPs were in noncoding regions. Among these, rs10757274 and rs496892 are included in the sequences of long interspersed elements (LINEs). Multiple binding sites for transcription factors (TFs) were included in the regions containing the studied SNPs ( S2 Fig.). The expression of ANRIL in the premature CAD subjects carrying the risk genotypes was lower than that in the control group (Fig. 3 ). For rs10757274, rs2383206, rs2383207, and rs10757278, the expression of exon 19 of ANRIL (a representative of long isoforms) was significantly downregulated in PCAD carriers of risk genotypes, however, detection of the first exon of ANRIL was significantly reduced in PCAD carriers of risk genotypes by rs496892 and rs10738605. For rs10757274, rs2383206, rs2383207, rs10757278, and rs10738605, the risk genotypes were GG compared to TT for rs496892. The Receiver Operating Characteristic (ROC) analysis revealed that ANRIL expression could differentiate premature CAD from non-CAD regarding rs10738605 and rs496892 The ROC analysis was performed to investigate whether subjects with the risk genotypes of the studied SNPs could be differentiated into premature CAD and non-CAD groups regarding ANRIL expression. The ROC curves (Fig. 3 G,H) revealed that in the subjects carrying the risk genotype of rs10738605, the expression of ANRIL was capable of separating premature CAD from non-CAD subjects with a sensitivity of 83%. Further, in the subjects carrying the risk genotypes for a combination of rs10738605 and rs496892, ANRIL expression predicted the patients correctly with 84% sensitivity (Fig. 3 H). The LD analysis ascertained two distinct LD blocks Haploview revealed strong LD between rs10757274, rs2383206, rs2383207, and rs10757278 and between rs496892 and rs10738605. Haploview demonstrated two distinct blocks: one for rs496894 and rs10738605 and another block for rs10757274, rs2383206, rs2383207, and rs10757278 (F i g. 3 I). The correlation analysis demonstrated strong links between some variables The correlation matrix is depicted as a heat map in Fig. 3 J. A negative correlation was detected between the status of premature CAD and age, triglyceride, fasting blood sugar, stroke history, T2D, and familial CAD. Positive correlations were observed between the expression of ANRIL using different primers (E1, E5-6, and E19), and between the status of the rs10757274, rs2383206, rs2383207, and rs10757278 genotypes as well. Some weaker positive correlations were discovered between the expression of ANRIL and the level of high-density lipoprotein and between familial CAD and familial myocardial infarction. Positive correlations were also indicated between the levels of triglyceride, cholesterol, fasting blood sugar, stroke history, and T2D (Fig. 3 J).
Discussion Atherosclerosis normally progresses in the form of late plaques in the fourth decade of life and involves older people. Nonetheless, in some cases, women aged below 55 and men younger than 45 experience CAD symptoms 3 . The principal cause of such incidence remains indefinite, although previous studies have reported the contribution of genetic factors in promoting atherosclerosis at younger ages 6 , 24 . The gold-standard diagnosis of CAD is angiography, which is an invasive technique. Developing noninvasive and low-cost screening methods to predict premature CAD occurrence in young adults is crucial since CAD has various socioeconomic effects on populations. According to a previous study, ANRIL expression exhibited post-treatment recovery. A negative correlation also existed between the number of diseased vessels and the expression of ANRIL 11 . The detection of terminal exons (E1 and E19) manifested promising potential to differentiate premature CAD from non-CAD, while the identification of medial exons (E5-6) lacked such a capability. This might be due to the differential expression of linear and circular isoforms in serum since internal exons are mostly enriched in circular variants 10 . Diabetes is the most common risk factor for premature CAD in the Iranian population 2 . It is associated with higher levels of ANRIL expression. Our CAD patients who suffered from T2D demonstrated lower expression of ANRIL , denoting the dominant effect of CAD over T2D. In contrast, our diabetic patients with no premature CAD symptoms expressed ANRIL to significantly higher extents. Our data chime in with the findings of a previous study 25 . A history of familial CAD was in keeping with the influence of CAD itself on the expression of ANRIL and demonstrated no independent association with CAD. A substantial body of evidence indicates that lncRNAs function through their distinct secondary structure, resulting in the dynamic accessibility of exons during cellular functions. The results of a study demonstrated a relationship between the presence of Alu elements in exons and their differential expression 10 . There are shreds of evidence that noncoding variants can affect expression as well as exonic ones. Intronic SNPs exert their functional role through their existence in exon–intron junctions, splicing branch points and enhancers, or through the creation of cryptic splice sites. Although most functional SNPs reside within a 30-bp distance from splice sites, SNPs in the middle of introns should not be underestimated 26 . While some of our investigated SNPs reside in the binding sites of transcription factors, some are included in LINEs, explaining the regulatory effect of these variants on ANRIL expression. LINEs can regulate gene expression by mediating chromatin remodeling, regulating transcription, and altering the stability of transcripts. They can function as promoters or enhancers as well 27 . Previous studies on the Iranian population have introduced rs10757274 and rs2383206 as SNPs associated with CAD 28 – 30 . Only a few investigations are available on premature CAD in the Iranian population 2 . There are few studies on ANRIL expression in the Iranian population. Previous studies in the Iranian population measured ANRIL expression in peripheral blood samples from patients with (aged) CAD 25 , 31 , whereas this study examined ANRIL expression in sera of premature CAD patients. Serum provides a higher sensitivity for biomarker detection and contains stable RNAs mostly enriched in extracellular vesicles 20 . As of yet, there are no serum expression data available for the Iranian population. Yari et al. 31 demonstrated no significant difference in the expression of NR_003529 transcript between CAD patients and the control group. However, we revealed a significant association between lower expression of ANRIL (NR_003529), the longest ANRIL transcript, and premature CAD in the Iranian population. Despite this discrepancy, their expression analysis of EU741058 transcript was consistent with ours regarding the first exon primer. Rahimi et al. 25 reported an upregulation in ANRIL expression in diabetic patients with CAD compared to non-CAD diabetic subjects. Here, we investigated ANRIL expression in premature CAD patients with and without diabetes. Serum expression of ANRIL showed a different pattern compared to peripheral blood mononuclear cells' expression. The correlation between ANRIL expression and the status of SNPs in its sequence has not yet been reported in Iranian patients with premature CAD. Here, we demonstrated that PCAD carriers of risk genotypes for rs496892 and rs10738605 showed lower expression of first exon of ANRIL . rs496892 and rs10738605, both are proximal to the first exons of ANRIL and revealed a strong linkage disequilibrium in our population. On the contrary, rs10757274, rs2383206, rs2383207, and rs10757278 reside in the 3' of ANRIL and correlate with lower expression of exon 19. These four SNPs share a haplotype block together supporting our findings on the effect of these SNPs on expression of terminal exons of ANRIL . Linking DNA variants to function is one of the major challenges in deciphering the influence of genetics on CAD. Unlike monogenic diseases, multifactorial disorders result from many small contributing factors. Therefore, large-scale studies are needed to validate the association between these variants and diseases 32 . Our ROC analysis demonstrated that ANRIL lacked predictive value as a biomarker for premature CAD per se (data not shown). Nevertheless, ANRIL expression could distinguish between premature CAD and non-CAD in carriers of rs496892 and rs10738605 risk genotypes (Fig. 3 G,H). In this group, the expression of ANRIL could accurately determine the status of premature CAD by 83% and 84% sensitivity (respectively for rs10738605 and a combination of rs10738605 and rs496892); still, the specificity values were slight (54% and 55%, respectively). Further research requires larger sample sizes. GWASs are incapable of discovering disease-causing SNPs solely. They only narrow down genetic variants to the most associated SNPs with the disease. Due to the population stratification in large-scale GWASs, the confirmation of the associations by replication studies in smaller and more homogenous populations is crucial 33 , 34 . Our findings did not replicate the association between the studied SNPs and premature CAD significantly. It might be a consequence of the small sample size, which was limited because of the exclusive age group of the included subjects. As proposed by the 1000 Genome Project, genetic variants are categorized into four groups: population-specific, continental area-specific, continental areas-shared, and all continents-shared 35 . The structure of the population also contributes to differences in the frequency of genotypes between sub-populations even in a shared geographic location 36 . It is essential to investigate the association between genetic variants and diseases in populations of various ancestries. Different expression and LD profiling among populations were convincing evidence to discover the expression, LD, and genetic association screening of the ANRIL region in the Iranian population. The interplay of SNP and expression has been reported to vary between populations as well 37 – 39 . The association between genetic variants and diseases in 1 population does not simply generalize to other populations. Besides ethnicity, age and other environmental factors affect the variation 40 . DNA variants and RNA expression lack predictive value when used solely, while a combination of both approaches is worthwhile. The last decade has witnessed a considerable fall in the costs of sequencing and expression analyses, raising the hopes of using genetics as a helping arm in clinics. Our principal notion is that a combination of DNA and RNA variants is capable of predicting premature CAD in young adults. Premature CAD is considered a burden on the health systems of countries: not only does it negatively impact a young workforce but also it contributes to societal challenges. Traditional screening methods lack the strength to precisely predict premature CAD 41 . What could substantially strengthen the armamentarium is the recruitment of artificial intelligence and machine learning.
Coronary artery disease (CAD) is the major cause of mortality in the world. Premature development of CAD can be attributed to women under 55 and men under 45. Many genetic factors play a part in premature CAD. Among them, ANRIL , a long noncoding RNA is located at the 9p21 risk locus, and its expression seems to be correlated with CAD. In the current study, premature CAD and control blood samples, with and without Type 2 Diabetes (T2D), were genotyped for six SNPs at the 9p21 locus. Additionally, ANRIL serum expression was assessed in both groups using real-time PCR. It was performed using different primers targeting exons 1, 5–6, and 19. The χ 2 test for association, along with t-tests and ANOVA, was employed for statistical analysis. In this study, we did not find any significant correlation between premature coronary artery disease and rs10757274, rs2383206, rs2383207, rs496892, rs10757278 and rs10738605. However, a lower ANRIL expression was correlated with each SNP risk genotype. Despite the correlation between lower ANRIL expression and CAD, Type 2 diabetes was associated with higher ANRIL expression. Altogether, the correlation between ANRIL expression and the genotypes of the studied SNPs indicated that genetic variants, even those in intronic regions, affect long noncoding RNA expression levels. In conclusion, we recommend combining genetic variants with expression analysis when developing screening strategies for families with premature CAD. To prevent the devastating outcomes of CAD in young adults, it is crucial to discover noninvasive genetic-based screening tests. Subject terms
Limitations of this study It is necessary to consider the limitations of the current study when interpreting its outcomes. Firstly, our small sample size precludes the generalizability of our results concerning the association of rs10757274, rs2383206, rs2383207, rs496892, rs10757278, and rs10738605 in Iranian patients with premature CAD. Secondly, our age threshold limited the number of participants in this study. Therefore, we strongly recommend replicating this study with a larger sample size of the Iranian population. To gain a better understanding of the independent effect of premature CAD occurrence on ANRIL expression, parallel studies on older individuals with CAD are strongly recommended. Supplementary Information
Supplementary Information The online version contains supplementary material available at 10.1038/s41598-024-51715-2. Author contributions E.T.B.: experiment design, lab work, data production, data interpretation, writing first draft of manuscript; A.Z., F.S., & M.M.: cardiovascular counselling, angiography results’ interpretation, data interpretation; H.B.: statistical analysis, data interpretation; S.J.M.: project design, data interpretation, manuscript edit; M.M.: project design, experiment design, data interpretation, manuscript final edit. All authors reviewed and confirmed the final draft. Funding This research was supported in part by a research grant awarded to Dr. Mahshid Malakootian from NIMAD (National Institute for Medical Research Development) [no. 943104] and Research Deputyship of Rajaie Cardiovascular Medical and Research Center [no. 96060]. Data availability The datasets generated during and/or analysed during the current study are available in the GenBank repository, sequences are accessible via OQ744705–OQ745198 accession numbers for CAD patients and OQ745199–OQ745577 for non-CAD subjects. The detailed list of accession numbers is available from the corresponding author. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1244
oa_package/2d/75/PMC10787829.tar.gz
PMC10787830
38218945
Introduction The presence of cell-in-cell (CIC) structures has been observed in a wide range of human cancers and is mainly associated with poor prognosis [ 1 – 7 ]. CIC formation involves a dynamic interaction between an outer and an inner cell. It can be the result of different types of mechanisms involving either homotypic or heterotypic interactions and initiated by either the outer (endocytic CIC) or the inner (invasive CIC) cell [ 8 ]. Entosis is defined as a homotypic invasion occurring between tumor cells of the same type where one living cell is internalized inside another one. After internalization by the outer cell, the inner cell is found in the entotic vacuole [ 9 ], where it will be degraded generally by autophagy, a non-apoptotic cell death involving the lysosome fusion. Entosis was discovered in vitro in breast cancer cells (MCF-7) cultured under different stress factors such as extracellular matrix detachment, glucose starvation, or ultraviolet radiation [ 10 – 12 ]. Entosis was also described in cells with prolonged aberrant mitosis to eliminate the aneuploid progenies by engulfment, maintaining thus the genome integrity [ 13 ]. Mechanistically, entosis was found dependent on cell-cell adhesion molecules such as cadherins and driven by imbalances in actomyosin contractility which is mainly under the control of the RhoA/ROCK pathway [ 10 , 14 ]. Rnd3, also known as RhoE, is an atypical RhoGTPase protein and a negative regulator of the Rho/ROCK pathway. Others and we previously described Rnd3 as a tumor suppressor in liver cancer [ 15 , 16 ]. Indeed, RND3 expression is downregulated in human hepatocellular carcinoma (HCC) samples compared to non-tumoral liver tissues and correlated with poor survival rates [ 17 , 18 ]. It is implicated in cancer development through the regulation of cell growth and invasion [ 17 , 19 ]. While working on the role of Rnd3 in HCC, we observed CIC structures, prompting us to study this phenomenon in liver cancer cells. Here, we reported that HCC cells are susceptible to form entosis after Rnd3 downregulation in cultured liver cancer cell lines and in xenografts in mice. Entosis is highly dependent on the RhoA/ROCK pathway, but not on E-cadherin. We also found that Rnd3 loss leads to Lamp-1 up-regulation, required for the internalization and degradation of the entotic cell. Finally, we related that entosis is rare in human HCC tissues, but associated with poor prognosis. Our results suggest that entotic engulfment induced by the loss of Rnd3 in HCC promotes liver tumor progression.
Materials/subjects and methods Cell culture The liver cancer cell lines (Hep3B, Huh7, HepG2, Huh6) and the breast adenocarcinoma cell line MCF-7 were cultured in Dulbecco modified Eagle’s medium (DMEM 1x glutamax, Fisher Scientific) supplemented with 10% heat-inactivated fetal bovine serum and incubated at 37 °C in a humidified 5% CO2 atmosphere. Cell line authentication was performed using short tandem repeat analysis, and the absence of mycoplasma contamination in cell culture media was tested every week. The nutrient deprivation was performed by culturing the cells in DMEM 1x glutamax with a very low concentration of glucose (Fisher Scientific) and supplemented with 10% of dialyzed heat-inactivated FBS for 10–13 h. For induction of cell suspension, the adherent cancer cells were trypsinized and cultured in non-treated plastic Petri dishes for 13 h. Stable cell lines with fluorescent nuclei or fluorescent F-actin were generated through transduction with lentiviruses expressing H2B-GFP (Addgene #25999), H2B-RFP (Addgene #26001) or LifeAct-mRuby. Transient knockdowns were done by transfection of small interfering RNAs (siRNAs) into cells using the lipofectamine RNAi max (Invitrogen) according to its manufacturer’s protocol. siRNAs targeting Rnd3 (SiRnd3 #1, SiRnd3 #2, SiRnd3 #3), p190RhoGAP-A (Sip190RhoGAP-A #1, Sip190RhoGAP-A #2), RhoA and E-cadherin (siE-cadherin #1) were purchased from Eurofins Genomics and the sequences are presented in Supplemental Table 5 . siE-cadherin #2 were purchased from Thermo Fisher Scientific (CDH1, cat#: 4427037, ID: s2768). Control siRNA corresponds to AllStars Negative control from Qiagen. Transient transfection of plasmids was performed using Lipofectamine TM 3000 reagent according to the manufacturer’s protocol (Thermo Fisher Scientific). The E-Cadherin-GFP construct was a generous gift from Dr. Peter Coopman (Montpellier). To induce a stable suppression of endogenous Rnd3 expression, we used Hep3B-shRnd3 and Huh7-shRnd3 cell lines with conditional, doxycycline-dependent, expression of shRnd3 as described previously [ 30 ]. Hep3B-shCtrl and Huh7-shCtrl cell lines, conditionally expressing the control shRNA targeting the firefly luciferase, were used as controls. Stable cell lines were cultured as described above and shRNA expressions were induced by 50 mg/mL of doxycycline. For ROCK inhibition, cells were treated with Y-27632 (Sigma Aldrich) at 5 μM for 24 h. Sorafenib (Bay 43–9006, Enzo Life Science) was used at two doses 8 or 10 μm to treat Huh7 cell lines. Antibodies, immunoblot analysis Cells were lysed in RIPA lysis Buffer (Sigma) supplemented with protease inhibitor cocktail (Roche Diagnostics) and protein concentration was determined using Bio-Rad Protein Assay (Lowry). 60 μg of proteins from each sample were separated on 10% polyacrylamide gel (Bio-Rad) and blotted onto nitrocellulose membranes (0.2 μm nitrocellulose, Bio-Rad) using Trans-Blot Turbo Transfer System (Bio-Rad). The membranes were blocked in Odyssey blocking buffer for 30 min and then incubated with each of the following specific primary antibodies: Mouse anti-RhoE/Rnd3 (1:1000, clone 4, Cell Signaling, #3664 S), Rabbit anti-actin (1:2000, Sigma Aldrich, #A2066), Mouse anti-HSP90 (1:1000, Santa Cruz, #sc-69703), Rabbit anti-Mypt-1 (1:500, Millipore, #07-672-I), Rabbit anti-phospho-Mypt-1 (Thr696) (1:500, Millipore, #ABS45), Mouse anti-RhoA (1:1000, Santa Cruz, #sc-418), Mouse anti-p190A (1:1000, BD Biosciences, #610149), Mouse anti-E-cadherin (1:1000; BD Biosciences, #610182), Mouse anti-HSP90 (1:1000, Santa Cruz, #sc-69703), Mouse anti-LAMP1 (1:1000, Santa Cruz, #sc-20011), Rabbit anti-GFP (1/1000, Abcam, #ab290) either 1 h at room temperature or overnight at 4 °C. After incubation with the specific secondary antibodies, all blots were analyzed with the Bio-Rad Chemidoc system. Immunofluorescence staining The following primary antibodies were used for immunofluorescence (IF): Mouse anti-beta-catenin (1:400; BD Biosciences, #610154), Mouse anti-E-cadherin (1:50; BD Biosciences, #610182), Mouse anti-LAMP1 (1:100, Santa Cruz, #sc-20011). IF was performed on adherent or suspended cells after cytospin using the Shandon Cytospin 2 centrifuge at 110 rpm for 5 min. Cells were fixed and permeabilized in 4% paraformaldehyde PFA (Electron Microscopy Science, #15710) and 1% Triton X-100 respectively for 10 min at RT, followed by PBS washing and blocking in PBS-5%BSA for 20 min. Cells were then incubated with primary antibodies for 45 min followed, after PBS washing, by a incubation with the secondary antibodies (Interchim) specific for primary antibodies. F-actin was stained using fluorescent phalloidin (Molecular Probes). Finally, the cells were counterstained with DAPI (Sigma, #D9542) and the coverslip mounting on the glass slide was done using Fluoromount-G medium (Interchim #FP-483331). Time-lapse microscopy Hep3B cells expressing H2B-GFP and Lifeact-mRuby were cultured on glass-bottom dishes 35 mm high (Ibidi) after Rnd3 silencing, and time-lapse microscopy was performed in 37 °C and 5% CO 2 live-cell incubation chambers. The fluorescence was acquired every 2 h for 48 h using the spinning-disk LiveSR confocal microscope. The image analysis and the video reconstitution were done using ImageJ program. Correlative light-electron microscopy (CLEM) Hep3B cells expressing H2B-GFP were transfected with siRNA targeting Rnd3 (SiRnd3 #1) and cultured on glass coverslips gridded and numbered (Delta microscopy, #72265-12). After selection of the area of interest containing the entotic events using fluorescent microscopy, cells were fixed with 2.5% glutaraldehyde (Electron Microscopy Science, #15960) for 30 min at RT followed by incubation at 4° for 1 h in the dark. After washing with Sorensen’s phosphate buffer 0.2 M, pH 7.4 (Electron Microscopy Science, #11601-10), cells were dehydrated in graded series of ethanol solutions, including 50%, 70%, 90% for 5 min and 100% for 2 times, 5 min at RT in the dark. Finally, the cells were dried with CO 2 and observed using a Zeiss GeminiSEM300 microscope. Entosis quantification The entotic event is defined by 1) the presence of a cell inside another one with a deformed nucleus and 2) the high concentration of actin staining around the inner cell. The entotic cells were observed with epifluorescence microscopy (Zeiss). Quantification was done by counting 300–400 of total cells for each condition. The percentage of entotic events was calculated by dividing the number of entotic events by the total. SILAC labeling, laser capture microdissection, and proteomic analysis To discriminate laser-captured proteins from undesirable exogenous contaminating proteins, Hep3B-H2B-GFP cells were first metabolically labeled using stable isotope labeling with amino acids in cell culture (SILAC) method, as previously published [ 22 ]. For that, Hep3B-H2B-GFP cells were cultured in DMEM medium (Dulbecco’s modified Eagle’s medium, Invitrogen) supplemented with 10% dialyzed fetal bovine serum, 200 mg/L L-proline, and 84 mg/L l -Arginine and Lysine. The incorporation of labeled amino acids was done after six cycles of cellular doubling. After transfection with siRnd3 #1, cells were seeded for 12–24 h on LamPen coated with collagen matrix allowing cells to adhere. Cells were fixed with PFA and the laser capture microdissection was performed using PALM type 4 micro-beam (Zeiss). SILAC-labeled cells were transfected with Rnd3-targeting siRNAs and entotic cells with crescent-shaped nuclei were microdissected. To obtain enough material for proteomic analysis, we manually collected 2000 cells in triplicate. With 2000 isolated entotic cells, we have identified a mean of 2880 13 C peptides corresponding to 406 proteins with at least 2 specific peptides. Given the small quantity of material analyzed, the first step guarantees the specificity of identifications by distinguishing labeled proteins from microdissected cells from unlabeled environmental contaminants (keratins, skin and hair proteins, etc...). In parallel, a standard range (500 ng to 5 ng) of protein quantity was done on cells labeled with SILAC and transfected with control siRNA (siCtrl) in order to compare the same quantity of entotic cells microdissected and obtained after transfection with siRnd3 to cells transfected with siCtrl. Three independent biological replicates on total protein extracts from SILAC-labeled cells were compared by label-free protein quantification. Proteins were loaded on a 10% acrylamide SDS-PAGE gel and proteins were visualized by Colloidal Blue staining. Migration was stopped when samples had just entered the resolving gel and the unresolved region of the gel was cut into only one segment. The steps of sample preparation and protein digestion by the trypsin were performed as previously described [ 31 ]. NanoLC-MS/MS analysis was performed using an Ultimate 3000 RSLC Nano-UPHLC system (Thermo Scientific, USA) coupled to a nanospray Orbitrap FusionTM LumosTM TribridTM Mass Spectrometer (Thermo Fisher Scientific, California, USA). Each peptide extracts were loaded on a 300 μm ID x 5 mm PepMap C 18 precolumn (Thermo Scientific, USA) at a flow rate of 10 μL/min. After a 3 min desalting step, peptides were separated on a 50 cm EasySpray column (75 μm ID, 2 μm C 18 beads, 100 Å pore size, ES903, Thermo Fisher Scientific) with a 4–40% linear gradient of solvent B (0.1% formic acid in 80% ACN) in 57 min. The separation flow rate was set at 300 nL/min. The mass spectrometer operated in positive ion mode at a 2.0 kV needle voltage. Data was acquired using Xcalibur 4.4 software in a data-dependent mode. MS scans ( m / z 375–1500) were recorded at a resolution of R = 120000 (@ m / z 200), a standard AGC target, and an injection time in automatic mode, followed by a top speed duty cycle of up to 3 s for MS/MS acquisition. Precursor ions (2 to 7 charge states) were isolated in the quadrupole with a mass window of 1.6 Th and fragmented with HCD@28% normalized collision energy. MS/MS data was acquired in the ion trap with rapid scan mode, a 20% normalized AGC target, and a maximum injection time in dynamic mode. Selected precursors were excluded for 60 s. Protein identification was done in Proteome Discoverer 2.5. Mascot 2.5 algorithm was used for protein identification in batch mode by searching against a UniProt Homo sapiens protein database (75796 entries, released September 3, 2020; https://www.uniprot.org/ website). Two missed enzyme cleavages were allowed for the trypsin. Mass tolerances in MS and MS/MS were set to 10 ppm and 0.6 Da. Oxidation (M), acetylation (K), SILAC modifications (K, R) were searched as dynamic modifications, and carbamidomethylation (C) as static modifications. Raw LC-MS/MS data were imported into Proline Studio for feature detection, alignment, and quantification. Protein identification was accepted only with at least 2 specific peptides with a pretty rank=1 and with a protein FDR value less than 1.0% calculated using the “decoy” option in Mascot. Label-free quantification of MS1 level by extracted ion chromatograms (XIC) was carried out with parameters indicated previously [ 31 ]. The normalization was carried out on the median of ratios. The inference of missing values was applied with 5% of the background noise. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository [ 32 ] with the dataset identifier PXD043640. Immunohistochemistry on HCC samples Mouse HCC samples were from Hep3B xenografts previously described [ 19 ]. This model consisted of subcutaneous inoculation of Hep3B-shCtrl or Hep3B-shRnd3 cells in immunodeficient NOG mice treated with doxycycline to induce or not Rnd3 knockdown. The analysis of entosis was performed on tumor sections stained with hematoxylin to mark the cytoplasm and nuclei of cells. Human samples came from resected or explanted livers of HCC patients treated in Bordeaux from 1992–2005. The characteristics of HCCs used for the IHC analysis (10 HCCs) are indicated in Table 1 . Formalin-fixed, paraffin-embedded sections of human or mouse HCC samples were used for Hematoxylin-Eosin or immunohistochemical staining. Staining of Rnd3 and pan-keratin/beta-catenin was performed as previously published [ 17 , 33 ]. Statistical analysis The frequency of entotic cells was represented as a percentage (mean ± SD) of the total counted cells for at least three experiments. Data were analyzed using GraphPad Prism 10. For all experiments, significance was determined with the Mann–Whitney U -test, * p < 0.05, ** p < 0.01 , *** p < 0.001.
Results Liver cancer cells are prone to perform entosis In order to investigate entosis in liver cancer, cultured cells (Hep3B, Huh7, Huh6, or HepG2) were subjected to stress conditions known to favor entosis in breast cancer cells such as nutrient deprivation [ 11 ] or matrix detachment [ 10 ]. MCF-7 human breast cancer cells were used as a positive control. Cell internalization was analyzed and quantified by confocal microscopy after immunostaining of nuclei, F-actin, and β-catenin to visualize nucleus deformation, shape, and cell periphery respectively. The outer cell shows a crescent-shaped nucleus, whereas the inner cell is rounded and retains its plasma membrane (Fig. 1A ). In normal conditions, entotic structures can be observed in about 1–3% of total adherent liver cancer cells. After nutrient deprivation, a significant increase in the percentage of entotic cells is observed in all four cell lines, with up to 10% of entotic cells in HepG2 and Huh6 cells, when compared to normal culture conditions (Fig. 1B–C ). Matrix detachment is also described as an inducer of entosis in MCF-7 cells [ 10 ]. In order to test this stimulus, cells were cultured in non-adherent conditions and deposited on a slide by cytospin for immunostaining. The results show an increase of entotic cells after matrix detachment in all liver cancer cells, even if the percentage of entotic cells remains lower in liver cancer cells (6–14%) when compared to MCF-7 cells (Fig. 1D ). As entosis is favored under stress conditions, we attempted to address whether hypoxia or drug treatment may trigger this mechanism in liver cancer cells. Hypoxic conditions were monitored by the increase in the mRNA expression of GLUT1 and VEGF , two target genes of Hypoxia-Inducible Factor (Hif1α) [ 20 ] (Supplementary Fig. 1A ). The quantification of entotic Hep3B and Huh7 cells under hypoxia did not show any significant change compared to basal levels (Supplementary Fig. 1B ). We also evaluated whether HCC treatments like Sorafenib could be an inducer of entosis. While Sorafenib treatment altered pERK/ERK ratio as expected, no significant difference was observed in the percentage of entosis in treated versus untreated HCC cells (Supplementary Fig. 1C ). Altogether, our results demonstrate that liver cancer cells are prone to perform entosis upon proper stimuli such as nutrient deprivation or matrix detachment. CIC formation is driven by the loss of Rnd3 expression While working on the role of Rnd3 in HCC progression [ 17 , 19 ], we noticed that, upon Rnd3 silencing, some cells appeared to be inside other cells, reminiscent of entosis. To examine whether the silencing of Rnd3 may trigger this mechanism, we first analyzed Rnd3 expression in liver cancer and MCF-7 cells. While Hep3B and Huh7 cells strongly express Rnd3 and Huh6 cells slightly less, Rnd3 is not expressed in HepG2 or MCF-7 cells (Supplementary Fig. 2A ). We thus choose to silence Rnd3 in Hep3B and Huh7 cells using two different approaches: transient transfection of siRNAs using three different siRNAs targeting Rnd3 (SiRnd3#1, #2, #3), or inducible Rnd3 knockdown (KD) in cell lines stably expressing shRNA [ 19 ]. We found that Rnd3 KD led to a significant increase of CIC events in Hep3B and Huh7 cells regardless of the approach used (Fig. 2A ). Neighboring cells engulfed each other, with about 10% of adherent cells containing engulfed neighbors. CIC events induced by the loss of Rnd3 show similar characteristics to stress-induced entotic cells, such as crescent-shaped nuclei and round-engulfed cells (Fig. 2B ). More complex cell structures were observed, with three or more cells involved in sequential engulfments (Fig. 2C ). Thus, Rnd3 loss-mediated CIC events share common features with entosis observed in nutrient-depleted or matrix-detached conditions. Moreover, the combination of stimuli, i.e., silencing of Rnd3 and starvation did not increase the entosis mechanism in Hep3B and Huh7 cells, suggesting the involvement of the same pathways to mediate cell engulfment (Supplementary Fig. 2B, C ). The loss of Rnd3 expression, and therefore function, appears to be a trigger of cell engulfment in HCC cells. We next aimed to identify entosis in vivo in tumor tissues using the xenograft mouse model described previously [ 19 ]. We noticed a significant increase in entosis percentage in tumors generated from Rnd3-KD cells when compared to control Hep3B cells (Fig. 2D ). Thus, Rnd3 loss favors HCC CIC events both in vitro and in vivo. Characteristics of entosis stages after silencing of Rnd3 To investigate the CIC events induced by the loss of Rnd3, Hep3B cells expressing GFP-H2B and LifeAct-mRuby were treated with siRnd3 #1 and analyzed by time-lapse microscopy (Video 1 ). Entosis was initiated by the contact between the two cells, and then the internalization was marked by a high concentration of actin at the border between the two cells. The nucleus of the outer cell acquired a change in its shape due to the nucleus of the inner cell that pushes and deforms it. The final stage was characterized by the degradation of the inner cell, visible by the disappearance of the inner cell nuclei (Fig. 3A ). Using correlative light-scanning electron microscopy (CLEM) allowing analysis of the same entotic cell by fluorescence microscopy and scanning EM (Fig. 3B ), we found that the internalized cell has a rounded shape and seemed to be denser than the outer cell, with apparent cytoskeletal filaments. The inner cell appeared to be inside a large vacuole, the so-called entotic vacuole. The nucleus of the host cell acquired a crescent-like shape and was pushed to the cell periphery. The analysis of CLEM demonstrated that CIC structures mediated by the loss of Rnd3 are similar to entosis, showing a complete internalization of one cell inside another. To further examine whether the loss of Rnd3 was required in inner cells, outer cells, or both, we generated Hep3B cells with green and red nuclei by expression of GFP-H2B (G) and mCherry-H2B (R), respectively (Supplementary Fig. 3A ). Hep3B-mcherry-H2B were then transfected with siRnd3#1 and co-cultured with untreated Hep3B-GFP-H2B (Fig. 3C ; Supplementary Fig. 3A, B ). 48 h after mixing, entotic cells were analyzed by fluorescence microscopy and counted for each combination (G in G, G in R, R in G, or R in R) (Fig. 3C ). Among all entotic cells, the percentage of events between wild-type cells (G in G) is about 10%, and it reaches 50% between those transfected with siRnd3#1 (R in R). This result demonstrated that the decrease of Rnd3 expression is important in both the inner and outer cells for internalization. We noted that the percentage of the combination R in G is higher than that G in R combination suggesting that the loss of Rnd3 in the inner cell could be sufficient to induce its internalization in the wild-type outer cell. Entosis mediated by the loss of Rnd3 is dependent on an active Rho/ROCK pathway and a decrease of E-cadherin expression The Rho/ROCK pathway was implicated in the entosis induced after matrix detachment and starvation. As Rnd3 is a known antagonist of this signaling pathway, we analyzed whether the RhoA/ROCK pathway was required for cell engulfment after Rnd3 silencing. We first combined RhoA and Rnd3 knockdowns in HCC cells to assess RhoA involvement (Supplementary Fig. 4A ). We demonstrated that the inhibition of RhoA reverses the effect of Rnd3 silencing by decreasing the percentage of entosis in Hep3B cells (Fig. 4A , left panel). We next used the Y-27632 compound to inhibit RhoA downstream effector, ROCK. We confirmed that Y-27632 treatment was able to inhibit MYPT1 phosphorylation induced upon Rnd3 KD in Hep3B cells (Supplementary Fig. 4B ). Moreover, similar to RhoA KD, we found that this treatment decreases the percentage of entosis induced by the silencing of Rnd3 (Fig. 4A , right panel). Like Rnd3, p190RhoGAP (p190A) is a negative regulator of the Rho/ROCK pathway. HCC cells were transfected with siRNA targeting p190A (Supplementary Fig. 4C ). Consistently the silencing of p190A increased the percentage of entotic cells in Hep3B and Huh7 cell lines, similar results to those obtained after Rnd3 inhibition (Fig. 4B ). The inhibition of p190A expression in Hep3B and Huh7 cells in the presence of siRNA targeting Rnd3 did not modify the percentage of entotic cells compared to the condition where Rnd3 expression was inhibited alone, confirming that Rnd3 and p190A act in the same pathway for the induction of entosis (Supplementary Fig. 5 ). Altogether, these data demonstrate that entosis mediated by the silencing of Rnd3 is dependent on the RhoA/ROCK pathway in HCC cells. Therefore, deregulation of the Rho/ROCK pathway alters the ability of HCC cells to perform entosis. Entosis was described to be dependent on E-cadherin-based cell-cell junctions [ 14 ]. However, we have previously found a decrease in E-cadherin expression upon silencing of Rnd3 in HCC cells [ 17 ], which raises the question of the involvement of E-cadherin in the entosis observed here. We thus used E-cadherin-targeting siRNAs and analyzed their impact on entosis induced by the silencing of Rnd3 in HCC cells (Fig. 4C ; Supplementary Fig. 6A ). As previously published, E-cadherin expression decreased upon Rnd3 silencing in both cell types (Fig. 4C ). Our results showed that inhibition of E-cadherin in Hep3B and Huh7 cells did not alter the percentage of entotic cells compared to Rnd3 knockdown alone (Fig. 4C ), demonstrating that entosis mediated by Rnd3 loss is independent on E-cadherin in HCC cells. These data prompted us to explore the impact of E-cadherin silencing alone on entosis in HCC cells. Surprisingly, in contrast to breast cancer cells [ 21 ], we found that the decrease of E-cadherin expression promoted entosis in HCC cells (Supplementary Fig. 6B ). We then asked whether overexpression of E-cadherin could rescue the suppression effects of Rnd3 on entosis. To do so, we overexpressed a GFP tagged-version of E-Cadherin in Rnd3-silenced Hep3B cells (Supplementary Fig. 7A ). Whereas overexpression of GFP alone did not alter Rnd3 KD-induced entosis, we observed that overexpression of E-Cadherin-GFP significantly decreased the percentage of entotic cells suggesting that both Rnd3 and E-cadherin down-expression are required to promote entosis in Hep3B cells (Supplementary Fig. 7B ). Identification of LAMP1 as an effector of Rnd3 loss-mediated entosis In order to better characterize the entosis mediated by Rnd3 loss, we sought to apply a global approach. However, isolation of entotic cells is not an easy task, and we failed to do so using flow cytometry. We, therefore, applied a pipeline combining isolation of entotic cells by laser microdissection in a Hep3B-H2B-GFP cell population transfected with siRnd3 #1 (Fig. 5A , left panel) and mass spectrometry-based proteomic analysis according to our published protocol [ 22 ]. We obtained 139 proteins that were underrepresented in the entotic proteome with an entosis/total proteome abundance ratio ≤0.5 and 55 enriched proteins with an entosis/total proteome abundance ratio ≥2 (Supplemental Tables 1 and 2 ). We used Gene Ontology (GO) to classify the identified proteins (Supplemental Tables 3 and 4 ). We highlighted that the underrepresented proteins were mainly involved in cellular metabolic processes with 63% ( n = 88) of proteins associated with these biological processes. Proteins associated with intracellular anatomical structures and organelles were also found significantly decreased in entotic samples, suggesting a disassembly of cellular elements. Notably, we found a decrease in the abundance of Lamin-B1 and Lamin-B2, proteins of the nuclear envelope (Supplemental Table 2 ). In parallel, we found an enrichment of RNA binding proteins (42%, 23 proteins) including ribosomal and/or translation proteins (Supplemental Table 3 ). Network analysis using the Ingenuity Pathways Analysis platform (IPA, Qiagen) of the upregulated proteins further showed an implication of these candidates mainly in cell death/survival (Fig. 5A , right panel). We indeed found an overrepresentation of ER-associated proteins involved in stress response and/or folding of proteins such as Prdx2, HSPA5 (Bip), or protein disulfide-isomerases (P4HB, PDIA6), and PPIA, suggesting the recognition of cellular intrusion as a stressful cellular event (Supplemental Table 1 ). Network analysis using IPA also revealed a significantly altered functional network linked to the LAMP1 protein in entotic cells, where proteasome, chaperone, and ER stress proteins were also present (Supplemental Fig. 8 A). LAMP1 is known to be implicated in the degradation of the inner cells by autophagy during the entotic mechanism. We confirmed the high expression of LAMP1 in the inner cells in the early stage of degradation (Fig. 5B ). To validate the involvement of LAMP1 in entosis induced by Rnd3 silencing, we inhibited Rnd3 in the presence or absence of siRNA targeting LAMP1 in Hep3B and Huh7 cells (Fig. 5C ). Whereas, as previously shown, the silencing of Rnd3 increased the percentage of entosis in both cell lines, this percentage returned to control levels when Rnd3 and LAMP1 were co-silenced (Fig. 5D ). These data demonstrated the involvement of LAMP1 not only in the final stage of inner cell degradation but also in the overall mechanism of entosis induced by silencing of Rnd3. Interestingly, using RNAseq analysis of Hep3B cells (data not shown), LAMP1 gene expression was found upregulated upon Rnd3 silencing (Supplemental Fig. 8 B). These data were confirmed at the protein level in both Hep3B and Huh7 cells (Fig. 5C , bottom panel). This up-regulation of LAMP1 upon Rnd3 silencing appears to be dependent on RhoA as RhoA-KD rescued LAMP1 protein level (Supplemental Fig. 8C ), suggesting that LAMP1 and RhoA pathway are related factors. We further analyzed whether the overexpression of LAMP1 is sufficient to promote entosis. The ectopic expression of LAMP1 in Hep3B cells only slightly increased the percentage of entotic cells, which remains very low compared to that obtained with Rnd3-silencing alone. However, the combination of overexpression of LAMP1 and Rnd3 inhibition significantly increased the entosis percentage compared to Rnd3-silencing alone (Fig. 5E ). All these results suggest that Rnd3 favors entosis through LAMP1 expression and function. Thus, LAMP1 is an important factor to be considered all over the entotic mechanism. Entosis in patient tumors correlates with invasive features We next investigated the presence of entotic cells in HCC patient tissues. Using membrane (pan-keratin/β-catenin) and nucleus staining, we found few entotic cells in tumor samples (Fig. 6A ). These events were rare, but visible in the tumor part, using an immunofluorescence approach on tissues. We then performed Rnd3 staining on HCC sections and looked for entotic cells in negative and positive Rnd3 areas. We noticed that the entotic cells are present in tumor sections with low expression of Rnd3 (Fig. 6B ) supporting our in vitro data concerning the implication of Rnd3 downregulation in the entosis process. Using a cohort of 10 patient samples (Table 1 ), we then correlated the number of entotic cells with the characteristics of patient tumors. Similar to the low level of Rnd3 expression [ 17 ], we found that the number of entotic cells was significantly higher in tumors with satellite nodules or vascular invasion, which are indicative of local invasion of HCC (Fig. 6C ). All these results suggest the association between the entotic events, loss of Rnd3 and tumor progression.
Discussion In this study, we characterized entosis in liver cancer cells, which resembles to that largely described in the literature in breast cancer cells [ 23 ]. We demonstrated that entosis could be induced by matrix detachment and nutrient deprivation in HCC cells. In addition, we identified that entosis can be efficiently triggered in HCC cells through the loss of RND3 expression. Others and we previously described the downregulation of this gene expression in human HCC [ 17 , 18 , 24 ]. Moreover, Rnd3 loss was associated with liver cancer cell proliferation, invasion, chemoresistance, and senescence [ 17 – 19 , 25 ]. We revealed here that Rnd3 loss also favors entosis in HCC tumors. Recently, Rnd3 has been involved in the p53-dependent entotic mechanism driven by transient mitotic arrest in breast cancer cells [ 26 ]. However, in contrast to our results, it was found that Rnd3 expression under the regulation of p53 was essential to promote entosis in breast cancer cells. Thus, the Rnd3 role appears to be different in breast and liver cancers. In this line, Rnd3 expression does not seem to be regulated by p53 in HCC cells as no correlation between RND3 expression and TP53 mutations could be established in HCC human samples [ 17 ]. Rnd3 is a negative regulator of the RhoA/ROCK pathway. Herein, we confirmed the strong dependency of the entotic mechanism on this pathway. We found that entosis induced by the loss of RND3 is reduced after the inhibition of the Rho/ROCK pathway. Mechanically, entosis depends on actomyosin contractility regulated by the RhoA GTPase activity in the invading cell. It was described that outer cells have lower actomyosin contractility, consistent with a more deformable status compared to invading cells. In our co-culture experiments, the percentage of wild-type cells inside Rnd3-negative cells was very low, suggesting that the penetrating cells require an increase in contractility. However, in liver cancer cells, the loss of Rnd3 is important in both the inner and outer cells, as a maximum of entosis was found when Rnd3 expression is decreased in both the inner and the outer cells. This result is consistent with the existence of a required threshold for Rnd3 expression to favor entosis. Thus, a knockdown recreating an imbalance in contractility, in agreement with the literature, the more rigid cell could be internalized by a more deformable cell [ 21 ]. Consistently, we observed a high concentration of actin and the presence of fibers in the inner cells by electronic microscopy. Several studies have shown the importance of E-cadherin in entosis induced by matrix detachment and starvation [ 10 , 11 , 21 ]. Interestingly, we showed that entosis induced by the silencing of Rnd3 occurs independently of E-cadherin, and is even altered upon E-Cadherin re-expression. Consistent with our published results showing the decrease of E-cadherin expression upon silencing of Rnd3 in HCC cells [ 17 ], we even found that loss of E-cadherin favors entosis in Hep3B and Huh7 cells. Thus, entosis induced by the loss of Rnd3 could be dependent on other transmembrane proteins that may trigger cell-cell interaction. A proteomic analysis of laser-captured entotic cells allowed us to identify an overrepresentation of LAMP1 and ER-associated proteins involved in stress response and/or folding of proteins. Statistical analyses by Gene Set Enrichment Analysis did not reveal significant enrichment of canonical pathways of cellular degradation or stress. However, a functional environment of the LAMP1 protein is significantly modified with key proteins of these pathways. It is possible that entosis pathways involve only certain proteins known to be implicated in degradation and stress and mobilize a specific protein set. Our data confirm that entotic cells express highly LAMP1, a protein described in the literature as implicated in the final stage of entosis, i.e., degradation of inner cells. Indeed, the membrane of the entotic vacuole surrounding the internalized cells recruits LAMP1 and LC3 independently of the autophagosome formation. The fusion between the entotic vacuole and the lysosome of outer cells allows to degrade the inner cells [ 27 ]. Herein, we highlighted that LAMP1 expression is upregulated at the mRNA and protein levels upon Rnd3 silencing, in a RhoA-dependent manner. Our results are consistent with an involvement of LAMP1 in inner cell degradation after internalization induced by the loss of Rnd3. However, we found that LAMP1 is involved not only in degradation but also in the whole mechanism of entosis. LAMP1 could be among the different signals necessary for internalization. Although LAMP1 primarily resides at the lysosomal membrane, its localization to cell surface expression was described to mediate cell-cell adhesion and favor melanoma cell invasion [ 28 ]. The degradation of entotic cells could serve to feed cells, supporting the survival of the outer tumor cells in stress conditions. Entosis may also contribute to tumorigenesis by inducing aneuploidy [ 29 ]. Others and we previously described Rnd3 as a potential metastasis suppressor as its down-expression was associated with poor prognosis in HCC patients. The results presented here support the hypothesis that loss of Rnd3 could participate in HCC progression through the promotion of entosis. Accordingly, entosis in HCC patient tissues correlates with the presence of satellite nodules and vascular invasion. Thus, targeting the entotic mechanism may be valuable as a novel therapeutic avenue to impair HCC progression.
Entosis is a process that leads to the formation of cell-in-cell structures commonly found in cancers. Here, we identified entosis in hepatocellular carcinoma and the loss of Rnd3 (also known as RhoE) as an efficient inducer of this mechanism. We characterized the different stages and the molecular regulators of entosis induced after Rnd3 silencing. We demonstrated that this process depends on the RhoA/ROCK pathway, but not on E-cadherin. The proteomic profiling of entotic cells allowed us to identify LAMP1 as a protein upregulated by Rnd3 silencing and implicated not only in the degradation final stage of entosis, but also in the full mechanism. Moreover, we found a positive correlation between the presence of entotic cells and the metastatic potential of tumors in human patient samples. Altogether, these data suggest the involvement of entosis in liver tumor progression and highlight a new perspective for entosis analysis in medicine research as a novel therapeutic target. Subject terms
Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s41419-024-06420-3. Acknowledgements We thank N. Dugot-Senant from the Histopathology platform (TBMCore, UMS005, Bordeaux). Flow cytometry, hypoxia, and qRT-PCR experiments were respectively performed at the FACSility, CellOxia, and OneCell facilities (TBMCore, UMS005, Bordeaux). Electron and photonic microscopies were done in the Bordeaux Imaging Center, a service unit of the CNRS-INSERM and Bordeaux University, a member of the national infrastructure France Bio-Imaging, with the help of Dr. Etienne Gontier and Isabelle Svahn. We thank Dr. Peter Coopman (Montpellier) for providing us with the E-Cadherin-GFP construct. We express our gratitude to Paul Divet, Cyril Dourthe, and Pr. Clotilde Billottet (BRIC, Bordeaux) for their help, with, respectively, evaluation of entotic cells, bubble plot representation of proteomic data, and collecting patient tumor data. SB was supported by Bordeaux University and a postdoctoral fellowship from the Fondation ARC/Région Nouvelle-Aquitaine. This work was supported by grants from the Fondation ARC, La Ligue contre le Cancer (comité regional, Gironde), and from Institut National du Cancer (PLBIO-INCa2014-182) to VM. VM’s team was supported by La Fondation pour la Recherche Médicale “Equipe labellisée 2018”. Author contributions Study design: SB, VM. Generation of experimental data: SB, LD, C Dantzer, SDT, JWD. Analysis and interpretation of data: SB, LD, AAR. Providing biological samples from HCC patients: PBS, C Desdouets. Writing of the manuscript: SB, VM. Critical reading of the manuscript: LD, JWD, AAR, FS. Supervision of the project: VM. Data availability The mass spectrometry proteomic data have been deposited to the ProteomeXchange consortium with the dataset identifier PXD043640. Additional data are available from the corresponding author upon reasonable request. Competing interests The authors declare no competing interests. Ethics approval and consent to participate All animal procedures were approved by the ethical committee of the University of Bordeaux according to the French government regulations. To minimize animal use, we retrospectively analyze samples from a previous study [ 19 ]. Human samples were acquired from resected livers of patients with HCC treated in Bordeaux from 1992 to 2005. Written informed consent was obtained from all patients according to French legislation.
CC BY
no
2024-01-15 23:42:00
Cell Death Dis. 2024 Jan 13; 15(1):46
oa_package/9b/d5/PMC10787830.tar.gz
PMC10787831
38218871
Introduction Ferroelectricity, characterized by a remanent and switchable polarization, just experiences the 100th anniversary of its first discovery in Rochelle salt in 1920 1 , 2 . The research and application of ferroelectrics were first limited to bulk ceramics, such as in the domains of actuators, piezoelectric transducers, and pyroelectric detectors. Since 1989 3 , with the development of thin-films, the switchable polarization by an electric field greater than the coercive field has attracted amounts of effort for the military and commercial applications of ferroelectric films in non-volatile memories 4 – 15 . The high switching speed at sub-nanosecond and the excellent endurance of over 10 12 make the ferroelectric memory a competitive candidate for replacing the current flash memory products which suffers from slow speed of ~10 −3 s and limited endurance of ~10 4 cycles 16 . Furthermore, the programmability of polarization renders the ferroelectric memory viable for in-memory computing 17 – 24 , which offers huge advantages in terms of computing power, latency, and energy efficiency by performing parallel multiply-accumulate (MAC) calculations directly with Ohm’s law for multiplication and Kirchhoff’s law for accumulation 25 – 28 . Recently, several breakthroughs in materials technology, such as Si-compatible hafnia-based ferroelectrics with 3D integration 29 – 33 , low-weight molecular ferroelectrics 34 – 36 and polar topology in ferroelectrics 37 – 40 , have further generated unprecedented enthusiasm for the development of ferroelectric memories. Currently, there are three basic ferroelectric memory structures namely capacitors 3 , 41 , tunnel junctions 42 , 43 and field-effect transistors 18 . The capacitor-type ferroelectric random access memory (FeRAM) shows outstanding endurance of over 10 15 and is now commercially available in a cell structure of one transistor and one capacitor (1T1C) 41 (Fig. 1a ). However, the charge reading of polarization reversal in the capacitor-type FeRAM not only asks for a large area for detectable charges but is also itself destructive and requires a rewrite process after each polarization-reversal reading operation, which hinders the direct analog MAC calculations for in-memory computing. To overcome this destructive readout, the ferroelectric tunnel junction (FTJ) varies its conductance through modulation of the tunnel barrier height by polarization reversal 44 , 45 . The simple structure of metal/ferroelectric/metal in FTJ allows a high memory density, but the few-nanometers-thick ferroelectric films for direct tunneling suffer from poor endurance (Fig. 1b ) 45 . The alternative three-terminal ferroelectric field-effect transistor (FeFET), where the gate dielectric is replaced by a ferroelectric layer in a standard metal-oxide-semiconductor field-effect transistor (MOSFET), encodes its memory states by modulating the conductance of the channel in the semiconductor by the gate polarization 46 . However, the lack of an epitaxial template results in mesoscopically disordered and polycrystalline ferroelectric on the semiconductor channel which intrinsically leads to uncontrolled device-to-device variation in nanoscale FeFET devices 46 (Fig. 1c ). This variation issue makes the commercialization of FeFET challenging. On the other hand, a much simple two-terminal memristor is preferred for a high-density hardware-level in-memory computing 47 , 48 . Therefore, it is extremely appealing to develop a novel memory technology that not only exploits all merits of fast switching speed, good endurance, non-volatile states, and low operation energy consumption of ferroelectrics, but also combines a simple structure and non-destructive readout mode for high-density memory and computing applications. In this work, we propose a two-terminal ferroelectric fin diode (FFD) in which a ferroelectric capacitor and a fin-like semiconductor channel are combined to share both top and bottom electrodes (Fig. 1d ). This FDD memory absorbs merits of both non-destructive conductance read mode as in FTJs and long endurance as in FeRAM while it allows ferroelectric directly on electrode other than semiconductor, eliminating the intrinsic source of device-to-device variation in a traditional FeFET. Technology computer-aided design (TCAD) simulations demonstrate that the Schottky barrier at the semiconductor/electrode interface twists the electric field distribution, adding a transverse electric field to the vertical semiconductor channel of the FFD. The remanent polarization plastically-reversible by the distorted electric field permits the FFD both digital and analog memory functionalities. Such a FFD operates using different ferroelectric materials (namely the organic P(VDF-TrFE) polymer and inorganic industrially used PZT compounds) which emphasizes its universal character. Compared to the vast family of the current nonvolatile memories, this FFD shows exhaustive properties of prior memories with high performances such as an endurance of over 10 10 cycles, an ON/OFF ratio of ~10 2 , a feature size of 30 nm, an operating energy of ~ 20 fJ and an operation speed of 100 ns. Benefiting from the simplicity for fabricating this FFD and its self-rectification ratio of ~ 10 4 , a passive crossbar array with 1.6 k units is constituted and used to demonstrate the in-memory computing of a simple pattern classification task. The high device-to-device uniformity is reflected by a small σ/μ value of ~0.023 for positive coercive voltage and ~0.019 for negative coercive voltage using a Gaussian distribution. This work opens a new avenue for efficient memories and emerging-computing architectures for big data and artificial intelligence applications.
Methods P(VDF-TrFE) based FFD The structure and fabrication process of the designed FFD are orderly presented in Fig. 2a . First of all, the striped 100-nm Pt films sputtered by DC magnetron sputtering instrument on the cleaned SiO 2 /Si were used as the gate and drain electrodes. The working power, chamber pressure, and Ar flow rate were 250 W, 0.3 Pa, and 50 sccm, respectively (step 1); Secondly, the P(VDF-TrFE) (70:30 mol%) ferroelectric polymer was dissolved in the diethyl carbonate with 2.5 wt%. The P(VDF-TrFE) is formed by spin coating the polymers in an initially homogeneous evaporative solution, followed by a thermal anneal at 135 °C for 4 h to facilitate the ferroelectric β -phase growth as shown in step 2; Thirdly, the ~20-nm Al was deposited on the 120-nm ferroelectric layer by thermal evaporation (the deposition rate is 10 Å/s); The crossbar structure between top electrodes and bottom electrodes are accomplished by switching the metal mask before depositing the electrodes. Next, the excess organic polymer (P(VDF-TrFE)) was removed by oxygen ion etching, and the P(VDF-TrFE) films covered by Al was retained. Finally, assisted by a metal mask, the 30-nm ZnO was sputtered onto the crossbar network using RF magnetron sputtering instrument at a chamber pressure of 0.8 pa, Ar flow rate of 50 sccm and working power of 80 W. PZT based FFD The structure and fabrication process of the designed FFD based on PZT are orderly presented in Fig. S 5 . Assisted by a metal mask, the ~300-nm Al electrodes with a periodic striped structure were deposited on the purchased PZT (200 nm)/Pt (50 nm)/SiO 2 /Si substrate by thermal evaporation (the deposition rate is 10 Å/s). The excess PZT was removed by argon ion etching, and the PZT covered by Al electrodes was retained. Finally, assisted by a metal mask, the 30-nm ZnO was sputtered to cover the electrodes boundary using RF magnetron sputtering instrument with a chamber pressure of 0.8 pa, Ar flow rate of 50 sccm, working power of 80 W. The reversed FFD The fabrication process of the designed reversed FFD based on P(VDF-TrFE) are orderly presented in Fig. S 11 . Firstly, assisted by lithography, the 100 nm-thick Al electrodes were deposited on the purchased SiO 2 /Si substrate by thermal evaporation (the deposition rate is 10 Å/s) (step1). Secondly, the P(VDF-TrFE) (70:30 mol%) ferroelectric polymer was dissolved in the diethyl carbonate with 2.5 wt%. The P(VDF-TrFE) ( ~ 360 nm, 6 layers) is formed by spin coating the polymers in an initially homogeneous evaporative solution, followed by a thermal anneal at 135 °C for 4 h to facilitate the ferroelectric β-phase growth as shown in step 2. The Pt electrode was prepared by means of photolithographic alignment as shown in Fig. S 11b–d and step 3. The Pt films are sputtered by DC magnetron sputtering instrument, and the working power, chamber pressure, and Ar flow rate were 250 W, 0.3 Pa, and 50 sccm, respectively. Thirdly, the excess P(VDF-TrFE) was removed by oxygen ion etching, and the P(VDF-TrFE) covered by Pt electrodes was retained (step 4). Finally, assisted by a metal mask, the 30-nm ZnO was sputtered to cover the electrodes boundary using RF magnetron sputtering instrument with a chamber pressure of 0.8 pa, Ar flow rate of 50 sccm, working power of 80 W (step 5). A FFD nano device The fabrication process of the designed FFD nano device based on P(VDF-TrFE) are orderly presented in Fig. S 12 . First of all, the striped 100-nm Pt films sputtered by DC magnetron sputtering instrument on the cleaned SiO 2 /Si were used as the gate and drain electrodes (step 1). The working power, chamber pressure, and Ar flow rate were 250 W, 0.3 Pa, and 50 sccm, respectively; Secondly, the P(VDF-TrFE) (70:30 mol%) ferroelectric polymer was dissolved in the diethyl carbonate with 2.5 wt%. The 60 nm-thick P(VDF-TrFE) (~60 nm, 1 layer) is formed by spin coating polymers in an initially homogeneous evaporative solution, followed by a thermal anneal at 135 °C for 4 h to facilitate the ferroelectric β -phase growth as shown in step 2; Thirdly, assisted by lithography, the ~1-μm photoresist (PMMA) with a periodic striped structure were prepared on the P(VDF-TrFE) (60 nm)/Pt (50 nm)/SiO 2 /Si substrate by spin coating and a thermal anneal (200 °C for 2 h) (step 3). Next, assisted by a metal mask, the 30-nm Al was deposited onto the crossbar network using thermal evaporation instrument (the deposition rate is 10 Å/s) (step 4). The excess Al was removed by argon ion etching, and the Al upright and close to the PMMA was retained as shown in step 5. Next, the remaining P(VDF-TrFE) was cleaned up by oxygen ion etching (step 6). Finally, assisted by a metal mask, the 30-nm ZnO was sputtered onto the crossbar network using RF magnetron sputtering instrument at a chamber pressure of 0.8 pa, Ar flow rate of 50 sccm and working power of 80 W (step 7). A 40 × 40 passive crossbar array Both the bottom electrode and the top electrode are assisted by striped mask plates arranged with 40 columns. The detailed preparation process is the same as that in the P(VDF-TrFE)-based FFD devices. STEM measurements The Pt/P(VDF-TrFE)/Al/ZnO STEM sample was fabricated using a focused ion beam machine (Hellios G4 UX). Structural characterization was conducted using a JEM- ARM200 scanning transmission electron microscope equipped with an ASCOR probe corrector operating at an accelerating voltage of 300 kV. Electrical measurements The I–V curves, programing speed and retention property of the FFD were measured, unless noted otherwise, under ambient conditions using a Keithley 4200A-SCS parameter analyzer with remote preamplifiers. The read voltage of 5 V (0.3 V) is used to obtain the retention characteristic of the FFD based on P(VDF-TrFE) (PZT). A HP4194A impedance analyzer with an ac amplitude of 0.04 V is used to measure the capacitance versus frequency ( C - f ) at various temperatures. The P – V hysteresis loops and the fatigue characteristic of the FFD were investigated using ferroelectric analyzer (TF Analyzer 3000). In fatigue measurements, the fatigue pulses with an amplitude of 25 V (10 V) and a frequency of 10 kHz (2 MHz) were used to reverse the ferroelectric domains in P(VDF-TrFE) (PZT) based devices. The checked P – V curves were measured using a triangular wave with the same amplitude and frequency. A quasi static I – V sweeping was performed after ferroelectric fatigue texts with different cycles to obtain the cycles-dependent I – V curves. For all temperature-dependent measurements, the temperature was changed at a rate of 1 K/min. Each temperature was maintained for 2 mins using a computer-controlled cryostat (MMR Tech, Inc.) before the following electrical measurements. The electrical performance of the FFD array was achieved by a multi-channel array test system (PXIe-4631) controlled by a labview program. TCAD simulation Silvaco TCAD simulator has been used to simulate the device we proposed. The structure is created by a 2D process simulation editor (Athena), where the size parameters of device modeling are consistent with the experiments. The contact between Pt and ZnO was defined as a Schottky contact with a barrier height of 1.02 eV, and the contact between Al and ZnO was set as an ohmic contact by default. A doping concentration in the ZnO is set as 10 20 cm -3 . The dielectric constant and the band gap of P(VDF-TrFE) are set as 10.7 and 6 eV, respectively. Physical models including mobility, recombination, and fermi-Dirac are employed in the simulation.
Results Structure of two-terminal FFD Ferroelectric thin films sandwiched between a top electrode and a bottom electrode are used to support a sidewall semiconductor, forming a vertical fin-like structure where the channel length defined by the thickness of the ferroelectric can be easily controlled at the nanoscale. As will be discussed below, by inducing a Schottky contact, for example between the bottom electrode and the fin-like semiconductor channel/bridge, the bottom electrode then plays a dual role i.e.; both the drain and the gate of a typical FeFET. While the fin-like semiconductor channel contributes most currents, the ferroelectric domain switching rearranges the electric field configuration at ferroelectric/semiconductor interface and results in a resistive switching. This combination structure of sandwiched ferroelectric and fin-like Schottky sidewall is morphologically called as ferroelectric fin diode (FFD). The FFD device can be easily fabricated using a common photolithography technique. Figure 2a presents the fabrication process of a FFD with the ferroelectric P(VDF-TrFE) and a n-type ZnO channel (please refer to the Methods section for details). The final heterostructure is examined by cross-sectional scanning transmission electron microscopy (STEM) (Fig. 2b, c ) that shows a clear stair-like device structure. According to the images of the energy dispersive X-ray spectrometry (EDS) mapping of Si, Pt, F, Al, and Zn elements, the expected FFD device schematized in Fig. 2a5 is indeed confirmed. The thickness of Pt/P(VDF-TrFE)/Al/ZnO layers is 150/100/20/30 nm, respectively. Note that the oblique ZnO channel with a deflection angle of 30° to the vertical may arise from the nonlinear etching process. Since the ZnO semiconductor and the P(VDF-TrFE) ferroelectric layers play an important role in this novel device, their topography and crystal structure are checked by atomic force microscope (AFM) and X-ray diffraction (XRD), respectively. Deposited on Pt electrode using magnetron sputtering method, the ZnO films show AFM images (inset of Fig. 2d ) with no macroscopic defects, pinholes or visible cracks and a root mean square roughness of 0.52 nm, which suggests that uniform and smooth ZnO films are obtained. The P(VDF-TrFE) films on Pt electrode have also a good quality and uniformity with a root mean square roughness of 2.01 nm (inset of Fig. 2f ). The XRD patterns show a Bragg peak at 34.1° for ZnO (Fig. 2d ) and 19.6° for P(VDF-TrFE) films (Fig. 2f ), which are specific to the wurtzite hexagonal phase (Fig. 2e ) and ferroelectric β phase 49 , respectively. The ferroelectricity of P(VDF-TrFE) films is further confirmed by the butterfly-shaped loop of the piezoelectric force microscope (PFM) amplitude and 180°-reversed loop of the PFM phase (Fig. 2g ). The reproducible ferroelectric domain switching is attested by the out-of-plane PFM images showing alternate of up and down polarization domains. Indeed, in Fig. 2h, i , stripe domains characteristics of ferroelectricity are written by alternately applying a +20 V and −20 V bias to the Pt electrode while the PFM tip is grounded. A clear 180° phase shift is observed between adjacent domains (Fig. 2i ) with depressed amplitude value at domain walls (Fig. 2h ). Polarization-driven resistive switching By quasi-statically sweeping a voltage from −20 V to +20 V on the bottom Pt electrode, the semi-logarithmic current versus voltage ( I – V ) curve shows an anticlockwise hysteresis with an ON/OFF ratio of ~10 2 (Fig. 3a ). Moreover, the I–V curve of the FFD further demonstrates that the structure of this new device can significantly reduce the reverse ( V < 0) current (Fig. S 1a ). To demonstrate whether the observed resistive switching is dominated by the polarization reversal, ferroelectric-phase-dependent resistive switching experiments are performed. As shown in the temperature-dependent dielectric permittivity of our P(VDF-TrFE) films (Fig. 3b ), the ferroelectric-to-paraelectric phase transition occurs at a Curie point T c of 380 K on heating while it drops to T c = 355 K on cooling, which agrees well with previous reports 50 . This thermal hysteresis of 25 K indicates that the ferroelectric transition of this copolymer is of first order. This phase transition is further attested by directly measuring the polarization versus voltage ( P – V ) loops. As shown in top panels in Fig. 3c , P – V loops, measured at 100 Hz, exhibit clear hysteresis with a remanent polarization ( P r ) of ~7.5 μC/cm 2 at room temperature. As the temperature increases, the P – V loop is altered at 373 K (both P r and the coercive voltage V c reduce) and disappears at 393 K i.e., above T c = 380 K evidenced on heating with temperature dependent dielectric permittivity, resulting in a linear behavior characteristic of a paraelectric state. When cooling back to 300 K, the P – V hysteresis loop recovers. Likely, the resistive switching (bottom panels in Fig. 3c ) behaves concomitantly with the P – V loop and vanishes when the P(VDF-TrFE) films are in the paraelectric phase. Note that the operating voltage can be significantly decreased by lowering the thickness of the ferroelectric layer and/or selecting ferroelectric materials with small coercive fields (Fig. S 1b–f ) to meet the CMOS technology requirements. Note that a tradeoff issue between operation voltage and conduction current, that is, a thinner film for lower operation voltage results in higher current, is usually suffered in 2-terminal resistive devices. In this view, it will be more energy efficient to decrease the operation voltage by using ferroelectric materials that possesses a much small coercive field. The direct correlation between ferroelectricity and resistive switching strongly supports that the resistive switching in the FFD device is dominated by the polarization reversal. To eliminate the influence of other possible mechanism such as defects or charging effect, we fabricate a referenced device sample in which the ferroelectric layer is replaced by a non-ferroelectric aluminum oxide (Al 2 O 3 ) (Fig. S 2a ). The thickness of Al 2 O 3 and ZnO layer in the referenced device is 27 nm and 30 nm respectively (Fig. S 3 ). The resistive switching in the FFD devices shows counterclockwise I – V curves in the first quadrant. However, when ferroelectric layer is replaced by dielectric of Al 2 O 3 , a small clockwise I – V curve is observed (Fig. S 2b ). This clockwise hysteresis can be understanded by the charge trapping effect 51 or the presence of impurities in the ZnO or interface. The opposite hysteresis in ferroelectric device with that in Al 2 O 3 dielectric confirms that the counterclockwise hysteresis in our ferroelectric device is not dominated by the charge trapping effect or impurities effect. A FeFET device is then fabricated and shows that the n-type ZnO channel can be effectively tuned by ferroelectric polarization (Fig. S 4 ), implying that the charge trapping or impurities effect are much weak and negligible compared with the field effect by ferroelectric polarization. It is worth mentioning that the I – V curves are asymmetric with a self-rectifying ratio of ~ 10 4 (Fig. 3a ). This self-rectifying characteristic results from the Schottky contact between the Pt electrode and n-type ZnO films. The work function of Al, Cu, Au and Pt metals is obtained from ultraviolet photoelectron spectroscopy (UPS) to be 3.80 eV, 4.63 eV, 4.86 eV and 5.04 eV respectively, while an affinity of 4.02 eV is obtained in ZnO semiconductor (Fig. S 5 ). With a clean interface contact, a higher work function than the affinity enables a Schottky contact at metal/ZnO interface, while a lower work function results in an ohmic one. The Pt (5.04 eV)/ZnO (4.02 eV) Schottky barrier and Al (3.80 eV)/ZnO (4.02 eV) ohmic contact are confirmed by their rectified and liner I – V curves respectively (Fig. S 6 ). The Schottky barrier of the Pt/ZnO interface is further confirmed by the linear relationship between ln( I / T 2 ) and V 1/2 (Fig. 3d ). By measuring the temperature-dependent slope of this linear relationship, a barrier height value of ~ 0.85 eV is obtained and agrees well with the energy band alignment when considering the energy level of the ZnO channel and Pt and Al electrodes (Fig. 3e ). To better understand how the polarization affects the electronic transport, a TCAD simulation is performed. Interestingly, because of the existence of Pt/ZnO Schottky barrier, there is lateral electric field pointing forward (backward) to the vertical ZnO channel when a voltage of +20 V (-20 V) is applied to the Pt electrode (Fig. 3 g and j). For example, when a voltage of -20 V is applied to the Pt electrode, the negatively-biased Schottky barrier suffers most voltage and results in curved electric potential in the ferroelectric layer (Fig. 3j ). The electric field, defined by minus the gradient of the electric potential, is represented with black arrows in Fig. 3 g and j. The lateral electric field, estimated to be of ~ 140 MV/m (near the Al/ZnO interface) or -200 MV/m (at most ZnO channel region) under the +20 V or -20 V bias, is high enough to reverse the polarization in ferroelectric copolymers 21 , 22 . Figure 3 h, k show the phenomenological model of polarization state after removing the voltage bias. By sweeping a negative voltage that is larger than the coercive voltage (- V c ~ -15 V) on the Pt electrode, the lateral electric field makes the polarization obliquely backward to the vertical ZnO channel. The remanent polarization with net negative bound charges at the ferroelectric/channel interface electrically weakens the carrier density of the n-type channel 12 , 17 and leads to a thicker Schottky barrier (Fig. 3i ), resulting in the low conductance state (LCS). These bound charges can be switched away by aligning the polarization upward, even forward to the vertical ZnO channel near the Al/ZnO interface, when applying a positive voltage higher than + V c (Fig. 3h ), which permits the Schottky barrier to recover its original state (Fig. 3f ) and gives the high conductance state (HCS). This model explains well the observed asymmetric resistive switching (Fig. 3a ) because of the non-uniform and amplitude-dependent lateral field in the ferroelectric layer (Fig. 3 g and j) due to the vertical ZnO channel in FFD devices. This contrasts with the symmetric behavior in more traditional FeFET (Fig. S 4 ) with expectedly vertical uniform field. Robustness performance and universality of FFD The fatigue behavior is a very important parameter, which is related to the service life of memory devices and the prospect of their potential industrialization. The ferroelectric and electrical fatigue characteristics of FFDs have been studied systematically. Figure 4a, b (up panels) and S 7a–d demonstrate the performance of a typical P(VDF-TrFE)-based FFD. The remanent polarization ( P r + and P r - ) and the corresponding P – V loops as a function of measured cycles are presented in Fig. 4a (up panel) and Fig. S 7c respectively. Interestingly, accompanying with the robust polarization reversal of up to 10 7 cycles, the device shows stable resistive switching throughout (Fig. 4b (up panel) and Fig. S 7d ). It is reasonable that the conductance sustains a better endurance than P r . The contribution to the resistive switching significantly involves the lateral region around the ferroelectric/semiconductor interface where the up-left switching (Fig. 3 g, h and j, k ) is more robust as it does not involve nucleation like in the up-down switching in the parallel capacitor region between bottom Pt electrode and top Al electrode, where P r is measured. To highlight the universal aspect of both these robust performances and the device structure itself, an industrialized inorganic ferroelectric, i.e. lead zirconate titanate (PZT) on a 4-inch Si wafer (Fig. S 8 ), is used to fabricate the FFD (Fig. S 9 ) and put in parallel to the one made with the organic P(VDF-TrFE) ferroelectric. As expected, an endurance of up to 10 10 cycles in both polarization reversal and resistive switching is obtained in a typical PZT-based FFD (Fig. 4a, b (bottom panels) and S 7e, f ). The same self-rectifying and counterclockwise characteristics of the resistive hysteresis in such devices based on both inorganic PZT and organic P(VDF-TrFE) indicates that the as-designed FFDs are already at high technology readiness level. Figure 4c shows the retention property of FFDs. Benefiting from the nonvolatility of ferroelectrics, both HCS and LCS show slight degradation, if any, in a period of ~10 4 s in either P(VDF-TrFE) (top panel) or PZT (bottom panel) based devices. In addition, the programming speed of FFDs is studied (Fig. S 10a, b ). The device is firstly written to LCS by applying a -28 V (-8 V) pulse with a width as long as 100 ms to the P(VDF-TrFE) (PZT) device. Then +35 V ( + 10 V) pulses with an increased width from 1 μs to 1 ms (100 ns to 100 μs) are used to write the P(VDF-TrFE) (PZT) device to HCS. Each programing voltage is followed by a read voltage pulse of +3.0 V (+0.2 V) with a width of 100 ms to check the conductance state. The achieved speed of P(VDF-TrFE)- and PZT-based devices is found to be 1 μs and 100 ns, respectively. The essential width of Al top electrode needed for the resistive properties is the limit of feature size in our devices. A reversed structure is used to verify the scalability of our memory devices (Fig. S 11 ). This reversed structure allows the design of across (I in Fig. S 11d, e ), just touch (inset of Fig. 4d and II in Fig. S 11d, e ) and separate (III in Fig. S 11d, e ) between vertical projection of Al electrode and Pt electrode where the just touching case is equivalent to infinitely reducing the width of Al. It is found that the resistive switching exists well when Al electrode and Pt electrode have across (Fig. S 11f ) and even just “zero” overlap (Fig. 4d ) within our photolithography error, implying a nanoscale scalability in our memory devices. Inspired by this, we further fabricated nano devices where the width of Al top electrode is only 30 nm (Fig. 4e and Fig. S 12 ). As shown in Fig. 4f , the resistive switching survives well in the 30-nm nano device. The energy consumption for each write operation of the ferroelectric fin diode is evaluated using the formula: E = UIt. For example, when an operating voltage with an amplitude of 20 V is used for a reversed FFD based on P(VDF-TrFE) (Fig. S 11f ), the energy consumption is estimated to be ~1 fJ and ~ 20 fJ for a reset operation and a set operation respectively. Device uniformity and analog storage for in-memory computing In a traditional ferroelectric transistor (Fig. 1c ), the ferroelectric layer has to be deposited on the semiconductor layer other than a seed electrode. The lack of an epitaxial template results in poor ferroelectric quality that is usually mesoscopically disordered and polycrystalline 46 . This leads to uncontrolled device-to-device variation in nanoscale devices. On the contrary, in a FFD (Fig. 1d ), the ferroelectric layer is directly deposited on a seed electrode (for example, Pt electrode for PZT films) and covered by a top electrode. This sandwiching metal/ferroelectric/metal structure, indeed the same as that of a commercial ferroelectric capacitor, possesses a high ferroelectric quality and good uniformity even in nanoscale. Note that during the following semiconductor deposition, the ferroelectric layer is protected by the top electrode. Thus, a good uniformity is expected in ferroelectric fin diodes. A 4 × 4 passive array with P(VDF-TrFE)-based fin diodes (Fig. 5a ) is fabricated on a Si/SiO 2 (300 nm thick) wafer. The current cross-talk issue in this passive crossbar array can be effectively eliminated by the self-rectifying characteristic in each device unit 48 . The device uniformity is checked by performing I – V curves in all 16 device units (Fig. 5b ) which shows that the dispersion of the response from one device unit to another is very small. The device-to-device variation is evaluated using the ratio of σ/μ in a Gaussian distribution function , where μ and σ are the mean value and standard deviation of the current, respectively. A good uniformity is found with a σ/μ value of ~0.18 for HCS and ~0.08 for LCS, respectively (Fig. 5c ). The electronic transport of a typical device unit is comprehensively measured. Five successive I – V curves between -20 V and +20 V are collected and show little deviation (Fig. S 13 ). Ferroelectric is well known for its nonvolatility and analog programing characteristic 52 . Figure 5d demonstrates the analog switching of polarization with multiple P r in amplitude-variant P – V curves of the FFD based on P(VDF-TrFE). Accordingly, by changing the sweeping amplitude of the positive voltage from 8 V to 20 V with a step of 1 V, clearly separated and nested hysteresis loops are observed (Fig. 5e ), implying multiple intermediate conductance states in the FFD. This is caused by the multiple intermediate polarization states in the ferroelectric layer as shown in Fig. 5d . The nonvolatility of these intermediate conductance states can be reflected by these open hysteresis loops. Six distinguishing states are also chosen to measure their retention characteristic and no degeneration is observed in a time scale of >100 s for all six states (Fig. 5f ). These persistent multiple conductance states provide the essential ingredients to emulate the so-called synaptic plasticity, plastic changes of synaptic weights, that is at the heart of the learning processes. The quasi-linear conductance potentiation (strengthening) and depression (weakening) with 25 discrete states can be achieved (Fig. 5g ) by repeatedly applying a voltage pulse sequence of 25 positive voltage pulses (+25 V/10 μs) followed by 25 negative voltage pulses (-18 V/10 μs), respectively (Fig. S 14 ). This analog behavior of conductance resembles the long-term potentiation/depression (LTP/D) process in synaptic devices. An artificial neural network constituted by crossbar array based on FFDs for digits recognition is simulated using the experimentally measured conductance states. Both the small image version (8 × 8 pixels) of handwritten digits from the “Optical Recognition of Handwritten Digits” (ORHD) dataset and the large image version (28 × 28 pixels) of hand-written digits from the “Modified National Institute of Standards and Technology” (MNIST) dataset are used to perform the back-propagation simulation. For the large (small) image version, a multilayer perceptron (MLP) neural network with 784 (64) input neurons, 300 (30) hidden neurons, and 10 output neurons is utilized by using Cross-Sim simulator (Fig. 5h ) 53 . In the simulation, the crossbar, regarded as part of a “neural core”, performs vector-matrix multiplication and outer-product update operations. Generally, the performance of a neural network is greatly influenced by the controllability (e.g., nonlinearity and write noise) of synaptic devices, which can be quantitatively analyzed using the probability distribution of the conductance change (Δ G ) induced by a write operation. The plots of Δ G versus initial Δ G 0 , derived from the cyclic conductance potentiation and depression, are presented in insets of Fig. 5i and Fig. S 15 , respectively. For small digits, the classification accuracy approaches 82.6% within the second training epoch and approaches 92.6% after 16 training epochs, which is close to 96.4% the theoretical limit of an ideal numeric training for small digits (Fig. 5i ). For large digits, our simulations show a classification accuracy of 82.9% (Fig. S 15 ). These results demonstrate that the analog characteristic of FFDs has the potential for in-memory computing applications. To further confirm the in-memory computing application of our FFD devices, we have fabricated a passive crossbar array with 1.6 k units (Fig. 6a, b ). The ferroelectricity is checked by transient I – V curves in 200 random-selected devices where transient current peaks correspond to ferroelectric coercive voltages (Fig. S 16 ). A uniformity with a σ/μ value of ~0.023 for positive coercive voltage and ~0.019 for negative coercive voltage in a Gaussian distribution is obtained (Fig. 6c ). The coercive voltage and diode characteristic together enable the intended programing in the FFD passive crossbar array (see supplementary note 1 , Fig. S 17 ). The resistive switching in 400 devices at cross points of alternate rows and alternate columns is carefully checked one by one. Stirringly, all units show resistive switching (Fig. S 18 ). These ON and OFF conductance states can be distinguished clearly (Fig. S 19 ) and the ON/OFF ratio at the read voltage of 3 V fluctuates around 10 (Fig. 6d ). We believe that the decay of ON/OFF ratio results from the sneak path issue while the fluctuation of electrical performances is due to the immature fabrication techniques. A simple pattern recognition task based the 1.6 k passive crossbar array is then demonstrated. Figure 6e–g display the schematic diagram for clarifying three 4 × 4-pixel images using a hardware-based artificial neural network (ANN). The images of “L”, “u” and “n” are trained using the Manhattan update rule (Fig. S 20 ) to obtain weights guiding to program the hardware ANN for recognizing the three letters. A region with 16 × 6 units are chosen to demonstrate the pattern recognition task where each weight is encoded by conductance difference between neighboring pair units (Fig. 6f, g ). Fig. S 21 shows the programming process and final conductance distribution in the 16 × 6 hardware ANN. During the inference progress, the content pixels are encoded into a voltage pulse with a width of 10 ms and vacant pixels into a voltage pulse with a width of 0 ms, that is absence of a voltage pulse. The patterns recognition is implemented successfully by collecting these column currents where the current difference ( I i + - I i - ) standing for the target image show much large value than others (Fig. 6h–j ). Images with noise are also tested to confirm the tolerance of our hardware ANN (Fig. S 22 ). As summarized in Fig. 6k , when zero, one, and two pixels are randomly flipped, the recognition accuracy is 100%, 50%, and 37.5%, respectively. The pattern recognition task mentioned above was automatically measured using multi-channel array test system as shown in Fig. S 23 . Comparison with state-of-the-art non-volatile memory devices Key parameters of state-of-of-the-art non-volatile memories including Not And logic gates (NAND Flash), phase change memory (PCM), FeRAM, resistive RAM (RRAM) and Magnetic RAM (MRAM) have been collected in recent review reports 41 . A comparison of the performance with other memories is summarized in Table S1 . Among the vast family of nonvolatile memories, our FFD memory cumulatively demonstrates very high performances with an endurance of over 10 10 cycles, a self-rectification ratio of ~10 4 , an operation speed of 100 ns, a feature size of 30 nm and cell size of 4 F 2 , and an ultralow energy consumption of ~20 fJ. The simple two-terminal structure and the high self-rectification ratio of ~10 4 permit to efficiently design passive crossbar arrays for high-density memories as well as emerging in-memory computing application.
Discussion A robust ferroelectric-based non-volatile memory with a novel FFD structure is proposed as a new building block for future electronic circuit architectures. The device absorbs merits of non-destructive read mode with resistive switching as in FTJs and long endurance as in FeRAM while it allows ferroelectric directly on electrode other than semiconductor, eliminating the intrinsic source of device-to-device variation in a traditional FeFET. Both digital and analog memory functionalities can be achieved in such device. It can operate with different ferroelectric materials illustrating its universal character. It demonstrates superior performances when compared to state-of-the-art nonvolatile memories with an endurance of over 10 10 cycles, an ON/OFF ratio of ~10 2 , a feature size of 30 nm and cell size of 4 F 2 , an operating energy of ~20 fJ and an operation speed of 100 ns. Analog storage using multiple conductance states is demonstrated showing such a device is suitable for synaptic learning plasticity. The simple two-terminal structure and its self-rectifying ratio of ~ 10 4 permit a passive crossbar array with 1.6 k units in which in-memory computing of a simple pattern classification task is accomplished. The high device-to-device uniformity is reflected by a small σ/μ value of ~0.023 for positive coercive voltage and ~0.019 for negative coercive voltage using a Gaussian distribution. This work paves a way to use this new electronic unit for designing passive crossbar arrays for either memories or in-memory computing applications.
Among today’s nonvolatile memories, ferroelectric-based capacitors, tunnel junctions and field-effect transistors (FET) are already industrially integrated and/or intensively investigated to improve their performances. Concurrently, because of the tremendous development of artificial intelligence and big-data issues, there is an urgent need to realize high-density crossbar arrays, a prerequisite for the future of memories and emerging computing algorithms. Here, a two-terminal ferroelectric fin diode (FFD) in which a ferroelectric capacitor and a fin-like semiconductor channel are combined to share both top and bottom electrodes is designed. Such a device not only shows both digital and analog memory functionalities but is also robust and universal as it works using two very different ferroelectric materials. When compared to all current nonvolatile memories, it cumulatively demonstrates an endurance up to 10 10 cycles, an ON/OFF ratio of ~10 2 , a feature size of 30 nm, an operating energy of ~20 fJ and an operation speed of 100 ns. Beyond these superior performances, the simple two-terminal structure and their self-rectifying ratio of ~ 10 4 permit to consider them as new electronic building blocks for designing passive crossbar arrays which are crucial for the future in-memory computing. Designing efficient high-density crossbar arrays are nowadays highly demanded for many artificial intelligence applications. Here, the authors propose a two-terminal ferroelectric fin diode non-volatile memory in which a ferroelectric capacitor and a fin-like semiconductor channel are combined to share both top and bottom electrodes with high performance and easy fabrication process Subject terms
Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s41467-024-44759-5. Acknowledgements B.T. would like to thank fundings of National Key Research and Development Program of China (No. 2021YFA1200700), National Natural Science Foundation of China (No. T2222025, 62174053 and 61804055), Open Research Projects of Zhejiang Lab (2021MD0AB03), Shanghai Science and Technology Innovation Action Plan (No. 19JC1416700, 21JC1402000 and 21520714100) and the Fundamental Research Funds for the Central Universities. Author contributions B.T. conceived the concept. B.T., Q.Z. and C.D. supervised the research. G.F. fabricated the devices. G.F., L.C. and B.T. performed the electrical and piezoelectric force microscope measurements. G.F., S.X. and Q.B. performed the ultraviolet photoelectron spectroscopy. B.T., G.F., X.Z., J.L., K.S. and J.J. performed the simulations and experimental classification tasks. Z.Y. and R.H. performed the STEM. X.L., B.T. and W.T. performed the TCAD. Q.Z., F.Y., H.P., X.T., X.G., J.W., A.J., B.D., J.C. and C.D. advised on the experiments and data analysis. G.F., B.T. and B.D. co-write the manuscript. All authors discussed the results and revised the manuscript. Peer review Peer review information Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available. Data availability The data that support the findings of this study are available from the corresponding author upon reasonable request. Code availability The codes that support the findings of this study are available from the corresponding author upon reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Nat Commun. 2024 Jan 13; 15:513
oa_package/09/5c/PMC10787831.tar.gz
PMC10787832
38218926
Introduction Industries that work with dyestuff, textiles, leather, paper, plastics, etc., typically drain synthetic dyes along with their effluent. The synthetic dye wastewater is mainly composed of organic components, which are complex and easy to show color in water, and usually slow or difficult to degrade in the natural environment 1 , 2 . Among the synthetic dyes, anionic azo dyes account for half of the dye synthesis and industrial application 3 . Due to the low coloring rate on natural fibers, anionic dyes account for a large proportion of the dye wastewater discharged by printing and dyeing factories. Methyl orange [(MO) dimethylaminoazobenzenesulfonate] is a common and typical azo anionic dye. This water-soluble organic synthetic dye has very high colorability and presents a bright orange color when dissolved in water. Azo dyes such as methyl orange contain aromatic and –N=N– groups in their molecules, which are highly toxic, carcinogenic and teratogenic 4 , 5 , and are harmful to the environment and organisms 6 , 7 . In addition, the dyes in the wastewater can lead to the deterioration of water quality 8 , so the wastewater containing dyes must be treated innocuously and the dye components need to be removed in order to discharge to the natural water environment or carry out secondary use 9 , 10 . The adsorption process is considered a valuable technique for the removal of dyes from wastewater, alongside many other physical and chemical processes 11 – 13 . Numerous research endeavours have been conducted with the aim of identifying cost-effective and efficient adsorbents for the purpose of reducing dye concentrations in aqueous solutions. The researchers incorporated several materials, such as activated carbon, peat, chitin, silica, among others 14 . Chitosan or its derivatives have been demonstrated to be an efficacious substance for the removal of anionic dyes 15 – 22 through its amino groups which have the ability to undergo cationization, resulting in a strong electrostatic interaction with anionic dyes when exposed to acidic conditions. In order to enhance the adsorption efficiency, previous studies have indicated that the adsorbent material underwent modification by the introduction of high chelating coordination of sulphur (S) and nitrogen (N) functional groups 23 – 26 . Nevertheless, the publications outlined a multitude of reaction steps required for polyamine design. Aminated Chitosan derivatives were synthesized by many authors using different approaches, however these studies used two or even three steps to achieve their goal 23 – 31 . In a recent study 32 , aminated Chitosan materials have been developed using a one-pot method. Further enhancement of the adsorption capabilities of Chitosan derivatives has been tried through crosslinking process using various cross-linking reagents which have dual purpose namely, stabilizing Chitosan in acid solutions, rendering it insoluble, and augmenting its mechanical qualities. The novelty of the current study focused on presents a newly developed one-step amination technique for Chitosan through coupling of the Chitosan (CS) with 2-chloroethylamine (ENH2) using click chemistry. The obtained Chitosan derivatives with an increased content of amine groups (CS-ENH2) have subsequently crosslinked using Glutaraldehyde, leading to the formation of amino-ethyl Chitosan Schiff bases. The CS-ENH2 Schiff bases were used as an adsorbent to remove Methyl Orange (MO) dye from aqueous solutions in batch mode process, and its performance was compared with the native Chitosan Schiff base. The impact of several factors, including adsorption time, initial dye concentration, adsorption temperature, adsorption pH, agitation speed, and adsorbent dosage, on the adsorption of Mo dye have been studied. These Schiff bases are further characterized using advanced analytical techniques such as Fourier Transform Infrared spectroscopy (FTIR), Thermal (TGA and DSC) analysis, and Scanning Electron Microscopy (SEM).
Materials and methods Materials Chitosan medium molecular weight (≥ 75% DD) and methyl orange dye (MO; 85%) were procured from Sigma-Aldrich (Germany). The compound 2-Chloro ethyl amine hydrochloride (99%) was acquired from Sigma-Aldrich (Germany). The other chemicals used in this study were obtained from El-Nasr Pharmaceutical Co for Chemicals (Egypt) included sulfuric acid (98%), sodium hydroxide (99%), and phenolphthalein (98%). The Glutaraldehyde (GA) used in this study was a 25.0 wt% aqueous solution acquired from ACROS Organics. Methods Preparation of amino-ethyl chitosan hydrogel A solution containing 2 grammes of CS was prepared by dissolving it in 40 millilitres of acetic acid with a concentration of 2%. Subsequently, different quantities of Cl-ENH2 (0, 6.4, 12.7, 25.5, and 51 mM) were added individually to the Chitosan solution and the obtained adsorbents were coded as CS, CS-ENH2-1, CS-ENH2-2, CS-ENH2-3, and CS-ENH2-4, respectively. The temperature was then increased to 70 degrees Celsius and the mixture was stirred for duration of 2 h. Subsequently, a volume of 0.5 ml of Glutaraldehyde solution with a concentration of 25% was introduced into the mixture while maintaining a consistent stirring motion for duration of one hour. The CS-ENH2 hydrogel was subjected to drying at a temperature of 60 °C for duration of one night. The dried samples underwent a milling process and were subsequently subjected to multiple washes using hot distilled water in order to eliminate any remaining unreacted components. Subsequently, the samples were subjected to a drying process and subsequently stored within desiccators to facilitate subsequent analysis and adsorption tests. Structural and morphological Characterization Infrared Spectrophotometric (FTIR) To confirm the change and establish the structure of the hydrogel, Fourier transform infrared spectroscopy was conducted using a Shimadzu FTIR-8400 S spectrophotometer from Japan. Thermal gravimetric analysis (TGA) The analysis of materials was conducted using a thermo gravimetric analyzer (Shimadzu TGA –50, Japan) with a Nitrogen flow rate of 30 ml/min. This analysis aimed to observe any structural changes resulting from the modification. The measurement of weight loss in the samples commenced at room temperature and continued up to 600 °C, with a heating rate of 10 °C per minute. Differential scanning calorimeter (DSC) Differential scanning calorimetry was performed on the samples using a Shimadzu DSC-60A apparatus from Japan. The analysis was conducted over a temperature range from ambient to 350 °C, with a heating rate of 10 °C/min and under a nitrogen flow of 30 ml/min. Scanning electron microscopic analysis (SEM) Prior to examination by scanning electron microscopy, the samples were subjected to a vacuum environment and coated with a thin layer of gold. The morphological changes on the surface of the samples were monitored using a secondary electron detector of a scanning electron microscope (SEM) model Joel Jsm 6360LA, manufactured in Japan. Physicochemical characterization Water uptake (%) The Water uptake behaviour of the prepared hydrogel was investigated using distilled water (pH 5.4). Accurately weighed amounts of hydrogels were immersed in water and allowed to swell for 24 h at R.T. The swollen hydrogel was periodically separated, and the moisture adhered to the surface of hydrogel was removed by blotting them gently in between two filter papers, immediately followed by weighing. The swelling degree of samples was determined according to the following formula 33 : where M t is the weight of the swollen hydrogel, and M0 is the initial dry weight. The ion exchange capacity A known weight of chitosan or schiff base hydrogels were added to the known volume of 0.1 M H 2 SO 4 solution, and the mixture was kept under shaking for three h. The mixture was filtered, and an aliquot was titrated against a standard solution of sodium hydroxide. Similarly, control titration without the addition of Chitosan was also run. From the difference in the volume of NaOH required for neutralization, the ionic capacity of chitosan samples was calculated using the following equation: where V 2 and V 1 are the volumes of NaOH required for complete neutralization of H 2 SO 4 in the absence and presence of chitosan membrane, respectively, A is the normality of NaOH and W is the weight of sample taken for analysis 34 . Batch equilibrium studies A stock solution of methyl orange (MO) dye with a concentration of 1g/L (1000 ppm) was made in distilled water. Subsequently, the desired concentrations, 10, 20, 25, 50, and 100 ppm, were achieved by diluting of 1, 2, 2.5, 5, and 10 mL of the stock solution with distilled water up to 100 mL using 100 mL flasks. The adsorption tests were performed using 100 mL flasks. A certain quantity of adsorbent was added to 25 mL dye solution with varying dye concentrations and pH values. The mixture was then agitated in an orbital shaker for a predetermined duration. The concentrations of MO in the initial and final aqueous solutions were determined by employing a UV–Vis spectrophotometer set to a wavelength of 465 nm. The quantity of dye adsorbed was determined by subtracting the initial concentration from the equilibrium concentration. The calculation of the percentage elimination value was determined using the following mathematical relationship: where; C 0 is the initial dye concentration and C t is the final dye concentration in supernatant.
Results and discussion Characterization Physicochemical characterization The confirmation of the immobilisation of excess amine into the chitosan backbone is achieved through the measurement of ion exchange capacity. Figure 1 illustrates a nearly linear augmentation in the ion exchange capacity as the quantity of reacted 2-chloroethyl amine in the modification procedure is elevated. This finding substantiates the observed elevation in the concentration of free amines inside the hydrogels that were synthesised. The water absorption capacity of the hydrogels that were created was assessed and is illustrated in Fig. 1 . A linear rise in water absorption is seen, providing confirmation of the immobilisation of more free amino groups. Furthermore, the incorporation of immobilised ethyl amine groups as grafted chains onto the Chitosan backbone serves several purposes. Firstly, it enhances the internal space within the hydrogel. Secondly, it reduces the crystallinity of the hydrogel. Lastly, it increases the size of the three-dimensional network, thereby promoting the diffusion of water molecules and consequently augmenting the amount of water uptake. Infrared spectrophotometric The characteristics’ spectrum of CS displays a strong absorption band at 3437 cm −1 due to OH and amine N–H symmetrical stretching vibration. A peak at 2921 cm −1 was due to symmetrical C-H stretching vibration attributed to pyranose ring. The sharp peak at 1383 cm −1 was assigned to CH3 in amide group. The broad peak at 1095 cm −1 was indicated C–O–C stretching vibration in CS 30 , peaks at 1649 and 1425 cm −1 were due to C=O stretching (amide I) and N–H stretching (amide II). The absorption band at 1153 cm -1 was assigned to the anti-symmetric stretching of C–O–C bridge and 1095 cm −1 , 1010 cm −1 were assigned to the skeletal vibration involving the C–O stretching. In the other hand, the infrared spectrum of ethylamine, wave numbers ~ 1500 to 400 cm −1 is considered the fingerprint region for the identification of ethylamine and most organic compounds. It is due to a unique set of complex overlapping vibrations of the atoms of the molecule of ethylamine. The most prominent infrared absorption lines of ethylamine at wavenumbers ~ 3500 to 3300 cm −1 is a broad band for N–H bond stretching vibrations, characteristic of amines. The hydrogen bonding interferes with the N–H stretching vibrations producing the broad band peaking at around ~ 3400 cm −1 . There are also characteristic bands due to N–H vibrations at wave numbers 1650–1580 cm −1 . Vibrations characteristic of C-N bonds in aliphatic amines like ethylamine occur at 1220–1020 cm −1 . Around 3000–2800 cm −1 are absorptions due to C–H stretching vibrations—they overlap with the N–H stretching vibrations. The similarity between most of the characteristic peaks of the Chitosan and 2-Chlorethyl amine, makes overlapping of both parent materials peaks. Figure 2 depict the Fourier-transform infrared (FT-IR) spectroscopic analysis of Chitosan and amino-ethyl Chitosan hydrogels. The provided charts depict the characteristic bands of functional groups in polysaccharides, including prominent band within the range of 3400 cm −1 , which correspond to the vibrational modes of hydroxyl and amine groups involved in starching. Chitosan shows band at a wave number of 3433.4 cm −1 . A shift of the wave number to 3435.33, 3439.2, 3437.26, and 3338.9 cm −1 of the CS-ENH2-1, CS-ENH2-2, CS-ENH2-3, and CS-ENH2-4, respectively, has been recognized with varied absorption intestines. The presence of bands at a wave number of 2938 cm −1 is indicative of the presence of aliphatic C–H bonds. A shift of the wave number to 2924, 2930, 2922, and 3014.84 cm −1 of the CS-ENH2-1, CS-ENH2-2, CS-ENH2-3, and CS-ENH2-4, respectively, has been recognized with varied absorption intestines. The amide groups of Chitosan were seen at a wave number of 1641 cm −1 . A shift of the wave number to 1637.62, 1653, and 1614.47 cm −1 of the CS-ENH2-2, CS-ENH2-3, and CS-ENH2-4, respectively, has been recognized with varied absorption intestines. The spectral bands observed within the range of 1000–1300 cm −1 are associated with the C–O stretching vibrations occurring in the glucose ring of Chitosan. Chitosan shows band at a wave number of 1068.6 cm −1 . A shift of the wave number to 1076.3, 1078.24, 1070.53, and 1074.39 cm −1 of the CS-ENH2-1, CS-ENH2-2, CS-ENH2-3, and CS-ENH2-4, respectively, has been recognized with varied absorption intestines. Thermal gravimetric analysis (TGA) Thermogravimetric analysis (TGA) was conducted on the hydrogel derivatives of CS and CS-ENH2, and the results are illustrated in Fig. 3 . The presented chart displays data pertaining to the thermal degradation process occurring in the presence of a Nitrogen atmosphere. The observed trend indicates a gradual decline in the measured weight samples values, commencing from the initial temperature of the surrounding environment and continuing until about 150 °C. The recorded samples weight loss percent values fall within the range of 10.87–12.79%, which can potentially be attributed to the reduction in moisture content inside the polymers, as suggested by previous studies. The occurrence of a secondary samples weight loss percent at an elevated temperature, ranging from 17.48 to 27.25%, can perhaps be attributed to the oxidative degradation of the pyranose ring within the chitosan backbone. During this phase, the occurrence of samples weight loss can be attributed to the decomposition of the amine groups inside the pyranose ring, leading to the formation of novel crosslinked fragments. The residue that was generated underwent a gradual decomposition process by the application of increased temperature within the range 35 – 37 . Differential scanning calorimetry (DSC) The differential scanning calorimetry (DSC) investigation was performed on the hydrogel derivatives of CS and CS-ENH2, as depicted in Fig. 4 . The initial endothermic peak observed in all studied materials, occurring between 50 and 120 °C, can be attributed to the increase in moisture content. The inclusion of hydrophilic functional groups, such as hydroxyl and amine groups, along the polymer chain imparts a strong attraction to water molecules. This property allows the polymer to effectively absorb and retain water from the surrounding atmosphere or during the production process. The second thermal event observed in the chart can perhaps be attributed to the disintegration of the glucose amine (GlcN) units inside the Chitosan hydrogels. This decomposition process is characterised by an exothermic peak occurring at a temperature of 210 °C 38 . Scanning electron microscope (SEM) and EDAX analysis The morphological study of Chitosan and amino-ethyl Chitosan hydrogel was conducted utilising a scanning electron microscope, as depicted in Fig. 5 a–e. The images demonstrate a marginal elevation in surface roughness and pore structure as the amine content of chitosan is augmented. The observed phenomenon can be attributed to the favourable interaction between the extended side chains containing terminal amine groups and the molecular structure of Chitosan. These chains offer an alternative active site for the process of crosslinking, as opposed to the original amine groups, perhaps resulting in increased space between polymer chains. The verification of the structural modifications of Chitosan to amino-ethyl Chitosan by introducing ethyl amine groups has been conducted using EDAX analysis (Fig. 5 f). The study reveals an increase in the mass percentage of carbon (C) and nitrogen (N) in comparison to the oxygen (O) mass percentage observed in the Chitosan blank sample. Figure 5 g, h present empirical support for the adsorption of MO on the Chitosan adsorbent, as indicated by the presence of S and Na elements and an increase in the mass percentage of C and N, accompanied by a corresponding decrease in the mass percentage of O. On the contrary, the adsorption of MO on the CS-ENH2-4 adsorbent (Fig. 5 j) exhibits an increase in the presence of S and Na elements compared to the CS-ENH2-4 adsorbent (Fig. 5 i). Additionally, there is a slight increase in the C mass percent, while the N and O mass percents experience a slight decrease. It is noteworthy to mention that the elements S and Na exhibit a higher mass percentage in comparison to Chitosan-MO, as depicted in Fig. 5 h. This observation provides evidence for the enhanced performance of the CS-ENH2-4 adsorbent in comparison to Chitosan. Adsorption process This study explores the sorption characteristics of methyl orange (MO) using Chitosan and amino-ethyl Chitosan hydrogels in the presence of artificially contaminated water solution. A series of experiments were conducted to examine the adsorption characteristics of hydrogels towards MO under various adsorption settings, including adsorption time, adsorption temperature, adsorption pH, agitation rate, adsorbent dosage, and starting dye concentration. Effect of the adsorption time Figure 6 a depicts the impact of adsorption time on the efficacy of MO removal percentage (%) using different Chitosan adsorbents. From the Figure, It is acknowledged that the percentage of dye removal exhibits a linear rapid progress as a first initial period for all the Chitosan adsorbents. That period varied with the amine content of the adsorbent. For Chitosan, it is recognized to be 90 min, which after almost no noticeable MO removal was noticed suggesting the consumption of all the amine active adsorption sites and attainment of equilibrium. In the other hand, it can be noticed that all the derivatives of amino-ethyl Chitosan exhibit a higher linear increase in the percentage of MO removal, compared to Chitosan, in the following order CS-ENH2-1, CS-ENH2-2, CS-ENH2-3, and CS-ENH2-4 correlated to the increment of the amine content adsorption sites. At this stage, all the active adsorption sites are free and accessable to MO molecules, in addition to the existence of high MO concentration gradient between the adsorbents solid phase and the MO liquid phase. These driving force leads to a high rate of MO removal compared to Chitosan counterpart. Subsequently, with a progress of adsorption time, that driving force was reduced as a result of consuming most of the adsorption active sites and reduced of the concentration gradient. A slower rate of increase is observed until reaching saturation at 240 min where the adsorption is controlled by the diffusion of MO molecules to the interior pores of the adsorbents. The performance of Chitosan is influenced by secondary key factor namely hydrophobic interactions between the hydrophobic aliphatic moieties of the adsorbents and the aromatic ones of the MO molecules. The hydrophobic interactions arise from the presence of the methyl group in the acetamide moiety (inside partial acetyl amine groups) and the –CH and –CH2 groups in the glucose ring. The adsorption process of MO, an anionic dye, onto Chitosan adsorbents was conducted using a combination of physical sorption behavior and chemisorptions, which occurred due to the electrostatic interaction between opposite charges in accordance with previously published data where Chitosan Schiff bases have used in the removal of MO dye from aqueous solution 39 – 41 . The Chitosan structure as a linear cationic polymer was modifie to improve its behaviour via preparing two different crosslinked Chitosan Schiff bases hydrogels using a glutaraldehyde crosslinker. The Chitosan was coupled with succinimide (Ch/Su) and 1-methyl-2-pyrrolidinone (Ch/Mp) 39 . Cross-linked Chitosan derivate Schiff bases obtained from the coupling of Chitosan with 1-vinyl 2- pyrrolidone [Schiff base (I)] and 4-amino acetanilide (Schiff base (II)) 40 . Other Chitosan Schiff base derivative was developed by reaction of Chitosan with 4-methoxybenzaldehyde to have Chitosan Schiff base (Cs/MeB) in the presence of Glutaraldehyde as a crosslinker 41 . In the case of the existing adsorbents, augmenting the degree of ethylamine substitution leads to heightened hydrophobic interactions and more advantageous active sites comprising amine groups. In order to investigate the effects of incorporating additional amine groups by fictionalization with 2-chloroethyl amine, the cationic exchange capacity of the amino-ethyl Chitosan derivatives that were synthesized was determined. This measurement was then associated with the percentage of MO removal, as shown in Fig. 6 b. The figure illustrates a notable linear correlation between the removal percentage of MO in CS and the concentration of CS-ENH2-4 in the barrel, as seen by the nearly twofold rise in removal percentage from 43.3 to 84.2% when the ion exchange capacity (IEC) of CS and CS-ENH2-4 increased from 7.4 to 12.8 meq/g. This accomplishment demonstrates the efficacy of using more amine groups to boost the removal effectiveness of MO, on one hand. On the contrary, this provides an indication of the prevailing electrostatic interaction between opposing charges, specifically the positive charge on CS-ENH2 derivatives and the negative charge on the MO molecules. Effect of temperature The study investigated the impact of variations in environmental temperature on the percentage of MO removal utilising cross-linked Chitosan and amino-ethyl Chitosan hydrogels. The temperature range examined spanned from 25 to 60 degrees Celsius, as depicted in Fig. 7 . Based on the data presented in the Figure, it is evident that there is a consistent linear increase in the percentage of MO removal by all of the adsorbents utilised, with a similar rate of growth observed. The elevation of the medium temperature amplifies the mobility of the big dye ions, so expediting their stochastic motion within the solution. Consequently, this augmentation leads to an increase in the frequency of collisions between the dye ions and the adsorbent surface. Furthermore, it induces an increase in volume inside the internal framework of the adsorbent, hence facilitating deeper penetration of the larger dye molecules 42 . Similar finding where Cross-linked Chitosan derivate Schiff bases obtained from the coupling of Chitosan with 1-vinyl 2- pyrrolidone [Schiff base (I)] and 4-amino acetanilide (Schiff base (II)) were used in the removal of MO dye 40 . The authors explained the effect of the temperature increment would increase the mobility of the large dye ions as well as produce a swelling effect on the internal structure of the Chitosan. That consequently facilitates the diffusion of the large dye molecules 40 . It is noteworthy to notice that the amino-ethyl Chitosan derivatives exhibited a larger boost in the percentage of MO removal at the lowest temperature (25 °C), with the CS-ENH2-4 sample achieving a 100% removal rate. On the contrary, it was observed that the increase in the elimination percentage of MO was only 48.5% at the greatest temperature (60 °C). Hence, the adsorption capacity is predominantly influenced by the chemical interaction occurring between the functional groups present on the internal surface of the adsorbent (namely, the surface of its pores) and the adsorbate. This capacity is expected to augment with increasing temperature. Effect of pH Figure 8 illustrates the impact of the pH value of the initial MO solutions on the efficiency of adsorption. The absorption of MO dye is significantly higher in acidic solutions compared to alkaline settings. Under acidic conditions, the presence of free amine groups along the chitosan backbone leads to their protonation, resulting in the formation of a positive charge on the surface of the hydrogel. This positive charge facilitates electrostatic interactions with the negatively charged sulfonate group of MO. Comparable findings have reported by other authors 43 , 44 . The adsorption capabilities of chitosan hydrogels are observed to decrease at alkaline pH levels. At the given pH, the surface charges of chitosan exhibited a negative polarity, hence impeding the adsorption process due to the electrostatic repulsion between the negatively charged dye molecules and the adsorbent (chitosan hydrogel). One intriguing finding in this study is the inverse relationship between the removal percentages of MO and the cation exchange capacity of amino-ethyl Chitosan samples. Notably, the CS-ENH2-4 sample exhibited the lowest reduction rate, with a rapid decrease in MO removal % found after reaching a pH of 7.0. On the contrary, it was observed that all the adsorbents exhibited similar percentages of MO removal in pH 10.0, ranging from 35 to 40%. This observation shows that the primary factor influencing the adsorption process is the hydrophobic-hydrophobic interaction 45 . Similar finding results where joint steady adsorption pH range from 6.0 to 8.0 of the Chitosan Schiff bases adsorbents can be explained by two reasons. The first is the reduced number of the free amine groups’ numbers affected by the deprotonation. The second is to increase the physical adsorption role of the chitosan Schiff bases adsorbents via hydrophobic-hydrophobic interaction, which is expected to be higher in the Ch/Mp Schiff base hydrogel containing heterocyclic ring with an attached methyl group 39 . On the other hand, the dye removal % by Cs/MeB was slightly affected by the increase of pH were decreased from 95% at pH 4.0 to 86% at pH 9.0. This behaviour confirmed the dominated hydrophobic-hydrophobic physical adsorption between the benzene rings and the methyl hydrophobic groups of the Cs/MeB adsorbent and the MO dye molecules in a wide range of pH; from 4.0 to 9.0, while the elimination of the cationic charge of the last free amine groups at pH 10.0 leads to the collapse of the Cs/MeB hydrogel structure with loss of its water content leading to reduce the pores volume and so the internal pores surface area. The high and almost constant MO removal % by Cs/MeB hydrogel in a wide pH range is a great advantage for its application in the treatment of industrial effluents contaminated with MO dye 41 . Effect of agitation rate The impact of agitation rate on the adsorption of MO was investigated by conducting experiments at various agitation rates ranging from 50 to 250 rpm, while keeping the kinetic parameters constant. Figure 9 presents an overview of the findings derived from the study. The adsorption of the MO is observed to exhibit a notable enhancement as the agitation rate is raised within the range of 50–200 rpm, after which it reaches a plateau until 250 rpm. These findings can be attributed to the observation that higher agitation speeds enhance the diffusion of MO towards the surface of the adsorbents. The relationship between intraparticle diffusivity and adsorption capacity, as well as the surface characteristics of adsorbents, has been suggested to be of significant importance. Increasing the agitation rate has the potential to surpass the thickness of the liquid layer and the resistance to mass transfer on the surfaces of the adsorbents being examined. Therefore, it can be inferred that a shaking rate of 200 rpm is adequate to facilitate the accessibility of all surface binding sites for the uptake of methyl orange 40 , 46 , 47 . Effect of the initial dye concentration The impact of the initial concentration of methyl orange on the process of adsorption was investigated over a range of concentrations (10, 20, 25, 50, and 100 ppm). The findings of this investigation are presented in Fig. 10 . The figure yields two primary observations. The initial observation pertains to the complete elimination of the lowest concentration (10 ppm) of MO by all hydrogel samples employed. This outcome can be attributed to the presence of an ample number of active sites on the Chitosan sample, thereby accommodating all MO molecules. Consequently, the aminated chitosan samples, which possess a greater number of active sites, do not exhibit any discernible impact due to the limited availability of MO. On the other hand, a notable increase in the percentage of removal of MO has been seen as the initial concentration of MO is increased up to 100 ppm. Specifically, the CS-ENH2-4 sample exhibits a four-fold higher removal percentage of MO compared to the CS sample. The second primary observation pertained to the decline in the percentages of MO removal as the starting MO concentration increased. The process of reduction consists of two distinct steps. The initial acute stage was noticed when the concentration of MO was increased from 10 to 25 ppm. This resulted in a decrease in the reduction percentage, in conjunction with the degree of amination of the aminated chitosan samples. Eventually, a nearly linear reduction rate was achieved with the CS-NH2-4 sample. The second stage of reduction, characterised by a nearly identical decrease in the rate, was observed for all samples of adsorbents, with a recognition threshold of up to 100 ppm. A similar finding was observed where the removal percentage linearly reduced with an increase in initial MO concentration using 4-dimethylamino benzaldehyde chitosan Schiff base and benzophenone chitosan Schiff base. This trend is due to the electrostatic repulsion between the dye molecules with increasing concentration, resulting in a competition between the dye molecules for the limited active sites in the adsorbent 48 . Effect of the absorbent dose Figure 11 illustrates the impact of varying doses of Chitosan adsorbents on the percentage of MO removal, while keeping the kinetic parameters constant. In general, it was observed that an escalation in the dosage resulted in a proportional rise in the elimination % of MO for both CS and CS-NH2-1 samples, exhibiting a nearly linear relationship. The clearance percentages of MO exhibited a gradual increase when lower rates were applied to the other aminated samples. Ultimately, the adsorption efficiency achieved with 0.3 g of adsorbents exhibits minimal variation, falling within the range of 90–100%. The CS-NH2-4 sample has a plateau effect, commencing at a dosage of 0.2 g, whereby it achieves a clearance percentage of 95%. This phenomenon may be attributed to the observation that, when the starting dye concentration remains constant, increases in the quantity of adsorbent material results in a larger surface area and a greater number of sorption sites 49 , 50 . Similar quite tendency have been reported using other sorbents reported in the previous work 51 . Reusability The sorption–desorption cycle, as depicted in Fig. 12 , can be utilised to estimate the recovery of MO absorbed from aqueous solution. The reusability of the CS-NH2-4 adsorbent for removing MO from aqueous solutions was investigated by analysing the sorption–desorption cycles. The cycle was repeated ten times using a sodium hydroxide solution. The figure clearly demonstrates a continuous, nearly linear reduction in MO removal. However, the decrease in MO removal efficiency was not significant; the percentage of MO removal reached 66% in the tenth cycle, compared to 80.1% in the first cycle. Only 18% of the MO removal efficiency was lost after ten cycles of sorption–desorption processes. The CS-NH2-4 adsorbent exhibits favourable sorption–desorption performance and can be confidently used without a noticeable decrease in its sorption capacity for MO removal. Comparative adsorption capacity study Table 1 presents a comparison of the highest adsorption capacity for MO on the CS-NH2-4 adsorbent in relation to other adsorbents documented in the literature 39 – 41 , 48 , 52 – 55 . Based on the tabular data, it can be observed that the adsorption capacity of the hydrogels composed of alginate and alginate/poly aspartate is comparatively lower than that of the hydrogels being analysed 55 . The example experienced a reversal upon the utilisation of Calcium alginate MWNTs 53 , resulting in a six-fold increase in adsorption capacity. In contrast, the adsorption capacity of CS-NH2-4 is comparatively lower when compared to other Chitosan and Chitosan derivatives 39 – 41 , 52 . Magnetic multi-walled carbon nanotubes (MWCNTs) and bottom ash have also demonstrated elevated adsorption capacity 53 , 56 . The modest adsorption capacity of the CS-NH2-4 adsorbent can be attributed to several factors, which can be summarised as follows: The scarcity of amine active sites that are protonated for the purpose of adsorption at a pH level of 7.0. The limited expansion of the material at a pH level of 7.0, resulting in a diffusion constraint for the MO (material of interest). The utilisation of the initially available active amine group sites during chemical crosslinking procedures involving Glutaraldehyde. It is advisable to conduct a more comprehensive investigation into the specific parameters of the crosslinking process in order to enhance the adsorption capacity.
Results and discussion Characterization Physicochemical characterization The confirmation of the immobilisation of excess amine into the chitosan backbone is achieved through the measurement of ion exchange capacity. Figure 1 illustrates a nearly linear augmentation in the ion exchange capacity as the quantity of reacted 2-chloroethyl amine in the modification procedure is elevated. This finding substantiates the observed elevation in the concentration of free amines inside the hydrogels that were synthesised. The water absorption capacity of the hydrogels that were created was assessed and is illustrated in Fig. 1 . A linear rise in water absorption is seen, providing confirmation of the immobilisation of more free amino groups. Furthermore, the incorporation of immobilised ethyl amine groups as grafted chains onto the Chitosan backbone serves several purposes. Firstly, it enhances the internal space within the hydrogel. Secondly, it reduces the crystallinity of the hydrogel. Lastly, it increases the size of the three-dimensional network, thereby promoting the diffusion of water molecules and consequently augmenting the amount of water uptake. Infrared spectrophotometric The characteristics’ spectrum of CS displays a strong absorption band at 3437 cm −1 due to OH and amine N–H symmetrical stretching vibration. A peak at 2921 cm −1 was due to symmetrical C-H stretching vibration attributed to pyranose ring. The sharp peak at 1383 cm −1 was assigned to CH3 in amide group. The broad peak at 1095 cm −1 was indicated C–O–C stretching vibration in CS 30 , peaks at 1649 and 1425 cm −1 were due to C=O stretching (amide I) and N–H stretching (amide II). The absorption band at 1153 cm -1 was assigned to the anti-symmetric stretching of C–O–C bridge and 1095 cm −1 , 1010 cm −1 were assigned to the skeletal vibration involving the C–O stretching. In the other hand, the infrared spectrum of ethylamine, wave numbers ~ 1500 to 400 cm −1 is considered the fingerprint region for the identification of ethylamine and most organic compounds. It is due to a unique set of complex overlapping vibrations of the atoms of the molecule of ethylamine. The most prominent infrared absorption lines of ethylamine at wavenumbers ~ 3500 to 3300 cm −1 is a broad band for N–H bond stretching vibrations, characteristic of amines. The hydrogen bonding interferes with the N–H stretching vibrations producing the broad band peaking at around ~ 3400 cm −1 . There are also characteristic bands due to N–H vibrations at wave numbers 1650–1580 cm −1 . Vibrations characteristic of C-N bonds in aliphatic amines like ethylamine occur at 1220–1020 cm −1 . Around 3000–2800 cm −1 are absorptions due to C–H stretching vibrations—they overlap with the N–H stretching vibrations. The similarity between most of the characteristic peaks of the Chitosan and 2-Chlorethyl amine, makes overlapping of both parent materials peaks. Figure 2 depict the Fourier-transform infrared (FT-IR) spectroscopic analysis of Chitosan and amino-ethyl Chitosan hydrogels. The provided charts depict the characteristic bands of functional groups in polysaccharides, including prominent band within the range of 3400 cm −1 , which correspond to the vibrational modes of hydroxyl and amine groups involved in starching. Chitosan shows band at a wave number of 3433.4 cm −1 . A shift of the wave number to 3435.33, 3439.2, 3437.26, and 3338.9 cm −1 of the CS-ENH2-1, CS-ENH2-2, CS-ENH2-3, and CS-ENH2-4, respectively, has been recognized with varied absorption intestines. The presence of bands at a wave number of 2938 cm −1 is indicative of the presence of aliphatic C–H bonds. A shift of the wave number to 2924, 2930, 2922, and 3014.84 cm −1 of the CS-ENH2-1, CS-ENH2-2, CS-ENH2-3, and CS-ENH2-4, respectively, has been recognized with varied absorption intestines. The amide groups of Chitosan were seen at a wave number of 1641 cm −1 . A shift of the wave number to 1637.62, 1653, and 1614.47 cm −1 of the CS-ENH2-2, CS-ENH2-3, and CS-ENH2-4, respectively, has been recognized with varied absorption intestines. The spectral bands observed within the range of 1000–1300 cm −1 are associated with the C–O stretching vibrations occurring in the glucose ring of Chitosan. Chitosan shows band at a wave number of 1068.6 cm −1 . A shift of the wave number to 1076.3, 1078.24, 1070.53, and 1074.39 cm −1 of the CS-ENH2-1, CS-ENH2-2, CS-ENH2-3, and CS-ENH2-4, respectively, has been recognized with varied absorption intestines. Thermal gravimetric analysis (TGA) Thermogravimetric analysis (TGA) was conducted on the hydrogel derivatives of CS and CS-ENH2, and the results are illustrated in Fig. 3 . The presented chart displays data pertaining to the thermal degradation process occurring in the presence of a Nitrogen atmosphere. The observed trend indicates a gradual decline in the measured weight samples values, commencing from the initial temperature of the surrounding environment and continuing until about 150 °C. The recorded samples weight loss percent values fall within the range of 10.87–12.79%, which can potentially be attributed to the reduction in moisture content inside the polymers, as suggested by previous studies. The occurrence of a secondary samples weight loss percent at an elevated temperature, ranging from 17.48 to 27.25%, can perhaps be attributed to the oxidative degradation of the pyranose ring within the chitosan backbone. During this phase, the occurrence of samples weight loss can be attributed to the decomposition of the amine groups inside the pyranose ring, leading to the formation of novel crosslinked fragments. The residue that was generated underwent a gradual decomposition process by the application of increased temperature within the range 35 – 37 . Differential scanning calorimetry (DSC) The differential scanning calorimetry (DSC) investigation was performed on the hydrogel derivatives of CS and CS-ENH2, as depicted in Fig. 4 . The initial endothermic peak observed in all studied materials, occurring between 50 and 120 °C, can be attributed to the increase in moisture content. The inclusion of hydrophilic functional groups, such as hydroxyl and amine groups, along the polymer chain imparts a strong attraction to water molecules. This property allows the polymer to effectively absorb and retain water from the surrounding atmosphere or during the production process. The second thermal event observed in the chart can perhaps be attributed to the disintegration of the glucose amine (GlcN) units inside the Chitosan hydrogels. This decomposition process is characterised by an exothermic peak occurring at a temperature of 210 °C 38 . Scanning electron microscope (SEM) and EDAX analysis The morphological study of Chitosan and amino-ethyl Chitosan hydrogel was conducted utilising a scanning electron microscope, as depicted in Fig. 5 a–e. The images demonstrate a marginal elevation in surface roughness and pore structure as the amine content of chitosan is augmented. The observed phenomenon can be attributed to the favourable interaction between the extended side chains containing terminal amine groups and the molecular structure of Chitosan. These chains offer an alternative active site for the process of crosslinking, as opposed to the original amine groups, perhaps resulting in increased space between polymer chains. The verification of the structural modifications of Chitosan to amino-ethyl Chitosan by introducing ethyl amine groups has been conducted using EDAX analysis (Fig. 5 f). The study reveals an increase in the mass percentage of carbon (C) and nitrogen (N) in comparison to the oxygen (O) mass percentage observed in the Chitosan blank sample. Figure 5 g, h present empirical support for the adsorption of MO on the Chitosan adsorbent, as indicated by the presence of S and Na elements and an increase in the mass percentage of C and N, accompanied by a corresponding decrease in the mass percentage of O. On the contrary, the adsorption of MO on the CS-ENH2-4 adsorbent (Fig. 5 j) exhibits an increase in the presence of S and Na elements compared to the CS-ENH2-4 adsorbent (Fig. 5 i). Additionally, there is a slight increase in the C mass percent, while the N and O mass percents experience a slight decrease. It is noteworthy to mention that the elements S and Na exhibit a higher mass percentage in comparison to Chitosan-MO, as depicted in Fig. 5 h. This observation provides evidence for the enhanced performance of the CS-ENH2-4 adsorbent in comparison to Chitosan. Adsorption process This study explores the sorption characteristics of methyl orange (MO) using Chitosan and amino-ethyl Chitosan hydrogels in the presence of artificially contaminated water solution. A series of experiments were conducted to examine the adsorption characteristics of hydrogels towards MO under various adsorption settings, including adsorption time, adsorption temperature, adsorption pH, agitation rate, adsorbent dosage, and starting dye concentration. Effect of the adsorption time Figure 6 a depicts the impact of adsorption time on the efficacy of MO removal percentage (%) using different Chitosan adsorbents. From the Figure, It is acknowledged that the percentage of dye removal exhibits a linear rapid progress as a first initial period for all the Chitosan adsorbents. That period varied with the amine content of the adsorbent. For Chitosan, it is recognized to be 90 min, which after almost no noticeable MO removal was noticed suggesting the consumption of all the amine active adsorption sites and attainment of equilibrium. In the other hand, it can be noticed that all the derivatives of amino-ethyl Chitosan exhibit a higher linear increase in the percentage of MO removal, compared to Chitosan, in the following order CS-ENH2-1, CS-ENH2-2, CS-ENH2-3, and CS-ENH2-4 correlated to the increment of the amine content adsorption sites. At this stage, all the active adsorption sites are free and accessable to MO molecules, in addition to the existence of high MO concentration gradient between the adsorbents solid phase and the MO liquid phase. These driving force leads to a high rate of MO removal compared to Chitosan counterpart. Subsequently, with a progress of adsorption time, that driving force was reduced as a result of consuming most of the adsorption active sites and reduced of the concentration gradient. A slower rate of increase is observed until reaching saturation at 240 min where the adsorption is controlled by the diffusion of MO molecules to the interior pores of the adsorbents. The performance of Chitosan is influenced by secondary key factor namely hydrophobic interactions between the hydrophobic aliphatic moieties of the adsorbents and the aromatic ones of the MO molecules. The hydrophobic interactions arise from the presence of the methyl group in the acetamide moiety (inside partial acetyl amine groups) and the –CH and –CH2 groups in the glucose ring. The adsorption process of MO, an anionic dye, onto Chitosan adsorbents was conducted using a combination of physical sorption behavior and chemisorptions, which occurred due to the electrostatic interaction between opposite charges in accordance with previously published data where Chitosan Schiff bases have used in the removal of MO dye from aqueous solution 39 – 41 . The Chitosan structure as a linear cationic polymer was modifie to improve its behaviour via preparing two different crosslinked Chitosan Schiff bases hydrogels using a glutaraldehyde crosslinker. The Chitosan was coupled with succinimide (Ch/Su) and 1-methyl-2-pyrrolidinone (Ch/Mp) 39 . Cross-linked Chitosan derivate Schiff bases obtained from the coupling of Chitosan with 1-vinyl 2- pyrrolidone [Schiff base (I)] and 4-amino acetanilide (Schiff base (II)) 40 . Other Chitosan Schiff base derivative was developed by reaction of Chitosan with 4-methoxybenzaldehyde to have Chitosan Schiff base (Cs/MeB) in the presence of Glutaraldehyde as a crosslinker 41 . In the case of the existing adsorbents, augmenting the degree of ethylamine substitution leads to heightened hydrophobic interactions and more advantageous active sites comprising amine groups. In order to investigate the effects of incorporating additional amine groups by fictionalization with 2-chloroethyl amine, the cationic exchange capacity of the amino-ethyl Chitosan derivatives that were synthesized was determined. This measurement was then associated with the percentage of MO removal, as shown in Fig. 6 b. The figure illustrates a notable linear correlation between the removal percentage of MO in CS and the concentration of CS-ENH2-4 in the barrel, as seen by the nearly twofold rise in removal percentage from 43.3 to 84.2% when the ion exchange capacity (IEC) of CS and CS-ENH2-4 increased from 7.4 to 12.8 meq/g. This accomplishment demonstrates the efficacy of using more amine groups to boost the removal effectiveness of MO, on one hand. On the contrary, this provides an indication of the prevailing electrostatic interaction between opposing charges, specifically the positive charge on CS-ENH2 derivatives and the negative charge on the MO molecules. Effect of temperature The study investigated the impact of variations in environmental temperature on the percentage of MO removal utilising cross-linked Chitosan and amino-ethyl Chitosan hydrogels. The temperature range examined spanned from 25 to 60 degrees Celsius, as depicted in Fig. 7 . Based on the data presented in the Figure, it is evident that there is a consistent linear increase in the percentage of MO removal by all of the adsorbents utilised, with a similar rate of growth observed. The elevation of the medium temperature amplifies the mobility of the big dye ions, so expediting their stochastic motion within the solution. Consequently, this augmentation leads to an increase in the frequency of collisions between the dye ions and the adsorbent surface. Furthermore, it induces an increase in volume inside the internal framework of the adsorbent, hence facilitating deeper penetration of the larger dye molecules 42 . Similar finding where Cross-linked Chitosan derivate Schiff bases obtained from the coupling of Chitosan with 1-vinyl 2- pyrrolidone [Schiff base (I)] and 4-amino acetanilide (Schiff base (II)) were used in the removal of MO dye 40 . The authors explained the effect of the temperature increment would increase the mobility of the large dye ions as well as produce a swelling effect on the internal structure of the Chitosan. That consequently facilitates the diffusion of the large dye molecules 40 . It is noteworthy to notice that the amino-ethyl Chitosan derivatives exhibited a larger boost in the percentage of MO removal at the lowest temperature (25 °C), with the CS-ENH2-4 sample achieving a 100% removal rate. On the contrary, it was observed that the increase in the elimination percentage of MO was only 48.5% at the greatest temperature (60 °C). Hence, the adsorption capacity is predominantly influenced by the chemical interaction occurring between the functional groups present on the internal surface of the adsorbent (namely, the surface of its pores) and the adsorbate. This capacity is expected to augment with increasing temperature. Effect of pH Figure 8 illustrates the impact of the pH value of the initial MO solutions on the efficiency of adsorption. The absorption of MO dye is significantly higher in acidic solutions compared to alkaline settings. Under acidic conditions, the presence of free amine groups along the chitosan backbone leads to their protonation, resulting in the formation of a positive charge on the surface of the hydrogel. This positive charge facilitates electrostatic interactions with the negatively charged sulfonate group of MO. Comparable findings have reported by other authors 43 , 44 . The adsorption capabilities of chitosan hydrogels are observed to decrease at alkaline pH levels. At the given pH, the surface charges of chitosan exhibited a negative polarity, hence impeding the adsorption process due to the electrostatic repulsion between the negatively charged dye molecules and the adsorbent (chitosan hydrogel). One intriguing finding in this study is the inverse relationship between the removal percentages of MO and the cation exchange capacity of amino-ethyl Chitosan samples. Notably, the CS-ENH2-4 sample exhibited the lowest reduction rate, with a rapid decrease in MO removal % found after reaching a pH of 7.0. On the contrary, it was observed that all the adsorbents exhibited similar percentages of MO removal in pH 10.0, ranging from 35 to 40%. This observation shows that the primary factor influencing the adsorption process is the hydrophobic-hydrophobic interaction 45 . Similar finding results where joint steady adsorption pH range from 6.0 to 8.0 of the Chitosan Schiff bases adsorbents can be explained by two reasons. The first is the reduced number of the free amine groups’ numbers affected by the deprotonation. The second is to increase the physical adsorption role of the chitosan Schiff bases adsorbents via hydrophobic-hydrophobic interaction, which is expected to be higher in the Ch/Mp Schiff base hydrogel containing heterocyclic ring with an attached methyl group 39 . On the other hand, the dye removal % by Cs/MeB was slightly affected by the increase of pH were decreased from 95% at pH 4.0 to 86% at pH 9.0. This behaviour confirmed the dominated hydrophobic-hydrophobic physical adsorption between the benzene rings and the methyl hydrophobic groups of the Cs/MeB adsorbent and the MO dye molecules in a wide range of pH; from 4.0 to 9.0, while the elimination of the cationic charge of the last free amine groups at pH 10.0 leads to the collapse of the Cs/MeB hydrogel structure with loss of its water content leading to reduce the pores volume and so the internal pores surface area. The high and almost constant MO removal % by Cs/MeB hydrogel in a wide pH range is a great advantage for its application in the treatment of industrial effluents contaminated with MO dye 41 . Effect of agitation rate The impact of agitation rate on the adsorption of MO was investigated by conducting experiments at various agitation rates ranging from 50 to 250 rpm, while keeping the kinetic parameters constant. Figure 9 presents an overview of the findings derived from the study. The adsorption of the MO is observed to exhibit a notable enhancement as the agitation rate is raised within the range of 50–200 rpm, after which it reaches a plateau until 250 rpm. These findings can be attributed to the observation that higher agitation speeds enhance the diffusion of MO towards the surface of the adsorbents. The relationship between intraparticle diffusivity and adsorption capacity, as well as the surface characteristics of adsorbents, has been suggested to be of significant importance. Increasing the agitation rate has the potential to surpass the thickness of the liquid layer and the resistance to mass transfer on the surfaces of the adsorbents being examined. Therefore, it can be inferred that a shaking rate of 200 rpm is adequate to facilitate the accessibility of all surface binding sites for the uptake of methyl orange 40 , 46 , 47 . Effect of the initial dye concentration The impact of the initial concentration of methyl orange on the process of adsorption was investigated over a range of concentrations (10, 20, 25, 50, and 100 ppm). The findings of this investigation are presented in Fig. 10 . The figure yields two primary observations. The initial observation pertains to the complete elimination of the lowest concentration (10 ppm) of MO by all hydrogel samples employed. This outcome can be attributed to the presence of an ample number of active sites on the Chitosan sample, thereby accommodating all MO molecules. Consequently, the aminated chitosan samples, which possess a greater number of active sites, do not exhibit any discernible impact due to the limited availability of MO. On the other hand, a notable increase in the percentage of removal of MO has been seen as the initial concentration of MO is increased up to 100 ppm. Specifically, the CS-ENH2-4 sample exhibits a four-fold higher removal percentage of MO compared to the CS sample. The second primary observation pertained to the decline in the percentages of MO removal as the starting MO concentration increased. The process of reduction consists of two distinct steps. The initial acute stage was noticed when the concentration of MO was increased from 10 to 25 ppm. This resulted in a decrease in the reduction percentage, in conjunction with the degree of amination of the aminated chitosan samples. Eventually, a nearly linear reduction rate was achieved with the CS-NH2-4 sample. The second stage of reduction, characterised by a nearly identical decrease in the rate, was observed for all samples of adsorbents, with a recognition threshold of up to 100 ppm. A similar finding was observed where the removal percentage linearly reduced with an increase in initial MO concentration using 4-dimethylamino benzaldehyde chitosan Schiff base and benzophenone chitosan Schiff base. This trend is due to the electrostatic repulsion between the dye molecules with increasing concentration, resulting in a competition between the dye molecules for the limited active sites in the adsorbent 48 . Effect of the absorbent dose Figure 11 illustrates the impact of varying doses of Chitosan adsorbents on the percentage of MO removal, while keeping the kinetic parameters constant. In general, it was observed that an escalation in the dosage resulted in a proportional rise in the elimination % of MO for both CS and CS-NH2-1 samples, exhibiting a nearly linear relationship. The clearance percentages of MO exhibited a gradual increase when lower rates were applied to the other aminated samples. Ultimately, the adsorption efficiency achieved with 0.3 g of adsorbents exhibits minimal variation, falling within the range of 90–100%. The CS-NH2-4 sample has a plateau effect, commencing at a dosage of 0.2 g, whereby it achieves a clearance percentage of 95%. This phenomenon may be attributed to the observation that, when the starting dye concentration remains constant, increases in the quantity of adsorbent material results in a larger surface area and a greater number of sorption sites 49 , 50 . Similar quite tendency have been reported using other sorbents reported in the previous work 51 . Reusability The sorption–desorption cycle, as depicted in Fig. 12 , can be utilised to estimate the recovery of MO absorbed from aqueous solution. The reusability of the CS-NH2-4 adsorbent for removing MO from aqueous solutions was investigated by analysing the sorption–desorption cycles. The cycle was repeated ten times using a sodium hydroxide solution. The figure clearly demonstrates a continuous, nearly linear reduction in MO removal. However, the decrease in MO removal efficiency was not significant; the percentage of MO removal reached 66% in the tenth cycle, compared to 80.1% in the first cycle. Only 18% of the MO removal efficiency was lost after ten cycles of sorption–desorption processes. The CS-NH2-4 adsorbent exhibits favourable sorption–desorption performance and can be confidently used without a noticeable decrease in its sorption capacity for MO removal. Comparative adsorption capacity study Table 1 presents a comparison of the highest adsorption capacity for MO on the CS-NH2-4 adsorbent in relation to other adsorbents documented in the literature 39 – 41 , 48 , 52 – 55 . Based on the tabular data, it can be observed that the adsorption capacity of the hydrogels composed of alginate and alginate/poly aspartate is comparatively lower than that of the hydrogels being analysed 55 . The example experienced a reversal upon the utilisation of Calcium alginate MWNTs 53 , resulting in a six-fold increase in adsorption capacity. In contrast, the adsorption capacity of CS-NH2-4 is comparatively lower when compared to other Chitosan and Chitosan derivatives 39 – 41 , 52 . Magnetic multi-walled carbon nanotubes (MWCNTs) and bottom ash have also demonstrated elevated adsorption capacity 53 , 56 . The modest adsorption capacity of the CS-NH2-4 adsorbent can be attributed to several factors, which can be summarised as follows: The scarcity of amine active sites that are protonated for the purpose of adsorption at a pH level of 7.0. The limited expansion of the material at a pH level of 7.0, resulting in a diffusion constraint for the MO (material of interest). The utilisation of the initially available active amine group sites during chemical crosslinking procedures involving Glutaraldehyde. It is advisable to conduct a more comprehensive investigation into the specific parameters of the crosslinking process in order to enhance the adsorption capacity.
Conclusion Novel amino-ethyl Chitosan derivatives (CS-ENH2) with varied and higher amine group content than Chitosan have been developed using one step click chemistry reaction of Chitosan (CS) with 2-chloroethylamine (ENH2) amination reagent, then crosslinked using Glutaraldehyde to form new amino-ethyl Chitosan Schiff bases. The introduced amine groups increased the amine content of Chitosan by 70% which corresponding to increase the cationic exchange capacity of Chitosan (CS) from 7.4 to 12.8 meq/g of CS-ENH2-4 sample. The developed CS-ENH2-4 adsorbent shown 300% adsorption capacity of Methyl Orange (MO) dye solution, 100 ppm, compared to native Chitosan one. The study of adsorption time show fast and linear rate in the first 60 min, then lower rate of adsorption was observed until equilibrium starts to reach at 90 min. The CS-ENH2-4 adsorbent shows an almost constant MO removal percentage over a pH range from 2.0 to 7.0 compared with linear decline of the Chitosan counterpart. Both of the adsorption temperature and agitation speed have the same trend which the MO removal percentage increased along with. The experimental findings indicated that the highest percentage of MO dye removal was achieved under the conditions of pH 2, a temperature of 60 °C, agitation speed of 250 rpm, and adsorption duration of 90 min. Furthermore, the CS-ENH2-4 adsorbent exhibits a favorable potential for reusability, as it only experienced a reduction of 18% of its adsorption effectiveness after undergoing 10 cycles of adsorption–desorption. The structural, morphological, and physiochemical characterization of the developed amino-ethyl Chitosan derivatives underwent by Fourier Transform Infrared spectroscopy (FTIR), Thermal analysis (TGA and DSC), Scanning Electron Microscopy (SEM), and Energy Dispersive X-ray Analysis (EDAX). The later proved the amination process and the adsorption of MO through the increase of the N% by about 70%, in accordance with the results of the ion exchange capacity; 73%, and appearance of new S and Na elements in the analysis of MO-CS-ENH2-4, respectively.
The present study introduces a new and straightforward method for the amination of Chitosan. This method involves coupling Chitosan (CS) with 2-chloroethylamine (ENH2) in a single step to produce an amino-ethyl Chitosan derivatives with increased amine group content (CS-ENH2) using click chemistry. The resulting derivatives were then crosslinked using Glutaraldehyde to form amino-ethyl Chitosan Schiff bases. The novel amino-ethyl Chitosan Schiff bases were subsequently utilized as adsorbents for the removal of Methyl Orange (MO) dye from aqueous solutions using a batch technique, and the performance of the produced Schiff bases was compared with that of the native Chitosan Schiff base. The CS-ENH2 adsorbents show improved adsorption capacity up to 300% of the native Chitosan Schiff base with almost double removal rate. The adsorption temperature has a positive impact in general while almost 100% of MO removed at 60 °C using CS-ENH2 adsorbents compared with 66% of the native Chitosan Schiff base adsorbent. The adsorption pH shows a negative impact on the MO removal percent. That effect reduced sharply using the CS-ENH2 adsorbents with higher amination degree while the MO removal percent almost being constant over a wide range of pH; 2.0–7.0. The agitation speed has the same positive effect over all the adsorbents. However, the rate of MO removal percent decreased with increase the agitation speed up to 250 rpm. The experimental findings demonstrated that the highest percentage of MO dye removal was achieved under the conditions of pH 2.0, a temperature of 60 °C, agitation speed of 250 rpm, and adsorption duration of 90 min. These Schiff bases were subsequently characterized using advanced analytical techniques including Fourier Transform Infrared spectroscopy, Thermal analysis (TGA and DSC), and Scanning Electron Microscopy. Subject terms Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB).
Author contributions Prof. Dr. A.M.O. and Prof. Dr. T.M.T. have proposed the point of research, design the experimental work, and follow the executive of the work. Prof. Dr. W.A.S. and Prof. Dr. R.A. follow the executive of the work, and supervise the writing of the first draft of the manuscript. Chem. M.M.A.-E. has executed the experiments, drawing the figures and tabulating the data, writes the first draft of the manuscript. Prof. Dr. M.S.M.-E. revises the final version of the manuscript and submits to the journal. Funding Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB). Data availability The datasets used and/or analysed during the current study available from the corresponding author on reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1284
oa_package/71/63/PMC10787832.tar.gz
PMC10787833
38218895
Introduction Atrial fibrillation (AF) is the most common sustained arrhythmia and it is a major public health problem worldwide associated with substantial morbidity and mortality 1 , 2 . Risk factors for atrial fibrillation include age, hypertension, coronary artery disease, diabetes mellitus (DM), chronic obstructive pulmonary disease (COPD), obstructive sleep apnea (OSA), valvular heart disease, and congestive heart failure 3 – 6 . Among these risk factors, hypertension is the most common cause of AF because hypertension is the most prevalent disease among these conditions 2 , 7 – 9 . In some studies, up to 90% of AF patients are observed to be hypertensive 10 . However, the mechanisms underlying the increased susceptibility to atrial fibrillation in hypertensive patients are not completely understood. Nonetheless, studies that illustrated the development of AF in various risk factors indicated the occurrence of pathophysiological changes in the atrium at a cellular and molecular level, leading to interstitial fibrosis in the atrial tissue. The remodeling of the atrial structure increases the propensity to re-entry atrial tachyarrhythmia 11 – 15 . The renin–angiotensin–aldosterone system (RAAS) may play a role in the occurrence of AF. Angiotensin-II has been shown to regulate cardiac cell proliferation and modulating myocyte ion channels 8 . Thus, the use of angiotensin converting enzyme (ACE) inhibitors and Angiotensin-II receptor blockers (ARBs) may be effective in the prevention of AF in patients with heart failure 8 , 16 , 17 . However, it is difficult to determine if these agents are effective in the prevention of AF in patients with hypertension. This study aims to investigate the relationship of LA size with the occurrence of AF in hypertensive patients.
Methods Study design This was a retrospective observational cross – sectional study that reviewed records of patients diagnosed with hypertension and admitted to the cardiology department of King Abdullah university hospital (KAUH) in Irbid, Jordan from 1/1/2021 to 31/12/2021. The diagnosis of AF and associated risk factors were obtained from patient’s electronic records. Records included the Electrocardiographic reports (ECG’s), laboratory data, echocardiographic reports and clinical data (history and progress notes). Left atrium (LA) size was measured by transthoracic echocardiography machine (Hp-Sonos 5500, USA), measuring anteroposterior dimensions by M-mode directed by two dimensional (2-D) (real-time) echo in long axis parasternal view measured from LA posterior wall leading edge to leading edge at the level of aortic sinuses. Study participants All patients with hypertension who were admitted to the cardiac units (CCU and IMCU) of KAUH during January to December of 2021 were included. Diagnosis of hypertension was documented from electronic charts based on history of hypertension and being on antihypertensive treatment. Presence of AF was documented from the electrocardiograms and retrieved from patients’ electronic charts and the progress notes of the patients during hospitalization or during the follow up in the outpatient clinics. Inclusion and exclusion criteria Only AF in setting of hypertension was included in this study (144 cases). AF due to rheumatic valvular heart disease (3 cases), pre-excitation (2 cases), AF associated with thyrotoxicosis (1 case), AF associated with chronic Cor pulmonale (2 cases) and AF associated with obstructive sleep apnea (OSA; 1 case) were excluded from the study population, because these cases were not having hypertension. Study variables Patients’ gender, age, and smoking status were reported. Further, patients’ records of average systolic blood pressure, diastolic blood pressure, left atrial size, left ventricular ejection fraction (%), left ventricular dimensions in diastole, left ventricular dimensions in systole, and serum creatinine (μmol/L) were included. Risk factors for AF (age, sex, diabetes mellitus, coronary artery disease, dilated cardiomyopathies (DCM), congenital heart disease, valvular heart disease, Cor-pulmonale and congestive cardiac failure) were also retrieved from electronic charts of the patients and registered on an excel sheet. Analysis The baseline characteristics of the patients with atrial fibrillation and those without were compared using students t test for continuous variables, and chi-square test (X 2 ) for categorical variables. Furthermore, binary logistic regression was used to predict the occurrence of atrial fibrillation based on age, left atrial size, coronary artery disease, left ventricular ejection fraction, left ventricular dimensions in systole and diastole, and heart failure with the occurrence of atrial fibrillation after controlling for gender, smoking, and diabetes. Analysis was performed using the statistical package (“SPSS” software version 23), and a p-value less than 0.05 was considered the level of significance. Ethics approval and consent to participate This study was approved by the institutional review board of Jordan University of Science and Technology (number 124-2022). The requirement for informed consent from the study subjects was waived by the IRB of (Jordan University of Science and Technology/Research committee) due to the retrospective study design. Only patients’ file number were extracted with the data and no names or identifiable information was included. In addition, the committee ensured that all methods used in this research was performed in accordance with relevant guidelines/regulations.
Results There was a total of 958 hypertensive patients included in the study. The mean age of study patients was 61.40 (± 11.46) years, most of them were males (65.4%), and 40.3% were current smokers. There was a high prevalence of coronary artery disease (CAD) and diabetes mellitus (DM) among participating patients (51.3% and 59.2%, respectively) and about 1 out of each 5 patients had HF (Table 1 ). Patients with AF represented 15% of the sample (n = 144). Among AF patients, there were 6 patients (4.2%) with valvular health diseases, 10 (6.9%) with Cor pulmonale or COPD, and 6 patients (4.2%) with obstructive sleep apnea (OSA). Those patients had only one of these condition, except 3 patients who had both Cor pulmonale and OSA. There was a significant statistical difference between patients with AF and patients without AF for the means of the following measured variables: LA size, LVEF%, LV dimensions in diastole and systole (P-values < 0.05). Serum creatinine mean levels were not statistically different between patients with AF and those without AF (Table 1 ). The binary logistic regression model demonstrated a significant relationship of age, LA size, CAD, and HF with the occurrence of AF after controlling for gender, smoking, and diabetes. For each year increase in age, the probability of AF occurrence increased by 4.7% [OR = 1.047, 95% CI 1.026–1.069, p = value < 0.0001]. In addition, the risk of AF increased 3.2 times more among patients with increased LA size [OR = 3.204, 95% CI 1.749–5.870, p = value < 0.0001]. There was an increased risk of AF by 16% for each 1 cm increase in LA size, as noticed in Table 2 . The reasons of admission to cardiac units (CCU and IMCU) during the year (January 1st to December 31st of 2021) are illustrated in Table 3 .
Discussion Atrial fibrillation is associated with several risk factors and predisposing conditions that increases its occurrence, including hypertension that accelerates the development and progression of AF 18 . The prevalence of AF in this current study was 15%. This rate is similar to a study that showed prevalence of AF in hypertensive patients was equal to 14.9%. In that study, there were 9474 hypertensive patients followed for a median period of 24.1 years and a total of 1414 cases of AF (14.9%) was identified during the follow-up period 9 . Most of the studies has identified hypertension as a major contributor to the development of AF, and therefore, the study population of this study were all hypertensive patients in order to identify other intrinsic cardiac conditions that could affect the occurrence of AF. The American Heart Association (AHA) listed factors that are associated with AF development, including advanced age, a prolonged uncontrolled hypertension, underlying heart disease such as valve problems or hypertrophic cardiomyopathy, family history, and other chronic conditions like diabetes or sleep apnea 19 . In the current study, age, heart disease, CAD, and LA size were significantly associated with a higher likelihood of AF occurrence among hypertensive patients. Similarly, the Framingham heart study reported that the incidence of AF increased with age, heart failure, coronary heart disease, among other factors. AF was almost doubled every 10-year increment in age. The significant relationship of age with AF was shown in other studies as well 20 , 21 . In addition, the relationship of coronary heart disease and heart failure with AF was well established in previous Framingham heart studies and other studies, which was significant even with adjustment for age and gender 18 . There are few studies that investigated the relationship of atrial or ventricular size with the occurrence or the progression of AF. However, Kannel et al. in 2 scoping review studies illustrated that echocardiographic predictors of AF included left atrial enlargement, left ventricular fractional shortening, left ventricular wall thickness, and mitral annular calcification 5 , 18 . Left atrial size was an independent predictor for persistent AF, even when factors like age and gender were controlled 7 , 22 , which is similar to the findings of the current study. Further, Gerdts et al. had demonstrated that left ventricular hypertrophy and LA enlargement attributed to higher rate of AF in hypertensive patients 23 . The occurrence of AF seems to be complex in nature due to the several risk factors that could play a role and impose a multifactorial effect on the disease. Moreover, the structural intrinsic changes of the heart that is likely to progress with the advancement of age contribute to the development of AF. While a patient with chronic conditions, such as diabetes or hypertension, advances in age, progressive changes manifest in LA anatomy and function, may promote AF 8 . Various inflammatory markers and mediators such as C-reactive protein (CRP), tumor necrosis factor (TNF-a), interleukin -2 (IL-2), interleukin – 6 (IL-6), interleukin – 8 (IL-8), and monocyte chemoattractant protein—I (MCP – I) have been linked to the development and outcome of AF 15 , 24 – 28 . Animal models and human studies revealed a strong relationship between atrial myopathy and incidence of AF. Atrial myopathy as characterized by atrial fibrotic remodeling leading to electrical and autonomic changes which facilitate the development of AF 29 . Atrial myopathy leads to structural and electrophysiological changes of the left atrium (i.e. inflammation, oxidative stress, stretch, and fibrosis) which in turn progress into electrical and autonomic remodeling and prothrombotic state 29 . Diagnosis of atrial myopathy entails the usage of several tools, such as electrical (ECG), echocardiography, laboratory tests (inflammatory biomarkers), tissue biopsy, and 4 – dimensional magnetic resonance imaging (4-D MRI). The clinical implications of diagnosing atrial myopathy by these means may assist clinicians to consider anticoagulation therapy, even before the onset of AF, for selected subgroups of patients with low risk of strokes. Therefore, Determining the LA size, echocardiographically, can be considered a clinical risk identifier in preclinical AF which should be included in routine comprehensive echocardiography evaluation 27 . Study limitations Since this study is a retrospective study, we could not identify paroxysmal AF from chronic permanent AF. Another limitation is that we were not able to include the left ventricular mass index because the diastolic posterior wall thickness was not measured in the majority of patients.
Conclusion Findings suggest that LA enlargement in hypertensive patients play a major role in the development of AF through structural and electrophysiological changes to the atrium and other possible mechanisms 21 . Although there are several risk factors associated with the development of AF, chronic conditions and advancement of age remains the most critical detrimental factors in the onset and progression of AF. Moreover, large scale randomized studies are needed to establish a definitive role of antihypertensive therapy in reducing AF incidence.
Atrial fibrillation (AF) is the most common sustained arrhythmia and it is a major public health problem worldwide. Hypertension is one of the major risk factors for the development of AF. This study is carried out to determine the prevalence and independent risk factors for atrial fibrillation (AF) in hypertensive patients and to evaluate the relationship of AF with left atrial size. This is a retrospective observational cross – sectional study that used a retrospective electronic chart review of all admitted patients to cardiology department at King Abdullah university hospital (KAUH) in Irbid, Jordan, with a diagnosis of hypertension along with various acute cardiac admissions, including AF during 1-year period (January 1st to December 31 of 2021). Risk factors for AF (age, sex, DM, coronary artery disease, valvular heart disease, Cor-pulmonale, obstructive sleep apnea, and congestive cardiac failure) were retrieved from electronic charts of the patients. A total of 958 patients were admitted to the coronary care unit (CCU) and intermediate care unit (IMCU) during a 1-year period. Among them, 276 had 2 or 3 admissions. The main reason of admission was acute coronary syndrome (n = 491), heart failure (n = 180), and AF (n = 144), indicating AF prevalence of 15%. However, there were 40 patients with combined causes. All patients in the study (n = 958) were diagnosed with hypertension, including patients with atrial fibrillation (n = 144). The mean age of patients was 61.4 (± 11.46) years, and approximately two thirds of them were males (65.4%). The binary logistic regression model demonstrated a significant statistical relationship of age, left atrial size, coronary artery disease, left ventricular ejection fraction, left ventricular dimensions in systole and diastole, and heart failure with the occurrence of AF after controlling for gender, smoking, and diabetes. Findings indicate that left atrial size plays a significant role in the development of AF in patients with hypertension. However, the prevalence of AF significantly increased with advancing age in both sexes because of increased left ventricular hypertrophy, which leads to increased left atrial size. Subject terms
Author contributions A.S. and R.S. conceptualization and study design. A.S. and B.J. methodology and supervised data collection. T.Z., S.G. and T.A. data collection and validation. R.S. analysis. A.S. and R.S. writing the manuscript. Data availability The datasets used and/or analyzed during the current study available from the corresponding author on reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1250
oa_package/2e/61/PMC10787833.tar.gz
PMC10787834
38218739
Conclusion and outlook There is ample evidence that efferocytosis by neutrophils plays an important role in the response to dead cell accumulation. During inflammation, they contribute to the clearance of aggregates of apoptotic neutrophils. In cancer, they participate in the removal of dead tumour cell aggregates. The neutrophil response after engulfing apoptotic cells contributes to the resolution of inflammation and tissue regeneration. However, in the case of cancer, this can be harmful. A meta-analysis of expression signatures from more than 18,000 human tumours found that neutrophils are the tumour-associated cell type linked with the worst prognosis [ 58 ]. Neutrophilic efferocytosis might contribute to this situation. As professional phagocytes, neutrophils have the full machinery for engulfment and express many receptors for the detection and binding of dead cells. However, it is also still unclear whether neutrophils distinguish between the different types of cell death from which their target cells have died. Although an increasingly clear picture is emerging on efferocytosis by neutrophils, there are still many unanswered questions awaiting exploration.
When a cell dies of apoptosis, it is eliminated either by neighbouring cells or by attracted professional phagocytes. Although it was generally believed that neutrophils also have the ability to perform efferocytosis, their contribution to the clearance of apoptotic cells was considered less important compared with macrophages. Therefore, this ability of neutrophils remained unexplored for a long time. Over the past decade, it has been shown that during inflammation, neutrophils contribute significantly to the clearance of apoptotic neutrophils that accumulate in large numbers at the site of tissue damage. This “neutrophil cannibalism” is accompanied by inhibition of pro-inflammatory activities of these cells, such as respiratory burst and formation of neutrophil extracellular traps (NETs). Furthermore, efferocytosing neutrophils secrete anti-inflammatory mediators and mitogens including hepatocyte growth factor (HGF), fibroblast growth factor 2 (FGF2), vascular endothelial growth factors (VEGF), and transforming growth factor beta (TGFβ). Thus, efferocytosis by neutrophils is involved in resolution of inflammation. Recent research indicates that it plays also a role in cancer. Many different solid tumours contain aggregates of dead tumour cells that have undergone spontaneous apoptosis. Their extent correlates with poor clinical outcome in most cancer types. These clusters of apoptotic tumour cells are strongly infiltrated by tumour-associated neutrophils (TANs) that acquired an anti-inflammatory and pro-resolving polarization state. This review summarizes the potential consequences discussed in the current literature. Although the picture of the role of efferocytosis by neutrophils in inflammation and cancer is becoming clearer, many questions are still unexplored. Subject terms
FACTS Apoptotic cells release “find-me” signals which attract predominantly neutrophils. Neutrophils accumulate in areas of tissue damage and apoptotic tumour cells. After engulfment of apoptotic cells, neutrophils block respiratory burst and NETosis. Efferocytosing neutrophils secrete a variety of soluble mediators such as cytokines, chemokines, and mitogens which create a pro-resolving and tumorigenic microenvironment. OPEN QUESTIONS Several of the molecules known to be involved in the process of efferocytosis in macrophages are also expressed in neutrophils, but for many of them there is still a lack of evidence that they also fulfil this function there. The exact mechanism by which neutrophils adopt a regenerative and tumourigenic phenotype after the uptake of apoptotic cells is still largely unexplored. Neutrophils in inflammation Neutrophils represent the first line of cellular innate immune response to infection and tissue damage. Recent evidence indicate that this short-lived myeloid cell population exhibits a great phenotypic and functional diversity [ 1 ]. It not only plays an important role in triggering inflammation in reaction to pathogens, but also contributes to its subsequent resolution after their clearance. Neutrophils accumulate quickly at the site of tissue damage through a multi-step process called “neutrophil swarming” [ 2 ]. Damage-associated molecular patterns (DAMPs) activate resident cells to release short-range chemoattractants for neutrophils. Pioneer neutrophils from around the damage site migrate to the tissue injury within minutes. The contact with pathogen-associated molecular patterns (PAMPs) stimulates them to deploy a plethora of antimicrobial weapons [ 3 ]. They form neutrophil extracellular traps (NETs) to entrap invading pathogens. They release a variety of antimicrobial and pro-inflammatory molecules from their granules and produce reactive oxygen species to kill bacteria. Finally, they clear pathogens by phagocytosis. Neutrophil-derived leukotriene B4 (LTB4) enhances the radius of recruitment of further neutrophils from distant tissue sites [ 2 ]. Notably, neutrophils also support the resolution of inflammation right from the start. They release anti-inflammatory, resolving and angiogenic mediators such as IL-10, transforming growth factor β (TGFβ), lipoxin 4A, resolvins, protectins, defensins, and vascular endothelial growth factor (VEGF) [ 4 ]. Neutrophils that have engulfed pathogens die through phagocytosis-induced apoptosis [ 5 ]. Cytokine receptors such as IL-1R on their surface, which are no longer functional, scavenge their pro-inflammatory ligands from the microenvironment [ 6 ]. Neighbourhood macrophages that phagocytose dying neutrophils adopt an anti-inflammatory, resolving and reparative M2-like phenotype [ 7 – 9 ]. Such resolving mechanisms begin to gain the upper hand as soon as the infection is pushed back. The accumulation of neutrophils at the site of tissue damage thus enables the restructuring of the extracellular matrix, the formation of dense aggregates that seal the wound tightly, and finally the initiation of tissue repair processes. It must be emphasized that signals of tissue damage are sufficient to trigger the attraction of neutrophils, which accordingly can also be observed in response to sterile injury without pathogens [ 3 ]. Efferocytosis by neutrophils Especially during the early phase of neutrophil swarming, the number of resident macrophages is still very low and probably insufficient to clear all apoptotic neutrophils. Kristina Rydell-Törmänen showed in a mouse model of sterile lung inflammation that almost 50% of neutrophils at the side of injury have phagosomes that contained material from other neutrophils [ 10 ]. The authors termed this process “neutrophil cannibalism”. The efferocytotic capacity of neutrophils is similar to that of blood derived DCs, but clearly lower as compared to blood-derived macrophages [ 11 ]. It increases in response to pro-inflammatory cytokines such as TNF-α, interferon-gamma (IFN-γ) and granulocyte-macrophage colony-stimulating factor (GM-CSF) and to ligands of TLR2 (Malp2, Pam3CSK4), TLR4 (LPS), TLR7/TLR8 (R848), and TLR9 (ODN 2006) [ 11 , 12 ]. The efferocytotic ability of neutrophils is not exclusively limited to apoptotic conspecifics but also includes remnants of other cell types. Detection of apoptotic cells Apoptotic cells in general release various “find-me” signals that specifically attract neutrophils, including CCL3, CXCL1, CXCL5, CXCL8/IL8, tyrosyl tRNA synthetase (TyrRS) and endothelial monocyte activating polypeptide II (EMAPII) [ 13 – 15 ] (Fig. 1 ). This suggests that neutrophils may be intentionally recruited to help clear apoptotic cell debris. It has to be noted that apoptotic cells release also lactoferrin, which inhibits neutrophil migration [ 16 ]. However, this “keep-out” signal seems not to be sufficient to antagonize other chemoattractants in vivo. Garg and co-workers induced apoptotic cell death in a lung carcinoma cell line before injecting them intradermally into the mice ear pinna [ 17 ]. Cells exposed to the immunogenic apoptosis inducer mitoxantrone stimulated rapid recruitment of neutrophils, which in comparison to other leucocyte subsets constituted the predominant immune cell population accumulating at sites of apoptosis. Similarly, neutrophils accumulate at sites of apoptotic hepatocytes of patients with hepatocellular carcinoma [ 18 ]. The predominant “eat me” signal on the surface of apoptotic cells is phosphatidylserine (PS). There is an extensive literature on the different mechanisms that macrophages use to detect PS-positive cell debris (reviewed in [ 8 , 19 ]). They include directly binding receptors such as adhesion G protein-coupled receptor B1 (ADGRB1), stabilin-2 or T-cell membrane protein 4 (TIM-4). Furthermore, PS is also detected indirectly via soluble “bridging factors” like growth-arrest-specific gene-6 (GAS6) or Milk fat globule-EGF factor 8 protein (MFG-E8), which bind to PS and are then themselves detected by specific receptors on the macrophage. The MFG-E8 receptor αVβ3 integrin is also highly expressed in neutrophils [ 20 ]. However, neutrophils do not express any direct PS receptor (ADGRB1, stabilin-2, or TIM-4) or any receptor of GAS6. Besides PS, there are also other ‘eat-me’ signals exposed on apoptotic cells, including calreticulin, annexin A1, thrombospondin 1 binding sites, and complement proteins C1q or C3b binding sites [ 21 ]. They are recognized by CD91, formyl peptide receptor 2, CD47, CD93, and CD35, respectively. All of them are highly expressed in neutrophils [ 22 – 26 ]. However, their functional role in efferocytosis by these cells is still unexplored. Clearance of apoptotic cells The subsequent events in efferocytosis comprise the engulfment of apoptotic cellular corpses, followed by the formation and maturation of the phagosome, culminating with the degradation of the cargo within the phagolysosomal compartment. The molecular mechanisms of this tightly regulated multi-step process have been investigated in detail in macrophages (reviewed in [ 27 ]). After activation of “eat-me” receptors, the submembranous actin cortex undergoes specific rearrangements which activates the Rho family of small GTPase RAC1 and promotes the formation of a phagosome [ 27 , 28 ]. The processing of engulfed cellular cargo requires a non-canonical LC3-asscociated phagocytosis (LAP) [ 29 ]. LAP represents a specialized mechanism that utilizes components of the autophagic machinery to enhance the degradation of phagocytosed material in an immunologically silent manner [ 30 ]. LAP is triggered by the recruitment of LC3 (microtubule-associated protein 1A/1B-light chain 3) proteins to the single-membrane phagosomes (or LAPosome). For that, PI3KC3 complex needs to be assembled, which consists of Rubicon, vps34, beclin-1, and vps15. This complex converts the LAPosome-bound phosphatidylinositol into the signalling lipid phosphatidylinositol 3-phosphate (PI3P) [ 27 , 28 ]. The PI3P-coated LAPosome stabilizes the NOX2 complex which is responsible for ROS generation, leading to LC3 ligation machinery activation and LC3-II recruitment to the LAPosome [ 31 ]. This last step facilitates the LAPosme-lysosome fusion, resulting in a rapid degradation of the cargo. Recently, Prajsnar et al. identified the LAP machinery in neutrophils, but unfortunately its activation upon efferocytosis of apoptotic cells has not been investigated [ 32 ]. Cunha et al. noted that engulfment of apoptotic corpses per se does not result in immunosuppression, but it is rather the subsequent accumulation of digested products which induces immune tolerance [ 33 ]. The phagocytes overload with lipids and cholesterol stimulates the activation of nuclear steroid receptors from the liver X receptors (LXRs) and peroxisome proliferator-activated receptors (PPARs) families [ 34 , 35 ]. Apart from mediating lipid homoeostasis, LXRs and PPARs induce the clearance of apoptotic cells via expression of phagocytic receptors and opsonins, resembling a positive feedback loop. The anti-inflammatory effects attributed to efferocytosis are also mediated by these pathways, by promoting upregulation of the anti-inflammatory cytokines TGFβ and IL-10 whereas the pro-inflammatory cytokines TNFα, IL-1β, and IL-6 are downregulated [ 19 ]. A similar response has been observed in neutrophils that had phagocytosed apoptotic neutrophils [ 11 , 36 ]. They showed an elevated expression of anti-inflammatory TGFβ and of neutrophil chemoattractants CXCL1 and CXCL8/IL8, and a lower secretion of pro-inflammatory cytokines TNFα and CXCL10/IP-10. Furthermore, they downregulate respiratory burst due to a reduced phosphorylation of p38 MAPK and PKCδ, the kinases involved in NADPH oxidase activation [ 37 ]. Incubation with anti-TGFβ1 antibodies restores respiratory burst [ 36 ]. The inhibitory effect of neutrophil cannibalism on respiratory burst is exploited by invading bacteria to their own advantage. For instance, Leishmania major -infected neutrophils acquire enhanced capacity to engulf apoptotic cells. The uptake of apoptotic cells inhibits respiratory burst, protecting thereby the bacteria [ 37 ]. Manfredi et al. found that MPO and elastase are translocated into phagolysosomes during the process of efferocytosis to facilitate cargo degradation, making these enzymes unavailable for participating in chromatin decondensation – a prerequisite for NET formation [ 38 ]. Thus, neutrophilic efferocytosis impedes NETosis and primes these cells towards a non-inflammatory and resolving phenotype. We could demonstrate recently that neutrophils engulf apoptotic cell-derived extracellular vesicles (aEV) from hepatocytes and several cancer cell lines [ 39 ]. This is associated with an increase of cell surface activation markers CD11b, CD16, CD45, CD66b, CD62L, and secretion of various mitogens, including hepatocyte growth factor (HGF), fibroblast growth factor 2 (FGF2), VEGF, and transforming growth factor alpha (TGFα). Neutrophils express HGF mRNA and store the active protein in secretory vesicles and gelatinase granules [ 40 ]. The release of HGF and other mitogens in response to aEV results in an elevated metabolic activity and proliferation of co-cultured hepatocytes [ 39 ]. This indicates that efferocytosing neutrophils induce tissue regeneration in response to an uptake of apoptotic cells. Strong corroboration for this hypothesis comes from an observation in patients undergoing partial hepatectomy, a surgical procedure that results in massive local apoptosis at the resection margins of the remaining liver lobes [ 39 ]. Free HGF as well as neutrophil-bound HGF in the circulation of these patients correlate with the degree of apoptosis. Notably, higher levels of HGF are associated with improved liver regeneration. Neutrophils in cancer Fridlender et al. identified two distinct populations of tumour-associated neutrophils (TANs): anti-tumourigenic N1 and pro-tumourigenic N2 TANs [ 41 ]. The latter type prevails in many human cancers [ 42 ]. It releases a variety of cytokines, chemokines, and growth factors that promote tumour cell survival and proliferation, such as prostaglandin E2 (PGE2), CCL17, interleukin-6 (IL-6), tumour necrosis factor-alpha (TNF-α), VEGF, and epidermal growth factor (EGF) (Fig. 2 ) [ 43 ]. N2 TANs also secrete collagenase (MMP8) and gelatinase B (MMP9), which facilitate the invasion of tumour cells by remodelling the extracellular matrix [ 44 ]. In addition, their arginase-1 (ARG1) degrades extracellular arginine, which dampens the proliferation of T cells [ 45 ]. Thus, N2 TANs resemble in many respects to neutrophils after uptake of apoptotic cells. Spontaneous apoptosis of single tumour cells can be observed in many treatment-naive patients. It was shown already more than 25 years ago in prostate cancer that an elevated frequency of tumour cell apoptosis correlates with a higher 5-years progression rate [ 46 ]. Similarly, colorectal cancer patients with a higher number of apoptotic cancer cells have a worse overall survival [ 47 ]. Table 1 summarizes numerous studies investigating the relationship between cancer apoptosis rate and clinical outcome in 18 different cancer types. A positive association between apoptosis and poor prognosis was found in most cancers types. Only thyroid carcinoma, neuroblastoma, and glioblastoma showed an inverse relationship. Thus, tumour cell apoptosis promotes the progression of remaining viable tumour cells. Apoptotic cells release the growth factor FGF-β, PGE2, and VEGF, which have a direct promoting effect on the proliferation of adjacent tumour cells (recently reviewed in [ 48 ]). However, there is strong evidence that the pro-tumourigenic effect of apoptosis is mainly mediated by the phagocyte response during apoptotic cell clearance [ 49 ]. Most studies focussed on macrophages, whereas the contribution of efferocytosing neutrophils to tumour growth is much less investigated. Dead tumour cells are not equally distributed throughout the tumour tissue. Many solid cancers show dense cribriform nests or pseudoluminal structures with central aggregates of disintegrated dead tumour cells [ 50 ]. We found recently in colorectal cancer patients that such massive dead cell accumulations stain positive for caspase-cleaved cytokeratin 18 and CXCL8/IL-8, indicating that they derive from apoptotic tumour cells, which release a neutrophil chemoattractant (Fig. 2 ) [ 15 ]. Accordingly, the great majority of aggregates is highly infiltrated with neutrophils and anti-inflammatory polarized TAMs. Blocking the apoptotic cell-derived CXCL8/IL-8 prevents neutrophil-induced anti-inflammatory macrophage polarization. These data fit to the above-proposed concept that neutrophils play a major role in efferocytosis in cases of massive accumulations of apoptotic dead cell remnants. Interestingly, also activation of LAP promotes tumour immune tolerance. LAP-sufficient tumour animal models revealed accumulation of M2 macrophages which support the pro-tumorigenic effects of tumour-associated macrophages (TAMs) [ 33 ]. Consequently, T cell differentiation is skewed towards regulatory T cells that support inflammation resolution [ 19 , 33 ]. Indeed, LAP-deficient TAMs trigger STING-mediated type I interferon responses inducing a pro-inflammatory gene expression and increasing CD8 + T cell function. Remarkably, the overexpression of Rubicon in cancer patients, which is required for LAP but not autophagy, have been suggested as a potential poor prognostic marker [ 51 ]. In line with that, evidence suggest that specifically targeting LAP within the tumour microenvironment through pharmaceutical means promotes an anti-tumour response in a T cell-dependent manner [ 33 ]. Hence, development of therapies targeting efferocytosis-related pathways, in macrophages as well as neutrophils, could present a promising approach for cancer treatment. Neutrophils support tumour growth and spreading not only in the tissue but also in the blood stream. They form heterologous clusters with circulating tumour cells (CTCs) which prolongs their half-life (Fig. 2 ) [ 52 ]. CTC-neutrophils clusters support cell cycle progression, proliferation and survival of tumour cells resulting in extended metastatic potential [ 53 ]. Patients with CTC–neutrophil clusters have poorer outcomes compared to those with homotypic CTC clusters [ 54 ]. CTC-neutrophil clusters may also include NETs, which promote adhesion and extravasation of CTCs at the site of metastasis [ 55 , 56 ]. However, sequencing analysis of CTC-associated neutrophils revealed a N2-like gene expression profile, indicating that not all neutrophils in the clusters form NETs [ 41 ]. N2 neutrophils offer CTCs protection from immune surveillance by inhibition of CD8 + T cells NK cells [ 57 ]. The blood-stream is a harsh environment for CTCs and single tumour cells may die quickly. They may be efferocytosed by adjacent neutrophils in the cluster, leading to mitogenic support of remaining tumour cell in the cluster. In summary, anti-inflammatory neutrophils support tumour growth and spreading in a variety of ways. Although there is growing evidence that efferocytosis contributes to tumorigenic TAN polarization, further research is needed to confirm this concept.
Acknowledgements This study was supported by the Medical University of Vienna. Author contributions CR and RO performed writing, review and revision of the paper. Both authors read and approved the final paper. Data availability Data sharing not applicable to this article as no datasets were generated or analyzed during the current study. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Cell Death Discov. 2024 Jan 13; 10:26
oa_package/5b/91/PMC10787834.tar.gz
PMC10787835
38218730
Introduction In a non-centrosymmetric material, light-matter interactions can generate a finite DC photocurrent under homogeneous illumination in absence of external bias and spatial inhomogeneity. This photovoltaic effect, governed by the intrinsic symmetry properties of materials, is referred to the intrinsic photovoltaic effect (IPVE) or bulk photovoltaic effect (BPVE) 1 – 4 . Hence, the unique physics of IPVE offers an effective approach to surpass the Shockley-Queisser limit in traditional photovoltaic devices 1 – 9 , which attracts growing attention recently. Initial studies on IPVE mainly focused on ferroelectric insulators, such as LiNbO 3 2 , BiFeO 3 10 and BaTiO 3 8 , 11 . Later, researchers found that reducing bandgap size and lowering dimensionality could further enhance the efficiency of IPVE 5 – 7 , 12 – 17 . For example, the IPVE photocurrents observed in narrow bandgap semiconductors (including one-dimensional/1D WS 2 nanotubes 5 ) and Weyl semimetals with broken inversion symmetry are orders of larger than those in wide bandgap ferroelectric insulators 6 , 7 , 12 – 15 . On the other hand, van der Waals (vdW) layered materials meet all merits for IPVE investigations due to its low dimensionality, tunable bandgap, flexibility, easy manipulation, and rich species 15 – 26 . For example, strain-gradient-engineered MoS 2 shows a strong IPVE with photocurrent density over 10 2 A cm −2 , which is comparable to that in 1D WS 2 nanotube 5 , 20 ; The external quantum efficiency of 3R-MoS 2 with spontaneous out-of-plane polarization shows the highest reported value of 16% 22 ; IPVE observed in the in-plane strained 3R-MoS 2 is over two orders of magnitude higher than the unstrained one 20 ; The non-centrosymmetric nano-antennas in centrosymmetric graphene can result in artificial IPVE 23 – 25 ; Moiré-pattern in twisted bilayer graphene and WSe 2 /BP interface can lead to the emergence of spontaneous IPVE 21 , 26 ; Low-dimensional vdW structures such as quasi-1D edges of Weyl semimetal WTe 2 can generate strong IPVE-induced photocurrents, attributing to the strong symmetry breaking and low dimensionality of edges 15 ; Robust IPVE-induced photocurrents are observed in topological insulator monolayer WTe 2 16 . Here, we introduce an alternative low-dimensional system, one-dimensional grain boundary (GB) with non-centrosymmetric crystalline structure, for IPVE investigations. Distinct from previous IPVE systems, GBs widely exist in all kinds of materials. For example, GBs have been uncovered in various vdW layered materials regardless of their crystalline symmetry, including graphene 27 , MoS 2 28 , ReS 2 29 – 33 , and MoSe 2 34 . 1 T′-ReS 2 GBs are ideal for IPVE investigations due to following reasons. (1) Anisotropic optical properties of ReS 2 allow to identify positions of GBs and subdomains simply using polarization-resolved optical microscopy; (2) GBs in ReS 2 have well-defined structures free of dangling bonds. In this work, we uncover strong and robust IPVE in 1D vdW GBs in ReS 2 . Symmetry analysis and experimental results demonstrate that inversion symmetry is broken near GBs, which results in a DC photocurrent that propagates along GBs without any voltage bias. We demonstrate that this IPVE-induced photocurrent is gate tunable and possesses a pronounced polarization-independent component. Furthermore, the IPVE-induced photocurrent densities in 1D ReS 2 GBs are among the highest values compared with reported material systems.
Methods Sample preparation ReS 2 samples were prepared on silicon substrate covered with 300 nm SiO 2 through standard mechanical exfoliation method. Angle-resolved polarized optical microscope is used to identify GBs and domains in ReS 2 . Electrodes (5/35 nm Cr/Au) were patterned via standard photolithography process (MicroWriter ML3, Durham Magneto Optics Ltd). STEM characterizations The STEM/HAADF images were obtained using a JEOL ARM200F equipped with a CEOS aberration corrector. The microscope featured a cold field emission gun and was operated at an accelerating voltage of 200 kV. The convergence angle was ~28 mrad. Optical characterizations The devices were characterized using a semiconductor parameter analyzer (FS-Pro) under vacuum (~10 −6 mbar) at room temperature. For short-circuit photocurrent measurements, 532 nm lasers were used as excitation sources with laser power of 200 μW, respectively. The lasers were focused by a 50× microscope objective lens (0.5 N.A.). The size of laser spot with a Gaussian profile was ~3 μm for 532 nm laser. Angle-resolved polarized Raman spectroscopy was performed using a 532 nm laser with a spectrometer (Andor SR-500i-D2). A linear polarizer and half-wave plate (Thorlabs) were used to adjust the orientation of the laser polarization. Theoretical calculation Density functional theory (DFT) calculations for structure optimization and electronic properties were performed using the Vinna ab initio simulation package (VASP) 49 . Exchange-correction functional was treated within the generalized gradient approximation of Perdew, Burke, and Ernzerhof 50 . The electronic wave functions were expanded using a planewave basis set with an energy cutoff of 300 eV, and the tolerance for the total energy was <10 −4 eV. A 1 × 10 × 1 k-mesh was utilized for self-consistent calculations of the supercell structure.
Results Characterization of 1D vdW GBs in ReS 2 Bulk 1 T′-ReS 2 is a vdW semiconductor with a centrosymmetric crystalline structure under the inversion symmetric space group of P 35 – 38 , as demonstrated by the scanning transmission electron microscope (STEM) image in Fig. 1a . Thus, IPVE is not allowed in thin-film ReS 2 . In addition, we find an abundance of 1D GBs in ReS 2 . The in-plane orientations, signified by the direction of Re chains, of the two neighboring subdomains form a 120° angle (see Fig. 1b and Supplementary Fig. 1 ), which aligns with prior STEM research findings 30 , 33 . Figure 1c shows a top view schematic of ReS 2 crystalline structure with GBs. Here, we use A and B to represent two adjacent subdomains. BA and AB GBs are denoted by “↑” and “↓” arrows, respectively. Using polarization-resolved optical microscopy (see Fig. 1d, e and Supplementary Fig. 2 ), we can clearly identify the positions of GBs in ReS 2 flakes due to the anisotropic optical reflection and different Re-chain directions of subdomains. As shown in Fig. 1e , the ReS 2 flake is separated by multiple GBs (along the y -direction) and forms multi-domain structures. Angle-resolved polarized Raman spectroscopy was further performed to identify the crystalline orientations of ReS 2 subdomains (see Fig. 1f ). The intensity of A g2 mode (212 cm −1 ) is maximum when the light polarization direction is parallel to the Re-chain direction of ReS 2 30 , 32 , 36 . Raman result indicates that there is ~117° difference between Re-chain directions of two adjacent subdomains, which is very close to the angle ~120° observed in STEM. The ~3° deviation of Raman characterization is within the permissible range of our instruments. IPVE theory in 1D vdW GBs The only symmetry present near GB regions is the two-fold rotation along the y -directional axis. Expanded analysis in Supplementary Note 1 and Supplementary Fig. 3 suggests that the GB region is characterized by a point group of C 2 and associated with a broken inversion symmetry, resulting in nonzero second-order nonlinear light-matter-interaction tensors , where w is the angular frequency of incident light, is the wave vector and l / j / k represents x -, y- , or z -directions. Under linearly polarized light, a finite DC photocurrent density along l -direction can be generated (only consider the -independent term) 20 , 39 where . Since only depends on the intrinsic physical properties of materials, this effect is called IPVE or BPVE. Here, we mainly focus on IPVE-induced photocurrent along GB ( y -direction) . For incident light normal to the two-dimensional plane of ReS 2 flake ( E z = 0), can be written as Equation 2 can be further simplified utilizing the rotation symmetry in GB region as shown in Fig. 1b (see Supplementary Note 2 for details). Under rotation symmetry ( x , y , z → - x , y , - z ), which makes . If we write E = [ E 0 sin θ , E 0 cos θ , 0], where θ is the angle between y -direction and light polarization direction, then we have Here, and are the polarization-dependent and polarization-independent terms, respectively. Furthermore, IPVE-induced photocurrents along two adjacent ↑ and ↓ GBs should have opposite directions restricted by their reversed orientations. Experimental observation of IPVE in 1D vdW GBs To study the IPVE in ReS 2 GBs, photodetectors with channel parallel to GBs are fabricated. Figure 2a shows the optical image of a device under polarized white light (the angle between light polarization and y -axis is ~30°). The optical images at other polarized angles are shown in Supplementary Fig. 2 . The thickness of ReS 2 flake is ~180 nm determined by atomic force microscope (AFM) (see Supplementary Fig. 4 ). The ↑ and ↓ GBs are denoted by blue and red dash lines, respectively. The detailed fabrication process can be found in the Method section. The device shows linear current-voltage ( I ds - V ds ) characteristic, indicating a good Ohmic contact between ReS 2 and metal electrodes (see Supplementary Fig. 5 ). A linearly polarized 532 nm laser with a diameter around 3 μm was focused on the channel and short-circuit photocurrents ( I ph ) were collected (the angle between laser polarization and y -axis is ~30°). As shown in Fig. 2b , I ph ( y ) at pristine region ( x = 0 μm) shows ordinary shape with vanishing value in the middle of channel. The finite photocurrents near electrodes can be attributed to extrinsic photovoltaic effect, such as built-in Schottky junction between ReS 2 and electrodes and photo-thermoelectric effect. This indicates that the pristine region of ReS 2 does not support IPVE due to the preservation of inversion symmetry. This observation is further confirmed in a device based on ReS 2 without GBs (see Supplementary Fig. 6 ). On the other hand, we observed very robust photocurrents in the middle regions of GBs with negative values at ↑ GB (along blue dashed line in Fig. 2a ) and positive values at ↓ GB (along red dashed line in Fig. 2a ) in sharp contrast to vanishing photocurrents in pristine regions. This phenomenon is reproducible in other samples (see Supplementary Fig. 7 ). Scanning photocurrent spectroscopy of total I ph further confirms the observation as shown in Fig. 2c . Moreover, we measured photocurrent along x -direction I ph ( x ) at fixed y position (indicated by the black dashed line in Fig. 2a ). As shown in Fig. 2d , consistent and robust valley and peak features are observed at ↑ and ↓ GBs, respectively. The above results show excellent agreement with IPVE theory. To further demonstrate the effectiveness of IPVE theory, we check if a large polarization-independent photocurrent term exists in ReS 2 GBs. Figure 2e shows the polarization-resolved photocurrents in middle of ↑ and ↓ GBs. The polarization-dependent term is complicated since it is influenced by both anisotropic properties of ReS 2 domains and IPVE. Hence, we mainly focused on the polarization-independent term . Besides, polarization-independent term is more appealing for energy harvesting applications due to the unpolarized nature of sunlight. We further extract and show the mapping results in Fig. 2f . The opposite directions of are observed in ↑ and ↓ GBs, consistent with IPVE theory. We then investigated the electrical tunability of IPVE in ReS 2 GBs using gate bias. Figure 3a shows a device based on few-layer ReS 2 with GBs. The thickness of ReS 2 flake is 8 nm. The few-layer ReS 2 phototransistor exhibits n-type characteristics (see Supplementary Fig. 8 ), consistent with previous reports 38 , 40 . Opposite directional photocurrents are observed at ↑ and ↓ GBs (see Fig. 3b ), showing good agreement with other samples. Moreover, IPVE-induced photocurrents can be effectively tuned by gate voltage with ~ 28% and 60% enhancement from −40 to 40 V for ↑ and ↓ GBs, respectively. To understand this gate tunability of IPVE-induced photocurrent, we first examine whether it simply originates from the tuned Schottky barrier at metal-ReS 2 interface which may affect the collection efficiency of carriers. As shown in Supplementary Fig. 9 , the 8 nm-thick ReS 2 device shows a good linear current-voltage characteristic at various gate voltages, indicating a good Ohmic contact between ReS 2 and metal electrodes. Hence, if there exists Schottky barrier, it would be very low which is unlikely to significantly affect the photocurrent intensities. On the other hand, IPVE-induced photocurrents have two contributions which are shift and ballistic currents 6 , 19 , 20 , 41 , 42 . Shift and ballistic currents strongly depend on the properties of nonequilibrium carriers excited by polarized lights. Thermalization of nonequilibrium carriers can be caused by electron-defect, electron-phonon, and electron-electron interactions. At different gate voltages, electron concentration changes which probably affects the thermalization processes of excited nonequilibrium carriers, such as their mean free bath length and mobility, and hence affects induced IPVE photocurrent densities. This is one plausible explanation. Further studies can be conducted to fully understand this phenomenon. We compared the strength of IPVE in ReS 2 GBs with other materials. Although structures of ReS 2 GBs are well defined, the effective width of GBs, which denotes the active region with strong inversion symmetry breaking for generating IPVE photocurrent, is unknown. Here, we give a photocurrent density range when effective width varies from 3 to 300 nm. As shown in Fig. 4 , the photocurrent densities in ReS 2 GBs are comparable to those in 1D WS 2 nanotube 5 , strained 3R-MoS 2 19 and MoS 2 20 and orders of magnitude higher than those in ferroelectric materials 3 , 4 , 17 , 43 , 44 .
Discussion To better understand the giant IPVE photocurrent densities in ReS 2 GBs and its underlying physical mechanism, the first-principles calculations of band properties are performed. Detailed information about calculations can be found in the Method Section and Supplementary Information (see Supplementary Figs. 10 , 11 ). As shown in Fig. 5a , ReS 2 near GBs has a lower conduction band minimum and higher valence band maximum compared with those of pristine ReS 2 . In addition, GBs have significant influence to the band structures of ReS 2 near GBs through introducing significant number of new states (see Supplementary Figs. 10 , 11 ). These new states might improve the light absorption and enhance the IPVE photocurrent. Importantly, a quantum-well structure is formed along x -direction (normal to GB direction) due to lower conduction band minimum and higher valence band maximum near GBs as shown in Fig. 5b . This indicates that carriers generated near GBs tends to be caught into the quantum well and transport along GBs (carrier collection direction of electrodes) is more preferred than other directions. This further enhances the IPVE photocurrent. Besides, the well-defined structures of GBs without any dangling bonds and the indirect bandgap of ReS 2 near GBs could further suppress scatterings and recombination of photo-excited carriers (see Supplementary Fig. 10 ). These are possible reasons that lead to the giant IPVE photocurrent density in ReS 2 GBs. As shown in Supplementary Fig. 12 , we still can observe pronounced IPVE photocurrent at ReS 2 GBs with channel length over 100 μm. In addition, we also studied IPVE at ReS 2 edges for comparison, since edges are non-centrosymmetric with broken periodic structures. We fabricated and measured three ReS 2 samples in which we did not found detectable IPVE-induced photocurrent at edges (see Supplementary Fig. 13 ). High density of defect states at edges, such as dangling bonds, might induce strong electron-defect scatterings and suppress the IPVE photocurrent 45 . Power- and wavelength- dependent photocurrents are measured to further clarify the physical mechanism of IPVE observed in ReS 2 GBs. As shown in Fig. 5c , the power-dependent photocurrent at GBs shows a transition from linear to square-root dependence when power increases which is consistent with the prediction of theoretical shift current model and previous experimental reports 5 , 19 , 20 , 42 , 46 . We theoretically calculated the shift current in ReS 2 GBs. Detailed calculation process and discussion are shown in Supplementary Information (see Supplementary Note 3 and Supplementary Fig. 14 ). As shown in Fig. 5d , our shift current model shows good agreement with experimental results at different excitation wavelengths. All above results suggest that shift current dominates the photocurrent generation process of IPVE in ReS 2 GBs. Finally, we conclude through discussing the distinctive aspects of GB-induced symmetry breaking and its potential implications relative to prior research. Firstly, GBs widely exist in all kinds of materials and have a variety of configuration, which provides a capacious platform for IPVE and physics investigations. Secondly, GBs are embedded in bulk materials and there is no symmetry requirement for the crystalline structure of bulk material to induce symmetry breaking in GBs. Thirdly, formation of the quantum-well structure makes GBs a good 1D/quasi-1D system for IPVE investigation which can effectively suppress carrier dissipation to other directions. Fourthly, compared with edges 45 , GBs with well-defined crystalline structures are free of dangling bonds. The reduced electron-defect scatterings in GBs with well-defined structures might suppress scatterings of photo-excited carriers and enhance IPVE photocurrent. Lastly, structures and densities of GBs can be generated and controlled through adjusting material growth conditions 47 , 48 . Other approaches, such as external strain, can also generate and control GBs in materials 29 . The ability to control formation and structures of GBs is important for making efficient optoelectronic devices. Hence, we believe the rich species and configurations, well-defined 1D/quasi-1D structures, and potential controllability make GBs a promising optoelectronic platform for novel physics and device applications.
The photovoltaic effect lies at the heart of eco-friendly energy harvesting. However, the conversion efficiency of traditional photovoltaic effect utilizing the built-in electric effect in p-n junctions is restricted by the Shockley-Queisser limit. Alternatively, intrinsic/bulk photovoltaic effect (IPVE/BPVE), a second-order nonlinear optoelectronic effect arising from the broken inversion symmetry of crystalline structure, can overcome this theoretical limit. Here, we uncover giant and robust IPVE in one-dimensional (1D) van der Waals (vdW) grain boundaries (GBs) in a layered semiconductor, ReS 2 . The IPVE-induced photocurrent densities in vdW GBs are among the highest reported values compared with all kinds of material platforms. Furthermore, the IPVE-induced photocurrent is gate-tunable with a polarization-independent component along the GBs, which is preferred for energy harvesting. The observed IPVE in vdW GBs demonstrates a promising mechanism for emerging optoelectronics applications. The intrinsic photovoltaic effect (IPVE) in noncentrosymmetric materials has the potential to overcome the limitations of traditional photovoltaic devices. Here, the authors report the observation of a strong and gate-tunable IPVE in 1D grain boundaries of a van der Waals semiconductor, ReS 2 . Subject terms
Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s41467-024-44792-4. Acknowledgements The work was financially supported by the National Natural Science Foundation of China (62275117, X.C.; 62261136552, J.M.; 52273279, X.Zhao), Shenzhen Excellent Youth Program (RCYX20221008092900001, X.C.), Open research fund of State Key Laboratory of Infrared Physics, Shanghai Institute of Technical Physics, Chinese Academy of Sciences (SITP-NLIST-YB-2022-05, X.C.), Shenzhen Basic Research Program (20220815162316001, X.C.), Natural Science Foundation of Guangdong Province (2023A1515011852, X.L.Y.), Guangdong Major Talent Project (2019QN01C177, X.C.; 2019CX01X014, T.W.), Fundamental Research Funds for the Central Universities (X.Zhao), and Beijing Natural Science Foundation (Z220020 X.Zhao). Author contributions X.C. conceived and supervised the projects. Y.Z. fabricated ReS 2 samples and devices with the assistance of Z.L. Y.Z. characterized photocurrent of devices with the assistance of Z.L. and T.W. X.L.Y. did the theoretical calculations. X.Zhou conducted the STEM characterizations with the assistance of X.Zhao, X.C., J.M. and Y.Z. proposed the IPVE/BPVE mechanisms. Y.Z. and X.C. drafted the manuscript with assistance of X.L.Y., X.Zhou and J.M. All authors discussed and commented the manuscript. Peer review Peer review information Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available. Data availability Relevant data supporting the key findings of this study are available within the article and the Supplementary Information file. All raw data generated during the current study are available from the corresponding authors upon request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Nat Commun. 2024 Jan 13; 15:501
oa_package/69/a9/PMC10787835.tar.gz
PMC10787836
38218955
Introduction Diabetes mellitus (DM) is a major public health problem worldwide and one of the major diseases threatening human health and survival. Diabetic retinopathy (DR) is the most common microvascular complication of diabetes and one of the leading causes of blindness in people with diabetes in clinical practice, affecting people’s quality of life 1 – 3 . The biochemical mechanisms underpinning the development of DR are poorly understood and may be related to metabolic abnormalities, cellular signaling pathway transduction, glucotoxicity, and oxidative stress 4 . Although accumulating epidemiological evidence has linked serum metal exposure to diabetes, the association between heavy metals and DR remains elusive. Due to the rapid development of industrialization and urbanization, the pollution concentration of lead (Pb), cadmium (Cd), mercury (Hg), selenium (Se), and manganese (Mn) in the environment is higher than that of other heavy metals 5 , 6 . Epidemiologic studies that have evaluated the relationship between these heavy metals and the risk of type 2 diabetes mellitus (T2DM) and its complications. An observational study showed that higher Mn intake was directly associated with a lower risk of type 2 diabetes in a population of postmenopausal women 7 . Previous study found a negative association between typical levels of serum Cd and diabetes risk in a United States of America population aged 20 years or older 8 . Kornhauser et al. reported that plasma Se was negatively associated with the severity of diabetic nephropathy in patients with type 2 diabetes 9 . There exists scientific evidence indicating that low levels of Cd exposure exacerbate the progression of kidney disease in diabetics 10 . However, the relationship between these heavy metals and diabetic retinopathy is poorly studied. Serum albumin (ALB) is a crucial physiological transporter of vital metal ions Pb 2+ , Cd 2+ , Hg 2+ , Se 2+ , and Mn 2+ in the bloodstream 11 , 12 . Nonetheless, the evidence as to whether heavy metals influence the incidence of DR by binding to ALB and thereby affecting the incidence of DR evidence on the relationship between serum ALB and heavy metals and DR is unclear. Therefore, we hypothesized that the relationship between blood levels of these five heavy metals and DR is nonlinear. The present sought to explore the association between blood levels of Pb, Cd, Hg, Se, Mn, and DR using data from a representative sample of people with T2DM who participated in the National Health and Nutrition Examination Survey (NHANES) between 2011 and 2020.
Methods Sample and overall The National Center for Health Statistics (NCHS) conducted a population-based cross-sectional NHANES to collect data on the nutritional and health conditions of children and adults in the United States of America via interviews, physical examinations, and laboratory tests. NHANES employs a categorized, multistage probability sampling design to select a sample of an unstructured population to evaluate health and nutritional status. NHANES procedures and protocols were approved by the NCHS Research Ethics Review Committee and written consent was obtained. Inclusion and exclusion criteria The study population was mainly over 30 years of age. This cross-sectional analysis used data from five NHANES survey cycles (each lasting two years) between 2011 and 2020. Of the 45,462 participants in this study, participants aged < 30 years (n = 23,484), those with unknown diabetes-related conditions, and those lacking information on the glycated hemoglobin (HbA1c) state or the HbA1c level was less than 6.5% (n = 18,955) were eliminated. Diabetic patients aged < 30 years (n = 910) were eliminated to lower the likelihood of having type 1 diabetes. Besides, participants lacking information on serum levels of heavy metals (n = 517) and DR (n = 13) were disqualified. We filled in the remaining missing values in using the average of the entire sequence in the spss software. Finally, 1583 participants were included in the analysis. The study design flow is depicted in Fig. 1 . Diagnosis of DR The following criteria were used to identify diabetes: (1) a prior diagnosis from a medical expert, (2) fasting blood glucose (FBG) levels of > 7.0 mmol/L, (3) HbA1c levels of > 6.5%, or (4) receiving diabetes medication 13 . DR was self-reported using a dichotomous classification, whereby the respondent had been informed by their physician that diabetes was affecting their eyes. Determination of heavy metals Whole blood specimens were collected by physicians at the NHANES Mobile Examination Center (MEC) via venipuncture in EDTA-coated tubes, centrifuged on-site, and stored frozen at − 30 °C for transportation to the Centers for Disease Control and Prevention (CDC) laboratory in California, where they were stored in a frozen state until further analysis. First, in the sample dilution step, a small amount of whole blood is taken from a larger sample and mixed to distribute the cellular components evenly. This mixing is critical because certain metals (e.g., lead) are primarily associated with red blood cells. Samples with clots or microclots are identified and excluded from analysis due to concerns about sample inhomogeneity. Diluted blood samples are prepared by mixing 1 part of the sample, 1 part of water, and 48 parts of diluent. The diluent contains chemicals that release metals from red blood cells, reduce ionization inhibition, prevent clogging, and enable the use of internal standards. The diluted sample is then passed through an inductively coupled plasma (ICP) ion source into the mass spectrometer. The liquid blood sample is converted into aerosol droplets that are vaporized, atomized, and ionized in the plasma region. The resulting ions enter the mass spectrometer with argon gas for analysis. The Dynamic Reaction Cell (DRC) plays a vital role in the selective reaction by removing interferences or enhancing the ion signal of specific elements. The ions passing through the DRC are electrically selected and directed to the analytical quadrupole. The electrical signal generated by the ion impact detector is processed into digital information to determine the elemental concentration. For the values under LODs, an imputed fill value (LOD/√2) was adopted to fill up the vacancy in data preprocessing. Quality assessment and quality control (QA/QC) processes for data collection from blood samples complied with the Therapeutic Laboratory Development Act of 1988 14 . Detail information of the NHANES laboratory procedure is available at https://www.cdc.gov/nchs/nhanes/index.htm . Covariates Data on demographics (age, gender, and race/ethnicity), academic achievement, poverty-to-income proportion, smoking, body mass index (BMI), waist circumference, serum ALB, FBG, HbA1c, triglycerides, and total cholesterol were collected using standardized questionnaires. Participants were divided into Caucasian, Black, Hispanic, Mexican American, and other race categories. Height (H, m) and weight (W, kg) were measured according to norms, and BMI was calculated as W/H 2 (kg/m 2 ). Smokers were defined as individuals who smoked ≥ 100 cigarettes 15 . Smoking status was divided into three categories: never smoked (< 100 cigarettes in their lifetime), gave up smoking (previously smoked > 100 cigarettes but no longer smoke), and still smoking (previously smoked > 100 cigarettes and still smoking). Hypertension was defined as systolic blood pressure > 140 mmHg or diastolic blood pressure > 90 mmHg, use of blood pressure drugs, or self-reported high blood pressure 16 . Statistical analysis Categorical variables were expressed as proportions and frequencies and were statistically analyzed using Chi-square analyses. Continuous variables were expressed as mean and standard errors and were analyzed using linear regression models. Multifactorial logistic regression analysis was used to examine the relationship between heavy metal exposure and DR. Additionally, subgroup analysis of heavy metal concentration and DR was performed to test whether the effect of serum levels of heavy metal on DR could be altered by age, sex, and BMI. All data analyses were performed using R (version 4.0.2) and SPSS software (version 24.0). p < 0.05 was considered statistically significant. Institutional review board statement Our study used five cycles of open NHANES database (2011–2020), National Center for Health Statistics granted the study procedures of Ethics Review Board. These data could be accessible at the following URL: https://www.cdc.gov/nchs/nhanes/irba98.htm#print . Ethics statement All survey participants signed a declaration of consent form after being made aware of the nature of the poll. The informed consent was approved after evaluation by the Committee of the National Center for Statistics on Health Ethics Assessment Board. To make the best use of these resources, all of the data is made publicly accessible after formal anonymization. These data could be accessible at the following URL: https://www.cdc.gov/nchs/nhanes/irba98.htm#print .
Results Participant characteristics This study included 1583 T2DM patients, of which 331 had concomitant DR and 1252 had no concomitant DR. The demographic details of participants with or without DR are shown in Table 1 . DR participants were older and had higher HbA1c levels and lower waist circumference, BMI, serum ALB, and serum Mn levels than non-DR participants. No statistically significant difference was observed between DR and non-DR groups in terms of race, gender, education, household income-to-poverty ratio, triglycerides, total cholesterol, smoking status, blood Pb, Cd, Hg, and Se, and hypertension status (p > 0.05). Relationship between serum level of heavy metal and DR Multifactorial logistic regression analysis was performed to evaluate the relationship between serum heavy metal exposure and DR, as shown in Table 2 . The data from Model 1 was left unadjusted, Model 2 was adjusted for gender, age, and race, and Model 3 was adjusted for education, household income-to-poverty ratio, smoking, hypertension, and BMI based on Model 2. The blood Mn exposure in Model 1 decreased the incidence of DR (p = 0.002). The blood Mn exposure was significantly negatively associated with DR in Models 2 and 3. However, no association was observed between blood Cd, Se, Hg, or Pb exposure and DR in all models (p > 0.05). Subgroup analysis A significant negative correlation between blood Mn content and DR was observed after adjusting for age, sex, ethnicity, and other factors. The subgroups were further stratified by age, sex, and BMI, as shown in Table 3 . In the age ≥ 60 years group, blood Mn was significantly negatively associated with DR prevalence without adjusting for confounders (p < 0.001) and even after adjusting for all potential variables. However, no significant association between blood Mn and DR in the 30–44 and 45–60 age categories (p > 0.05). After adjusting for all variables, the data were stratified by gender, and the results showed there was a significant negative association between blood Mn concentration and DR prevalence in the male population compared with the female population (p = 0.013). When stratified by BMI, no significant correlations were observed between blood Mn and DR in 0–25 and 25–30 kg/m 2 groups after adjusting for confounders. Additionally, a significant inverse correlation was found between blood Mn and DR in the BMI ≥ 30 kg/m 2 group (p = 0.014). Effect of mediating factors on the association between serum Mn and DR The Baron and Kenny causal step method combined with multiple linear regression models was utilized to analyze the mediating role of serum ALB in the association between serum Mn and DR. As shown in Fig. 2 , after adjusting for Model 3, serum Mn had a direct impact on the prevalence of DR (p = 0.010), which was partially mediated by ALB (p = 0.008). Serum ALB mediated 12.80% of the association between blood Mn and DR.
Discussion The present study aimed to elucidate whether there was an association between serum levels of heavy metals and DR among individuals with T2DM in the United States of America. We found a substantial negative relationship between blood Mn and DR, which was partially mediated by serum ALB. Only individuals aged 60 years exhibited a significant inverse association between blood Mn levels and DR after age stratification. Under sex stratification, a significant negative association between blood Mn levels and DR prevalence was observed in the male population compared with the female population. Moreover, a statistically significant association between blood Mn and DR was observed in the BMI ≥ 30 kg/m 2 group under BMI stratification. DR, a common microvascular complication of diabetes with a global prevalence of 34.6%, is a major cause of blindness 17 . The onset and evolution of DR are influenced by various mechanisms, such as inflammatory processes, high blood pressure, abnormal lipid metabolism, and insulin resistance 18 – 20 . Only a few studies have examined the association between exposure to heavy metals and DR, and the findings are controversial. Pb, Hg, Se, and Cd are widespread environmental heavy metal toxicants. Handan et al. found that Se exerts a protective effect on retinal pigment epithelium (ARPE-19) and primary human retinal microvascular endothelial (ACBRI 181) cells against high glucose (HG)-induced oxidative stress and apoptosis 21 . A previous retrospective study found that blood Cd levels were higher in the DR group than in the DM group 22 . These findings are inconsistent with our results, perhaps due to different populations, races, disease states, and health conditions that affect the absorption, transport, and metabolism of Se and Cd in the body, affecting the association between Se and Cd and the disease. Mn is one of the essential components of the Mn super antioxidant dismutase (MnSOD), which scavenges reactive oxygen species (ROS) under mitochondrial oxidative stress. MnSOD genes and Mn levels may impact MnSOD activity 23 . It was previously demonstrated that Mn administration improved MnSOD activation and shielded people from T2DM and related consequences 24 . Kowluru et al. demonstrated that Mn-SOD overexpression prevented glucose-induced increased oxidative stress and apoptosis of retinal endothelial cells, suggesting a protective role for Mn-SOD in the pathogenesis of diabetic retinopathy 25 . On the other hand, previous studies have found that insulin synthesis and secretion in the pancreas are impaired, and insulin degradation and glucagon release are accelerated in Mn-deficient mice, whereas other studies have found that Mn reduces oxidative stress through antioxidant enzyme systems and non-enzymatic pathways to protect pancreatic islet cells from reactive oxygen species (ROS) 26 , 27 . The current study suggested that blood Mn may act as a barrier against the onset of DR. This is consistent with our findings. However, whether Mn can influence the development of DR through other metabolic pathways has yet to be investigated. In plasma, approximately 80% of the oxidized state of Mn 2+ is bound to ALB and globulin and transported to the liver, kidney, small intestine, endocrine glands, pancreas, brain, bone, muscle, and hair 12 , 28 . ALB is a glycosylated protein synthesized and secreted by hepatocytes, which has anti-oxidative stress and anti-inflammatory 29 . Previous studies have shown reduced serum ALB levels are associated with DR risk 30 , 31 . The present study found that serum ALB significantly mediated the association between blood Mn and DR. This implies that blood manganese combined with serum albumin may reduce the development of DR through its antioxidant and anti-inflammatory effects. In this study, we found a negative correlation between blood Mn and DR only in people aged ≥ 60 years. This may be because aging is linked to a decline in Mn levels. This generally agrees with a large cross-sectional study 32 , which found that serum Mn levels in people aged ≥ 60 years may contribute to preventing and controlling prediabetes and diabetes. Compared with females, males in the current study exhibited an adverse correlation between serum Mn levels and DR. According to a survey of dietary intake of Mn, females received considerably more Mn from their diets than males 33 . The survey also showed that the biological half-life of Mn was substantially lower in women than in men. These results imply that Mn absorption and metabolism differ between males and females, which may help to explain some of the observed gender disparities in our study. Our study is the first study to explore the relationship between blood levels of heavy metals and DR in T2DM patients older than 30 in the United States of America. Nonetheless, the current study has some limitations. First, we could not demonstrate the causal link between specific molecular mechanisms in blood Mn and DR due to the cross-sectional design of the study, which should be further validated in a prospective cohort study and fundamental research. Second, the outcome variable was based on self-reported history of DR, which may not be entirely accurate. Third, although some of the confounders were adjusted in this study, there were still many confounders affecting DR, such as drug use and family history of dyslipidemia, that we were unable to obtain. Finally, because the NHANES-based study population was used, it was difficult to assess the validity of the findings from various ethnic studies.
Conclusions In summary, the present study found a significant negative association between blood Mn levels and DR, suggesting that an appropriate increase in Mn intake may delay the onset and development of DR.
The present study utilized the National Health and Nutrition Examination Survey (NHANES) database to examine the relationship between serum levels of heavy metals and Diabetic retinopathy (DR) in individuals aged over 30 years with type 2 diabetes mellitus (T2DM) in the United States. A cross-sectional analysis was conducted on 1583 individuals with T2DM from the NHANES 2011–2020, including 331 individuals in the DR group and 1252 individuals in the non-DR group. We collected data on serum levels of heavy metals, DR, and serum albumin for descriptive statistics, linear regression, and logistical regression analysis. After adjusting for age, gender, race and other factors, there was no statistically significant association between blood cadmium, selenium, mercury, or lead and DR. However, serum manganese (Mn) and DR had a significant negative association (β = − 0.2045, 95% CI = − 0.3484, − 0.0606). Serum albumin partially modulated the indirect influence of serum Mn on the incidence of DR, accounting for 12.80% of the association between serum Mn and DR. There was a negative association between serum Mn levels and the prevalence of DR in people with T2DM. Mn intake at least in this study has a little influence on the onset and development of DR. Subject terms
Author contributions This study’s data gathering, curation, analysis of statistics, and writing of the publication were all jointly completed by Y.Z. Input from X.K.L. was used in the statistical analysis. W.X. oversaw the research and helped with the editing and revision of the manuscript. All authors contributed to the article and approved the submitted version. Funding This work was supported by generous grants from the Xuzhou Medical University Affiliated Hospital Development Fund (XYFZ2020008), National Nature Science Foundation of China (NSFC-81870534), the China International Medical Exchange Foundation Endocrinology and Metabolism Elite Research Fund (2021-N-03). Data availability The data used in the present research were obtained from publicly accessible sources. These data could be accessible at the following URL: https://www.cdc.gov/nchs/nhanes/ . Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1268
oa_package/b6/a5/PMC10787836.tar.gz
PMC10787837
38218935
Introduction Acute lung injury (ALI) and its more severe form, acute respiratory distress syndrome (ARDS) are common, life-threatening critical illnesses that lead to significant morbidity and mortality [ 1 ]. Despite prominent breakthroughs in the pathophysiology of ALI/ARDS, the hospital mortality rate of these disorders remains high (46.1%), and effective pharmacological treatments are still lacking [ 2 ]. ALI/ARDS is characterized by dysregulated lung parenchymal inflammation, leading to diffuse alveolar damage and edema, ultimately contributing to acute hypoxemic respiratory failure [ 3 ]. Uncontrolled local or systemic inflammation is believed to be the predominant cause of ALI/ARDS [ 4 ]. Activated macrophages, especially recruited circulating monocyte-derived macrophages further release pro-inflammatory cytokines, which give rise to an inflammation cascade [ 5 ]. The NOD-like receptor pyrin domain-containing protein 3 (NLRP3) inflammasome is excessively activated in macrophages during ALI/ARDS progression [ 6 ]. The NLRP3 inflammasome, which consists of a sensor (NLRP3), an adaptor apoptosis-associated speck-like protein containing a caspase-recruitment domain (ASC), and an effector caspase (caspase-1), is involved in the production of pro-inflammatory cytokines, interleukin (IL)-1β and IL-18 [ 7 ]. NLRP3 inflammasome activation reportedly involves two steps: priming and activation [ 8 ]. The priming step of NLRP3 inflammasome activation is regulated via transcriptional and post-translational mechanisms [ 9 ]. NF-κB signaling induces the transcriptional activation of NLRP3 priming by upregulating the gene expression of NLRP3 inflammasome components [ 10 , 11 ]. Post-translational modifications (PTMs) of NLRP3, such as ubiquitination, phosphorylation, and SUMOylation, may stabilize NLRP3 in an auto-suppressed inactive state [ 12 ]. The activation step is induced by various pathogen-associated molecular patterns (PAMPs) and damage-associated molecular patterns (DAMPs), including extracellular ATP, pore-forming toxins, RNA viruses, and particulate matter [ 13 ]. However, post-transcriptional regulation of NLRP3 inflammasome activation in ALI/ARDS remains unclear. N 6 -methyladenosine (m 6 A) modification, the most abundant modification of messenger RNA (mRNA), may reversibly regulate target genes at the post-transcriptional level, thereby affecting almost all crucial biological processes [ 14 ]. This dynamic and reversible process is primarily regulated by the m 6 A methyltransferase complex, which contains methyltransferase-like 3 (METTL3), methyltransferase-like 14 (METTL14), Wilms tumor suppressor-1-associated protein (WTAP), and demethylases, including fat mass and obesity-related protein (FTO) and ALKB homolog 5 protein (ALKBH5) [ 15 ]. Meanwhile, RNA-binding proteins that identify and bind to m 6 A sites, such as the YT521-B homology domain (YTHD) family, and the insulin-like growth factor 2 mRNA-binding protein (IGF2BP) family, serve as m 6 A readers and direct the fate of target RNAs by influencing alternative pre-mRNA splicing, RNA stability, and translation efficiency [ 16 ]. m 6 A facilitates the progression of several inflammatory diseases, such as non-alcoholic fatty liver disease, autoimmune diseases, and infections [ 17 – 20 ]. Evidently, global m 6 A levels are significantly increased in alveolar epithelial cells, mediated by the upregulation of METTL3, which is closely associated with ALI [ 21 ]. Nonetheless, the effects of METTL14-regulated m 6 A methylation in ALI/ARDS remain unclear and the precise molecular targets of METTL14 in ALI/ARDS remain to be elucidated. Therefore, we sought to determine the functional role of METTL14 and its target in ALI/ARDS. Herein, we first elucidated that RNA m 6 A modification in macrophages is involved in the progression of ALI/ARDS, and then verified the role of METTL14 in this process. Further mechanistic studies revealed that NLRP3 is the methylated target of METTL14 and that IGF2BP2 stabilizes NLRP3 mRNA during NLRP3 inflammasome activation in ALI/ARDS. These results indicated that METTL4 together with IGF2BP2 may be promising therapeutic targets in ALI/ARDS.
Materials and methods Animals Male specific-pathogen-free C57/BL6 mice (8–10 weeks old) were obtained from the Guangdong Medical Laboratory Animal Center (Guangzhou, China). All mice were housed under controlled, pathogen-free conditions at the Laboratory Animal Center of Sun Yat-sen University Cancer Center. All animals were housed in separate cages in a temperature- (24 ± 1 °C) and humidity-controlled (50–60%) room under a 12/12-h light/dark cycle. All experiments were conducted in accordance with the guidelines defined by the Sun Yat-sen University Cancer Center. The study was approved by the Animal Care and Ethics Committee of Sun Yat-sen University Cancer Center (Permit Number: 2021-000043). Animal models and treatments A mouse LPS-induced ALI model was established as previously described [ 47 ]. Briefly, mice from the ALI groups were treated with a single intraperitoneal dose of 15 mg/kg LPS obtained from Escherichia coli 055: B5 (Sigma-Aldrich, St. Louis, MO, USA) in saline, whereas mice injected with the same volume of saline served as controls. After 24 h, the mice were killed and the lung lobes were harvested for further analysis. This in vivo study was performed via six series of experiments. Mice in the first series were randomly divided into control and ALI groups. Mice in the second series were randomly assigned to receive the following treatments: (1) control + si-NC, (2) control + si-METTL14, (3) control + MCC950, (4) ALI + si-NC, (5) ALI + si-METTL14 and (6) ALI + MCC950 (an NLRP3 inflammasome inhibitor, 50 mg/kg, i.p., Selleck, Shanghai, China). Mice in the third series were grouped as follows: (1) control + si-NC, (2) control + si-IGF2BP2, (3) ALI + si-NC, or (4) ALI + si-IGF2BP2. Each group received a dose of 20 nmol siRNA (either si-NC, si-METTL14 or si-IGF2BP2) in 200 μl of saline via the tail vein 2 d before being challenged with LPS or saline. Mice in the fourth series were grouped as follows: (1) control + AAV-GFP, (2) control + AAV-METTL14, (3) ALI + AAV-GFP, or (4) ALI + AAV-METTL14. Each group received a dose of 50 μl viral solution (either AAV-GFP or AAV-METTL14 with 10 12 vg/ml titer) via intranasal instillation 4 weeks before being challenged with LPS or saline. Mice in the fifth series were grouped as follows: (1) ALI + AAV-GFP + DMSO, (2) ALI + AAV-METTL14 + DMSO, (3) ALI + AAV-GFP + MCC950, or (4) ALI + AAV-METTL14 + MCC950. Mice in the sixth series were grouped as follows: (1) ALI + AAV-GFP, (2) ALI + AAV-METTL14, (3) ALI + AAV-METTL14 + si-NC, or (4) ALI + AAV-METTL14 + si-IGF2BP2. The treatments of each group were performed according to the mentioned above. Histopathological analysis Left lung lobes were fixed in 4% paraformaldehyde for 48 h, dehydrated, embedded in paraffin, and sliced into 5-μm-thick sections. The sections were then stained with hematoxylin and eosin (H&E) according to the manufacturer’s instructions, to evaluate lung histopathology. The damage to the lung tissues was scored using a previously described semiquantitative scoring system [ 47 ]. Images were captured using a microscope (NIKON Eclipse Ni-U; NIKON, Tokyo, Japan). Lung wet/dry ratio Any blood present on the isolated right lungs was blotted with filter paper before the weights of these lungs were recorded as wet weight. Then, the lungs were then stored in an incubator at 60 °C for 48 h, following which the weight of the lungs was recorded as dry weight. The lung wet/dry (W/D) ratio was used to evaluate the degree of pulmonary edema. Bronchoalveolar lavage At the time of lavage, the mice were anesthetized with an i.p. injection of 1% pentobarbital sodium (50 mg/kg). The chest cavity was exposed, then the trachea was intubated, and a whole lung lavage was performed by employed sterile PBS (1 mL). The collected lavage fluid was centrifuged at 1000× g for 10 min at 4 °C, and the cell-free supernatants were harvested and stored at −80 ◦ C for further analysis. The total protein concentration of bronchoalveolar lavage fluid (BALF) was measured using the BCA Protein Assay Kit (Thermo Fisher Scientific). Cell culture and treatments RAW264.7 cells, a mouse macrophage cell line, was obtained from ATCC and cultured in Dulbecco’s modified Eagle’s medium (Gibco from Thermo Fisher Scientific, Waltham, MA, USA) with 10% fetal bovine serum (Gibco) in an incubator at 37 °C and 5% CO 2 . To establish an NLRP3 inflammasome activation model in vitro, RAW264.7 cells were stimulated with LPS (1 μg/mL) for 6 h, then treated with nigericin (10 μM, InvivoGen, San Diego, CA, USA) for 30 min. For transient transfection purpose, cells were seeded at 30–40% confluence and cultured overnight, following which si-METTL14, si-IGF2BP2, and negative control (si-NC) were diluted in Opti-MEM ® medium (Thermo Fisher Scientific, Waltham, MA, USA) and transfected using Lipofectamine 3000 transfection reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. After 48 h of transfection, the cells were treated with or without LPS (1 μg/mL) for 6 h. Three siRNA sequences targeting METTL14 and IGF2BP2 were designed and synthesized by RiboBio (Guangzhou, China) and these were listed in Supplemental Table 1 . Specific siRNA with the best knockdown effect was used for further research and in vivo assays. ELISA analysis and myeloperoxidase activity Mice were anesthetized with an intraperitoneal injection of 1% pentobarbital sodium (50 mg/kg). Blood samples were collected from the retro-orbital sinus after the mice lost consciousness. Subsequently, blood samples were allowed to clot by leaving them undisturbed at 25 °C for 30 min. The clots were then removed to obtain serum via centrifugation at 1000 × g for 10 min at 4 °C. Part of the right lung from each mouse was homogenized with ELISA buffer and centrifuged to obtain lung tissue supernatants. Samples of murine serum, lung tissue supernatants, and cell culture supernatants were used to quantify the concentrations of IL-1β (R&D System, Minneapolis, MN, USA) and IL-18 (R&D System) by using murine ELISA kits, according to the manufacturer’s instructions. The MPO activity in the lung tissue was assessed by an MPO assay kit (R&D System). RNA m 6 A dot blot assay Total RNA and poly-A RNA were isolated from the lung tissue or RAW264.7 macrophages using a RNeasy mini kit (Qiagen, Düsseldorf, Germany) and a Dynabeads ® mRNA purification kit (Ambion, Austin, TX, USA), according to the manufacturer’s instructions. RNA was quantified using a Nanodrop, and equal amounts of RNA were crosslinked onto Hybond-N discs (Cytiva, USA) using a UV crosslinker (Spectroliner, Long Island, NY, USA). The membrane was quickly washed and blocked using 5% nonfat dry milk in 0.1% phosphate-buffered saline with Tween-20 (PBST) supplemented with RNaseOUT (Thermo Fisher Scientific). The membrane was incubated overnight at 4 °C with rabbit anti-m 6 A antibody (1:500, cat#A-1802-100, EpiGentek, Farmingdale, NY, USA), followed by incubation with horseradish peroxidase (HRP)-conjugated secondary anti-rabbit antibody. Membranes were washed and visualized using an enhanced chemiluminescence detection system. Images were acquired using a ChemiDocTM Touch Imaging System (Bio-Rad, Berkeley, CA, USA). Finally, the membranes were stained with methylene blue as a loading control. The signal intensity of the dot blot was analyzed using ImageJ software (NIH, Bethesda, MD, USA). RNA m 6 A modification quantification The levels of m 6 A in lung tissues and RAW264.7 macrophages were quantified using an EpiQuik m 6 A RNA Methylation Quantification Kit (EpiGentek), according to the manufacturer’s recommendations. Western blotting Mouse lung tissues or cells were lysed using RIPA lysis buffer (Beyotime, Shanghai, China) containing a protein inhibitor cocktail (Roche, Mannheim, Baden Württemberg, Germany). The total protein concentration was quantified using a BCA kit (Thermo Fisher Scientific). Samples were denatured at 100 °C for 10 min and separated on 10–12% SDS-PAGE gels with a molecular weight standard. Proteins from SDS-PAGE gel were transferred onto PVDF membranes (Merck Millipore, Darmstadt, Germany), blocked using 5% non-fat milk for 2 h, and incubated at 4 °C with the following primary antibodies overnight: METTL3 (1:1000, 15073-1-AP, Proteintech, Wuhan, China), METTL14 (1:1000, A8530, Abclonal, Boston, MA, USA), METTL16 (1:1000, 17676 S, Cell Signaling Technology, Danvers, MA, USA), WTAP (1:1000, 60188-1-Ig, Proteintech), FTO (1:1000, 45980 S, Cell Signaling Technology), ALKBH5 (1:1000, 16837-1-AP, Proteintech), NLRP3 (1:1000, AG-20B-0006-C100, AdipoGen, San Diego, CA, USA), Caspase-1 (1:1000, AG-20B-0042-C100, AdipoGen), IL-1β (1:500, AF-401-NA, R&D System), IGF2BP1 (1:1000, A1517, Abclonal), IGF2BP2 (1:1000, 11601-1-AP, Proteintech), IGF2BP3 (1:1000, A23295, Abclonal). After three washes, the membranes were incubated with the corresponding HRP-conjugated secondary antibody (1:1000, Abcam, Cambridge, UK) at room temperature for 1 h. Protein bands were detected using ECL and visualized using a ChemiDocTM Touch Imaging System (Bio-Rad). The band intensities were analyzed by using ImageJ software. Immunofluorescence Left lung lobes were fixed with 4% paraformaldehyde for 48 h, dehydrated, embedded in paraffin, and sectioned into 5-μm slices. After deparaffinated, dehydration, and antigen recovery, sections were incubated in blocking solution (Beyotime) for 1 h at room temperature and then incubated with primary antibodies overnight at 4 °C, followed by incubation at 25 °C for 1 h with fluorescently labeled secondary antibodies. Nuclei were stained for 10 min with DAPI. Images from six representative non-overlapping high-power fields (HPFs) of individual mice were taken on a fluorescent microscope (Leica, Wetzlar, Germany) in a blinded manner. The following antibodies were used: anti-METTL14 (1:200, A8530, Abclonal), anti-CD68 (18985-1-AP, 1:100, Proteintech), anti-F4/80 (18985-1-AP, 1:100, Proteintech), anti-Siglec F (18985-1-AP, 1:100, Proteintech), and anti-CK7 (16001- 1-AP, 1:100; Proteintech). Quantitative real-time RT-PCR Total RNA was extracted using TRIzol reagent (Invitrogen), and subsequently reverse-transcribed to cDNA according to the instructions of manufacturer of the qPCR transcription kit (EZ Bioscience, Roseville, NM, USA). Quantitative PCR was performed using SYBR Green Mix (EZ Bioscience) and a CFX96 Real-Time PCR Detection System (Bio-Rad). Target mRNA expression was calculated via the 2 -ΔΔCt method using GAPDH as an endogenous control. Primer sequences are listed in Supplemental Table 2 . RNA stability assay RAW264.7 macrophages were cultured in six-well culture plates until they reached 80% confluence. Actinomycin D (Abmole, Houston, TX, USA) was added at a final concentration of 5 μg/ml. Cells were collected at 0, 0.5, 1, 2, 4, and 6 h. Total RNA was extracted, RT–qPCR was performed as described above, and GAPDH was used as the loading control for normalization. Then, the RNA half-life was calculated. RNA immunoprecipitation assay RIP was performed according to the manufacturer’s instructions (Merck Millipore). Briefly, lung tissues of equal weight were mechanically sheared into a single-cell suspension using a homogenizer and resuspended in RIP lysis buffer containing protease inhibitor and RNase inhibitors. The mixture was centrifuged at 14 000 rpm at 4 °C for 10 min to obtain the supernatant, which was then divided into three fractions: Input, IP, and IgG. Each fraction was incubated overnight with the corresponding primary antibody at 4 °C, followed by protein A/G bead incubation at room temperature for 30 min. After six washes, the beads were incubated with 150 μl proteinase K buffer at 55 °C for 30 min. Total RNA was extracted, and the expression of related genes was detected via RT-qPCR. m 6 A RNA Immunoprecipitation assay The MeRIP assay was performed using a Magna MeRIP m 6 A Kit (Merck Millipore). Briefly, total RNA was extracted from lung tissues and RAW264.7 macrophages using TRIzol reagent. Approximately 20 μg of purified RNA was incubated with RNA fragmentation buffer. Then, 1 μg of the fragmented mRNA was used as input, while the remaining RNA was incubated overnight with anti-m 6 A antibody (Synaptic Systems, Gottingen, Germany) or anti-IgG antibody in 500 μl of IP buffer at 4 °C. The following procedure was performed in the same manner as that described for the RIP assay. RNA pull-down assay An RNA pull-down assay was performed using a Magnetic RNA-Protein Pull-down Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. The 3’-end Biotin-TEG modified-DNA probes against NLRP3 were synthesized by RiboBio. The biotinylated NLRP3 probe (50 pmol) was incubated with streptavidin beads to generate probe-coated beads. Lung tissue homogenates with probe-coated beads was incubated at 4 °C overnight. After three washes, proteins bound to the beads was boiled and used for the immunoblotting. Statistical analysis All sample size information is shown (figure legends). Statistical analyses were conducted using GraphPad Prism 8.0 (GraphPad software Inc, San Diego, CA, USA). Quantitative data were assessed for normality test and presented as the means ± SD. Differences between two normally distributed groups were analyzed using a two-tailed unpaired Student’s t test. Multiple comparisons of parametric data were performed using one-way ANOVA. Exact P values are indicated in all figures, and statistical significance was set at P < 0.05.
Results Global m 6 A levels and METTL14 expression are increased in ALI mice To confirm the role of m 6 A modification in ALI/ARDS, we evaluated the global m 6 A levels in the lung tissues of control and ALI mice. Both dot blot assay and colorimetric quantification showed that global m 6 A levels in the total RNA were significantly increased in ALI lungs compared to the control group (Fig. 1A–C ). We then detected the expression of m 6 A methyltransferases (METTL3, METTL14, METTL16, and WTAP) and demethylases (FTO and ALKBH5) in lung tissues. The expression of METTL14 mRNA and METTL14 protein was markedly upregulated in ALI mice, whereas no significant differences were found in the expression of other regulators (Fig. 1D–F ). These results indicated that METTL14-mediated m 6 A methylation may play a functional role in ALI/ARDS. Subsequently, we employed immunofluorescence staining to identify the specific cell types involved in ALI/ARDS that express METTL14. Our findings revealed co-localization of METTL14 not only with CD68-labeled macrophages but also with CK7-labeled pulmonary epithelial cells. Interestingly, compared with sham group, ALI lungs exhibited an increased number of METTL14-expressed CD68 + macrophages, while the number of METTL14-expressed CK7 + epithelial cells did not show a significant change (Figs. 1G, H and S1A, B ). To clarify the origins of elevated METTL14 + macrophages, quantitative analysis unveiled that METTL14 + / F4/80 + , rather than METTL14 + /Siglec-F + (a marker for resident alveolar macrophage) cells increased in ALI mice compared with the corresponding sham mice (Fig. 1I–L ). Collectively, these findings indicated that the expression level of METTL14 is elevated in LPS-induced ALI model, particularly in recruited circulating monocyte-derived macrophages within the lung. Global m 6 A levels and METTL14 expression are increased in LPS-activated macrophages The NLRP3 inflammasome is a crucial factor in triggering the activation of macrophages during this pathogenesis [ 22 ]. We subsequently assessed the expression level of METTL14 within a NLRP3 inflammasome activation model in RAW264.7 macrophages. As expected, treatment of RAW264.7 macrophages with LPS and nigericin significantly increased the release of NLRP3-inflammasome-dependent cytokines, including IL-1β p17, Caspase-1 p20, IL-1β and IL-18 (Fig. 2A, E, F ). The mRNA and protein levels of METTL14 and NLRP3 in RAW264.7 macrophages were observably enhanced following stimulation with LPS whether combined with nigericin or not, indicating METTL14 may participate in the priming step (Fig. 2A–D ). Both dot blot assay and colorimetric quantification showed that global m 6 A levels of total RNA in activated macrophages were notably elevated (Fig. 2G–I ). Our findings pointed towards a possible role for METTL14-mediated m 6 A methylation in the process of NLRP3 inflammasome activation. Knocking down METTL14 inhibits the activation of NLRP3 inflammasome and alleviates lung injury in vitro and in vivo To determine the function of METTL14, we knocked down METTL14 expression in RAW264.7 macrophages using small interfering RNA (siRNA). Western blotting, real-time PCR, and colorimetric m 6 A quantification were used to verify the knockdown effect (Fig. 3A–D ). Considering si-METTL14 #3 exhibited the superior knockdown efficiency, it was selected for further experiments. Compared with the negative control (si-NC) cells, knockdown of METTL14 exhibited a significant decrease in NLRP3 protein expression, as well as a reduced release of IL-1β and IL-18 cytokines in LPS-stimulated macrophages (Fig. 3E–H ). Interestingly, METTL14 knockdown only downregulated the mRNA expression of NLRP3, but not that of IL-1b or IL-18 in LPS-treated cells (Fig. 3I–K ). These results suggested that METTL14 may mediate NLRP3 inflammasome activation via regulating NLRP3. We next employed METTL14 siRNA to determine the in vivo function of METTL14 in ALI. Total m 6 A levels in ALI mice were reduced after METTL14 was knocked down (Fig. 3L ), suggesting that m 6 A modification occurring in ALI/ARDS was mainly mediated by METTL14. Both METTL14 siRNA and MCC950 (NLRP3 inhibitor) administration decreased the lung wet/dry ratio in ALI mice, indicating an alleviation of pulmonary edema associated with ALI/ARDS. (Fig. 3M ). Compared with ALI group, the total protein concentrations in BALF and myeloperoxidase (MPO) activity in lung tissues of si-METTL14 + ALI and MCC950 + ALI group were notably lower (Fig. 3N, O ). Similarly, H&E staining showed relatively intact alveolar structure and less inflammatory cell infiltration in si-METTL14 + ALI and MCC950 + ALI group than those in the ALI group (Fig. 3P, Q ). Consistent with the in vitro results, we found that METTL14 knockdown inhibited the activation of NLRP3 inflammasome in ALI mice via regulating the mRNA levels of NLRP3, rather than IL-1b and IL-18 (Figs. 3R–T and S2A–D). Collectively, these results suggested that METTL14 knockdown may inhibit NLRP3 inflammasome activation and alleviate lung injury in vitro and in vivo, confirming that METTL14 plays a vital role in NLRP3 inflammasome activation in ALI/ARDS. Elevated METTL14 promotes the activation of NLRP3 inflammasome and aggravates lung injury in vitro and in vivo To further elucidate the function of METTL14 in ALI, we performed gain-of-function assay by overexpressing METTL14 in RAW264.7 macrophages (Fig. 4A–C ). METTL14 overexpression in RAW264.7 macrophages increased m 6 A levels (Fig. 4D ) and activated NLRP3 inflammasome in macrophages by upregulating the mRNA levels of NLRP3, rather than IL-1b and IL-18 (Fig. 4E–K ). We subsequently explored the in vivo function of METTL14 in ALI using AAV9 that expressed full-length METTL14. AAV-GFP was used as a control. As shown in Fig. 4L , a marked increase in the level of m 6 A modification was detected in lung tissue of ALI mice. AVV-METTL14 aggravated pulmonary edema of ALI mice, as revealed by lung wet/dry ratio (Fig. 4M ). METTL14 overexpression also increased the total protein concentrations in BALF and MPO activity (Fig. 4N, O ). Likewise, H&E staining showed thicker alveolar walls and more inflammatory infiltration in AVV-METTL14 + ALI group than those in the ALI group (Fig. 4P, Q ). Indeed, METTL14 activated NLRP3 inflammasome via upregulating the mRNA levels of NLRP3 (Fig. 4R–V ). Taken together, these data supported that the elevation of METTL14 contributes to the activation of the NLRP3 inflammasome and the exacerbation of lung injury in vitro and in vivo. NLRP3 is the direct target of METTL14-mediated m 6 A modification NLRP3 is present in low concentrations under resting conditions, which is insufficient to activate the inflammasome [ 11 ]. Based on previous results showing that METTL14 regulated the mRNA expression of NLRP3 both in vivo and in vitro, we surmised that NLRP3 may be a direct target of METTL14. To validate the role of m 6 A methylation modulated by METTL14 in NLRP3 transcript, we analyzed potential m 6 A targeting motifs using SRAMP ( http://www.cuilab.cn/sramp ). A total of 24 RRACH m 6 A-binding motifs were identified in NLRP3 mRNA sequence (Fig. 5A and Supplemental Table 3 ). The m 6 A RNA immunoprecipitation (MeRIP) assays confirmed that NLRP3 mRNA m 6 A modification was enhanced in ALI mice and LPS-treated macrophages (Fig. 5B, D ), but significantly decreased after METTL14 knockdown (Fig. 5C, E ). Next, we used RNA pull-down and RNA immunoprecipitation (RIP) assays to determine whether there is a direct interaction between NLRP3 mRNA and METTL14. RNA pull-down assays showed that METTL14 interacts with the NLRP3 transcript, and that this interaction was enhanced in ALI mice (Fig. 5F ). RIP analysis with the METTL14 antibody further confirmed the interaction between METTL14 and NLRP3 mRNA both in vivo and in vitro (Fig. 5G, H ). Moreover, rescue assays were performed by using MCC950, an NLRP3 inhibitor, in AAV-METTL14 mice. Our data revealed that the extent of lung injury in AAV-METTL14 ALI mice was restored by MCC950 treatment, as revealed by lung wet/dry ratio, BALF protein content, MPO activity and histological injury score (Fig. 5I–M ). The over-release of IL-1β and IL-18 in lung tissues and serum in AAV-METTL14 ALI mice was reversed by MCC950 treatment (Fig. 5N–Q ). These findings indicated that NLRP3 is a direct and functional target of METTL14-mediated m 6 A modification during NLRP3 inflammasome activation in ALI/ARDS. IGF2BP2 is upregulated and enhances the stability of NLRP3 mRNA in ALI Considering that METTL14 induces NLRP3 mRNA m 6 A methylation and that the loss of m 6 A in NLRP3 mRNA mediated by METTL14 knockdown leads to a decrease in NLRP3 mRNA and protein expression in ALI mice, we sought to determine whether METTL14-mediated m 6 A modification affects the NLRP3 mRNA stability. We treated RAW264.7 macrophages with the transcription inhibitor actinomycin D (ActD) and found that NLRP3 decay in si-METTL14-treated macrophages was faster than that in corresponding controls when stimulated with LPS (Fig. 6A ), suggesting that METTL14 regulates NLRP3 expression via an m 6 A-dependent mRNA decay mechanism. Therefore, we identified m 6 A readers that may participate in the regulation of NLRP3 mRNA stability. The IGF2BP family regulates the stability of methylated mRNA by acting as m 6 A readers [ 23 ]. First, we detected the protein expression of IGF2BP1, IGF2BP2, and IGF2BP3 using western blotting. We found that protein expression of IGF2BP2 was distinctly upregulated in ALI mice (Fig. 6B, C ), which was consistent with its mRNA expression (Fig. 6D ). Similarly, the protein and mRNA expression levels of IGF2BP2 were notably augmented in LPS-treated RAW264.7 macrophages (Fig. 6E–G ). To further validate the direct interaction between NLRP3 mRNA and IGF2BP2, we performed an in vivo RNA precipitation assay using a biotinylated NLRP3 probe. RNA pull-down assay detected that specific binding of IGF2BP2 was enhanced in ALI mice (Fig. 6H ). RIP analysis with the IGF2BP2 antibody further confirmed that their interaction was facilitated in vivo and in vitro during ALI (Fig. 6I, J ). These results implied that the stability of NLRP3 mRNA might be regulated by IGF2BP2 via METTL14-mediated m 6 A modification. IGF2BP2 knockdown decreases NLRP3 mRNA stability and inhibits NLRP3 inflammasome activation in LPS-activated macrophages To examine whether IGF2BP2 regulates NLRP3 expression, we used siRNA to knockdown IGF2BP2 in RAW264.7 macrophages (Fig. 7A–C ), and si-IGF2BP2 #2 with the best knockdown effect was selected for further experiments. As shown, IGF2BP2 knockdown significantly downregulated the mRNA expression of NLRP3 (Fig. 7D ), but not IL-1b (Fig. 7E ) or IL-18 (Fig. 7F ), in LPS-treated RAW264.7 macrophages. To further determine the mechanism underlying IGF2BP2-induced regulation of NLRP3, we examined the effect of IGF2BP2 knockdown on the lifetime of NLRP3 mRNA. We found that the stability of NLRP3 mRNA in LPS-treated RAW264.7 macrophages was reduced by IGF2BP2 knockdown (Fig. 7G ). As expected, IGF2BP2 knockdown inhibited NLRP3 expression (Fig. 7H, I ) and the release of IL-1β and IL-18 (Fig. 7J, K ) in LPS-treated RAW264.7 macrophages. We proceeded to silence IGF2BP2 in METTL14-overexpressing cells. Our data showed that IGF2BP2 knockdown reduced NLRP3 mRNA lifespan (Fig. 7L ) and restored the over-release of IL-1β and IL-18 (Fig. 7M, N ) in METTL14-overexpressing macrophages. These results confirmed that IGF2BP2 participates in METTL14-mediated NLRP3 inflammasome activation by enhancing the stability of NLRP3 mRNA in macrophages. Knocking down IGF2BP2 inhibits the activation of NLRP3 inflammasome and alleviates lung injury in ALI mice We further determined the therapeutic potential of IGF2BP2 against mouse ALI models by applying siRNA to knock down IGF2BP2 in vivo. Compared to mice treated with control siRNA, the si-IGF2BP2 group showed significantly alleviated lung wet/dry ratio in ALI mice (Fig. 8A ). IGF2BP2 inhibition also decreased the total protein levels in BALF and MPO activity in ALI lung (Fig. 8B, C ). Similar effects of IGF2BP2 knockdown on alleviating lung injury in ALI mice were revealed H&E staining (Fig. 8D, E ). Disruption of IGF2BP2 downregulated the mRNA expression of NLRP3 in the lung tissues of ALI mice (Fig. 8F ). The dramatic increase in the IL-1β and IL-18 levels were efficiently diminished in ALI mice after treated with si-IGF2BP2 (Fig. 8G–J ). We further performed IGF2BP2 inhibition in AAV-METTL14 mice. The deterioration of lung wet/dry ratio, BALF protein content, MPO activity in AAV-METTL14 ALI mice were restored by IGF2BP2 knockdown (Fig. 8K-M ). IGF2BP2 inhibition also reduced the upregulation of IL-1β and IL-18 levels in lung tissues and serum (Fig. 8N–Q ) and NLRP3 mRNA (Fig. 8R ) in AAV-METTL14 ALI mice. These results manifested that IGF2BP2 knockdown may relieve ALI via inhibiting the NLRP3 inflammasome activation in vivo.
Discussion In this study, we discovered that the contents of m 6 A and METTL14 in lung tissues of ALI mice subjected to LPS were enhanced. METTL14-mediated NLRP3 mRNA m 6 A modification increases NLRP3 mRNA stability in injured lungs in an IGF2BP2-dependent manner. Thus, knocking down METTL14 or IGF2BP2 may play a protective role in ALI by inhibiting NLRP3 inflammasome activation. This finding may provide new pathophysiological insights into potential therapeutic strategies for ALI/ARDS. N 6 -Methyladenosine (m 6 A) is the most abundant epigenetic mRNA modification and exerts different biological effects in various diseases via post-transcriptional regulation [ 24 – 26 ]. Emerging evidence indicates that m 6 A may play an indispensable role in some inflammatory diseases [ 27 ]. RNA m 6 A modification is mediated by m 6 A writers (methyltransferases), erasers (demethylases), and readers. METTL14, a key component of the m 6 A methyltransferase complex, stabilizes the structure of METTL3 and enhances its enzymatic activity by binding to RNA, which ultimately increases m 6 A level [ 28 ]. Our study showed that m 6 A modification and the m 6 A methyltransferase METTL14 were increased in ALI mice. Further analysis confirmed METTL14 is mainly elevated in recruited circulating monocyte-derived macrophages of ALI mice. Interestingly, some studies have shown that neutrophil extracellular traps (NETs) induced ferroptosis in alveolar epithelial cells of cecal ligation and puncture (CLP)-mouse model by activating METTL3, while a few studies showed METTL3-mediated m 6 A modification alleviated ALI via inhibiting endothelial injury, indicating that m 6 A may exert different effects on ALI/ARDS owing to cell types and challenges [ 29 – 31 ]. Alveolar macrophages (AMs) consist of two subpopulations, including resident AMs and recruited AMs [ 32 ]. The resident AMs serve as an immunosuppressive subpopulation and mainly present the M2 phenotype, whereas the recruited AMs, which are derived from circulating monocytes, prefer to differentiate into pro-inflammatory M1 phenotype [ 33 , 34 ]. Consistent with the enrichment of METTL14 in recruited macrophages, we found the m 6 A levels and METTL14 expression were also increased in a RAW264.7 macrophage NLRP3 inflammasome activation model. This finding was in line with recent studies showing that METTL14 activated M1 polarization of macrophages in ischemic stroke and coronary heart disease, indicating that METTL14 may play a vital role in the functional regulation of macrophages [ 35 , 36 ]. Uncontrolled inflammatory responses mediated by pulmonary macrophages are indeed crucial in the pathogenesis of ALI/ARDS [ 37 ]. In the present study, we found that METTL14 knockdown alleviated lung injury via inhibiting NLRP3 inflammasome activation in macrophages, consistent with the result in sepsis-associated myocardial dysfunction [ 38 ]. The NLRP3 inflammasome, which acts as the core of the inflammatory response, mediates caspase-1 activation and the secretion of proinflammatory cytokines, IL-1β/IL-18 [ 39 ]. Enhanced activation of the NLRP3 inflammasome in alveolar macrophage is involved in the pathogenesis of ALI/ARDS caused by various pathogenic factors [ 40 , 41 ]. Inhibition NLRP3 inflammasome using the specific inhibitor MCC950 has achieved satisfactory therapeutic results not only in ALI model but also other inflammatory conditions including autoimmune diseases [ 42 ]. However, the liver toxicity of MCC950 was found in a phase II clinical trial for rheumatoid arthritis, which casts a shadow over the future clinical application of MCC950 [ 43 ]. In our study, we found that the protective effects of METTL14 knockdown were similar to MCC950 in ALI. Therefore, it is promising to develop a specific inhibitor of METTL14 for treatment on ALI/ARDS and other inflammatory diseases. Although some studies have revealed an association between METTL14 and NLRP3 inflammasome activation [ 44 , 45 ], whether METTL14 plays a direct role in regulating NLRP3 expression remains unclear. NLRP3 in low concentrations is inadequate for initiating inflammasome activation under resting conditions [ 9 ]. Our results revealed that METTL14 knockdown markedly downregulated the mRNA expression of NLRP3, but not that of IL-1b or IL-18, both in vivo and in vitro. Therefore, we suspected that NLRP3 mRNA may be the m 6 A methylation target of METTL14 in ALI/ARDS. Hence, we analyzed potential m 6 A targeting motifs in SRAMP and identified 24 RRACH m 6 A-binding motifs in NLRP3 mRNA. We further confirmed that the loss of METTL14 abolished the increase in m 6 A methylation levels of NLRP3 mRNA in ALI mice and RAW264.7 macrophages treated with LPS. RNA pull-down and RIP assays confirmed that METTL14 directly interacted with NLRP3 mRNA, and that such binding was enhanced in ALI mice. These results indicated that NLRP3 mRNA is a direct m 6 A methylation target of METTL14 during NLRP3 inflammasome activation in ALI/ARDS. The biological functions of m 6 A modifications rely on m 6 A readers, which regulate RNA metabolism, including translation, splicing, export, and degradation [ 16 ]. Elevated m 6 A modification mediated by METTL14 increases NLRP3 mRNA and protein expression in ALI mice. Furthermore, an ActD RNA stability assay showed that the half-life of NLRP3 transcripts had decreased following METTL14 knockdown, indicating that NLRP3 expression was modulated via an m 6 A-dependent mRNA decay mechanism. The m 6 A readers, IGF2BP1/2/3, are involved in regulating the stability of methylated mRNA [ 46 ]. Based on our data indicating that only IGF2BP2 was upregulated in ALI mice and LPS-treated RAW264.7 macrophages, we hypothesized that IGF2BP2 may act as the potential binding protein of NLRP3 mRNA via an m 6 A-dependent mRNA decay mechanism. Indeed, RNA pull-down and RIP assays confirmed that IGF2BP2 directly binds to NLRP3 transcripts. Moreover, our findings suggested that IGF2BP2 knockdown may decrease the NLRP3 mRNA stability and inhibit NLRP3 inflammasome activation in ALI mice and LPS-treated RAW264.7 macrophages, thereby alleviating lung injury. Collectively, these results suggest that IGF2BP2 specifically binds to the NLRP3 transcripts and enhances NLRP3 mRNA stability in an m 6 A-dependent manner during ALI/ARDS. This study has some limitations. We investigated the role of METTL14/IGF2BP2 in NLRP3 inflammasome activation in ALI mice and RAW264.7 macrophages. However, the role of METTL14/IGF2BP2 in clinical patients of ARDS remains to be elucidated. In addition, although we verified that m 6 A modification of NLRP3 mRNA was mediated by METTL14, the specific motif of NLRP3 transcripts methylated by METTL14 has not yet been elucidated and may have to be confirmed in future research. Third, although we found that METTL14 and IGF2BP2 were upregulated in ALI mice and RAW264.7 macrophages, upstream mechanisms underlying this process have not yet been explored and require further investigation via a follow-up study. Overall, our study provides robust in vitro and in vivo evidence supporting the critical roles of METTL14/IGF2BP2 in NLRP3 inflammasome activation during ALI/ARDS. Mechanistically, METTL14-catalyzed NLRP3 mRNA m 6 A methylation enhances the stability of NLRP3 mRNA in an IGF2BP2-dependent manner in ALI/ARDS. Our findings indicate that METTL14/IGF2BP2 shows potential as therapeutic targets in the treatment of ALI/ARDS.
Acute lung injury (ALI) as well as its more severe form, acute respiratory distress syndrome (ARDS), frequently leads to an uncontrolled inflammatory response. N 6 -methyladenosine (m 6 A) modification was associated with the progression of several inflammatory diseases. However, the role of methyltransferase-like 14 (METTL14)-mediated m 6 A methylation in ALI/ARDS remains unclear. Here, we reported an increase in overall expression levels of m 6 A and METTL14 in circulating monocyte-derived macrophages recruited to the lung following ALI, which is correlated with the severity of lung injury. We further demonstrated the critical function of METTL14 in activating NOD-like receptor pyrin domain-containing protein 3 (NLRP3) inflammasome in vitro and in mouse models of ALI/ARDS, and validated NLRP3 as the downstream target of METTL14 by the m 6 A RNA immunoprecipitation (MeRIP) and RIP assays. Mechanistically, METTL14-methylated NLRP3 transcripts were subsequently recognized by insulin-like growth factor 2 mRNA-binding protein 2 (IGF2BP2), an m 6 A reader, which stabilized NLRP3 mRNA. Furthermore, we observed that IGF2BP2 knockdown diminished LPS-induced ALI in mice by downregulating NLRP3 expression. In summation, our study revealed that the molecular mechanism underlying the pathogenesis of ALI/ARDS involves METTL14-mediated activation of NLRP3 inflammasome in an IGF2BP2 dependent manner, thereby demonstrating the potential of METTL14 and IGF2BP2 as promising biomarkers and therapeutic targets for ALI/ARDS treatment. Subject terms
Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s41419-023-06407-6. Acknowledgements We thank our colleagues for technical help and stimulating discussion. This work was supported by grants from the National Natural Science Foundation of China (81870878, 8217102207, 82101348), Guangdong Basic and Applied Basic Research Foundation (2019B151502010, 2022B1515120026, and 2021A1515220117). Author contributions JX, WZ, and FC conceived and designed research; FC, GC, YX, and XW performed research; FC, YX, and JX performed writing, review, and revision of the paper; XT analyzed data; WZ, XS, and XY provided technical and material support. All authors read and approved the final paper. Data availability Data are available from the corresponding author on reasonable request. Competing interests The authors declare no competing interests. Ethical approval The animal study was approved by the Institutional Animal Care and Use Committee of Sun Yat-Sen University Cancer Center and carried out under the guidelines of the Guide for the Care and Use of Laboratory Animals of the China National Institutes of Health.
CC BY
no
2024-01-15 23:42:00
Cell Death Dis. 2024 Jan 13; 15(1):43
oa_package/52/7f/PMC10787837.tar.gz
PMC10787838
38219004
Introduction One of the most important evolutionary characteristics of human beings is their social nature. The ability to develop social bonds in the environment in which we live largely depends on the ability of individuals of the same species to recognize and accurately interpret the emotional state of other group members 1 , 2 . The ability to correctly identify emotions, as well as to demonstrate emotions through facial expressions, is crucial for communication 3 , 4 . Facial expressions are, therefore, a major contributor to social connections, as they not only express emotional states but also convey intentions 5 , 6 . This was demonstrated by Bradley et al. 7 , in which pictures of families and babies were rated as highly pleasant, and were associated with the activation of neural systems that lead to closeness and evoked greater activity of the zygomatic major muscle (related to a smile) compared to the activities evoked by neutral or unpleasant pictures. Likewise, social bonding stimuli, such as pictures of people interacting, are perceived as more pleasant (high valence) and more arousing (greater activation) than pictures depicting the same people not making direct social contact 8 . Pictures with social interaction cues were recently shown to promote larger EMG activation of the zygomatic major muscle in healthy individuals compared to that for pictures without social interaction. Pictures with social interaction cues also increased the expectation of closeness and reduced fear of rejection in these individuals 9 . People with depression have difficulties in social interaction processes 10 – 13 and have differentiated muscle activity in response to emotional stimuli. Evidence indicates that individuals with depressive symptoms express fewer facial expressions than those without these symptoms, except when the expression is related to sadness 13 , 14 . Low facial responsiveness occurs mainly in the activity of the zygomatic major muscle associated with happy facial expressions 4 , 15 – 18 . Low facial expressiveness, together with deficits in the recognition of facial emotions, can compromise interaction initiation and subsequent development 5 , 6 . Consequently, individuals with depression have fewer interactions and greater social rejection by non-depressed individuals. Thus, these individuals may assess themselves and be judged by others to be less socially competent 13 . Although individuals with depression show reduced zygomatic activity when viewing pleasant pictures in general 4 , 15 – 18 , studies have not yet investigated the differences in the EMG reactivity of the zygomatic major in these individuals with pleasant pictures in a context of social interaction. Therefore, we investigated whether the smiling facial expression of individuals with high levels of depressive symptoms differed from those of individuals without depressive symptoms when viewing pictures with and without social interaction clues, with the same valence and activation, when the only difference between pictures is the presence or absence of social interaction cues. We also evaluated whether, as observed in individuals without depression, social interaction cues were relevant enough to increase the emotional states of sociability and altruism in individuals with depression. We believe that this is the first study to compare the reactivity of the zygomatic major muscle, affiliative state and altruism behavior evoked by visual stimuli that portray the same people, in the same background scene and environment, with valence and activation pairing, where the only difference between the stimuli is the context of social bond (pictures with and without social interaction) in individuals with and without depressive symptoms. Therefore, we aimed to investigate whether social interaction pictures differently modulated the facial EMG reactivity of the major zygomatic muscle, affiliative state, and altruistic behavior between individuals with and without depression. We hypothesized that the group with depression would exhibit less zygomatic muscle reactivity and lower levels of closeness expectation and altruism when viewing affiliative pictures compared to those in the group without depression. We also hypothesized that the group with depression would not show altered fear of rejection, while the group without depression would show a lower level when viewing interaction pictures. Our general hypothesis was that individuals with depression are less sensitive to the social interaction cues displayed in pictures of social interaction stimuli (affiliative pictures) compared to non-social interaction (control pictures) and neutral image stimuli and, therefore, do not exhibit changes in the expression of smile, emotional state and altruism behavior compared to individuals without depression.
Methods Participants The cohort comprised 85 individuals (66 women; 77%) aged between 18 and 35 years (average = 23.9, SD = 3.7). The exclusion criteria were: a medical diagnosis of psychiatric diseases; history of peripheral or central facial paralysis; and use of continuous medications (except for contraceptives). The participants answered the Beck Depression Inventory II (BDI-II) 56 , and based on the final score, the participants were separated into groups without symptoms of depression (non-depressive group: n = 69) and with symptoms of depression (depressive group: n = 16). The study followed the recommendations of the Declaration of Helsinki and the experimental protocol was approved by the Research Ethics Committee of the Federal University of Ouro Preto – Brazil (CAAE 90012318.10000.5150). All participants provided written informed consent. Data were collected before the COVID-19 outbreak. Visual stimuli The participants were exposed to three blocks containing 28 pictures each (neutral, affiliative and control). The neutral block consisted of neutral pictures taken from the International Affective Picture System (IAPS) 25 (valence: mean = 4.96; SD = 0.49; activation: mean = 2.76; SD = 0.8). The pictures of the affiliative and control blocks were extracted from an image 8 and their valence and activation values were classified as described by Ref. 25 . These blocks depicted pairs composed of an adult and a child, or two children. For each image in the affiliative block depicting two individuals interacting socially, there was an image in the control block that portrayed the same pair of individuals, in the same scenario, nevertheless without direct social interaction between them. The affiliative and control blocks had approximately the same number of smiles (affiliative: 87.5% and control: 71.4%) and were paired by valence (affiliative: average = 7.17; standard deviation = 0.39; control: average = 7.02; standard deviation = 0.38; t = 1.39; p = 0.17) and activation (affiliative: average = 3.69; standard deviation = 0.50; control: average = 3.92; standard deviation = 0.59; t = − 1.58; p = 0.41). The block order was fixed across participants to reduce the number of sequences and because we previously tested the block difference in a similar experimental design 9 . The pictures taken from the IAPS catalog were: 2880; 5510; 5520; 6150; 7000; 7002; 7004; 7006; 7009; 7010; 7025; 7050; 7080; 7130; 7090; 7150; 7170; 7175; 7207; 7211; 7217; 7233; 7235; 7490; 7550; 7595; 7705; and 7950. For examples of pictures affiliative and control block, see Refs. 8 and 9 . Psychometric assessments The participants completed the following questionnaires: BDI-II 56 , 57 : This 21-item self-report measure assesses depressive symptoms in the cognitive, affective, somatic, and motivational dimensions. The BDI-II scores range from 0 to 64, with the choice of cut-off point depending on the sample and study objectives 58 , 59 . In clinically undiagnosed samples the term “depression” should be used only for scores above 20, preferably with a concomitant clinical diagnosis 59 – 61 . Thus, in the present study, individuals with scores above 21 were included in the depressive group 59 , 60 . Affiliative status scale 62 . This scale comprises 27 adjectives that are divided into two subscales: 13 adjectives form the expectation of approximation subscale, which assesses the individual’s motivational state in performing social interactions, and 14 adjectives form the fear of rejection subscale, which predicts the motivation to be kept away from contact with others. The scores on the two subscales range from 13 to 52 points for the expectation of closeness and 13 to 52 points for the fear of rejection. Altruistic behavior scale 63 . This scale includes 16 items, 8 of which assess the individual’s altruistic behavior towards a friend and eight which assess behavior towards a stranger. The score for each subscale ranges from 0 to 32 points. The final score is the sum of the two subscales, defined as the total altruism score, where the higher the index, the higher the individual’s level of altruism. Display of visual stimuli and signal processing Visual stimuli were displayed on a 23′′ S23C550H Samsung TV monitor placed 94 cm in front of the participants. All pictures were viewed full-screen. E-Prime version 2.0 (Professional Psychology Software Tools Inc., Pittsburgh, PA), was used to generate the visual stimuli shown on the screen and to generate the pulses related to the beginning of these stimuli. The triggers generated by E-Prime were sent by parallel cable to the EMG signal acquisition system. A BIOPAC MP100 amplifier (Biopac Systems, Inc., Goleta, CA; https://www.biopac.com/ ) was used, with a sampling rate of 1000 Hz and a gain of 1000 for the EMG100C electromyographic module. The EMG signal was filtered online using a 10-Hz high-pass filter and a 500-Hz low-pass filter. The amplifier was connected directly to Acknowledge software ( https://www.biopac.com/ ) to acquire the raw electromyographic signals in real time. The raw signals were pre-processed off-line in MATLAB version 7.0 (MATricesLABoratory). We applied a 2th-order Butterworth filter with a cutoff frequency of 20 Hz followed by a constant detrending to remove artifacts in the EMG signal. After we used the full-wave rectified signal. Spreadsheets with the signs of each volunteer were prepared in Microsoft Excel 2016. The total window analysis of zygomaticus major muscle was defined from − 2 to 6 s. Mean amplitude of activation of EMG muscle activity in the interval between − 2 s and time zero (beginning of the stimulus) was used as the baseline and was applied for baseline correction. Mean amplitude of activation of EMG muscle activity in the interval between 0 and 6 s (picture visualization) in bins of 500 ms was used as the picture windows analysis, totaling 12 intervals. Processing was performed as described by Refs. 9 , 20 , 21 . Experimental procedure All procedures were performed in a room with controlled lighting, acoustic isolation, and controlled temperature (21° C). The participants remained seated in a comfortable armchair positioned in front of a table upon which was placed a computer monitor. To preserve privacy and avoid discomfort when completing the scales, the participants remained alone in the room while completing the questionnaires and viewing the pictures. The participants started the experiment by completing the BDI-II 56 . Afterward, each participant was asked to clean their face with running water and neutral soap and were advised to use the bathroom and drink water, if desired. Then, the area on which the electrodes were placed on the face of the participants was cleaned with 70% alcohol and a small exfoliation with a paper towel was performed. To collect the EMG signal, two 4-mm silver chloride (Ag–AgCl) electrodes were placed on the zygomatic major muscle on the left side of the face, as described by Ref. 64 . The participants received brief instructions on the dynamics of the experiment and were informed that all guidance regarding the procedures would appear on the computer screen. To ensure that the participants’ attention was maintained on the stimuli, the participants were instructed to keep their gaze fixed on the central point and carefully observe the pictures without looking away. Each block contained 28 pictures and each picture was displayed for 6 s, separated by a black screen, with a fixation point, which was displayed for a random duration (4–5 s). The experiment began with the presentation of the 28 pictures of the neutral block, followed by the blocks of interest; namely, the affiliate and control blocks. Between each block was an interval during which the participants responded to the mood state scales (affiliative state and altruistic behavior scales). After completing these scales, the display of pictures were continued, thus totaling 3 blocks and three applications of the mood scales. All three blocks had the same time settings. The duration of each of the three blocks varied between 280 s (4 min 40 s) and 308 s (5 min 8 s). At the end of the experimental session, the electrodes were removed and each participant was thanked. See Fig. 5 for the experimental sequence. Statistical analyses Statistical analyses were performed using the web-based development environment of SAS software (SAS Institute Inc., 2015), version 9.04, and the graphs were plotted using GraphPad Prism 6.0.1 (GraphPad Software, Inc.). For sample characterization, the two groups were compared using t-tests (age and symptoms of depression) and chi-squared test (gender). The EMG data and the Mood State Scale data were analyzed using repeated measures models (ANOVA) and residual dependence was modeled considering a compound symmetry structure, which is typical of repeated measures data analysis. In the case of the Mood State Scale data, a model was used for each scale (expectation of approximation, fear of rejection, altruistic behavior to friends, altruistic behavior to strangers) considering group (depressive and non-depressive) as the between-subjects factor, and block (neutral, affiliative and control) as the within-subjects factor. For the EMG data, a model was used considering group (depressive and non-depressive) as the between-subjects factor, and block (neutral, affiliative and control) and time (12 bins of 0.5 s) as the within-subjects factor. In this repeated measures model, interactions among these three factors were considered, and, since the number of times within a block was relatively high (12), it was found more appropriate to model residual dependence by means of an autoregressive structure of order 1. Under this structure, the residual covariance between the two time points tends to be lower as the time points are farther apart. This and other structures of residual dependence are easily handled in the 'Mixed' procedure of the SAS software. Moreover, for EMG data, the Box-Cox transformation 65 was applied, using the boxcox function of the MASS package 66 of the R software. For a given statistical model, the Box-Cox method estimates the best power transformation such that the distribution of residuals approaches a normal distribution.In both repeated measures models (for Mood State Scale and EMG data), whenever a pertinent null hypothesis was rejected (of an interaction, or of a single factor that does not interact with other factors), a post hoc multiple comparison test was carried out using the Tukey–Kramer method of adjustment. This method controls the familywise error rate, considering all pairs of means to be compared as a family of comparisons 67 . The level of significance adopted in this study was α = 0.05.
Results Cohort characterization The participants were divided into the non-depressive (n = 69; 55 women; mean age = 24.29; SD = 3.4) and depressive (n = 16; 11 women; mean age = 25.6; SD = 4.08) groups. The groups were similar in age (t = 0.69; p = 0.48) and sex (p = 0.11). The depressive group showed higher levels of depression compared to the non-depressive group (p < 0.001). The values for the self-depreciation, affection and cognition, and somatic dimensions of the depression scale in the depressive group were significantly higher than those in the non-depressive group (Table 1 ). Social interaction pictures increase zygomatic major muscle activation in non-depressive but not in depressive individuals The ANOVA showed a main effect for pictures block (F(2, 345) = 20.83, p < 0.0001), and time (F(11, 1683) = 11.74, p < 0.0001), but not for group (F(1, 125) = 2.69, p = 0.10). It also showed an interaction between group and block (F(2, 337) = 6.74, p = 0.001), group and time (F(11, 1,694,345) = 1.81, p = 0.047), and block and time (F(22, 1618) = 2.98, p < 0.0001), but it did not show a triple interaction among group, block and time (F(22, 1614) = 1.10, p = 0.34). The Tukey–Kramer post hoc tests of the interaction between group and block showed that the non-depressive group had greater activation of the zygomatic major muscle during visualization of the affiliative block than the neutral (p < 0.0001) and control (p < 0.0001) blocks. In addition, zygomatic major muscle activation was greater in the non-depressive group than in the depressive group during visualization of the affiliative block (p = 0.02) (Fig. 1 ). With regard to the interaction between group and time, the post hoc tests showed greater activation of the zygomatic major muscle in the non-depressive group during the second time compared to the first (p = 0.02) and during the third time compared to the second (p < 0.001) (Fig. 2 ). Finally, in the interaction between block and time, the Tukey–Kramer tests showed that the affiliative block of pictures was associated with a greater activation of the zygomatic major muscle during the second time compared to the first (p = 0.049) and similarly when comparing the third time to the second (p = 0.002) (Fig. 3 ). Social interaction pictures reduce the fear of rejection in non-depressive individuals and increase it in depressive individuals The ANOVA with the scale of fear of rejection showed a main effect of group (F(1, 71) = 15.41, p = 0.0002) and block (F(2, 142) = 3.53, p = 0.0 3) and an interaction between group and block (F(2, 142) = 4.92, p = 0.009). The Tukey–Kramer post hoc tests showed that fear of rejection marginally reduced after the affiliative block compared to the neutral block in the non-depressive group (p = 0.06). The fear of rejection was significantly greater in the depressive group than in the non-depressive group after the affiliative (p = 0.002) and control (p = 0.0004) blocks (Fig. 4 ). The ANOVA with the scale of expectation of approximation showed a main effect of group (F(1, 71) = 9.87, p = 0.03), but no main effect of block (F(2, 142) = 2.68, p = 0.07) or interaction between group and block (F(2, 142) = 2.01, p = 0.14). The non-depressive group had more expectation of approximation than the depressive group (p = 0.03) independently of the block of pictures. The ANOVA with the scale of altruistic behavior to friends showed a main effect of group (F(1, 69) = 4.00, p = 0.049), but no main effect of block (F(2, 138) = 0.40, p = 0.67) or interaction between group and block (F(2, 138) = 0.06, p = 0.94). The non-depressive group had more altruistic behavior to friends than the depressive group (p = 0.49) independently of the block of pictures. The ANOVA with the scale of altruistic behavior to strangers showed no main effect of group (F(1, 69) = 1.90, p = 0.17) or block (F(2, 138) = 1.95, p = 0.15) nor for the interaction between group and block (F(2, 138) = 0.15, p = 0.86).
Discussion The present study evaluated the differences in EMG reactivity of the zygomatic major muscle and emotional states in depressive and non-depressive individuals upon exposure to visual stimuli that were neutral, with social interaction (affiliative), and without social interaction (control). An important control performed in the present study was that the blocks of pictures with and without social interaction were pairs of pictures showing the same people and background scene, with the same mean valence and activation. The only difference between the blocks was the context of social interaction. When viewing the affiliative pictures, the non-depressive individuals showed greater zygomatic EMG reactivity compared to the control and neutral pictures, thus demonstrating increased smile expression. In contrast, the depressive group showed no difference in EMG reactivity while viewing the three image blocks. In addition, greater sociability (less fear of rejection) was observed in the non-depressive group compared to those in the depressive group. These results suggested that individuals with depression have decreased smile and sociability to social interaction scenes compared to that in individuals without depression, which demonstrated their difficulty in reacting to pleasant stimuli with social content. Higher valence stimuli increase the EMG activity of the zygomatic major muscle, while lower valence stimuli show lower activation of this musculature 7 , 19 . The results observed in non-depressive individuals corroborate those reported previously and are consistent with those of a study showing increased zygomatic activity for social interaction pictures (preceded or not by a priming text) compared to control pictures also pared by valence and arousal 9 . Additionally, the temporal analysis during image visualization showed that the affiliative pictures promoted increased smile expression from 0.5 s onwards compared to the control pictures, a difference that persisted until the end of the image visualization. Thus, the social interaction clues not only promoted greater smile expression but also induced a sustained positive emotional mood 9 , 20 , 21 . Among the high-valence visual stimuli, those that evoked sociability, such as pictures of people interacting socially and pictures of babies, increased zygomatic EMG activity by activating brain circuits that “prepare” individuals for social interaction 22 – 25 . However, most studies on psychophysiological reactions using social interaction scenes are carried out in healthy individuals and do not present stimuli matched for valence and activation 4 , 26 . Therefore, when observing the increase in zygomatic EMG activity during exposure to pictures with scenes of social interaction (affiliative block) compared to pictures with the same pairs of people, but without social interaction (control block), the results of the present study suggested that the context of social interaction promoted a smile expression, which acted as a facilitator for social interactions in healthy people 9 , 27 . However, this effect did not occur in people with depression. Depressed individuals show less smile expression than control individuals 28 , 29 . A reduction in zygomatic EMG activity was also observed during the visualization of high-valence pictures depicting happy faces, indicating that social pictures of high pleasantness provoke negative responses in individuals with depression compared to those in control individuals 4 , 30 . A review of 39 studies showed alterations in emotional facial expressions across different mental disorders. The majority of studies point towards decreased facial emotional expressivity in individuals with depression, specifically, the decrease in facial expression is mainly evident for positive stimuli 31 . Our results corroborate these findings and additionally show that individuals with depression are not responsive to highly pleasant social stimulus. In the present study, the affiliative and control image sets had similar valence and arousal, number of people and background scene; the only difference between the blocks was the social content. Thus, affiliative pictures promote changes in relevant facial expressions that drive social interactions 9 but not in individuals with depressive symptoms. Facial expressions have an important influence in social contexts, especially the smile, which can evoke feelings of affiliation. Zygomatic activity has been proposed to act as a social facilitator 32 indicating a willingness to make connections 33 – 35 . We suggest that individuals with depression have low responsiveness to the social context, as reflected by apathy in their facial expressions. Therefore, this factor may be a determinant for poorer social interactions and avoidance of contact with other people among individuals with depression 13 . Studies on the social interactions have reported sadness or anguish in the facial expressions of depressed individuals, including low zygomatic activity to avoid arouse sympathy, help, and proximity to those around them 36 . However, other studies emphasize that the interpersonal behaviors of individuals with depression lead to their rejection by the people with whom they live. The behaviors of people with depression are categorized as aversive, which, in turn, reinforces the cycle of loneliness and further accentuates the depression. While some studies claim that people with depression have low accuracy in correctly identifying happy faces, others 1 , 4 , 28 , 36 observed that depressed individuals could differentiate pictures with happy expressions from pictures from other categories (for example, pictures with sad expressions), with similar precision to non-depressed individuals. However, the picture blocks in the present study were matched for valence and activation and had similar proportions of smiles (affiliative block 87.5%, control block: 71.4%); the only differences between them were the subtle cues of social interaction. Thus, we supposed that for individuals with depression, social interaction is not a pleasant enough stimulus, resulting in the lack of changes in smile expression. Another component that could influence the facial expressiveness in depressed individuals is facial mimicry. Mimicry facilitates the creation and maintenance of sociability and plays an important role in social interactions seeing that it creates empathy, linking and affiliation between people 37 , 38 . Hess and Fisher suggested that mimicry functions as a social regulator 39 , 40 ; thus, when mimicry is hindered or disturbed, it can impair emotion recognition 41 and lead to elevated stress reactions in the interaction partner 42 . In a review, Kampf 43 showed that mimicry is decreased in depressive states 44 – 46 , whereas mimicry is increased in positive mood states 46 . Research has shown that patients with depression show less mimicry of pictures of happy and sad faces compared to the non-clinical control group 47 , and symptom severity of depression is associated with fewer affiliative and higher non-affiliative facial expressions 29 . Moreover, acutely depressed patients compared to remitted patients and non-clinical participants showed less mimicry of happy faces and were less accurate in recognizing happy faces, yet reduced mimicry did not mediate these deficits 14 . This suggests that people with affective disorders might show less mimicry behavior during depressive episodes, which may in turn influence their social relationships 39 . After exposure to the affiliative block of pictures, the non-depressive group showed less fear of rejection compared to neutral block. No modulation was observed in the depressive group. Therefore, in the non-depressive group, the affiliative pictures motivated social interactions, increasing states of sociability (reducing the fear of rejection scores), an effect not observed in the depressive group. These results corroborate those described in the literature, which showed a reduced fear of rejection after viewing affiliative pictures in non-depressive individuals 9 , 23 . It is known that people with depression often have the expectation that they might be rejected or that it is too exhausting to engage with others 48 , 49 , which may also lead to less mimicry behavior and less states of sociability. However, to the best of our knowledge, this result has not been demonstrated in people with high levels of depressive symptoms during the visualization of positive social stimuli. Our results can also be explained by the “social risk hypothesis” of depression 29 , 50 . This hypothesis proposes that people with high levels of depressive symptoms tend to distance themselves from other people to protect themselves from rejection, contempt, and social exclusion. This happens through the signaling of submission in socially competitive environments, in addition to distancing in exchange-oriented social contexts, where requests for help could be ignored or ridiculed 28 . As depressive symptoms disappear, individuals send more signals that indicate their willingness to interact socially 50 . Our results are consistent with this hypothesis, in which the affiliative pictures did not reduce the fear of social exclusion, in contrast to the observations in the non-depressive group. Furthermore, according to the social risk hypothesis, help-seeking behavior is reserved for contexts oriented towards reciprocity with friends and family, who are more likely to provide the requested help 29 , 50 . Since our stimuli comprised of pairs of pictures depicting individuals unknown to the volunteers, we can assume that this may also have contributed to the non-modulation of affiliative behaviors in individuals with depression, as they tend to direct their affiliative behaviors to known people in whom the risk of exclusion is minimized 29 , 50 . Adaptations in future experiments should be considered to assess the responsiveness of individuals with depression exposed to visual stimuli from people in their social circle, as described by Ref. 51 in a non-depressed sample. Altruistic behaviors are defined as voluntary actions performed without interest in receiving internal or external rewards and intended to improve the well-being of others 52 , 53 . We expected that non-depressive group would increase the altruistic behavior after visualizing the affiliative block. But in the non-depressive and depressive group, no changes were observed in altruistic behavior towards friends or strangers. The present study has some limitations. First, our sample, despite high scores on the BDI-II, is not a clinical sample and the participants were not diagnosed with depression based on the diagnostic and statistical manual of mental disorders, Fifth Edition (DSM-5) by a psychiatry 54 . Therefore, future studies in a clinical sample should be considered. Secondly, although the groups had the same internal proportions of men and women, the sample mostly comprised women. Additional studies with balanced sex distributions in the sample overall are needed. Thirdly, our groups are unbalanced. We have much more non-depressive participants. Fourthly, our visual stimuli had a low ethnic diversity. As the feeling of belonging to a group favors facial mimicry 55 , visual stimuli that cover all ethnic diversity are needed to increase participant sense of belonging in the experimental context. We concluded that individuals with high levels of depressive symptoms are not susceptible to emotional modulation promoted by social interaction cues; that is, they are not able to promote the sustained expression of a smile and to increase their feelings of sociability. Through self-report and facial EMG measurements, we showed that affiliative pictures increased the somatic responses of smile expression and reduced the fear of being rejected in individuals without depressions, in contrast to individuals with high levels of depressive symptoms, who did not present these responses.
Individuals with severe depressive symptoms present diminished facial expressions compared to healthy individuals. This reduced facial expression, which occurs in most depressive patients could impair social relationships. The current study sought to investigate whether pictures with social interaction cues could elicit different modulations of facial expressions and mood states in individuals with depressive symptoms compared to healthy individuals. A total of 85 individuals were divided into depressive and non-depressive groups based on their beck depression inventory scores. Participants viewed pictures containing neutral (objects), affiliative (people interacting socially), and control (people not interacting) scenes. Electromyographic signals were collected during the entire period of visualization of the blocks, and emotional questionnaires were evaluated after each block to assess sociability and altruism (prosocial states). In non-depressed individuals, affiliative pictures increased the activity of the zygomatic muscle compared to both neutral and control pictures and reduced fear of rejection compared to neutral pictures. During the visualization of the affiliative block, zygomatic major muscle activation was higher and fear of rejection was lower in the non-depressive individuals than in the depressive. These effects reflected the low expressions of smiling and sociability to affiliative pictures in depressive individuals. These findings highlight the importance of smiling and prosocial states in social interactions, especially in these individuals. Subject terms
Acknowledgements This work was supported by the National Council for Scientific and Technological Development, Brazil ( Conselho Nacional de Desenvolvimento Científico e Tecnológico —CNPq); Coordination for the Improvement of Higher Education Personnel, Brazil ( Coordenação de Aperfeiçoamento de Pessoal de Nível Superior —CAPES); Research Foundation of the State of Minas Gerais ( Fundação de Amparo à Pesquisa do Estado de Minas Gerais —FAPEMIG); and the Federal University of Ouro Preto—Brazil ( Universidade Federal de Ouro Preto —UFOP). Author contributions K.C.D.L. Developed the study, conducted data acquisition, wrote the manuscript draft, analyzed and interpreted the data and prepared the figures. F.C.O.S. Conducted data acquisition and critically reviewed the final manuscript. C.R.A. and B.E.F.M. Contributed with data analysis and interpretation and critically reviewed the final manuscript. P.M.G, W.B. and L. V. critically reviewed the final manuscript. E.B analyzed the data and critically reviewed the final manuscript. G.G.L.S. Developed the study concept and study design, and supervised and administered the project. Contributed with data analysis and critically reviewed the final manuscript. All authors contributed to and have approved the final manuscript. All authors have approved the final version of the manuscript. Data availability The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1266
oa_package/d2/d2/PMC10787838.tar.gz
PMC10787839
38218918
Introduction Bacterial biofilms pose a formidable global health challenge, exhibiting resilience against antibiotics, phagocytosis, and environmental stressors, perpetuating chronic infections, particularly by Staphylococcal and Pseudomonas strains 1 . Biofilm formation entails an intricate, multistep process where bacteria adhere to surfaces, leading to heightened antibiotic resistance. The ensuing microbial consortium releases extracellular polysaccharides (EPS), evading immune surveillance and instigating pathogenesis 2 – 4 . Recent research reveals a profound connection between biofilms and various cancers, going beyond conventional links to colorectal and gastric cancers. In vivo, malignant cells rely on a complex tumor microenvironment (TME) for crucial support, where communication within the TME leads to significant cellular and genetic diversity. This dynamic environment triggers changes in the extracellular matrix (ECM), epithelial-to-mesenchymal transition (EMT), and immune modulation, creating a microenvironment conducive to cancer progression 5 . Concurrently, cancer cells and biofilms undergo metabolic adaptations, notably the Warburg effect, optimizing nutrient distribution for ATP generation and heightened glutamine levels essential for growth. Biofilms manifest analogous metabolic strategies, efficiently metabolizing glucose for sustained energy in oxygen-limited conditions 6 , 7 . The coming together of these common metabolic processes presents a crucial challenge with significant implications for treatment approaches. Biofilms impact cancer through various ways, including triggering inflammation, altering immune responses, producing carcinogenic toxins, and changing host metabolism. The developing concept of the tumor microbiome within the tumor microenvironment closely links with cancer progression, highlighting the need to understand bacterial interactions in cancer and uncover the mechanisms of biofilms for improved diagnostic accuracy and innovative therapeutic strategies 8 . Current scientific investigations seek innovative approaches for precise therapeutic agent delivery, enhancing bacterial mortality and impeding cancer cell viability. Employing nanoparticles is a key method to target specific tissues, such as tumor tissues and biofilms, due to their small size and unique properties. This enables direct drug delivery to biofilm-infected tumor sites, overcoming a major challenge in biofilm treatment—limited drug penetration. Nanotechnology involves engineering atoms and molecules to synthesize nanoparticles (NPs), solid particles typically 1 to 100 nm in diameter and 1 to 1000 nm in length. These NPs, with amorphous or crystalline structures, serve as versatile carriers for liquids or gases, bridging bulk materials and molecular structures 9 , 10 . The nanoparticle size varies based on synthesis methods: chemical, physical, and biological. It is crucial to note that chemical and physical methods may pose environmental and organismal risks due to potential toxicity. Hence, the adoption of a green synthesis approach, utilizing biological components like plant extracts, bacteria, yeast, algae, and fungi, is increasingly favored. Biological elements, particularly alkaloids and flavonoids, play pivotal roles in reducing NPs during green synthesis, rendering it more environmentally friendly than chemical and physical methods 11 , 12 . NPs fall into two main categories: monometallic nanoparticles (MNPs) with a single metal and bimetallic nanoparticles (BNPs) with two different metals. Recent research highlights the superior attributes of BNPs, driven by their increased surface area, making them particularly significant. BNPs leverage the distinct properties of both metals, resulting in unique combined attributes and diverse applications 13 – 16 . Silver (Ag) and copper (Cu) NPs, synthesized from biological sources like plants, fungi, bacteria, and algae, exhibit antibacterial, anticancer, anti-biofilm, and antioxidant activities. They emerge as promising candidates in various fields, especially biomedicine, demonstrating potential as carriers for diagnostics, hydrophobic medicines, hyperthermia, and cancer therapeutics. The combination of Ag and Cu into bimetallic NPs has piqued interest for its antimicrobial efficacy against various pathogens and preferential toxicity to cancer cells, positioning it as a valuable tool in combatting multidrug resistance and advancing cancer treatment 17 – 20 . This study marks the initial step in elucidating a sustainable methodology for synthesizing Ag–Cu bimetallic nanoparticles (NPs) using A. lanata plant extract. The NPs undergo comprehensive characterization, encompassing optical, morphological, and elemental analyses. Of specific interest is the assessment of their biological functionalities, namely antibacterial and antioxidant activities, with a targeted focus on elucidating their pronounced antibiofilm and cytotoxic activities.
Materials and methods Chemicals and materials Silver nitrate and copper sulfate (analytical grade) were obtained from Sigma Aldrich Bangalore, India. All bacteriological media components such as crystal violet, tryptone, yeast extract, glucose, and sodium chloride, and media such as nutrient both were procured from Hi-Media Laboratories Pvt. Ltd., Mumbai, India. The HeLa & HEK293 cell lines utilized in the experiment was acquired from American Type Culture Collection (ATTC, LG Promochem, Barce-lona, Spain). Preparation of Aerva lanata plant extract The Aerva lanata plant was sourced from the I-AIM herbal garden and nursery in Bengaluru, India, following the relevant guidelines and regulations set by CHRIST (Deemed to be University). To ensure the reproducibility of the study, the voucher representing the specimen has been deposited within the herbarium at the Department of Life Sciences, Christ (Deemed to be University), and is assigned the identification number CULS_AS_001 . The collected plant material, representing the entire plant, was washed with double distilled deionized water and subsequently shade-dried for a period of 5 days at room temperature. The dried plants were then finely chopped and coarsely powdered. To prepare the extract, 5 g of the whole plant powder was boiled with 100 mL of double distilled deionized water at 70 °C for 20 min. This solution was further filtered using Whatman No.1 filter paper and centrifuged at 6000 rpm for 5 min and the supernatant was stored at 4 °C for experimental use. Synthesis of bimetallic Ag–Cu NP’s Silver nitrate and copper sulphate were used as precursor salts for the biosynthesis of bimetallic Ag–Cu NPs. The synthesis process was performed by mixing a stock solution of both the precursor salts (AgNO 3 and CuSO 4 ) in an equal volume of 60 mL with the defined molar concentration of 0.01 M and boiled 10 min at 80 °C. About 30 mL of plant extract was added. With constant mixing on magnetic stirrer at 40 °C for 1 h. Variation in color was observed from green to dark brown and then finally to black precipitate of bimetallic Ag–Cu NPs. These NPs were collected by centrifuging at 10,000 rpm and the pellet was washed thoroughly with double-distilled deionized water, followed by the final rinse with ethanol. Nanoparticles were dried in a hot air oven at 80 °C and stored at 4 °C for further use 21 . Characterization Various analytical tools were employed to investigate the key properties of bimetallic NPs synthesized using A. lanata . Figure 1 depicts that the schematic study plan and objective. To determine the crystal structure, the Ag–Cu NPs were analyzed using Bragg–Brentano geometry with a fine focus of cu-anode working at 40 kV and 30 mA, utilizing a Shimadzu model XRD 6000 X-ray diffractometer. The spectrum of the bimetallic Ag–Cu NPs was recorded in the 2θ range, with the detector positioned from 10 to 80 degrees in a step scan mode of 0.02°. Fourier-transform infrared spectroscopy (FT-IR) was employed to study the functional groups present in the green-synthesized Ag–Cu NPs. The KBr pellet method was adopted for sample preparation to avoid spectrum adsorption in the IR region and enable the distinction of various functional groups. The FTIR spectrum was measured in the wave number range of 4000–400 cm –1 . The surface morphology and elemental composition of the synthesized NPs were examined using Field Emission-Scanning Electron Microscope (FE-SEM, model JEOL JSM-6390) and EDX (EDX Oxford Instrument, INCA PentaFETX3). To prevent agglomeration, the Ag–Cu NPs were suspended in a suitable solvent and subjected to sonication. A silicon wafer substrate was then immersed in the colloidal suspension and covered with a spin-coating process, ensuring secure attachment of the particles for SEM and EDX analysis. High Resolution—Transmission Electron Microscope (HR-TEM) analysis was conducted using a Philips TEM (CM200; Eindhoven, The Netherlands) operating at a potential of 120 kV to determine the average particle size and diffraction pattern. The NPs were sonicated and diluted, and the diluted sample was placed on a copper grid and dried at room temperature, HR-TEM images were captured at different magnification powers. Determination of minimum inhibitory and minimum bactericidal concentration (MIC and MBC) The MIC and MBC of the Ag–Cu NPs were tested against Staphylococcus aureus and Pseudomonas aeruginosa , and results were recorded as per Clinical and Laboratory Standards Institute guidelines. The assay was performed by diluting the concentration of NPs with sterile LB broth to finally obtain series of concentration (15, 30, 60, 120, 240 μg mL −1 ) inoculated with S. aureus and P. aeruginosa adjusted to the turbidity of 5 × 10 5 CFU/mL. These incubated broths were incubated on shaker incubator maintained at 37 °C for 24 h. The concentration at which there was no cell density or turbidity was recorded as MIC 22 . Following the Ag–Cu NPs MIC measurement, 30 μL aliquots from the tubes that were plated on MH agar plates and incubated overnight at 37 °C overnight. The lowest concentration of NPs that reported bacterial killing was reported to be the MBC value. Well diffusion assay Agar well diffusion technique was adopted to study the anti-microbial activity of the Ag–Cu NPs. This assay was performed with two bacterial species S. aureus and P. aeruginosa . A revived mid log phase bacterial cultures were spread plated on Muller Hinton agar. Followed by punching of the wells in the solidified media using gel puncher that resulted in wells with a diameter of 5 mm. Under aseptic conditions the wells were loaded with different concentrations of Ag–Cu NPs from 15, 30, 60, 120 μg mL −1 and positive control ampicillin. The sample loaded culture plate was then incubated at 37 °C for overnight to observe and record zone of clearance in the plates and the diameter of the zones was measured 23 . Effect of Ag–Cu NPs on biofilm formation Biofilm formation assay The test organism in LB broth was inoculated and allowed to grow until reaching a constant turbidity of 0.5 McFarland standards (5 × 10 5 CFU/mL). Different concentrations (15, 30, 60, 120, 240 μg mL −1 ) of Ag–Cu NPs were added to separate tubes containing the bacterial suspension and incubated at 37 °C for 24 h. A negative control without the addition of NPs was also maintained. After incubation, the media was removed, and the tubes were washed three times with phosphate buffer saline (PBS, pH 7.2). Subsequently, the tubes were stained with 0.1% crystal violet dye for 30 min. Excess stain was washed off with deionized water, and the tubes were dried at room temperature. The biofilm formation was assessed by observing the presence of a thin blue film on the walls of the tubes 24 . To quantitatively estimate the effectiveness of Ag–Cu NPs in inhibiting biofilm formation, a method proposed by Kalishwaralal et al. was employed using a 96-well microtiter plate 25 . Briefly, 10 μL of overnight grown and diluted (1:100) culture was added to sterile wells containing 180 μL of LB broth. Ag–Cu NPs were then added at concentrations of 15, 30, 60, 120, 240 μg mL −1 . After incubation at 37 °C for 24 h, the LB broth was discarded, and the wells were washed three times with phosphate buffer saline (PBS, pH 7.2) to remove any unbound cells. The biofilm adhered to the well walls was fixed with 2% w/v sodium acetate and stained with 0.1% crystal violet dye for 10 min. The stained cells were washed with sterile double distilled deionized water, and the crystal violet stain on the surface of the biofilm was solubilized by adding 30% acetic acid and incubating at room temperature for 10–15 min. The solubilized crystal violet was transferred to a new 96-well plate, and the absorbance at 570 nm (OD 570) was measured using an ELISA plate reader. The percentage of biofilm inhibition was calculated using the provided formula: Effect of Ag–Cu NPs on the production of extracellular polymeric substance (EPS) The inhibitory influence of Ag–Cu NPs on the production of EPS was quantitatively analyzed slightly modified assay 26 . An overnight culture of P. aeruginosa and S. aureus aseptically inoculated in 50 mL sterile LB broth and each of these cultures were subjected to different sub-inhibitory MIC concentrations of NPs (15, 30, 60, 120, 240 μg mL −1 ) and one without NPs was maintained as negative control. All the tubes were incubated at 37 °C for 24 h in a shaker incubator at 140 rpm . After incubation period, the control and the treated samples were centrifuged at 6000 rpm for 15 min at 4 °C and the supernatant was collected. Supernatant obtained was treated with double the volumes of ice-cold ethanol followed by refrigeration at 4 °C to facilitate the precipitation of EPS. This precipitation was separated and concentrated by centrifugation at 6000 rpm for 20 min at 4 °C. The obtained pellet was used for the quantification by measuring carbohydrates utilizing anthrone method 27 . The percentage of EPS inhibition was calculated using Eq. ( 1 ). Analysis of antioxidant activity of the synthesized bimetallic Ag–Cu NPs Reducing power assay The reducing power of the NPs was determined according to the method of Alavi and Karimi with some modifications 28 . The concentrations of the bimetallic NPs from 15, 30, 60, 120, 240 μg mL −1 were prepared using 0.2 M of phosphate buffer of pH 6.4. The aliquots were mixed with 2.5 mL of 10 mL –1 concentration of potassium ferricyanide and incubated at 50 °C for 30 min. After the incubation, 2.5 mL of 100 mL -1 concentration of trichloro acetic acid was added and centrifuged at 10,000 rpm for 12 min. The supernatant was mixed with equal volume of distilled water (3 mL), 0.6 mL of 1.0 mL concentration of ferric chloride was added. With BHT as control setup, the solution was read for absorbance at 700 nm. The results indicate that more the absorbance, the higher is its reducing power. Hydrogen peroxide scavenging assay The hydrogen peroxide scavenging assay was performed by modifying the methods of Vilas et al. 29 , with ascorbic acid as control, various concentration of the bimetallic NPs (15, 30, 60, 120, 240 μg mL −1 ) was aliquoted and added with 100 μL of 3 mM hydrogen peroxide solution. The solution was incubated at room temperature for 30 min and the absorbance was measured at 610 nm. The percentage of scavenging was calculated using Eq. ( 1 ). Cytotoxicity assay MTT assay for testing cell viability The effect of biogenic Ag–Cu NPs on HeLa and HEK293 cell lines was investigated using the MTT assay (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide) at various concentrations. Briefly, the metabolic rate and viability of the treated cells were assessed by measuring the reduction of the yellow tetrazolium salt MTT to a blue formazan product by mitochondrial dehydrogenase. The cells were seeded in a 96-well plate at a uniform density of 1 × 10 4 cells/well and allowed to adhere and proliferate overnight in a 5% CO 2 incubator. After a 24 h incubation period, the monolayer was washed with fresh medium, and the adhered cells were exposed to Ag–Cu NPs at concentrations ranging from 0 to 240 μg mL –1 for an additional 24 h. To monitor cell viability, 10 μL of MTT prepared with PBS was added to each well. The plates were then incubated in darkness at 37 °C in a humidified atmosphere containing 5% CO 2 for 4 h. The supernatant was mixed with 100 μL of DMSO to solubilize the formazan product. The absorbance of the plates was measured at 550 nm 30 . A blank well containing medium without bimetallic nanoparticles at the corresponding concentrations was included. The percentage of inhibition was calculated using a specific formula, and the concentration of nanoparticles required to inhibit cell growth by 50% (IC 50 ) values was determined from the dose–response curves for each cell line using GraphPad Prism 7.0. Statistical analysis The experiments were conducted in triplicate and analyzed using Student’s t test and two-way ANOVA, with statistical significance defined as p-values < 0.01, utilizing GraphPad Prism software (GraphPad Software Tools, Inc., La Jolla, CA, USA).
Results and discussions Synthesis of Ag–Cu NPs The plant chosen for this research belongs to the Amaranthaceae family and is found extensively in tropical and subtropical regions around the world. Due to its rich composition of phytochemicals, including alkaloids, terpenoids, sterols, flavonoid glycosides, and polyphenols, it holds significant importance in ayurveda and is utilized for various medicinal purposes such as antimicrobial, antiasthmatic, anti-urolithiasis and anti-hyperglycemic 31 , 32 . In recent studies, numerous investigations have confirmed the viability of utilizing A. lanata plant extract for the synthesis of different metal nanoparticles. The presence of phytochemicals in the plant extract acts as an essential reducing or capping agent, facilitating the synthesis of metallic nanoparticles 33 – 36 . Upon combining the plant extract with the precursor salt, the color of the solution undergoes a discernible change from green to black after an hour of incubation. This color transformation serves as confirmation of the reduction process facilitated by the plant's metabolites. Eventually, the resulting precipitate settles down undisturbed, indicating the successful reduction and formation of bimetallic Ag–Cu NPs. Characterization of bimetallic Ag–Cu NPs X-ray diffraction The phase—purity and the crystallinity of the prepared bimetallic Ag–Cu NPs were studied using XRD technique. The XRD pattern of the prepared sample is depicted in Fig. 2 . It is observed that, XRD pattern exhibits sharp, well defined high peaks. In detail, the peaks observed at 38.3°, 44.1° and 64.1° in 2θ are assigned to the (111), (200) and (220) planes which corresponds to the JCPDS card no. 04 - 0783 belongs to the face centered cubic (fcc) crystal structures of Ag 37 . The crystalline peaks observed at 43.0° and 50.1° are attributed to the (111) and (200) planes of fcc cubic structured Cu NPs corresponds to JCPDS card no. 04 - 0836 38 . In addition to the Ag and Cu crystalline peaks, certain unassigned peaks (marks with *) are also observed. The existence of these peaks could be due to the crystalline organic matter induced during the synthesis via plant extract. These findings show that Cu–Ag NPs produced using A. lanata extract results in the nano crystalline particles. The average crystallite size was estimated using Debye – Scherrer equation; D = kλ/βcosθ; where β is the full width at half maximum and λ is the x-ray wavelength (1.5418 Å) and k is the constant. All the peaks obtained in the study were used to estimate the size crystallites and the average size of the crystallite found be 14.6 nm that is in correspondence to previously reported Cu–Ag NPs 39 . FT-IR The FT-IR study was conducted to analyze the functional group of the tethered capping biomolecule that aided in the synthesis and stabilization of the bimetallic Ag–Cu NPs. Both the extract and the NPs were subjected to IR radiation within the range from 400 to 4000 cm –1 . FT-IR spectrum of A. lanata extract exhibited absorption bands at 3345, 1639, 1224, 1033, 925, 844, 702 cm –1 which corresponds to stretching vibration of medium N–H stretching (aliphatic primary amine), bending of –OH, C=O stretching of strong alkyl aryl ether, C–N variable stretching, O–H stretching and N–H wag 39 . For the spectrum of Ag–Cu/Al, the characteristic peaks of the A. lanata extract were found together with a new medium peak of 615 cm –1 correlating the C–H bond indicating nanoparticles of Ag–Cu and validating that Ag–Cu NPs were capped with phytoconstituents of A. lanata . Furthermore, when the A. lanata extract and Ag–Cu NPs were compared (Fig. 3 ), the peaks of hydroxyl, carbonyl, and amino groups were found to slightly shifted in the NPs, confirming the interaction and participation of bioactive molecules in the production of Ag–Cu NPs. These findings corroborated the utilization of bioactive A. lanata components on the surface of Ag–Cu NPs as capping, reducing, and stabilizing agents 11 , 40 . Field emission—scanning electron microscope (FE-SEM) and EDX The SEM images, presented in Fig. 4 , depict the bimetallic Ag–Cu NPs that were synthesized using a green method. The images reveal that these NPs have a varied size distribution and form semi-spherical agglomerated clusters with a combination of different shapes. Additionally, a significant portion of the NPs exhibit rough surfaces. The EDAX spectra were used to perform a semi-quantitative elemental mapping of the bimetallic Ag–Cu NPs. The results from EDAX confirm the presence of both Ag and Cu particles, indicating their bimetallic nature 41 . High resolution—transmission electron microscope (HR-TEM) The HR-TEM technique was employed to analyze the size distribution and morphology of the bimetallic NPs. The HR-TEM images, along with the corresponding size distribution histogram, are presented in Fig. 5 . Upon examining the HR-TEM images (Fig. 5 a,b), it can be observed that the particles are uniformly dispersed and exhibit a wide size distribution. The particle sizes were measured using Image J software, and the corresponding size distribution histogram is shown in Fig. 5 c. The histogram reveals that the particle sizes range from 5 to 30 nm, with the majority falling within the 7 nm to 12 nm range, this was further confirmed by dynamic light scattering analysis Fig. 5 d. The average size of the particles is approximately 9.5 nm, which aligns well with the crystallite size estimated from XRD analysis. It is worth noting that the bimetallic Ag–Cu NPs synthesized through the green method exhibit a spherical shape. The presence of larger particles may be attributed to the coagulation or overlapping of smaller particles, a phenomenon that has been reported in similar studies conducted by Al Tamimi, S. et al. 42 . Determination of minimum inhibitory concentration The MIC and MBC of the biosynthesized NPs against the desired organisms was evaluated. It was shown that the MIC and MBC of Ag–Cu NPs against P. aeruginosa and S. aureus was 240 and 120 μg mL −1 respectively. The findings demonstrate that the effectiveness of Ag–Cu NPs was more potent for S. aureus than P. aeruginosa . This is due to the presence of positive charged silver in the bimetallic NPs that trap and block the lipopolysaccharides core component of bacterial cell wall of Gram-negative bacteria P. aeruginosa , finally them susceptible for bimetallic NPs. Along with this morphology and surface-to-mass ratio of NPs influence their reactivity against microbes 43 . It was also implied that, due to its biological and chemical reactive edges, uneven and irregular particles have various binding features with the microbial surface. The plant-based synthesis resulted in a wide range of particle sizes. The findings of this study are in moderate correlation with NPs XRD results. A prominent peak of 111 facets can be seen in the bimetallic NPs. The XRD pattern is closely related to TEM analysis, emphasizing that that majority of synthesized NPs are spherical shaped nanoparticles in the sample with high 111 facets, which enhanced their antibacterial activity 44 . Agar well diffusion method The anti-bacterial activity of the green synthesized Ag–Cu NPs was qualitatively evaluated using well diffusion method against Gram positive ( S. aureus ) and Gram negative ( P. aeruginosa ) organisms. The results of the antibacterial activity are presented in Fig. 6 A, Table 1 and Figs. S1 , S2 . Figure 6 A illustrates that the highest inhibitory activity of the Ag–Cu NPs was observed at a concentration of 120 μg mL –1 against both the organisms. When compared to the control antibiotic, Ampicillin, the activity of the NPs against P. aeruginosa was found to be more effective, while against S. aureus , it was recorded as moderate. In comparison to the antibacterial activity of the plant extract as reported by Al-Ansari et al., as well as the monometallic Ag and Cu NPs reported by Naidu et al. and Thanganadar Appapalam & Panchamoorthy et al. our bimetallic nanoparticles exhibited superior efficacy 45 – 47 . They demonstrated significantly larger zones of inhibition at lower concentrations. This enhanced activity attributed to its unique property of large surface area ratio (less particle size) of the NPs which enables accumulation and penetration of the NPs through the external barrier cell wall followed by the generation of reactive oxygen species (ROS) a stress marker which could damage the cell integrity by the disruption of the cell membrane leading to the leakage of membrane proteins and cell lysis 48 (Fig. 6 B). According to Slavin et al. the copper ions have more affinity towards the gram-positive organisms mainly because of projected functional groups such as carboxyl and amine functional groups on the cell surface 49 . Crystal violet blue assay The objective of this assay was to investigate the impact of Ag–Cu nanoparticles (NPs) on biofilm formation, both quantitatively and qualitatively, using S. aureus and P. aeruginosa as test microbes. Sub-inhibitory concentrations of Ag–Cu NPs were used, and the results demon-started their inhibitory effect on biofilm formation. A comparative study was conducted, showing that the presence of bimetallic NPs inhibited biofilm formation without affecting the control group, and this effect was concentration-dependent. Qualitative evaluation involved observing the presence of a thin layer of biofilms on the walls of the culture tubes after staining with crystal violet. The findings indicated that S. aureus did not exhibit visible biofilm formation in the culture tubes containing Ag–Cu NPs concentrations of 120 μg/mL and higher. Similarly, P. aeruginosa showed no visible biofilm formation at concentrations of 240 μg/mL (Table 2 ). The quantitative assessment of biofilm inhibition was conducted using a microtiter plate assay at concentrations ranging from 15 to 240 μg/mL, with the inhibition percentage re-ported in Fig. 7 . These results align with a study conducted by Ghosh, S. et al. which examined the effect of AucoreAgshell NPs synthesized using Dioscorea bulbifera plant extract on biofilm production by P. aeruginosa 50 . The Au core Ag shell NPs showed an efficiency of 18.93% inhibition of biofilms at a concentration of 100 μg/mL. Effect of Ag–Cu NPs on the production of extracellular polymeric substance (EPS) Carbohydrates are considered as the major constituents for EPS in a pure culture that can mediate the process of biofilm formation. Current study data obtained clearly suggests that the bimetallic Ag–Cu NPs have potent antibiofilm activity by disrupting the EPS matrix. These Ag–Cu NPs exhibited potent EPS degradation against both the bacterial strains in a dose de-pendent manner (15, 30, 60, 120, and 240 μg mL –1 ). The qualitative observation was done by observing the formation of precipitate after the incubation time of 48 h. At a higher concentration of 120 and 240 μg mL −1 Ag–CuNPs had manifested a prominent inhibitory effect on the EPS formation with S. aureus moderate effect observed with P. aeruginosa however the highest concentration of 240 μg mL −1 manifested the disruption of EPS against P. aeruginosa as well. Anthrone method was found to be a reliable and highly sensitive method to estimate EPS (carbohydrate) and to interpret the inhibitory effect of Ag–Cu NPs (Figs. 8 a,b). The ratio of exopolysaccharide was comparatively less after the exposure of the bacterial cultures with the sub-inhibitory concentrations of NPs. From Table 3 it was shown that the highest percentage reduction of EPS was observed against S. aureus and P. aeruginosa with respect to the concentration of 120 and 240 μg mL −1 which is considered to be crucial for biofilm formation. In contrast to our findings, Borcherding et al. reported that when P. aeruginosa culture was treated with super paramagnetic iron oxide nanoparticles it demonstrated a significant increase in biofilm biomass 51 . This was attributed to the nutritional role of the iron NPs which supports microbial growth leading to the development of increased microbial biofilm. This demands extensive research to understand the differential effects of various NPs towards cell density and biofilm. Assessment of Ag–Cu Nps cytotoxic activity The cytotoxicity of A. Lanata aqueous extract and Ag–Cu NPs was investigated on HeLa and HEK293 cell lines, with doxorubicin serving as the positive control, and untreated cells as the negative control, using the MTT assay. The resulting MTT assay results are graphically depicted in Fig. 9 . Markedly, the IC50 values of HeLa cells were 15.86 μg mL −1 for the extract and 17.63 μg mL −1 for Ag–Cu NPs, respectively. In stark contrast, significantly higher IC50 values of 35.89 μg mL −1 and 21.78 μg mL −1 were obtained from the treatment of normal human embryonic kidney (HEK293) cells with the respective materials. Meanwhile, similar in range IC50 values were observed among the different cell lines treated with Doxorubicin, with values of 12.52 and 16.490 μM mL −1 for HeLa and HEK293 cell lines, respectively. This is consistent with the microscopic examination data, through which the cytotoxic activity could be distinguished by features such as membrane blebbing, cell swelling, or shrinkage, as shown in Figs. S3 , S4 . This study strongly suggests that the synergistic effect arising from the combined presence of silver and copper in the nanoparticles significantly contributes to their efficacy. When comparing the response of HeLa and normal HEK 293 cells, it was observed that the bimetallic nature of the nanoparticles exhibited lower toxicity towards non-cancerous cells, particularly at a concentration of 120 μg mL −1 . This phenomenon can be attributed to the antioxidant properties of the NPs, resulting in higher toxicity towards cancerous cells and lower toxicity towards healthy cells, possibly through the expression of apoptotic molecules 52 , 53 . Further comprehensive investigations are warranted to gain a deeper understanding of the underlying mechanisms responsible for the anticancer activity of the synthesized bimetal NPs. Antioxidant activity Antioxidant activity is a complex process influenced by various mechanisms and factors, making it challenging to fully comprehend using a single methodology. To capture the diverse mechanisms of antioxidant action, it is necessary to employ multiple evaluation methods for assessing antioxidant capacity. In this study, two complementary tests, namely the reducing power assay and the hydroxyl radical scavenging assay, were utilized to evaluate the anti-oxidant activity of bimetallic Ag–Cu NPs. The reducing power assay measures antioxidant activity by reducing ferric cyanide complex (Fe 3+ ) to the ferrocyanide form (Fe 2+ ). The presence of antioxidative properties in the NPs enables them to donate an electron, resulting in a color change from green to blue, which indicates their scavenging ability 29 . The results of this study revealed that the NPs exhibited a reduction ability, as evidenced by the color change to Perl's Prussian blue observed at 700 nm. The scavenging activity was assessed in comparison to ascorbic acid, a well-known standard antioxidant. Notably, the nanoparticles exhibited the most substantial reduction in Fig. 10 A and B when employed at a concentration of 240 μg mL −1 . Importantly, it is worth highlighting that, in accordance with Al-Ansari et al. research, the plant extract exhibited 58.5% antioxidant activity at its highest concentration (100 μg), whereas the synthesized Ag–Cu nanoparticles achieved ~ 60 ± 5% of activity at a lower concentration (60 μg mL −1 ) 46 . These findings indicates that the nanoparticles are capable of converting the ferric cyanide complex (Fe 3+ ) into the ferrocyanide form (Fe 2+ ) and effectively neutralizing hydroxyl (OH–) free radicals 54 . Moreover, the effectiveness of this scavenging activity is influenced by the dosage of the nanoparticles administered.
Results and discussions Synthesis of Ag–Cu NPs The plant chosen for this research belongs to the Amaranthaceae family and is found extensively in tropical and subtropical regions around the world. Due to its rich composition of phytochemicals, including alkaloids, terpenoids, sterols, flavonoid glycosides, and polyphenols, it holds significant importance in ayurveda and is utilized for various medicinal purposes such as antimicrobial, antiasthmatic, anti-urolithiasis and anti-hyperglycemic 31 , 32 . In recent studies, numerous investigations have confirmed the viability of utilizing A. lanata plant extract for the synthesis of different metal nanoparticles. The presence of phytochemicals in the plant extract acts as an essential reducing or capping agent, facilitating the synthesis of metallic nanoparticles 33 – 36 . Upon combining the plant extract with the precursor salt, the color of the solution undergoes a discernible change from green to black after an hour of incubation. This color transformation serves as confirmation of the reduction process facilitated by the plant's metabolites. Eventually, the resulting precipitate settles down undisturbed, indicating the successful reduction and formation of bimetallic Ag–Cu NPs. Characterization of bimetallic Ag–Cu NPs X-ray diffraction The phase—purity and the crystallinity of the prepared bimetallic Ag–Cu NPs were studied using XRD technique. The XRD pattern of the prepared sample is depicted in Fig. 2 . It is observed that, XRD pattern exhibits sharp, well defined high peaks. In detail, the peaks observed at 38.3°, 44.1° and 64.1° in 2θ are assigned to the (111), (200) and (220) planes which corresponds to the JCPDS card no. 04 - 0783 belongs to the face centered cubic (fcc) crystal structures of Ag 37 . The crystalline peaks observed at 43.0° and 50.1° are attributed to the (111) and (200) planes of fcc cubic structured Cu NPs corresponds to JCPDS card no. 04 - 0836 38 . In addition to the Ag and Cu crystalline peaks, certain unassigned peaks (marks with *) are also observed. The existence of these peaks could be due to the crystalline organic matter induced during the synthesis via plant extract. These findings show that Cu–Ag NPs produced using A. lanata extract results in the nano crystalline particles. The average crystallite size was estimated using Debye – Scherrer equation; D = kλ/βcosθ; where β is the full width at half maximum and λ is the x-ray wavelength (1.5418 Å) and k is the constant. All the peaks obtained in the study were used to estimate the size crystallites and the average size of the crystallite found be 14.6 nm that is in correspondence to previously reported Cu–Ag NPs 39 . FT-IR The FT-IR study was conducted to analyze the functional group of the tethered capping biomolecule that aided in the synthesis and stabilization of the bimetallic Ag–Cu NPs. Both the extract and the NPs were subjected to IR radiation within the range from 400 to 4000 cm –1 . FT-IR spectrum of A. lanata extract exhibited absorption bands at 3345, 1639, 1224, 1033, 925, 844, 702 cm –1 which corresponds to stretching vibration of medium N–H stretching (aliphatic primary amine), bending of –OH, C=O stretching of strong alkyl aryl ether, C–N variable stretching, O–H stretching and N–H wag 39 . For the spectrum of Ag–Cu/Al, the characteristic peaks of the A. lanata extract were found together with a new medium peak of 615 cm –1 correlating the C–H bond indicating nanoparticles of Ag–Cu and validating that Ag–Cu NPs were capped with phytoconstituents of A. lanata . Furthermore, when the A. lanata extract and Ag–Cu NPs were compared (Fig. 3 ), the peaks of hydroxyl, carbonyl, and amino groups were found to slightly shifted in the NPs, confirming the interaction and participation of bioactive molecules in the production of Ag–Cu NPs. These findings corroborated the utilization of bioactive A. lanata components on the surface of Ag–Cu NPs as capping, reducing, and stabilizing agents 11 , 40 . Field emission—scanning electron microscope (FE-SEM) and EDX The SEM images, presented in Fig. 4 , depict the bimetallic Ag–Cu NPs that were synthesized using a green method. The images reveal that these NPs have a varied size distribution and form semi-spherical agglomerated clusters with a combination of different shapes. Additionally, a significant portion of the NPs exhibit rough surfaces. The EDAX spectra were used to perform a semi-quantitative elemental mapping of the bimetallic Ag–Cu NPs. The results from EDAX confirm the presence of both Ag and Cu particles, indicating their bimetallic nature 41 . High resolution—transmission electron microscope (HR-TEM) The HR-TEM technique was employed to analyze the size distribution and morphology of the bimetallic NPs. The HR-TEM images, along with the corresponding size distribution histogram, are presented in Fig. 5 . Upon examining the HR-TEM images (Fig. 5 a,b), it can be observed that the particles are uniformly dispersed and exhibit a wide size distribution. The particle sizes were measured using Image J software, and the corresponding size distribution histogram is shown in Fig. 5 c. The histogram reveals that the particle sizes range from 5 to 30 nm, with the majority falling within the 7 nm to 12 nm range, this was further confirmed by dynamic light scattering analysis Fig. 5 d. The average size of the particles is approximately 9.5 nm, which aligns well with the crystallite size estimated from XRD analysis. It is worth noting that the bimetallic Ag–Cu NPs synthesized through the green method exhibit a spherical shape. The presence of larger particles may be attributed to the coagulation or overlapping of smaller particles, a phenomenon that has been reported in similar studies conducted by Al Tamimi, S. et al. 42 . Determination of minimum inhibitory concentration The MIC and MBC of the biosynthesized NPs against the desired organisms was evaluated. It was shown that the MIC and MBC of Ag–Cu NPs against P. aeruginosa and S. aureus was 240 and 120 μg mL −1 respectively. The findings demonstrate that the effectiveness of Ag–Cu NPs was more potent for S. aureus than P. aeruginosa . This is due to the presence of positive charged silver in the bimetallic NPs that trap and block the lipopolysaccharides core component of bacterial cell wall of Gram-negative bacteria P. aeruginosa , finally them susceptible for bimetallic NPs. Along with this morphology and surface-to-mass ratio of NPs influence their reactivity against microbes 43 . It was also implied that, due to its biological and chemical reactive edges, uneven and irregular particles have various binding features with the microbial surface. The plant-based synthesis resulted in a wide range of particle sizes. The findings of this study are in moderate correlation with NPs XRD results. A prominent peak of 111 facets can be seen in the bimetallic NPs. The XRD pattern is closely related to TEM analysis, emphasizing that that majority of synthesized NPs are spherical shaped nanoparticles in the sample with high 111 facets, which enhanced their antibacterial activity 44 . Agar well diffusion method The anti-bacterial activity of the green synthesized Ag–Cu NPs was qualitatively evaluated using well diffusion method against Gram positive ( S. aureus ) and Gram negative ( P. aeruginosa ) organisms. The results of the antibacterial activity are presented in Fig. 6 A, Table 1 and Figs. S1 , S2 . Figure 6 A illustrates that the highest inhibitory activity of the Ag–Cu NPs was observed at a concentration of 120 μg mL –1 against both the organisms. When compared to the control antibiotic, Ampicillin, the activity of the NPs against P. aeruginosa was found to be more effective, while against S. aureus , it was recorded as moderate. In comparison to the antibacterial activity of the plant extract as reported by Al-Ansari et al., as well as the monometallic Ag and Cu NPs reported by Naidu et al. and Thanganadar Appapalam & Panchamoorthy et al. our bimetallic nanoparticles exhibited superior efficacy 45 – 47 . They demonstrated significantly larger zones of inhibition at lower concentrations. This enhanced activity attributed to its unique property of large surface area ratio (less particle size) of the NPs which enables accumulation and penetration of the NPs through the external barrier cell wall followed by the generation of reactive oxygen species (ROS) a stress marker which could damage the cell integrity by the disruption of the cell membrane leading to the leakage of membrane proteins and cell lysis 48 (Fig. 6 B). According to Slavin et al. the copper ions have more affinity towards the gram-positive organisms mainly because of projected functional groups such as carboxyl and amine functional groups on the cell surface 49 . Crystal violet blue assay The objective of this assay was to investigate the impact of Ag–Cu nanoparticles (NPs) on biofilm formation, both quantitatively and qualitatively, using S. aureus and P. aeruginosa as test microbes. Sub-inhibitory concentrations of Ag–Cu NPs were used, and the results demon-started their inhibitory effect on biofilm formation. A comparative study was conducted, showing that the presence of bimetallic NPs inhibited biofilm formation without affecting the control group, and this effect was concentration-dependent. Qualitative evaluation involved observing the presence of a thin layer of biofilms on the walls of the culture tubes after staining with crystal violet. The findings indicated that S. aureus did not exhibit visible biofilm formation in the culture tubes containing Ag–Cu NPs concentrations of 120 μg/mL and higher. Similarly, P. aeruginosa showed no visible biofilm formation at concentrations of 240 μg/mL (Table 2 ). The quantitative assessment of biofilm inhibition was conducted using a microtiter plate assay at concentrations ranging from 15 to 240 μg/mL, with the inhibition percentage re-ported in Fig. 7 . These results align with a study conducted by Ghosh, S. et al. which examined the effect of AucoreAgshell NPs synthesized using Dioscorea bulbifera plant extract on biofilm production by P. aeruginosa 50 . The Au core Ag shell NPs showed an efficiency of 18.93% inhibition of biofilms at a concentration of 100 μg/mL. Effect of Ag–Cu NPs on the production of extracellular polymeric substance (EPS) Carbohydrates are considered as the major constituents for EPS in a pure culture that can mediate the process of biofilm formation. Current study data obtained clearly suggests that the bimetallic Ag–Cu NPs have potent antibiofilm activity by disrupting the EPS matrix. These Ag–Cu NPs exhibited potent EPS degradation against both the bacterial strains in a dose de-pendent manner (15, 30, 60, 120, and 240 μg mL –1 ). The qualitative observation was done by observing the formation of precipitate after the incubation time of 48 h. At a higher concentration of 120 and 240 μg mL −1 Ag–CuNPs had manifested a prominent inhibitory effect on the EPS formation with S. aureus moderate effect observed with P. aeruginosa however the highest concentration of 240 μg mL −1 manifested the disruption of EPS against P. aeruginosa as well. Anthrone method was found to be a reliable and highly sensitive method to estimate EPS (carbohydrate) and to interpret the inhibitory effect of Ag–Cu NPs (Figs. 8 a,b). The ratio of exopolysaccharide was comparatively less after the exposure of the bacterial cultures with the sub-inhibitory concentrations of NPs. From Table 3 it was shown that the highest percentage reduction of EPS was observed against S. aureus and P. aeruginosa with respect to the concentration of 120 and 240 μg mL −1 which is considered to be crucial for biofilm formation. In contrast to our findings, Borcherding et al. reported that when P. aeruginosa culture was treated with super paramagnetic iron oxide nanoparticles it demonstrated a significant increase in biofilm biomass 51 . This was attributed to the nutritional role of the iron NPs which supports microbial growth leading to the development of increased microbial biofilm. This demands extensive research to understand the differential effects of various NPs towards cell density and biofilm. Assessment of Ag–Cu Nps cytotoxic activity The cytotoxicity of A. Lanata aqueous extract and Ag–Cu NPs was investigated on HeLa and HEK293 cell lines, with doxorubicin serving as the positive control, and untreated cells as the negative control, using the MTT assay. The resulting MTT assay results are graphically depicted in Fig. 9 . Markedly, the IC50 values of HeLa cells were 15.86 μg mL −1 for the extract and 17.63 μg mL −1 for Ag–Cu NPs, respectively. In stark contrast, significantly higher IC50 values of 35.89 μg mL −1 and 21.78 μg mL −1 were obtained from the treatment of normal human embryonic kidney (HEK293) cells with the respective materials. Meanwhile, similar in range IC50 values were observed among the different cell lines treated with Doxorubicin, with values of 12.52 and 16.490 μM mL −1 for HeLa and HEK293 cell lines, respectively. This is consistent with the microscopic examination data, through which the cytotoxic activity could be distinguished by features such as membrane blebbing, cell swelling, or shrinkage, as shown in Figs. S3 , S4 . This study strongly suggests that the synergistic effect arising from the combined presence of silver and copper in the nanoparticles significantly contributes to their efficacy. When comparing the response of HeLa and normal HEK 293 cells, it was observed that the bimetallic nature of the nanoparticles exhibited lower toxicity towards non-cancerous cells, particularly at a concentration of 120 μg mL −1 . This phenomenon can be attributed to the antioxidant properties of the NPs, resulting in higher toxicity towards cancerous cells and lower toxicity towards healthy cells, possibly through the expression of apoptotic molecules 52 , 53 . Further comprehensive investigations are warranted to gain a deeper understanding of the underlying mechanisms responsible for the anticancer activity of the synthesized bimetal NPs. Antioxidant activity Antioxidant activity is a complex process influenced by various mechanisms and factors, making it challenging to fully comprehend using a single methodology. To capture the diverse mechanisms of antioxidant action, it is necessary to employ multiple evaluation methods for assessing antioxidant capacity. In this study, two complementary tests, namely the reducing power assay and the hydroxyl radical scavenging assay, were utilized to evaluate the anti-oxidant activity of bimetallic Ag–Cu NPs. The reducing power assay measures antioxidant activity by reducing ferric cyanide complex (Fe 3+ ) to the ferrocyanide form (Fe 2+ ). The presence of antioxidative properties in the NPs enables them to donate an electron, resulting in a color change from green to blue, which indicates their scavenging ability 29 . The results of this study revealed that the NPs exhibited a reduction ability, as evidenced by the color change to Perl's Prussian blue observed at 700 nm. The scavenging activity was assessed in comparison to ascorbic acid, a well-known standard antioxidant. Notably, the nanoparticles exhibited the most substantial reduction in Fig. 10 A and B when employed at a concentration of 240 μg mL −1 . Importantly, it is worth highlighting that, in accordance with Al-Ansari et al. research, the plant extract exhibited 58.5% antioxidant activity at its highest concentration (100 μg), whereas the synthesized Ag–Cu nanoparticles achieved ~ 60 ± 5% of activity at a lower concentration (60 μg mL −1 ) 46 . These findings indicates that the nanoparticles are capable of converting the ferric cyanide complex (Fe 3+ ) into the ferrocyanide form (Fe 2+ ) and effectively neutralizing hydroxyl (OH–) free radicals 54 . Moreover, the effectiveness of this scavenging activity is influenced by the dosage of the nanoparticles administered.
Conclusion The present study successfully demonstrates the development of an eco-friendly, rapid, and cost-efficient method for synthesizing Ag–Cu NPs using Aerva lanata extract, fabricated through a phytofabrication process, exhibit great potential for applications in therapeutics as antibiofilm treatments. The FT-IR analysis indicates the presence of hydroxyl, carbonyl, and amino groups in the A. lanata extract, which play a significant role as capping agents for the Ag–Cu NPs. The green synthesis of Ag–Cu NPs resulted in a face-centered cubic crystal structure with an average particle size of 9.5 nm. Moreover, these nanoparticles exhibit promising activities in terms of cytotoxicity, antibacterial, and antioxidant properties, attributed to the intrinsic hydroxy and phenol capping, thus suggesting their potential for broader applications. These bimetallic nanoparticles hold promise for applications in combating bio-films. which are crucial factors in various fields such as healthcare and industry. Additionally, the Ag–Cu NPs demonstrate remarkable anti-cancer, antibacterial, and antioxidant activities, making them potential candidates for further research and development in the pursuit of novel therapeutic interventions.
In this study, we demonstrate the green synthesis of bimetallic silver-copper nanoparticles (Ag–Cu NPs) using Aerva lanata plant extract. These NPs possess diverse biological properties, including in vitro antioxidant, antibiofilm, and cytotoxic activities. The synthesis involves the reduction of silver nitrate and copper oxide salts mediated by the plant extract, resulting in the formation of crystalline Ag–Cu NPs with a face-centered cubic structure. Characterization techniques confirm the presence of functional groups from the plant extract, acting as stabilizing and reducing agents. The synthesized NPs exhibit uniform-sized spherical morphology ranging from 7 to 12 nm. They demonstrate significant antibacterial activity against Staphylococcus aureus and Pseudomonas aeruginosa , inhibiting extracellular polysaccharide secretion in a dose-dependent manner. The Ag–Cu NPs also exhibit potent cytotoxic activity against cancerous HeLa cell lines, with an inhibitory concentration (IC 50 ) of 17.63 μg mL −1 . Additionally, they demonstrate strong antioxidant potential, including reducing capability and H 2 O 2 radical scavenging activity, particularly at high concentrations (240 μg mL −1 ). Overall, these results emphasize the potential of A. lanata plant metabolite-driven NPs as effective agents against infectious diseases and cancer. Subject terms Open access funding provided by North-West University.
Supplementary Information
Supplementary Information The online version contains supplementary material available at 10.1038/s41598-024-51647-x. Acknowledgements All the authors are thankful to their respective universities and institutes for their support. The work was supported by the researchers supporting project number (RSP2024R261) King Saud University, Riyadh, Saudi Arabia. Author contributions This research article was produced through collaboration between the authors. Conceptualization, V.V.L., K.R.R.R., B.B.; writing-original manuscript, G.T., A.N; Methodology, data curation, and formal analysis, G.T., J.A.G., A.N., P.N., S.N., M.P., A.M.; Project ad-ministration and Supervision, V.V.L., A.N; Review and editing, B.B., A.M.A., and K.R.R.R.; Interpretation, and review/revision, V.V.L., B.B., K.R.R.R. and A.M.A. All authors have read and agreed to the published version of the manuscript. Funding Open access funding provided by North-West University. Data availability The data presented in this study are available on request from the corresponding authors. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1270
oa_package/62/2f/PMC10787839.tar.gz
PMC10787840
38218875
Introduction Glioblastoma (GBM) is the most aggressive brain tumor with extensive angiogenesis and poor clinical outcome [ 1 ]. Abnormally increased vascularization results in tumor progression and recurrence. Abundant pro-angiogenic factors in the tumor environment contributed greatly to endothelial cell (EC) proliferation. Among these factors, vascular endothelial growth factor (VEGF) is the most well-known and effective cytokine. Targeting blood vessels has now been an alternative treatment approach in GBM, especially recurrent GBM. Bevacizumab, the first anti-angiogenesis agent approved for glioma, neutralizing VEGF, prolongs progression-free survival (PFS) in GBM patient, however, showed minimal efficacy for overall survival (OS) [ 2 ]. These evidences indicate that further exploration in GBM angiogenesis mechanism is urgently needed. A subpopulation of tumor cells, termed GBM stem cells (GSCs), were verified to be responsible for GBM progression, vascularization and recurrence [ 3 ]. GSCs serve as progenitors for self-renew and proliferation in glioma tissue [ 4 ]. There is emerging evidence indicating that cancer stem cells (CSCs) might support tumor progression by enhancing tumor angiogenesis. The proliferation of adjacent blood vessels or the migration of BM-derived stem cells to the tumor result in tumor endothelial cells [ 5 ]. There are considerable evidences that GSCs would promote pro-angiogenic factors in tumor microenvironment. For instance, glioma stem cells produce high level of stromal-derived factor 1, which eventually promotes local endothelial activity and systemic angiogenic processes involving bone marrow-derived endothelial progenitor cells (EPC) [ 6 ]. A study reported by Ping and et al. revealed that glioma cells co-expressing CD133 and CXCR4 promotes angiogenesis by producing VEGF [ 7 ]. By activating histamine H1 receptor (H1R) and Ca2 + -NF-κB axis, GSCs-secreted histamine promotes angiogenesis and the progression of GBM [ 8 ]. Notably, several markers such as Sox2 and Nestin indicated that stem cells are increased in bevacizumab-resistant patients [ 9 ]. These results indicated glioma stem cell as a potential target in tumor angiogenesis. Understanding molecular mechanism involved in cancer stem cell-mediated angiogenesis is instrumental for anti-angiogenic therapy in glioma. Interferon-inducible transmembrane proteins (IFITM) comprise a family of interferon-induced molecules, consisting primarily of IFITM1, IFITM2, IFITM3 and IFITM5 [ 10 ]. During endothelial cells sprouting process in vitro, IFITM proteins were induced expeditiously and were essential for lumenized vessel formation [ 11 ]. IFITM 1–3 have been implicated in viral pathogen restriction [ 12 ]. IFITM3 was firstly identified to restrict influenza A virus infection and influence the adaptive immune response [ 13 ]. Recent findings have indicated IFITM3 as a key role in a number of pathological process, particularly in neoplasms. IFITM3 in prostate cancer cells promotes tumor progression and bone metastasis by activating TGFβ pathways [ 14 ]. Latest research indicated that IFITM3 phosphorylation accounts for amplification of PI3K signaling, facilitating malignant transformation of B cells [ 15 ]. In fact, human ESCs (hESCs) express high levels of IFITM1 and their expression decreases with differentiation into hepatocyte-like cells [ 16 ]. And Fragilis 2 (encoding IFITM1 in mouse) was reported to be expressed in murine pluripotent embryonic stem cells and a vital downstream mediator of Wnt/β-catenin [ 17 ]. These evidences suggest that IFITM family might play a crucial role in stem cells, including cancer stem cells. Considering these evidences that IFITM proteins are critical for cell stemness and angiogenesis, we attempted to explore the role of IFITM3 in GSCs and glioma angiogenesis and understand the potentially underlying mechanism. In the present study, we report a GSCs-mediated angiogenesis manner, in which IFITM3 expressed in GSCs regulates JAK/STAT3 signaling pathway that preferentially stimulates bFGF production, leading to tumor angiogenesis. Blocking IFITM3 expression in GSCs substantially suppresses angiogenesis, especially when in combination with anti-VEGF agent.
Materials and methods Glioblastoma specimen and cell culture Twenty-eight paraffin-embedded samples from human glioma patients (WHO I-IV) with corresponding clinic-pathological data were collected in the Department of Neurosurgery at Zhujiang Hospital from 2017 to 2019. Informed consent was obtained from all of patients. Both study protocol and informed consent were approved by the Ethical Committee of Zhujiang Hospital. For GBM stem cell culture, tumor samples were collected from consenting patients diagnosed as GBM. GBM tissues were treated as previously reported [ 18 ]. Digested cells were cultured in DMEM/F12 (Gibco, USA) medium supplemented with EGF (20 ng/ml, Peprotech, USA), bFGF (20 ng/ml, Peprotech) and B27 (1:50, Gibco). GSCs were cultured and expanded in two methods, suspension culture of neuro-spheres in low-attached wells, and adherent culture on Laminin (Corning Biosciences, USA) -coated plates [ 19 ]. Low passage (2–4) GSCs were used in this study. U87 and U251 cells were purchased from Chinese Academy of Sciences Cell Bank (Shanghai, China). Glioma cell lines were cultured in high glucose DMEM medium (Gibco) containing 10% fetal bovine serum (FBS, Gibco). Human brain micro-vessel endothelial cells (hBMECs) were purchased from Procell Life Science & Technology (Wuhan, China) and cultured in supplemented endothelial growth medium (EGM-2, Lonza, Walkersville, MD, USA). Reagents, shRNA and transfection shRNA and lentiviral vector for IFITM3 were designed and constructed by Genechem (Shanghai, PR China). The shRNA sequences include IFITM3 shRNA-1 (5′-GCTTCATAGCATTCGCCTACT-3′), IFITM3 shRNA-2 (5′-CCTGTTCAACAC CCTCTTCAT-3′) and shRNA Scrambled control (5′-GATATGTGCGTACCTAGCA T-3′). Glioma cells were transfected with 50 nmol/L shRNAs by using Lipofectamine (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. Bio-informatical analysis Gene expression profiles and corresponding clinical data in Glioma/GBM data were obtained from TCGA and CGGA databases. Primary glioma patients with complete survival data, mRNA sequencing data, and WHO Grade classification data were collected as including patients. Gene expression based on tumor grade and survival analysis were performed using R studio, GEPIA2, and Gliovis [ 20 , 21 ]. GBM samples from TCGA of CGGA were divided into high and low expression groups according to the median IFITM3 expression level. Differentially expressed genes (DEGs) and Gene set enrichment analysis (GSEA) were examined using R packages DESeq2 and clusterProfiler to pick out the significantly enriched pathways based on Gene ontology terms. The summary information for the patients is presented in supplementary table S2 . Cell Counting Kit-8 (CCK-8) and EdU cell proliferation assay CCK-8 and EdU assays were performed to examine endothelial cell proliferative capacity. For CCK-8 assay, endothelial cells were seeded onto 96-well plates (Costar, Cambridge, MA, USA) and cultivated for 7 days. Viable cells were analyzed using Cell Counting Kit-8 (Dojindo, Kumamoto, Japan) according to the manufacturer’s instructions. For Edu assay, endothelial cells to be stained were added with EdU solution, followed by incubation, fixation, permeabilization and Edu staining (Abcam, USA). Cell invasion assay Cell invasion assay was conducted using cell culture insert with 8-um pores in 24-well plates (Costar, USA). Insert was pre-coated with Matrigel (Corning) and the bottom chamber was filled with 0.5 mL medium containing 10% FBS. Cells (1 × 10 5 ) suspended in 100 μl DMEM medium were placed at the upper chamber and incubated for 24 h. Migrated cells on the underside of the insert were stained with crystal violet. Tube formation assay Tube formation assay was carried out as previously stated [ 22 ]. 96-well plate was coated with Growth factor reduced Matrigel and incubated at 37 °C for 30 min. Endothelial cells were seeded at 2 × 10 4 cells/well and incubated for 24 h. Tube quantification was examined with ImageJ software. Endothelium spheroid-based sprouting angiogenesis assay Sprouting assay was performed according to protocol reported by Korff with minor modification [ 23 ]. Dissociated endothelial cells were suspended in complete EGM-2 medium containing 0.25% methylcellulose and seeded on low attachment 96-well plates (Corning). About 800–1000 cells per well bond together within 48 h to form a single spheroid, which were then transferred to 24-well plate and cultivated in Collagen I solution (Enzo Life Sciences, USA) supplemented with 20% FBS. Sprouting capacity was measured via counting the total sprout number using NeuronJ plugin of ImageJ software. Immunoblotting and immunological analysis The immunoblotting assay was performed as previously described [ 24 ]. Human ELISA kit (eBioscience) was used to measure the concentration of bFGF from culture medium according to manufacturer’s protocol. Cells were cultured in 6-well plates for 72 h and supernatants were collected for further analysis. Experiments were performed in triplicate. Antibodies used in the present study were listed in the supplementary table S3 . Tissue Immunohistochemical (IHC) and Immunofluorescence staining (IF) Specimens sections of surgical GBM tissues and intracranial mice xenografts were prepared in a routine procedure [ 25 ] and immuno-stained with targeted proteins using a protocol as previously described [ 26 ]. In vivo xenograft assay BALB/C nude mice in the present research were procured from Experimental Animal Center of Southern Medical University. Orthotopic xenograft model was established using Balb/c nude mice (6-week-old) with GBM cells (1 × 10 5 cells in 0.1 ml PBS) stably transfected with mCherry-LUC vector based on Ozawa’s protocol [ 27 ]. Intracranial tumor growth was examined by in-vivo imaging system (IVIS Lumina II, Caliper, USA). The protocol has been registered and approved by the Animal Care and Use Committee of Southern Medical University. Statistical analysis All experiments in the study were carried out at least 3 time and analyzed using Prism 8 (GraphPad Software Inc., USA) or R software. Results were displayed as Mean ± SD. Statistical significance was determined via using Student’s t -test or one-way ANOVA with Bonferroni correction for multiple comparisons. P < 0.05 was considered statistically significant.
Results IFITM3 expression is elevated in human GBM tissue To examine the expression of IFITM3 in GBM, we investigated IFITM3 gene expression level in glioma datasets from The Cancer Genome Atlas (TCGA) and Chinese Glioma Genome Atlas (CGGA). Results showed a markedly elevated expression of IFITM3 in GBM, compared to lower grade gliomas (grade II, III) (Fig. 1A ). Data from GEPIA2 web tool also revealed an obviously higher IFITM3 expression in GBM sample than that in normal brain tissue (Fig. 1B ). Mesenchymal subtype of GBM has been associated with therapy resistant and more aggressive features than other subtypes [ 28 ]. Our data suggested that GBM sample with mesenchymal subtype displayed a markedly upregulated IFITM3 expression (Supplementary Fig. 1a ). IFITM3 also correlated with IDH1 (Supplementary Fig. 1b ) and MGMT (Supplementary Fig. 1C ) status in GBM samples. And upregulated IFITM3 significantly correlated with poor outcome in both GBM and LGG samples (Fig. 1C ). We examined IHC staining with IFITM3 from human protein atlas (HPA) and observed a strong intensity in high grade glioma, while a weak intensity was documented in low-grade glioma (Fig. 1D ). Furthermore, we collected 28 glioma specimens with clinicopathologic data. IHC staining of glioma samples indicated a positive relation between IFITM3 expression and tumor grade (Fig. 1E, F ). Moreover, patients were divided into two groups based on IFITM3 expression. Surprisingly, it revealed that IFITM3 expression was only significantly related to tumor grade, rather than gender, age and tumor size (Fig. 1G ). Taken together, these data indicated that IFITM3 expression was elevated in glioma samples, especially in GBM sample. IFITM3 is enriched in GBM stem cells Cancer stem cells (CSCs), a subpopulation inside the heterogeneous tumor tissues, are characterized by multi-lineage differentiation potential [ 29 ]. Evidences from various studies indicated that CSCs were essential for tumor progression, recurrence and metastasis [ 30 ]. In the present study, GBM stem cells (GSCs) were derived from fresh GBM samples and cultured in vitro. GSCs possessed the sphere-forming capacity (Fig. 2A ) and neural stem cell markers (Fig. 2B ) as reported previously. Immunoblotting confirmed that GSCs within the culture universally express IFITM3, despite some variations in levels between cells (Fig. 2C ). While GSCs were enriched for IFITM3, GSC serum-differentiated cells (GSDCs) exhibited reduced expression (Fig. 2D ), indicating that IFITM3 was solely expressed in neural stem cells or cancer stem cells. Moreover, we observed a positive association between IFITM3 and Nestin expression in samples from GBM patients (Fig. 2E ). Likewise, in CGGA GBM dataset, IFITM3 expression closely correlated with stem cell markers Nestin and Sox2 (Fig. 2F ). Sufficient evidence has established a role for hypoxia environment in stem cell maintenance [ 31 ]. Interestingly, low oxygen level induced increased IFITM3 and HIF1α expression in GSCs, rather than glioma cell lines (Fig. 2G ). When GSCs were exposed to 2% oxygen, protein level of IFITM3 gradually increased over time (Fig. 2H ). When exposed to varying levels of O 2 for 24 h, GSCs exhibited rising IFITM3 expression (Fig. 2H ). Similarly, GSC spheres were cultured under hypoxic environment, IFITM3 and HIF1α expression were elevated (Fig. 2I ). And in CGGA and Gravendeel GBM datasets, IFITM3 was significantly correlated with HIF1α mRNA levels (Fig. 2J ). IFITM3 in GBM stem cells regulates endothelium proliferation and sprouting To further determine the role of IFITM3 on biological function of GSCs. We examined sphere-forming competency of GSCs and found that downregulation of IFITM3 had no impact on GSC sphere-forming capacity (Fig. 3A ). Since sufficient angiogenesis is essential for solid tumor expansion [ 32 ], we co-cultured GSCs with brain microvascular endothelial cells (hBMECs), and examined cell proliferation and sprouting capacity (Fig. 3B ). It suggested that IFITM3 knockdown in GSCs reduced tube formation and sprouting capacity of hBMECs (Fig. 3C ). Moreover, proliferative ability of endothelial cells was impaired when IFITM3 was downregulated (Fig. 3D, E ). We established intracranial xenografts in nude mice through implanting GBM stem cells to evaluate the effect of IFITM3 on angiogenesis in vivo. Results indicated that IFITM3 downregulation led to reduced vessel density as assessed by CD34, while exhibited no effect on stem cell population (Fig. 3F ). Together, these findings suggested that IFITM3 in GBM stem cells would be responsible for increased angiogenesis in tumor microenvironment. IFITM3 induces JAK/STAT3 activation in GSCs GBM patients from TCGA dataset were categorized into high and low expression groups by using median expression value of IFITM3 as cut-off point. Differentially gene expression with RNA-seq data was analyzed (Fig. 4A ). Gene set enrichment analysis (GSEA) was utilized for assessment of key pathways between two groups (Fig. 4B ). Among these enriched pathways, angiogenesis-related pathways including PI3K/Akt, JAK/STAT and Notch were selected (Supplementary Table 1 ). GSCs were treated with scrambled shRNA targeting IFITM3, and key protein expressions of these pathways were determined. Immunoblotting assay indicated that IFITM3 in GSCs regulate key protein expression in JAK/STAT3 signaling pathway (Fig. 4C ). Next, GSCs were transfected with lentiviral vector encoding IFITM3, and co-incubated with endothelial cells. In vitro angiogenic assays displayed enhanced tube formation and sprouting capacities in hBMECs co-cultured with IFITM3-GSCs (Fig. 4D ). Moreover, when WP1066 (JAK inhibitor) was added into GSCs culture medium, IFITM3-induced angiogenesis was significantly attenuated (Fig. 4D ). In human GBM specimen, we observed elevated p-STAT expression in robust IFITM3-staining region (Fig. 4E ). When fractionated IFITM3 + or IFITM3 - GSCs were injected orthopically into nude mice, IFITM3 expression was also consistent with p-STAT3 expression (Fig. 4F ). GSC-derived IFITM3 contributes to angiogenesis via producing bFGF The GO term molecular function (MF) analysis indicated that cytokine activity was significantly enriched in IFITM3-regulated genes (Supplementary Fig. 2 ). Secreted cytokines have powerful influence on angiogenesis either in normal tissue or malignant tissue. To identify the angiogenic factors involved in IFITM3-mediated angiogenesis, an angiogenesis array was performed. The expression of bFGF and TGFβ1 markedly decreased in shIFITM3 GSCs as compared with shCon-GSCs (Fig. 5A, C ). When GSC#2 cells were fractionated into two groups with high or low IFITM3 expression. Culture media from high-IFITM3 cells showed increased bFGF level (Fig. 5B, C ). In various GSCs, IFITM3-regulated bFGF production was further validated (Fig. 5D, E ). Interestingly, a co-cultivation of GSCs with hBMECs revealed that blocking bFGF would substantially mitigate pro-angiogenic effect by IFITM3 (Fig. 5F ). To explore IFITM3-mediated angiogenesis in vivo, we established intracranial mice model by implanting Luc labeled GSC#1 into nude mice. Tumor growth was monitored using in vivo Imaging System. Results indicated that IFITM3 knockdown exhibited attenuated tumor growth (Fig. 5G ) and declined bFGF secretion (Fig. 5H ). In human GBM samples, our data showed that IFITM3 expression was associated with bFGF production and enhanced angiogenesis (Fig. 5I ). These findings indicated that GSC-derived IFITM3 contributed significantly to angiogenesis in vivo via regulating bFGF.
Discussion In the present study, we describe a novel mechanism of GSCs-induced GBM angiogenesis through expression of IFITM3 and bFGF (Fig. 5J ). IFITM3 is a interferon-induced protein mediating antiviral activity via suppressing the entry of viruses [ 33 ]. Interestingly, it has been gradually unraveled that IFITM3 correlates with tumor progression, including proliferation [ 34 ], invasion [ 35 ], chemoresistance [ 36 ], metastasis [ 37 ], as well as angiogenesis [ 38 ]. A Recent finding indicated that TGFβ-regulated IFITM3 expression facilitates glioma cell invasion [ 39 ]. However, further investigation regarding molecular mechanism, by which IFITM3 influences glioma progression remains unclear. Conclusive evidence is now provided that IFITM3 functions as an angiogenesis inducer. Angiogenesis is the formation of new blood vessels including multiple components, such as the endothelial cell proliferation, cell migration, cell adherence and in vitro tube formation [ 40 ]. In the present research, we focused our investigations on GBM stem cell-induced alterations on endothelial cells. we utilize a co-cultivation system to demonstrate that IFITM3 is capable to stimulate endothelial cell proliferation, migration and sprouting through interaction between GSCs and endothelial cells. The observation that conditioned media from GSCs, as well as physically separated GSCs modulates human brain endothelial cell phenotypes indicates that GSCs release pro-angiogenic cytokines in a IFITM3-dependent method. It has gradually become clear that cancer stem cells (CSCs) exist in various types of tumors tissues, even though the controversy about exact markers to identify and isolate CSCs [ 41 ]. Recent researches have unveiled that GSCs share similar characteristics. Both cell types express stem cell markers and the ability of self-renewal [ 42 ]. CD133 and Nestin have been widely considered as markers in GSCs [ 43 ]. However, there is also evidence that CD133-negative cells cultured from GBM patients harbored a self-renewal ability [ 44 ]. The present study demonstrates that IFITM3 is preferentially expressed in GSCs and decreased markedly in differentiated glioma cells, yet has no significant impact on GSC self-renewal ability (Fig. 3A ). In fact, IFITM3 has been reported to exert no effect on cell proliferation and invasion in glioma cell lines [ 45 ], indicating that IFITM3 had no direct effect on glioma cells. These evidences suggest that IFITM3 might not be an indispensable factor in stemness maintenance of cancer cells, despite the close correlation between protein expression and stemness status. Furthermore, hypoxic situation led to a dramatic elevation of IFITM3 expression (Fig. 2G, H ). Hypoxia is known to affect the CSCs maintenance and functions. Stem cells tend to residence in hypoxic regions within tumors, which probably benefit the preservation of tumor stemness [ 46 ]. And hypoxia could regulate a variety of targets related to tumor stemness and invasiveness. Our data is consistent with a prior study by Cai that high-IFITM3 expression correlated with high hypoxia score in the bladder cancer using bio-informatic method [ 47 ]. Rajan et al. demonstrated that gene expression of IFITM3 was elevated in microglia when rat brain suffered a transient cerebral ischemia [ 48 ]. Data from a study by Harmon and colleagues showed that ischemic injury in aged brains following stroke resulted in the induction of IFITM3 proteins [ 49 ]. Therefore, IFITM3 would act as a hypoxia-induced intermediate factor in GSCs, and subsequently the affects the tumor microenvironment surrounding GSCs. Besides from tumor initiation, CSCs are shown to interact with endothelial cells and be closely linked to vasculature formation in perivascular niche [ 50 ]. Recent findings indicate that GSCs contributed to tumor vasculature through direct differentiation or releasing pro-angiogenic cytokines. GSCs cultured from GBM patients differentiated into vascular endothelial cells and pericytes when induced by serum in vitro [ 51 , 52 ]. These GSCs-derived vascular cells harbored the tube formation capacity and could be identified in human GBM tissues [ 53 ]. In our study, IFITM3 high GSCs confers endothelial cells proliferation, migration and sprouting through paracrine. Through human angiogenesis antibody array, we identified bFGF as a key motivator in GSC-mediated angiogenesis. High bFGF level has been reported in individuals with various categories of neoplasms and predicted a poor prognosis [ 54 ]. Elevated bFGF levels have been identified in human glioma, suggesting its importance in tumor growth progression [ 55 ]. Besides, one of the most prominent reason for the resistance to anti-VEGF treatment is the compensatory mechanism of other growth factors, with the bFGF being at the top of the list [ 54 ]. These results showed that IFITM3-bFGF axis between GSCs and endothelial cells might be the compensatory angiogenic approach when VEGF-dependent angiogenesis is interrupted. In summary, the present study discloses a functional role IFITM3 in regulating GBM angiogenesis and identifying the downstream effectors, which are important for GBM progression. And bFGF is shown to be a downstream target of IFITM3 that modulates brain endothelial cell proliferation, migration and sprouting. Our findings establish the rationale for developing anti-vascular therapy based on the targeted disruption of IFITM3.
Interferon-induced transmembrane protein 3 (IFITM3) has been previously verified to be an endosomal protein that prevents viral infection. Recent findings suggested IFITM3 as a key factor in tumor invasion and progression. To clarify the role and molecular mechanism of IFITM3 in Glioblastoma multiforme (GBM) progression, we investigated the expression of IFITM3 in glioma datasets culled from The Cancer Genome Atlas (TCGA) and Chinese Glioma Genome Atlas (CGGA). Primary GBM stem cells (GSCs) were cultured and identified in vitro. Loss-of-function and gain-of-function experiments were established by using shRNAs and lentiviral vectors targeting IFITM3. Co-culture system of GSCs and vascular endothelial cells was constructed in a Transwell chamber. Tube formation and spheroid-based angiogenesis assays were performed to determine the angiogenic capacity of endothelial cells. Results revealed that IFITM3 is elevated in GBM samples and predictive of adverse outcome. Mechanistically, GSCs-derived IFITM3 causes activation of Jak2/STAT3 signaling and leads to robust secretion of bFGF into tumor environment, which eventually results in enhanced angiogenesis. Taken together, these evidence indicated IFITM3 as an essential factor in GBM angiogenesis. Our findings provide a new insight into mechanism by which IFITM3 modulates GBM angiogenesis. Subject terms
Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s41419-023-06416-5. Acknowledgements This work was supported by the National Natural Science Foundation of China (No. 82002631, No. 82072762, and No. 82203135) and President Foundation of Zhujiang Hospital, Southern Medical University (No. yzjj2022ms07). Author contributions Conceptualization: YL and YK; Methodology: XX, ZX, and HC; Data curation: ZX, XX, YZ, and TC; Original draft preparation: ZX and XX; Supervision: YL; All authors have read and agreed to the published version of the manuscript. Data availability All data generated in the present study are included either in the main article or in the supplementary information files. Competing interests The authors declare no competing interests. Ethics approval and consent to participate Animal study was performed with the permission of the Animal Care and Use Committee of Southern Medical University. And study protocol and informed consent were approved by the Ethical Committee of Zhujiang Hospital.
CC BY
no
2024-01-15 23:42:00
Cell Death Dis. 2024 Jan 13; 15(1):45
oa_package/b5/05/PMC10787840.tar.gz
PMC10787841
38218746
Introduction Since the beginning of the twenty-first century, industrial big data has experienced rapid development with the improvement of data collection and processing capabilities 1 . The centralized big data processing model centered around cloud computing can no longer support industrial data analysis. Its various drawbacks have become evident, such as difficulty integrating heterogeneous data from multiple sources, handling high broadband loads, and dealing with limited resources 2 – 4 . This is particularly critical for equipment management systems demanding highly real-time data processing. Suppose faults within the equipment are not detected at the earliest opportunity. In that case, it diminishes product processing quality and leads to even more significant losses across the entire industrial production line. In recent years, edge computing technology, based on industrial-grade intelligent hardware, has become a hot research field. Establishing a data bridge between production equipment and cloud-based systems achieves rapid sensing of equipment operating statuses in the industrial Internet of Things (IIoT) and enables intelligent adjustments. This advancement has propelled significant developments in intelligent systems and smart manufacturing 5 . The operating scope of edge computing technology includes downstream data from cloud services and upstream data from the Internet of Things services 6 . It is a novel computing model that performs computations at the network edge 7 . Edge computing can be traced back to the content distribution network proposed by Akamai in 1998 8 . In 2013, the American scholar Ryan La Mothe first proposed “edge computing” in an internal report 7 . In May 2016, Professor Shi Weisong and his team from Wayne State University in the United States formally defined edge computing 9 . In the same year, China established the Edge Computing Consortium (ECC), which Huawei Technologies Co. Ltd. and the Shenyang Institute of Automation of the Chinese Academy of Sciences founded 7 . The consortium covers various fields, such as scientific research institutions and industrial manufacturing. In the case of the Industrial Internet of Things (IIoT), edge computing meets its requirements for real-time control and edge device security and privacy in practical applications, making it a direction for developing the IIoT industry. In their study, Shi Weisong et al. 6 summarized the current situation and prospects of edge computing and provided a summary from the industrial Internet of Things perspective. Edge computing can address the real-time control of networked production and processing, edge device security and privacy, and localized processing of production data faced by the development of industrial IoT, and has advantages in improving performance, ensuring data security and privacy, and reducing operating costs in practical applications 6 , 7 . Currently, research on equipment management mainly focuses on three directions: predictive maintenance 10 – 12 , fault diagnosis 13 , 14 , and quality prediction, as shown in Fig. 1 . Equipment predictive maintenance refers to collecting operational data and environmental data during the operation of equipment, using big data and machine learning methods to predict the service life and damage of important components of the equipment, avoiding excessive maintenance of the equipment, reducing the failure rate of the equipment, and lowering the manufacturing cost of products. The main approach to equipment fault diagnosis is to establish the mechanism of equipment faults, study the relationships between various causes of faults, fault characterizations, and fault signals, in order to diagnose equipment faults quickly when they occur. Quality prediction is an essential means to reduce the probability of product quality problems and improve the qualification rate by analyzing the operating parameters of the equipment, obtaining quality characteristics based on equipment parameters, and monitoring and controlling the parameters of the equipment processing process 15 . Product quality management is an essential issue in intelligent factory information services, and many scholars have elaborated on different aspects such as reliability 16 , helpful life 17 , and retrievability 18 . Among them, the prediction of product quality is also a hot topic. Product quality often requires specialized, expensive, and complex testing equipment, and the testing process can take a long time. Therefore, rapid, effective product quality prediction is significant for providing decision-making services to factory managers.
Active control method for quality prediction This section first analyzed the product’s quality characteristics and selected criticalquality-related parameters with correlation coefficients greater than the set threshold based on the correlation coefficients of industrial product quality inspection results and quality-related parameters. Established the SMOTE-XGBoost quality prediction model and optimized the hyperparameters. Finally, the active control method for prediction. Analysis of quality characteristics In product quality issues,this paper abstracts the product processing process as a manufacturing processing unit and the process of changing the product quality state as process characteristic data of processing quality. Additionally, it analyzesthe process parameter data during equipment operation. As shown in Fig. 5 . In the manufacturing processing unit, represents the product state before the execution of the manufacturing processing unit; represents the product state after the execution of the manufacturing processing unit; From the perspective of quality data, refers to the resource processing data received by the manufacturing processing unit; represents the product quality state data before the manufacturing processing unit processes it; refers to the output product quality state data processed by the manufacturing processing unit; represents the difference between the actual qualified rate of the calculated output product and the qualified rate of the industrial product containing the predicted results, and is the threshold. When the value exceeds a certain threshold, the edge computing layer will generate corresponding process adjustment control instructions ,and send them to the relevant processing equipment, such as adjusting the spindle speed and feed rate 15 . From the perspective of task execution, process refers to the process of transforming the quality characteristics of a product from state to state through a series of processing methods. From the perspective of quality characteristics, the current quality characteristic is the result of the current process equipment processing the quality characteristic in the current environment 15 . The process of changing quality characteristics is the process of transforming input data into output data through its processing mechanism. As the manufacturing processing continues, manufacturing quality-related parameter data is collected one by one at a fixed frequency. The type of equipment process data parameter set for collection is , and each set of equipment quality-related parameter data collected is represented by an array, as shown in Eq. ( 1 ). In Eq. ( 1 ), represents an array of equipment quality-related parameters collected at a certain moment. represents the -th parameter of array . As time passes and the processing progresses, more and more data is collected, forming a matrix of quality-related parameter data as shown in Eq. ( 2 ). Selection of quality-related parameters Throughout the production process of industrial goods, a large amount of data related to their quality is collected through the equipment perception layer, including quality inspection results and corresponding quality-related parameters. Including quality inspection results and corresponding quality-related parameters. Based on the quality inspection results and corresponding quality-related parameters, important rules for selecting quality-related parameters can be established, as described in section “ Equipment process parameters ”. This article selects the quality-related parameters that affect the indirectly dynamic equipment process data. The selection rule of quality-related parameters mainly refers to selecting the key quality-related parameters with a correlation coefficient greater than a set threshold through the correlation analysis between the quality inspection results and the quality-related parameters in industrial product manufacturing. Formula for calculating the correlation coefficient between the quality inspection results and the quality-related parameters in industrial product manufacturing is: where represents the -th quality-related parameter, represents the number of nonconforming industrial product manufacturing quality inspection results, and represents the number of conforming industrial product manufacturing quality inspection results. represents the joint distribution of and represents the joint distribution of and . , and are the probability distributions of variables , , and ,respectively. and are adjustment coefficients for data imbalance, with a sum of 1, generally determined based on the quality of the data samples obtained. According to the correlation coefficient between the quality inspection results and the quality-related parameters in industrial product manufacturing, the importance of the features is sorted. Obtain a feature set , where represents the -th feature value. SMOTE-XGBoost algorithm for quality prediction Data preprocessing based on SMOTE According to the factory survey results, in the stable production line of brake discs, majority of the final quality is qualified, and only a small number of products have quality problems (unqualified products). The brake disc production line produces more than 1000 products per day, of which over 95% are qualified products. From a data mining perspective, this means that the input labels of the prediction model are imbalanced. Imbalanced label data is a common type of data that is widely present in various industrial fields. This article adopts the Synthetic Minority Oversampling Technique (SMOTE) algorithm to address the issue of imbalanced data. The core idea of the algorithm is to perform interpolation on the minority class samples in the dataset based on the k-nearest neighbor rule (as shown in Fig. 6 below). Generating more minority class samples as a result 39 . As the production dataset of brake discs is imbalanced, SMOTE is used in this chapter to balance the dataset. The main steps of the algorithm are as follows: The dataset of brake disc processing collected by the production line is an imbalanced dataset. Based on the number of minority class samples and majority class samples , the required number of synthesized samples is calculated: For each unqualified product data sample (minority class) , where . Select nearest neighbors ( is usually set to 5) of the minority class sample randomly with the Euclidean distance as the measurement standard. Assuming the selected neighboring point is , the new synthetic sample point is generated according to the following formula. where represents a random number between 0 and 1. Generate new minority class samples, merge them with the original data set to get a balanced data set. Then input them into XGboost for identification. Predictive model based on XGboost XGBoost is ensemble learning model framework based on gradient boosting algorithm, which was proposed by Dr. Tianqi Chen and his colleagues 40 . Compared with the traditional Gradient Based Decision Tree (GBDT), both are based on decision trees. However, XGboost effectively controls the complexity of the model and greatly reduces the variance of the model by using second-order Taylor expansion and adding regularization terms. The trained model is also simpler and more stable 41 . Assuming that the input samples are , The output of the XGboost model can be represented as the sum of weak learner outputs: where represents the output of the -th weak learner. The model’s bias and variance determine the prediction accuracy of a model. The loss function represents the bias of the model, and to reduce the variance, a regularization term needs to be added to the objective function to prevent overfitting. The objective function comprises the model’s loss function and a regularization term to suppress model complexity. The objective function to minimize in function space is: Here, represents the loss function, represents the regularization function, is the number of leaf nodes, and is the weight value of leaf nodes. In the XGBoost model, most weak learns are based on Classification and Regression Trees (CART). Therefore, each round of optimization only focuses on the objective function of the -th classification and regression tree based on the previous models. Next, perform second-order Taylor expansion on the loss function of XGboos: And in the above equation: In which, and are the first-order and second-order derivatives of each sample on the loss function, respectively. Therefore, the optimization of the objective function can be transformed into the process of finding the minimum value of a quadratic function. SMOTE-XGBoost with jointly optimized hyperparameters (1) SMOTE-XGBoost model The hyperparameter optimization methods mainly include grid search, random search, heuristic algorithms, and so on 42 . This article used the gridsearch method to optimize the above three hyperparameters, in order to obtain the optimal predictive model. The Smote algorithm and the XGboost algorithm both have hyperparameters that need to be set before training the algorithm. The setting of hyperparameters affects the performance of predictive models. Previous research has mainly focused on the hyperparameters in classification or regression models. Therefore, consider Smote and XGboost as a whole and propose a joint optimization method for hyperparameters, called SMOTE-XGboost, to improve the performance of quality prediction models. Specifically, this paper focuses on the optimization of the hyperparameters in SMOTE (Number of nearest neighbors for selecting samples), in XGboost (Number of decision trees), and in XGboost (Number of leaf nodes). Selecting the maximum score as the optimization objective to obtain the best hyperparameters. The principle of joint hyperparameter optimization is as follows: Train the original SMOTE-XGboost model on historical data, which can be represented as: where represents the number of nearest neighbors selected in SMOTE. represents the number of decision trees in XGBoost, and represents the number of leaf nodes in XGBoost. Training process of the SMOTE-XGboost prediction model described in this article includes: To optimize the hyperparameters of the SMOTE-XGboost model with the goal of obtaining the maximum AUC score, the following formula is used: In the expression: represents the true quality result; represents the predicted quality result; represents a quality prediction function; is a non-analytic function of the decision variable . is the scoring formula; represents the first data points in the test set. (2) Model evaluation indicators To effectively evaluate the reliability of predictive models, comparative experiments of different algorithms are conducted using the coefficient of determination ( ) and the AUC as evaluation metrics to assess the relationship between predicted values and true values of the models. AUC is defined as the area enclosed by the coordinate axis under the ROC curve. It is a comprehensive performance classification indicator, which is commonly used to measure classification performance 31 , 43 . The higher the AUC, the better the algorithm performance. Scoring formula for : In this expression, represents the true value, represents the predicted value, represents the sample mean, and represents the sample size. A higher value indicates better performance. When the predictive model makes no errors, achieves the maximum value of 1. Scoring formula for : In this expression, represents the ranking number of positive samples in the data set, represents the number of positive samples in the data set, and represents the total number of samples in the data set. (3) Active control methods In the actual production process, manufacturing process data of industrial products is first transmitted to edge computing nodes through Ethernet. The edge computing nodes use important quality-related parameter selection rules to filter and reduce data, and make real-time quality predictions for products as qualified or non-qualified based on the quality active prediction model deployed on the edge computing nodes. Active control methods refer to calculating the difference between the actual qualified rate of the produced product and the predicted qualified rate of products. If this difference is greater than a certain threshold, the edge computing layer will generate corresponding process adjustment control instructions and send them to the relevant processing equipment, such as adjusting spindle speed, feed rate, etc. Edge computing-based proactive control method for industrial product manufacturing quality prediction, It characteristics lie in the calculation formula for the difference between the actual qualified rate of the produced product and the qualified rate of products with prediction results, which is as follows: In the formula, and respectively represent the number of qualified products in the actual output, and the number of qualified products with prediction results; and respectively represent the total number of products in the actual output, and the total number of products with prediction results.
Experiments and results Qualitys correlation parameter selection Some studies 44 , 45 have pointed out that the equipment process data obtained from the processing equipment, including spindle power (P), spindle current (I), spindle speed (S), feed speed (F), and clamping force (N), were related to the changes of product quality characteristics in the processing process. In order to validate the effectiveness of the proposed quality prediction method, historical data sets from the edge server were collected as the data source for overall quality prediction analysis. The data set includes 5 quality characteristics and 1 final quality label (qualified or fault product). The quality characteristics are all continuous random variables. Table 3 shows the specific quality characteristics of the partial samples. There are 1844 samples in the data set, including 1778 samples of qualified products and 66 samples of fault products. The imbalance ratio of the data set is about 26.9:1. As per the calculation method described in section “ Selection of quality-related parameters ”, calculated the importance of each quality feature and sorted them in descending order, as shown in Fig. 7 , ultimately selected four quality features, including spindle speed (S) and feed speed (F), spindle power (P), and spindle current (I), and clamping force (N), to construct the prediction model. Classification results with other machine learning methods This paper conducted comparative experiments among different algorithms to validate the effectiveness of the proposed quality prediction model. The relationship between predicted and actual values was evaluated using coefficients ( ) and AUC as assessment metrics. All experiments in this study were deployed in a python3.6 environment and run on a desktop computer with an Intel Core i7 processor, 3.6 GHz, and 16 GB RAM. First, the data set was extracted based on the sorted quality features. The data description of the training set and test set is shown in Table 4 . Apply the SMOTE oversampling strategy only in the training set to avoid over-optimism 38 , 46 . The data after SMOTE processing is shown in Table 5 . Then, this text used the training set to build the SMOTE-XGBoost prediction model and used grid search to jointly optimize the hyperparameters of the brake disc quality prediction model (the hyperparameter optimization range is shown in Table 7 ). The final optimal values for each hyperparameter of the SMOTE-XGBoost were determined to be k = 6, e = 100, and T = 3. The optimized quality prediction model is named SMOTE-XGboost_t, and its prediction results on part of the test data set are shown in Fig. 8 . This paper designed comparative experiments from the perspectives of classification algorithms and hyperparameter optimization to highlight the superiority of the proposed method. (1) Comparison experiment of classification algorithms. To verify the classification performance of the proposed method compared to other classification methods under the same criteria, this study used the same SMOTE method and compared the proposed method with other mainstream machine learning classification methods (Support Vector Machine, SVM; Logistic Regression, LR; Decision Tree, DT; Random Forest, RF). The experimental results are shown in Table 6 , based on the table, as can be seen that the proposed SMOTE-XGboost_t method has slightly higher and AUC values compared to other classifiers in the experiment using the same SMOTE method. Moreover, the ROC curves of the model’s indicators are shown in Fig. 9 . AUC is defined as the area enclosed by the coordinate axis under the ROC curve. From the figure, as can be seen that the AUC value of the proposed SMOTE-XGboost_t method is as high as 0.916, which indicates that the proposed method can effectively identify unqualified products and thus better predict the quality of brake discs. (2) Hyperparameter optimization comparative experiment In addition, to investigate the impact of hyperparameter optimization on the model, this study conducted four different experiments:In the model named SMOTE-XGboost, the default values were used for the hyperparameters without any hyperparameter optimization; The hyperparameter ( Number of nearest neighbors for selecting samples) in SMOTE was optimized in the model SMOTE-XGboost_s; In the model SMOTE-XGboost_x, only the hyperparameters (Number of decision trees) and (Number of leaf nodes) in XGBoost were optimized; The last experiment involved joint optimization of the hyperparameters (Number of nearest neighbors for selecting samples), (Number of decision trees), and (number of leaf nodes) in both SMOTE and XGboost using grid search in the SMOTE-XGboost_t model. The optimal hyperparameters and optimization ranges for the predictive models in the four experiments are shown in Table 6 , and the experimental comparison results are shown in Table 8 . Based on Table 7 , see that in the SMOTE-XGboost_t model, the optimal value is 6 instead of the default value of = 5. This indicates that when integrating oversampling algorithms with traditional machine learning classification algorithms, there may be uncertainties in the prediction results due to the hyperparameters of the sampling model and the classification model. Therefore, optimizing the hyperparameters in both the SMOTE sampling algorithm and the XGboost classification model is beneficial to improve the quality prediction performance. Analysis of the ROC curves for the four experiments based on AUC values is shown in Fig. 10 . It can be observed that the SMOTE-XGboost_t and SMOTE-XGboost methods are slightly better than the other methods. SMOTE-XGboost_t had the best performance with an AUC value of 0.916. The analysis from Table 8 shows that the proposed method performs better than other methods in terms of AUC and scores, indicating that the quality prediction model has a strong ability to identify the quality of brake discs after joint optimization of hyperparameters. Based on the actual operation of the factory, factory managers are more concerned about defective products than the large quantity of qualified products. Therefore, the method proposed in this paper has strong comprehensive prediction ability.
Discussion In terms of imbalanced data, Table 6 and Fig. 9 demonstrate that the SMOTE and XGBoost combination outperforms the combination of SMOTE with other classification algorithms Fig. 7 displays the importance of quality features; selecting these features is crucial for predicting and analyzing quality issues. Additionally, simultaneous investigation of hyperparameters in joint optimization included k (the number of nearest neighbors in SMOTE), e (the number of decision trees in XGBoost), and T (the number of leaf nodes in XGBoost). Table 8 and Fig. 10 indicate that the SMOTE-XGBoost method with jointly optimized hyperparameters can enhance classification performance. This result also indicates that our proposed method contributes to addressing imbalanced data classification issues. and AUC are two widely used metrics in various classification problems. Additionally, AUC is a comprehensive metric that considers both qualified and defective products. Therefore, AUC is a more critical metric in unbalanced quality prediction scenarios and is widely used in various imbalanced classification problems. Existing traditional industrial product manufacturing quality has long relied on passive analysis methods such as statistical monitoring. This method primarily involves testing the product quality using quality inspection equipment after the production and processing of the product. The limitations of this method lie in two aspects. Firstly, specific products require particular quality inspection equipment, which takes considerable time and involves expensive equipment. Secondly, it is impossible to forecast whether the product quality will be up to standard. When faults occur in equipment affecting product quality, there is no timely feedback for adjusting the equipment. So, rapid and efficient quality prediction methods can potentially replace specialized equipment, saving on equipment costs and testing time.
Conclusion This article proposes an Edge computing-based proactive control method for industrial product manufacturing quality prediction, addressing the issue of imbalanced data in the manufacturing process. Firstly, an edge computing-based framework for quality prediction in industrial product manufacturing was proposed. Secondly, a method for selecting quality-related parameters was designed, this provides insights into quality analysis problems. Finally, a SMOTE-XGboost quality forecasting active control method based on joint optimization hyperparameters is proposed to solve the problem of manufacturing quality forecasting of industrial products under category imbalance (Table 8 ). This paper compared prediction algorithms based on five different classification methods under specific experimental conditions. The experimental results indicate that the proposed SMOTE-XGboost_t method slightly outperforms the other four classifiers in terms of and AUC metrics. This indicates that the proposed method has good performance in predicting the manufacturing quality of industrial products and detecting faulty products. Finally, the optimal values for each hyperparameter of SMOTE-XGboost were determined to be = 6, = 100, and = 3, and the prediction results were better than those obtained through single hyperparameter optimization. The research in this article enhances the capability for product quality control and provides intelligent information services for enterprises. However, there are still some issues that need further study. This paper only considered the product quality prediction results after processing in a single processing unit. Therefore, future research will focus on predicting product quality for multi-stage processing. Additionally, since the process-related data during manufacturing is incremental, another research direction involves addressing the issue of the source database of the quality prediction model in the edge computing scenario updating over time in the production line. This involves devising an incremental data training strategy for obtaining performance updates by training incremental data on the existing model.
With the emergence of intelligent manufacturing, new-generation information technologies such as big data and artificial intelligence are rapidly integrating with the manufacturing industry. One of the primary applications is to assist manufacturing plants in predicting product quality. Traditional predictive models primarily focus on establishing high-precision classification or regression models, with less emphasis on imbalanced data. This is a specific but common scenario in practical industrial environments concerning quality prediction. A SMOTE-XGboost quality prediction active control method based on joint optimization hyperparameters is proposed to address the problem of imbalanced data classification in product quality prediction. In addition, edge computing technology is introduced to address issues in industrial manufacturing, such as the large bandwidth load and resource limitations associated with traditional cloud computing models. Finally, the practicality and effectiveness of the proposed method are validated through a case study of the brake disc production line. Experimental results indicate that the proposed method outperforms other classification methods in brake disc quality prediction. Subject terms
Related work The current research on quality prediction methods is mainly divided into two categories: model-based prediction methods and data-driven prediction methods. The main difference between the two methods lies in whether the design of the controller is based on the system model or only on the I/O data. In other words, whether the design of the controller involves the dynamic model of the system or not. If the system model is involved in the design of the controller, it is a model-based prediction method; otherwise, it is a data-driven prediction method 19 . From this perspective, it can be concluded that certain prediction methods, such as those reliant on neural networks, fuzzy control prediction techniques, and various other intelligent control prediction methods, are founded upon data-driven predictive approaches 20 . Many scholars have conducted extensive research and exploration on quality prediction. Table 1 summarizes the relevant papers. The current research on quality prediction methods is mainly divided into two categories: model-based prediction methods and data-driven prediction methods. The main difference between the two methods lies in whether the design of the controller is based on the system model or only on the I/O data. In other words, whether the design of the controller involves the dynamic model of the system or not. If the system model is involved in the design of the controller, it is a model-based prediction method; otherwise, it is a data-driven prediction method 19 . From this perspective, some prediction methods based on neural networks, fuzzy control prediction methods, and many other intelligent control prediction methods are based on data-driven prediction methods 20 . Many scholars have conducted extensive research and exploration on quality prediction. Table 1 summarizes the relevant papers. From the existing research perspective, improving data acquisition and processing capabilities provides a foundation for data-driven quality control. It provides research ideas for the analysis of equipment operating data. This includes a model identification algorithm, proposing a multi-degree-of-freedom torsional vibration model for transmission systems, serving as a digital twin model for monitoring the remaining useful life of transmission system components 30 . Additionally, a method for predicting the quality of purifier carrier products is developed based on improved principal component analysis (PCA) and enhanced support vector machine (SVM). Other researchers have studied the mixed manifold learning and support vector machine algorithm based on optimized kernel functions (KML-SVM). They use support vector machines to classify and predict low-dimensional embedded data and optimize the kernel function of the support vector machine to maximize classification accuracy 31 . Using random forests for dimensionality reduction and analyzing key quality characteristics 32 . The principle of quality improvement in mechanical product development based on the Bayesian network can be used for the principle-empirical (P-E) model of quality improvement. It provides a method for learning the structure of the P-E model, and the quality characteristic (QC) relationship is determined by empirical data 32 , 33 . By analyzing the relationship between manufacturing resources and product quality status 34 , proposed a real-time quality control system (RTQCS) based on manufacturing process data, establishing the relationship between real-time product quality status and machining task processes 35 . A single-board computer and sensors were used to construct an edge device that can collect, process, store, and analyze data. Based on this, they developed a machine fault detection model using long short-term memory recurrent neural networks. Additionally, it is crucial to consider a real-time selection of the best model. In many cases, a simple probabilistic model can outperform more complex ones. Beruvides and colleagues achieved good drilling quality measurement and control results by employing the wavelet packet analysis method and fitting a statistical regression model 36 . Cruz and others proposed a two-step machine learning method for dynamic model selection, achieving favorable outcomes in predicting surface roughness during micro-machining processes and addressing complex cutting phenomena 37 . These scholars have significantly contributed to quality prediction, but there are also some issues. Firstly, on a stable production line, the quantity of qualified products far exceeds the number of faulty products (imbalanced product quality labels). Therefore, the quality prediction problem becomes an imbalanced data classification issue. Secondly, the equipment environment during the production process is complex, with numerous equipment process parameters affecting the quality characteristics of the processed products. Selecting equipment process parameters helps reduce the dimensionality of prediction models. Thirdly, some cloud-based quality prediction methods may result in issues such as delay, high broadband load, and resource limitations. To overcome these shortcomings, this paper initially introduces edge computing into product quality prediction to ensure shorter response times and higher reliability. Then, a method for selecting quality-correlated parameters is designed. Finally, addressing imbalanced data classification problems is achieved by employing the Synthetic Minority Oversampling Technique (SMOTE) and Extreme Gradient Boosting (XGBoost). The scientific-technical contribution of this article: Explored an edge computing-based framework for predicting the manufacturing quality of industrial products, offering guidance for flexible handling of industrial data. The proposed is an active control method for quality prediction using SMOTE-XGBoost based on joint optimization of hyperparameters, applied in predicting manufacturing quality for industrial products to address the imbalanced data classification issue within product quality prediction. The experimental results validated the superiority of the proposed method. Based on this paper’s proposed active control method for quality prediction, a selection and analysis of equipment process parameters for the brake disc production line was conducted using quality-correlated parameter selection, providing guidance and reference for the actual production and processing of brake disc products. For modern manufacturing, ensuring reliable industrial product quality has always been crucial in enterprise manufacturing process control. Guided by data-driven proactive quality control, modern manufacturing enterprises can gather vast amounts of industrial product manufacturing process data and apply it across various models. However, these models must operate at sufficiently high processing speeds to meet the practical production needs. Hence, the introduction of edge computing technology plays a pivotal role. Deploying models to the edge of the production line according to the actual industrial environment and establishing an edge-side IoT platform allows for more effective processing and application of. The remaining sections of this paper are organized as follows: section “ Industrial product manufacturing quality prediction frame work ” presents an edge computing-based framework for industrial product quality prediction. Section “ Active control method for quality prediction ” introduces a SMOTE-XGboost quality prediction active control method based on joint optimization hyperparameters. Following this, in section “ Case study ”, an experimental analysis of the processing quality of the brake disc production line is conducted based on the proposed quality prediction method, confirming the superiority of this approach and providing guidance and reference for actual brake disc production. Finally, section “ Conclusion ” provides a conclusion. Industrial product manufacturing quality prediction frame work This section constructed an edge-computing architecture for the industrial Internet of Things and analyzed the application methods of existing architectures. This explains the necessity of deploying industrial product quality prediction models using edge computing methods and introduces the quality prediction method proposed in this study. Industrial internet of things for industrial production lines To better manage the production line’s equipment operation status and product quality of the production line, and achieve real-timeproduct quality prediction, an industrial IoT architecture for the production line is established, as shown in Fig. 2 . This is the basis for implementing industrial intelligence services. This framework consists of four layers: perception layer, edge layer, central layer, and application layer. Quality prediction activecontrol method Figure 3 shows an example of industrial product manufacturing quality prediction based on edge computing. The data from the equipment side includes historical and real-time data and analyzes and describes its specific applications, while also analyzing the process parameter data during equipment operation. Historical data Historical data is mainly used for training the prediction model. The collected data is uploaded to the central layer through the perception layer for training the quality prediction model using machine learning algorithms. However, as the production process of products advances, the operating state changes over time. Therefore, the quality diagnosis and prediction model based on historical data is difficult to adapt to current production requirements. Some articles have also studied the update mechanism of predictive models 34 , 38 . The complexity of manufacturing systems has led to the development of prediction methods that combine historical data and real-time measurement data, which are in line with the characteristics of edge computing technology. Real-time data Real-time data collected by the perception layer is transmitted to the edge layer, which undergoes preprocessing operations on the real-time data. Filtering the collected real-time data based on quality characteristic and then using the prediction model deployed on the edge device to make real-time judgments on product quality. Simultaneously, the preprocessed data from edge devices is transmitted to the cloud center through the perception layer. As the data volume is reduced after preprocessing, it alleviates the bandwidth pressure and accelerates the transmission speed. For the received data, the central layer can update existing quality prediction models over time using incremental learning methods, addressing the issue of database updates in a time series. Equipment process parameters In existing research 15 , divided equipment process parameters into static process data, direct dynamic process data, and indirect dynamic process data based on their impact on the quality characteristics of the processed products to facilitate the application of equipment process parameters. Among them, static equipment process data refers to the type of equipment process data that generally does not change during the product processing process; direct dynamic process data refers to the equipment process data that changes dynamically during the product processing process, and the numerical changes directly reflect the product quality characteristics; Indirect dynamic process data refers to the equipment process data that changes dynamically during the processing process, but its changes do not directly reflect the product quality characteristics. Table 2 presents an example classification result of equipment process data 15 . Indirect dynamic equipment process data is the focus of this study. Introduction to the method A proactive control method for quality prediction based on historical data is proposed, comprising two components: quality prediction and proactive control. The Active control methods refer to calculating the difference between the actual qualified rate of the produced product and the predicted qualified rate of products. If this difference exceeds a certain threshold, the edge computing layer will generate corresponding process adjustment control instructions and send them to the relevant processing equipment. Figure 4 presents the workflow of this method. Firstly, indirect dynamic process data from production equipment is collected, and crucial quality-related parameters are computed using mutual information. These parameters are then selected based on their importance, followed by splitting the dataset into training and testing sets using stratified sampling. Subsequently, the SMOTE algorithm obtains a balanced dataset fed into the eXtreme Gradient Boosting (XGboost) for quality classification. Furthermore, a grid search method is applied for joint optimization of the hyperparameters of SMOTE and XGboost. Ultimately, the optimal quality prediction model is derived and utilized for product quality prediction. The details of this method are described in section “ Active control method for quality prediction ”. Case study This section takes the brake disc production line as an example to verify the practicality and effectiveness of the proposed method. The experiment consists of two parts: the selection of quality-related parameters and the classification results of the proposed method. Finally, the experimental results were analyzed. Experimental background The data was obtained from a brake disc production line in a certain enterprise, which is mainly used to provide high-quality brake disc products for CRH (China Railway High-speed), urban rail transit, locomotives, and world-leading railway trains. In recent years, with the demand for low energy consumption and lightweight trains, the brake disc production line has undertaken the trial production tasks of new aluminum-based silicon carbide brake discs and carbon-ceramic composite brake discs, realizing the flexible switching between mass production and trial processing to adapt to R&D innovation and new market demands. The brake disc is a component of the brake system that generates braking force to hinder the movement or motion trend of the vehicle. The surface of the brake disc requires high precision and must meet the qualified performance standards. The final quality inspection of the brake disc is tested by specialized magnetic particle inspection equipment and dynamic balancing equipment to determine whether it is qualified or not. This process takes a long time and the equipment is expensive. Therefore, using data-driven methods to predict the quality of brake disc products has the potential to replace specialized equipment, which can save equipment costs and inspection time. The entire production line of brake disc machining includes production and processing equipment, inspection equipment, and each equipment is equipped with a data collection gateway, which collects data to the edge-side server for data storage and computing power. The historical data stored in the edge server is uploaded to the private cloud, and the proposed quality prediction model is trained in the private cloud center. The trained model is then deployed on the edge server, and real-time unmarked data is transmitted to the edge server via protocols such as OLE for Process Control Unified Architecture (OPC UA). The data is preprocessed on the edge server, such as removing abnormal values, and then the quality label is obtained through the quality prediction model on an industrial computer.
Author contributions Formal analysis, L.L.; data curation, M.C.; writing original draftpreparat-ion, M.C.; supervision, Z.W.; format adjustment K.Z. All authors have read and agreed to the publi-shed version of the manuscript. Funding The support of this work by the National Natural Science Foundation of China (No. 51975386), Liaoning Province “Unveiling and Commanding” technology projects (2022020630-JH1/108), and Science and Technology Research and Development Program of China National Railway Group Corporation (N2022J014) are gratefully acknowledged. Data availability The datasets generated during and/or analysed during the current study are not publicly available due to [Information related to product processing] but are available from the corresponding author on reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 14; 14:1288
oa_package/9d/db/PMC10787841.tar.gz
PMC10787842
38219006
Introduction Online health consultation is a novel form of doctor-patient communication that has emerged alongside the rapid advancements in Internet technology 1 . Patients can employ mobile phones, computers and other terminal devices to engage in various communication modes, such as pictures, texts, voice or phone consultation on online health consultation platforms. Similarly, physicians can also choose an appropriate method to provide professional advice for patients 2 . In traditional offline health-seeking behavior, patients face limitations in accessing physicians’ information when choosing a healthcare provider 3 , because beyond readily identifiable information such as the hospital and professional title of physicians, obtaining nuanced information concerning personal characteristics and communication styles of physicians poses a frequent obstacle for patients. While online health consultation platforms can provide a wealth of information, such as physicians’ professional title, service attitude, communication styles, response speed and so on. This multifaceted information help patients form cognitive judgments and affective attitudes towards a physician, so as to make further decisions 4 , 5 . As mentioned above, in traditional offline health-seeking behavior, patients have limited access to physicians’ information 3 . This leads to a scenario where patients’ preferences for physicians tend to converge, resulting in a chaotic pursuit of top-tier hospital renowned physicians irrespective of major or minor ailments 6 . So now online health consultation platforms can provide comprehensive physicians’ information for patients 4 , 7 , do patients still exhibit uniform preference for physicians? That is, what specific type of physician information do patients rely on to make decisions when choosing a physician on online health consultation platform? Do they continue to prioritize cognitive stimulation information that can reflect the physician’s professionalism, such as the hospital and professional title of physicians 2 , or do they favor affective stimulation information, which reflects the physician’s care for the patient, such as the physician's attitude and communication skills 2 ? Thus, the primary objective of this study is to explore the preferences of patients for different types of physician information during the process of choosing physicians on online health consultation platforms. Previous studies have shown that the influence of information on individual decision-making behavior is regulated by environmental and contextual factors. These factors encompass the characteristics of users, products or services 8 , of which involvement is an important variable that modulates this influence relationship 9 – 11 . Involvement is originally a psychological concept, referring to the perceived significance and relevance that individuals attribute to specific entities 12 . Diverse manifestations of involvement emerge based on the objects under consideration, including but not limited to product involvement, shopping involvement, brand involvement and advertising involvement 13 . In the field of health decision-making, the importance and relevance of different disease types to patients exhibit distinctions. Therefore, this study defines different disease types as health involvement 14 . Results from previous studies indicate that patients’ behavior is modulated by health involvement, and patients with different disease types exhibit distinct preferences for physicians’ information. For example, Lu and Wu 8 found that the lower the risk of disease, that is, the lower the health involvement, patients’ choice behavior becomes more susceptible to affective stimulus information, such as physicians’ attitude. Thus the first objective of this study is to investigate the moderating effect of health involvement (disease type) on the patients’ preferences for physicians’ information on online health consultation platforms. However, it is evident that the same disease varying degrees of significance among distinct patients. For example, individuals deeply preoccupied with their health status or even have health anxiety 15 , they may perceive a common ailment such as a headache as a potentially grave threat to their well-being. This perception could subsequently influence their health information searching behavior and even health decision-making 16 , 17 . Conversely, individuals with lower levels of health anxiety may regard the same symptom as less severe, deeming it necessitating only minor recuperation. Consequently, during the follow-up experiments, deliberate efforts were made to minimize the potential influence of health anxiety on the experimental conditions pertaining to health involvement. Individual preference or adoption of different types of information is actually a process in which an individual makes a value evaluation on the received information and then makes decision according to the value ranking of different alternatives 18 – 20 . Hence, the neural mechanism of individual information value ranking process can refer to value-based decision theory. Value-based decision theory posits that the varied preferences and attitudes of individuals towards the same entity are termed subjective value. Subjective value enables individuals to integrate diverse and complex alternatives into a common dimension for comparative evaluation 21 , 22 . On the neural level, extensive research has shown that the ventromedial prefrontal cortex (VMPFC) and ventral striatum (VS) are crucial brain regions associated with subjective value. These regions exhibit notable activation during individual evaluations of subjective value 20 – 23 . Therefore, the second objective of this study is to reveal the neural mechanism of the moderating effect of health involvement on patients’ preferences for physicians’ information by analyzing the activation differences of VMPFC and VS brain regions when patients process information under different health involvement. According to the above, to achieve our two objectives in this paper, we first employed behavioral experiment to identify patients’ preferences for different types of physicians’ information (cognitive stimulation information and affective stimulation information) under different health involvement, and then used functional magnetic resonance imaging (fMRI) experiment to analyze the brain activation mechanism of patients under different health involvement from the neural level, thereby providing neural/physiological evidence elucidating the influence of health involvement on the information preferences of patients in online health consultation services.
Methods Manipulation check To divide the abundant physician information provided on the online health consultation platform into two types: cognitive stimulation information and affective stimulation information, this paper referred to the physician-related information provided on an actual online health consultation platform, and extracted seven types of physician attributes: (1) physician rank (indicating both the hospital rank and the physician’s professional title); (2) professional knowledge (signifying the specific expertise of the physician); (3) treatment effect; (4) service attitude; (5) communication skills; (6) response speed; (7) service commitment (the presence of physician on the online health consultation platform means that the physician promises to provide services for users online. The indicator of the number of replies in recent two weeks on the actual platform can reflect whether physicians have enough time and energy to provide services for users online, so we use service commitment to represent the number of replies in recent two weeks). 53 participants (M age = 28.98 years) were recruited via the sample service provided by the website wjx ( https://www.wjx.cn/ ). The participants were stratified into two groups: the high satisfaction group (experimental condition: physician information + 100% favorable rating, N = 26) and the low satisfaction group (experimental condition: physician information + 75% favorable rating, N = 27) to assess the seven physician attributes mentioned above. Participants were required to assess both cognitive and affective trust perceptions after reviewing a physician’s information. If the information about a physician differed solely in the cognitive trust scores of the participants, without affecting affective trust, it categorizes as cognitive stimulation information; otherwise it is considered affective stimulation information (the measurement scales see Supplementary 1. Table S2 ). Behavioral experiment Given the objective of this study is to identify the physician preferences of patients with different health involvement. 94 participants with prior experience in online health consultations were recruited for a mixed-design experiment using the sample service provided by the website wjx ( https://www.wjx.cn/ ). The participants were required to engage in this experiment by recalling their most recent online health consultation experience. Subsequently, participants were categorized into disease types corresponding to their perception of the severity of their previous consultation illness, distinguishing between mild and acute cases. Specifically, 48 participants were low health involvement in mild diseases, 46 participants were high health involvement in acute diseases. The mean age of the participants was 29.28 ± 7.84 years. During the experiment, participants rated the physician’s willingness to choose according to a combination of a cognitive stimulation information + favorite rating and an affective stimulation information + favorite rating. These sets of information were randomly drawn from the physician’s information selected from the manipulation check. Each participant encountered scenarios involving high cognitive trust & high affective trust (Hh: cognitive stimulation information + 100% favorable rating & affective stimulation information + 100% favorable rating), high cognitive trust & low affective trust (Hl: cognitive stimulation information + 100% favorable rating & affective stimulation information + 75% favorable rating), low cognitive trust & high affective trust (Lh: cognitive stimulation information + 75% favorable rating & affective stimulation information + 100% favorable rating), low cognitive trust & low affective trust (Ll: cognitive stimulation information + 75% favorable rating & affective stimulation information + 75% favorable rating) these four experimental conditions, corresponding to four different physicians. Participants were explicitly informed that the four physicians shared identical conditions, differing only in the ratings assigned to the experimental materials. Participants were then instructed to evaluate the willingness to choose for each physician based on the information provided by the experimental materials (the measurement scales see Supplementary 1. Table S2 ). Imaging experiment The experimental task in the imaging experiment was consistent with the behavioral experimental task. However, to generate a stronger BOLD signal in the imaging experiment, the low satisfaction rating was adjusted from 75% favorable rating to 25% favorable rating. The experimental paradigm employed in this study was rapid-presentation (jittered) event-related design, and the order and timing of events were generated with the help of optseq2 tool 35 . We recruited 33 undergraduate and graduate students as participants for a within-subject design experiment via online promotional channels, including WeChat groups and campus forums. One participant was excluded due to exceeding the specified head movement range of 2 mm, other 32 participants were included in the final analysis (M age = 23.97 ± 2.74 years). All the participants were right-handed, physically healthy, devoid of mental related diseases, and exhibited either normal or corrected vision. Upon completion of the experiment, each participant got RMB 150 yuan as a reward for their participation. Scanning procedure The participants first closed their eyes to receive a T1 structural image scan with a duration of about 6 min, and then performed two functional image scans with a duration of about 11 min. Participants were allowed to take a brief respite between each scan at their discretion. The two functional image scans were two tasks, corresponding to the two types of diseases with different health involvement. The imaging experiment was a simulation in which participants were required to hypothetically possess a specific disease. Drawing upon the severity and urgency of the disease, and building on the findings of Li et al. 36 , this study explicitly defined the disease under low health involvement experimental conditions as a cold, while designating acute abdominal pain for high health involvement conditions. In each task, participants were initially presented with an experimental instruction with a duration of 12 s, followed by a fixation point with a duration of 3 s, and then a formal experimental stimulus with a duration of 7 s. Each task contained 80 trails, including four experimental conditions: Hh, Hl, Lh, and Ll. Each condition repeated 20 times within a single task. Participants were instructed to provide keystroke responses during the duration of the formal experimental stimulus. Because the right-handed single-handed keypresses were used, the participants used the Likert 5-level scale to evaluate the physician’s willingness to choose in response to the experimental stimulus (where 1 indicated very unwilling, 5 indicated very willing). The experimental stimulus presentation program is E-prime, and the presentation procedures is shown in Fig. 3 . Data collection The imaging data was collected on a Siemens PRISMA 3T fMRI scanner at magnetic resonance imaging research center of Peking University. The scanning parameters of T1 structure images were: Voxel size = 0.5 mm × 0.5 mm × 1.0 m, Field of View = 256 mm, Slice thickness = 1.00 mm, TR = 2530.0 ms, TE = 2.98 ms, Flip angle = 7°, Matrix = 256 × 256. The functional image was scanned by echo planar imaging (EPI), and the scanning parameters were: Voxel size = 2.0 mm × 2.0 mm × 2.0 mm, Slices = 62, Field of View = 224 mm, Slice thickness = 2.0 mm, TR = 2000 ms, TE = 30.0 ms, Flip angle = 90, Matrix = 112 × 112. Ethics approval and consent to participate All study procedures and methods were performed in accordance with the relevant guidelines and regulations and approved by Ethical Review Board of School of Psychological and Cognitive Sciences, Peking University. Written Informed Consent was obtained from all participants before data collection.
Results Manipulation check results Through the manipulation checks, this paper selected four types of physician information from the seven types of physician information extracted on the online health consultation platform for formal experiment. The cognitive stimulation category included: physician rank (p < 0.001) and professional knowledge (p < 0.001), while the affective stimulation category encompassed service attitude (p = 0.012) and communication skills (p = 0.033). The results are shown in Table 1 . Behavioral results This paper used 2 (cognitive trust high/low) × 2 (affective trust high/low) behavioral experiment to explore the influence of cognitive trust and affective trust on patients’ willingness to choose under two different levels of health involvement (high/low) through two-way repeated ANOVAs, so as to identify patients’ preference to cognitive and affective stimulation information under different levels of health involvement. The results showed that health involvement had a significant impact on patients’ preferences for different types of physicians’ information. When health involvement was low, the interaction between cognitive trust and affective trust on participants’ willingness to choose was not significant (df = 47, F = 0.323, p = 0.572; Partial η 2 = 0.007). Conversely, under conditions of high health involvement, the interaction between cognitive trust and affective trust on participants’ willingness to choose was significant (df = 45, F = 13.049, p = 0.001; Partial η 2 = 0.225). The results are shown in Table 2 . Further post-hoc test found that (see Table 3 and Fig. 1 ), when the health involvement was low, both cognitive and affective stimulus information could influence patients' decision-making behavior. The results indicated that an enhancement in either cognitive trust or affective trust significantly heightened patients’ willingness to choose the physician. When health involvement was high, the isolated improvement of either cognitive stimulus information or affective stimulus information did not elicit any discernible impact on patients' decision-making behavior. The outcomes illustrated that enhancing either cognitive trust or affective trust independently did not significantly improve participants’ willingness to choose. Notably, it was only when both aspects were improved simultaneously that participants’ willingness to choose would be significantly improved. Imaging results To examine the activation differences of VMPFC and VS brain regions when patients process information under different health involvement conditions, this paper first used the general linear model (GLM) in SPM12 to generate the design matrix for the individual-level analyses (Specify 1st-level), calculated the task activation regions under four different experimental conditions of Hh, Hl, Lh, Ll for each participant, and then performed group-level analyses (Specify 2nd-level). On the group-level, we computed the contrast “CT [(Hh + Hl)−(Lh + Ll), AT [(Hh + Lh)−Hl + Ll)]” (for additional experimental parameters and design matrix used in the SPM analysis, see Supplementary 1. Fig. S1 ). Within our priori ROIs hypotheses-driven, we focused on VMPFC (6-mm sphere centered at the MNI coordinate: 2/46/− 16) 24 and VS (6-mm sphere centered at the MNI coordinate: − 16/6/− 12) 25 these two regions of interest (ROIs), extracted the contrast value of these two ROIs to analyze their differences in brain activation under different health involvement conditions. The results are shown in Fig. 2 (for whole-brain results, see Supplementary 1. Table S1 ), in the condition of low health involvement, activation in the VMPFC differed between CT contrast and AT contrast (t = 3.232, df = 31, p = 0.003; Cohen’s d = 0.572), CT contrast in the VMPFC was significantly activated (small volume FWE corrected at p = 0.001 and k = 26), but not AT contrast. Activation in the VS also differed between CT contrast and AT contrast (t = 2.800, df = 31, p = 0.009; Cohen’s d = 0.498), AT contrast in the VS was significantly activated (small volume FWE corrected at p = 0.003 and k = 17), while CT contrast was not significantly activated. In the condition of high health involvement, activation in the VMPFC (t = 0.762, df = 31, p = 0.452; Cohen’s d = 0.134) and VS (t = 0.699, df = 31, p = 0.489; Cohen’s d = 0.123) was not significantly different between CT contrast and AT contrast, and neither brain region was significantly activated in either contrast.
Discussion The results of behavioral experiment indicate that the preferences of patients in choosing physician online varied based on their levels of health involvement. When the health involvement was low, physicians with professionalism or good attitude can be preferred by patients. This study observed a noteworthy correlation between the improvement of cognitive trust or affective trust and patients’ willingness to choose physicians, which means that both cognitive stimulation information and affective stimulation information can influence the patient’s decision-making, and the effects of the two can be mutually substituted/compensated. When the health involvement was high, patients exclusively favored physicians possessing both good professionalism and attitude. This suggests that an increase in either cognitive trust or affective trust alone did not substantially improve patients’ willingness to choose. Instead, a significant improvement in the willingness to choose only occurred when both aspects were enhanced simultaneously. This underscores the crucial role of combining cognitive and affective stimulus information in influencing patients’ choice behavior. The results have also been confirmed at the neural level. Imaging experiment revealed that patients with low health involvement exhibited significant activation in the VMPFC when evaluating cognitive stimulation information and in the VS when evaluating affective stimulation information, and previous studies have demonstrated the close association of both VMPFC and VS with subjective value evaluation 20 – 23 . This implies that when health involvement is low, patients integrate cognitive and affective stimulation information within the same dimension for comparative evaluation to make decisions 21 , 22 . In essence, patients with low health involvement exhibit a preference for both cognitive and affective stimuli information. Conversely, when health involvement was high, the results indicated the absence of activation in related brain regions when patients evaluated cognitive and affective stimulation information. According to previous studies, the activation degree of related brain regions can serve as a reflection of cognitive resources allocated by individuals 26 . Therefore, the results of the imaging experiment may provide a theoretical explanation for the modulation of health involvement on patients’ preferences for different types of physicians’ information. According to the Elaboration Likelihood Model (ELM), a theoretical framework delineating individual information processing, two distinct routes exist for individuals when processing information, one is the central route and the other is the peripheral route 27 . When an individual uses the central route to process information, the individual's attitude will change after a comprehensive evaluation of the detailed information in a cognitive / rational way, which requires more cognitive resources. When an individual adopts the peripheral route, the individual will engage in a less refined processing of received information, allocating fewer cognitive resources 28 – 30 . This study employed imaging experiment to reveal the information processing processes of patients. The results indicated that when the health involvement was low, patients had a higher activation degree of brain regions, allocated more cognitive resources, which means they preferred to use the central route to process information. When the health involvement was high, patients allocated less cognitive resources and preferred to use peripheral route to process information. Although the results in this paper indicated that health involvement had an impact on patients' information processing routes, that is, patients with low health involvement used the central route, while those with high health involvement used peripheral route. However, this finding is inconsistent in the field of consumption decision-making. Previous studies have shown that when the product or service is more important to consumers (i.e., high involvement), they allocate more cognitive resources to evaluate information through the central route. Conversely, under low involvement, consumers resort to the peripheral route for information processing 31 – 33 . We posit that this discrepancy arises due to the distinct nature of health decision-making compared to product purchases. In the field of consumption decision-making, individuals freely choose their information processing route without restricting cognitive resources. However, in the field of health decision-making, the priority of the disease itself affects the patient's energy. When the disease is milder, patients possess more energy to comprehensively and carefully evaluate all obtained physician information when choosing physicians. Therefore, in instances of low health involvement, patients are more inclined to adopt the central route for information processing. Conversely, under high health involvement, the overwhelming nature of "care is chaos" may deprive patients of sufficient time and energy to comprehensively evaluate all information, leading them to rely on the peripheral route for decision-making. To sum up, this paper first found the impact of health involvement in the field of online health-seeking on patients' physician preferences through behavioral experiment, and then used imaging experiment to record the brain reaction process of patients’ decision-making to reveal the neural mechanism of this impact, provided neural/physiological evidence to support the derived conclusions. The findings expand the application of individual information processing theories in the field of health decision-making, and establish an effective connection between individual psychological function and decision-making behavior. Furthermore, the relevant conclusions can also provide insights to practical applications. From the behavioral results, it can be found that patients exhibiting diverse disease types employ distinct ways in processing physician information, thereby expressing preferences for physicians of different categories. Consequently, in practical applications, the platform has the potential to formulate a physician recommendation mechanism rooted in these observations. This mechanism could tailor recommendations based on the preferences of patients with specific disease types, fostering a targeted and efficacious alignment between patients and physicians. For example, with respect to patients with mild diseases, the platform can not only recommend physicians demonstrating high professional competence but also those general physicians skilled in cultivating affective trust with patients. This strategic approach mitigates the occurrence of a distorted phenomenon wherein all patients converge toward high-quality medical resources, thereby alleviating the strain on such resources and preventing the idle wastage of ordinary medical resources. Moreover, this approach bolsters the pivotal role of online health consultation platforms in the equitable distribution of doctors and efficient channeling of patient flows. Consequently, it contributes to the optimization of grassroots resource utilization and facilitates the judicious allocation of medical resources. The study has certain limitations. Firstly, all the experimental methods employed in this study were situational experiments. While situational experiments manage interference from irrelevant variables, a disparity exists between the experimental setting and the real-world scenario. Therefore, future research could employ field experiments or integrate virtual reality technology to enhance the fidelity of the experimental setting, bridging the gap between the experimental and real-world scenarios for more precise results. Secondly, as mentioned earlier, health anxiety can influence patients’ perception of health involvement. Although in the behavioral experiment, participants were categorized into disease types corresponding to their perception of the severity of their previous consultation illness, distinguishing between mild and acute cases. In the imaging experiment, we chose two clearly distinguishable diseases, cold and acute abdominal pain, to represent conditions with low and high health involvement, thereby mitigating potential influences. However, the control of health anxiety constitutes a limitation of this study. Future research should contemplate measuring participants’ health anxiety during the experiment to control this impact. In addition, the participants of the imaging experiment are mainly university students. Despite the predominant users of the online health consultation platform being young people 34 , the significant health demands of the aging population must not be overlooked.. To enhance the effective utilization of online health consultation services among the elderly, future research should prioritize investigating the behavior patterns of this demographic.
In traditional offline health-seeking behavior, patients consistently exhibit a preference for similar types of physicians due to limited access to physicians’ information. Nevertheless, with the advent of online health consultation platforms offering comprehensive physicians’ information for patients, raises the question: do patients continue to exhibit uniform preference for physicians? To address this issue, we first employed a behavioral experiment to discern patients’ preferences for different types of physicians’ information under different health involvement, and then conducted a functional magnetic resonance imaging (fMRI) experiment to furnish neural/physiological evidence. The results showed that health involvement modulates patients’ preferences, when health involvement was low, patients had diverse preferences for physicians, that is, different types of physicians’ information could individually impact patients’ choice and could serve as substitutes for each other. When health involvement was high, patients’ preference for physicians were uniform, highlighting that the collective influence of different types of physicians’ information on patients’ choice behavior. From the neural level, an explanation for the results was that the ventromedial prefrontal cortex (VMPFC) and ventral striatum (VS) brain regions, two key brain regions reflecting individual cognitive resource allocation, had different activation levels under different health involvement, indicating that patients allocated different cognitive resources. Subject terms
Supplementary Information
Supplementary Information The online version contains supplementary material available at 10.1038/s41598-024-51519-4. Acknowledgements This work was supported by the Open Project Funding from the Shanghai Key Laboratory of Brain-Machine Intelligence for Information Behavior, Shanghai International Studies University, Shanghai, China (2023KFKT007), National Natural Science Foundation of China (71874018, 71942003), and Science and Technology Research Program of Chongqing Municipal Education Commission (KJQN202300634). Author contributions Y.Z., Y.W. and H.R. participated in the research design and project implementation. Y.Z. and Y.W. participated in data collection and analysis. Y.W. and H.R. guided manuscript drafting. Y.Z. drafted the manuscript. All of the authors read and approved the final manuscript. Data availability The datasets used and/or analysed during the current study available from the corresponding author on reasonable request. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:00
Sci Rep. 2024 Jan 13; 14:1269
oa_package/f5/55/PMC10787842.tar.gz
PMC10787843
38218957
Introduction Advances in computational materials science have enabled the accurate prediction of novel materials possessing exceptional properties. Remarkably, these computational advancements have facilitated the successful experimental synthesis of materials that exhibit the anticipated properties. Some predicted materials, such as near-room-temperature superconductors, have been successfully synthesized under high-pressure conditions, with their superconducting temperatures in accordance with density functional theory (DFT) calculations 1 , 2 . To achieve accurate predictions, a priori knowledge of plausible molecular and crystal structures play a vital role in both theoretical and experimental studies. Several algorithms, such as evolutionary algorithms, swarm particle optimization, random sampling method, and etc., have been employed for structure prediction 3 – 5 . These algorithms rely on identifying local minima on the potential energy landscape obtained from, for example, DFT calculations and machine learning-driven methods 6 – 8 . In the case of crystal structures, where atoms are arranged in a three-dimensional space with periodic boundaries, additional criteria are necessary to enforce crystal symmetry constraints 5 . Recent approach for structure prediction employs denoising diffusion models to perform probabilistic inference. These models sample molecular and crystal structures from a probability distribution of atomic coordinates and types 9 – 12 , bypassing the computationally intensive DFT calculation that tediously determines the potential energy landscape. By leveraging sufficiently large datasets containing various compounds, this method enables the generation of diverse compositions and combinations of elements simultaneously. Furthermore, the models allow for the control of desired physical properties of the generated structures through conditional probability sampling 13 – 15 . These machine learning-based algorithms also hold promise for solving inverse problem efficiently, resolving structures from experimental characterizations, e.g., x-ray absorption spectroscopy and other techniques, a challenging problem in materials science 16 – 18 . There are two primary types of denoising diffusion models: score matching approach and denoising diffusion probabilistic models (DDPM) 19 – 21 . These two models can denoise (reverse) a normal distribution such that the distribution gradually transforms into the data distribution of interest. The score matching approach estimates the score function of the perturbed data directing the normal distribution toward the data distribution and employing large step sizes of variance. In contrast, DDPM gradually denoises the random noise through a joint distribution of data perturbed at different scales of variance. Both approaches have been utilized for generating molecular structures 10 – 12 . However, models based on DDPM tend to sample molecules with higher diversity and energy closer to the ground truth than models based on the score matching approach 11 . Since atomic positions in crystal structures are periodic and can be invariant under some rotation groups depending on their crystal symmetry, the core neural networks should favourably possess roto-translational equivariance 22 – 24 . Xie et al. 9 has proposed a model for crystal prediction by a combination between variational autoencoder (VAE) 25 and the denoising diffusion model, called crystal diffusion VAE (CDVAE). The model employs the score matching approach with (annealed) Langevin dynamics to generate new crystal structures from random coordinates 19 . The neural networks for an encoder and the diffusion model are roto-translationally equivariant. As a result, CDVAE can generate crystal structures with realistic bond lengths and respect crystal symmetry. Because of the periodic boundary condition imposed on the unit cell, gradually injecting sufficiently strong noises (in the forward process) to the fractional coordinates can lead to the uniform distribution of atomic positions at late times, the consequence of ergodicity in statistical mechanical sense. Rather than beginning with a Gaussian distribution and denoising it as in the original CDVAE formulation, Jiao et al. 26 perturbed and sampled atomic positions beginning with a wrapped normal distribution which satisfies the periodic boundary condition. With this approach, the reconstruction performance has been significantly improved. Other circular (periodic) distributions, e.g., the wrapped normal and von Mises distributions, are not natural for DDPM framework since there is no known analytical method to explicitly incorporate such distributions into the framework. There, one needs to resort to an additional sampling procedure to construct the DDPM 27 . In this work, we introduce a crystal generation framework called diffusion probabilistic CDVAE (DP-CDVAE). Similar to the original CDVAE, our model consists of two parts: the VAE part and the diffusion part. The purpose of the VAE part is to predict the lattice parameters and the number of atoms in the unit cell of crystal structures. On the other hand, the diffusion part utilizes the diffusion probabilistic approach to denoise fractional coordinates and predict atomic coordinates. By employing the DDPM instead of the score matching approach, the DP-CDVAE model shows reconstruction and generation task performances that are statistically comparable to those obtained from original CDVAE. Importantly, we demonstrate the significantly higher ground-state generation performance of DP-CDVAE, through the distance comparison between generated structures and those optimized using the DFT method. We also analyze the changes in energy and volume after relaxation to gain further insights into models’ capabilities.
Methods Diffusion probabilistic model In the diffusion probabilistic model, the data distribution is gradually perturbed by noise in the forward process until it becomes a normal distribution at late times. In this study, the distribution of the fractional coordinate ( ) is considered since their values of every crystal structures distribute over the same range ,i.e., . The Markov process is assumed for the forward diffusion such that the joint distribution is a product of the conditional distributions conditioned on the knowledge of the fractional coordinate at the previous time step: where the data distribution of the fractional coordinate, t is the discretized diffusion time step, T is the final diffusion time, is a noise schedule with a sigmoid scheduler 44 , and the conditional is a Gaussian kernel due to the Markov diffusion process assumption. Then can be expressed in the Langevin’s form through the reparameterization trick as where , and . This update rule does not necessitate to remain in ; however, we can impose the periodic boundary condition for the fractional coordinate so that Then, . In the reverse diffusion process, if the consecutive discretized time step is small compared to the diffusion timescale, the reverse coordinate trajectories can be approximately sampled also from the product of Gaussian diffusion kernels as where The reverse conditional distribution can be trained by minimizing the Kullback-Leibler divergence between and , the posterior of the corresponding forward process 21 . We use GemNetT for the diffusion network to train the parametrized noise 45 . Then, the coordinate in the earlier time can be sampled from , whose corresponding reverse Langevin’s dynamics reads where . Crucially, we empirically found that the final reconstruction performance is considerably improved when we impose the periodic boundary condition on the fractional coordinate at every time step such that and in the first term of Eq. ( 6 ) is replaced by . Namely, in our modified reverse process, the coordinate is sampled from An illustration of denoising atomic coordinates with Eq. ( 7 ) is demonstrated in Fig. 2 . The model performance using Eq. ( 6 ) is shown in Table S1 in SI, whereas the performance using Eq. ( 7 ) is shown in Table 1 . Graph neural networks Graph neural networks architecture facilitate machine learning of crystal graphs , graph representations of crystal structures. and are sets of nodes and edges, respectively, defined as where n and m are indices of atoms in a crystal structure, is a vector of M features of an atom in the unit cell, is a translation vector, and is a lattice matrix that converts a fractional coordinate into its atomic Cartesian coordinate . The atomic features, fractional coordinates, and atomic Cartesian coordinates of the crystal structure are vectorized (concatenated) as , , and . Three graph neural networks implemented in this work are DimeNet 28 , GINE 29 , and GemNetT 45 . DimeNet and GINE are employed for encoders, and GemNetT is used for a diffusion network. DimeNet and GemNetT, whose based architecture concerns geometry of the graphs, are rotationally equivariant. GemNetT has been devised by incorporating the polar angles between four atoms into DimeNet . This development grants GemNetT a higher degree of expressive power compared to DimeNet 46 . Furthermore, GINE has been developed to distinguish a graph isomorphism, but not graph geometry nor the distance between nodes, which is important for our study. Thus we supplement the edge attributes into GINE with the distances between atoms, i.e. . DP-CDVAE’s architecture The forward process of DP-CDVAE model is illustrated in Fig. 1 . The model is a combination of two generative models, which are VAE and diffusion probabilistic model. The pristine crystal structures consist of the fractional coordinate ( ), the lattice matrix ( ), ground-truth atomic type ( Z ), and the number of atoms in a unit cell ( ). For crystal graphs of the encoders, their node features are . The number of atoms in a unit cell is encoded through multilayer perceptron before concatenated with the latent features from other graph encoders. They are encoded to train and where is a learnable parameter of the encoders. The latent variables ( ) can be obtained by where . Then, will be decoded to compute the lattice lengths and angles, which then yield the lattice matrix ( ), , and . In the original CDVAE, is the probability vector indicating the fraction of each atomic type in the compound and is used to perturb Z by where is a multinomial distribution, is a one-hot vector of ground-truth atomic type Z , and is the variance for perturbing atomic types at time t , which is distinct from used for perturbing the atomic coordinates. Similar to the original CDVAE, is selected from the range of [0.01, 5]. For the diffusion network, the input structures are constructed from , , and where the (Cartesian) atomic coordinates at time t are computed by . These are then utilized by the crystal graphs for the diffusion network, whose node features are where is a Fourier embedding feature of (see SI). As proposed by Ho et al. 21 , we use the simple loss to train the model such that Since the diffusion model is trained to predict both and , so the loss of the diffusion network is where is the cross entropy loss, is a loss scaling factor, and where . In this work, t is randomly chosen for each crystal graph and randomly reinitialized for each epoch in the training process. The total loss in the trainig process is shown in Eq. S1 in SI. In the reverse diffusion process, we measure the model performance of two tasks: reconstruction and generation tasks. For the former task, is obtained from Eq. ( 8 ) by using the ground-truth structure as an input of the encoders. For the latter task, , which is then used to predict , , , and concatenate with the node feature of the crystal graph in the diffusion network. At the initial step, , is sampled from the highest probability of , and the final-time coordinate is obtained from sampling a Gaussian distribution, i.e. The coordinates can be denoised using Eq. ( 7 ), and the predicted atomic types are updated in each reversed time step by . DFT calculations The Vienna ab initio Simulation Package (VASP) was employed for structural relaxations and energy calculations based on DFT 47 , 48 . The calculations were conducted under the generalized gradient approximation (GGA), which is Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional, and the project augmented wave (PAW) method 49 , 50 . The thresholds for energy and force convergence were set to eV and eV/Å, respectively. The plane-wave energy cutoff was set to 800 eV, and the Brillouin zone integration was carried out on a k-point mesh of created by the Monkhorst-Pack method 51 , 52 .
Results The performances of DP-CDVAE models are herein presented. There are four DP-CDVAE models, differed by the choice of the encoder (see Fig. S1 in Supplemental Information (SI)). DimeNet has been employed for the main encoder for every DP-CDVAE models 28 . We then modify the encoder of DP-CDVAE to encode the crystal structure by two additional neural networks: a multilayer perceptron that takes the number of atoms in the unit cell ( ) as an input, and a graph isomorphism network (GINE) 29 . Their latent features are combined with the latent features from DimeNet through another multilayer perceptron. The is encoded such that the model can decode the accurately, and GINE encoder is inspired by GeoDiff 11 whose model is a combination of SchNet 30 and GINE which yields better performance. Three datasets, Perov-5 31 , 32 , Carbon-24 33 , and MP-20 34 , were selected to evaluate the performance of the model. The Perov-5 dataset consists of perovskite materials with cubic structures, but with variations in the combinations of elements within the structures. The Carbon-24 dataset comprises carbon materials, where the data consists of carbon element with various crystal systems obtained from ab initio random structure searching algorithm at pressure of 10 GPa 33 . The MP-20 dataset encompasses a wide range of compounds and structure types. Reconstruction performance The reconstruction performance is determined by the similarity between reconstructed and ground-truth structures. The similarity can be evaluated using StructureMatcher algorithm from pymatgen library 35 . The algorithm takes a pair of crystal structures and performs Niggli reduction to reduce their cells 36 , 37 . They are then compared by determining an average displacement between the two structures. If it falls within the error tolerence, the two structures are matched. The reconstructed and ground-truth structures are similar if they pass the criteria of StructureMatcher which are stol=0.5, angle_tol=10, ltol=0.3. Match rate is the percentage of those structures passed the criteria. If the reconstructed and ground-truth structures are similar under the criteria, the root-mean-square distance between their atomic positions is computed and then normalized by , where V is the unit-cell volume, and is the number of atoms in the unit cell. An average of the distances of every pair of structures ( ), computed using the StructureMatcher algorithm, is used as the performance metric. Table 1 presents the reconstruction performance of different models for three different datasets: Perov-5, Carbon-24, and MP-20. Note that Fourier-transformed crystal properties (FTCP) model is presented as a baseline model, which is based on VAE. It encodes and decodes both the real-space and reciprocal-space features of crystal structures 38 . For the Perov-5 dataset, the DP-CDVAE model achieves a match rate of 90.04%, indicating its ability to reconstruct a significant portion of the ground-truth structures. This performance is 7.48% lower than the CDVAE model but still demonstrates the effectiveness of our model. In terms of , the DP-CDVAE model achieves a value of 0.0212, comparable to the FTCP model 38 , but slightly higher than the CDVAE model. Similarly, for the Carbon-24 and MP-20 datasets, the DP-CDVAE model performs well in terms of both match rate and . It achieves match rates of 45.57% and 32.42% for Carbon-24 and MP-20, respectively. The corresponding values for Carbon-24 and MP-20 are 0.1513 and 0.0383, respectively, comparable to the CDVAE model. Regarding the DP-CDVAE+ model, the additional encoding of into the model leads to improved match rates for all datasets, with an increase of 2–5%. This enhancement can be attributed to the accurate prediction of . However, in terms of , only the Perov-5 dataset shows an improvement, with a value of 0.0149. On the other hand, for the Carbon-24 and MP-20 datasets, the values are higher compared to the DP-CDVAE model. For the DP-CDVAE+GINE and DP-CDVAE+ +GINE models, the additional encoding of GINE into the models leads to a substantial drop in match rates compared to the DP-CDVAE model, particularly for the Perov-5 and Carbon-24 datasets. In contrast, there is a moderate increase in the match rates for the MP-20 dataset. The values for the Perov-5 and Carbon-24 datasets are comparable to those of the DP-CDVAE model. However, for the MP-20 dataset, the is noticeably higher in the models with GINE encoder compared to the DP-CDVAE model. Overall, while the reconstruction performance of the DP-CDVAE model may be lower than the CDVAE model in terms of match rate, it still demonstrates competitive performance with relatively low . The match rate can be enhanced by additionally encoding the , but the performance is traded off by the increase in . Generation performance We follow the CDVAE model that used three metrics to determine the generation performance of the models 9 . The first metric is the Validity percentage, which encompasses two sub-metrics: Structural Validity (Struc.) with a criterion that ensures the distances between every pair of atoms are larger than 0.5 Å, and Compositional Validity (Comp.) with a criterion that maintains a neutral total charge in the unit cell. The second metric is called coverage (COV), which utilizes structure and composition fingerprints to evaluate the similarity between the generated and ground-truth structures. COV-R (Recall) represents the percentage of ground-truth structures covered by the generated structures. COV-P (Precision) represents the percentage of generated structures that are similar to the ground-truth structures, indicating the quality of the generation. The third metric is the Wasserstein distance between property distributions of generated and ground-truth structures. Three property statistics are density ( ), which is total atomic mass per volume (unit g/cm ), formation energy ( , unit eV/atom), and the number of elements in the unit cell (# elem.). A separated and pre-trained neural network is employed to predict E of the structures where the detail of the pre-training can be found in Ref. 9 . The first and second metrics are computed over 10,240 generated structures, and 1000 structures are randomly chosen from the generated structures that pass the validity tests to compute the third metric. The ground-truth structures used to evaluate the generation performance are from the test set. In Table 2 , the DP-CDVAE model achieves a validity rate of 100% for the Perov-5 dataset and close to 100% for the Carbon-24 and MP-20 datasets in terms of structure. The validity rate for composition is comparable to that of the CDVAE model. The DP-CDVAE model also demonstrates comparable COV-R values to the CDVAE model across all three datasets. Furthermore, the DP-CDVAE models with and/or GINE encoders exhibit similar Validity and COV-R metrics to those of the DP-CDVAE model. However, for COV-P, all DP-CDVAE models yield lower values compared to CDVAE. On the other hand, our models show significant improvements in property statistics. In the case of the MP-20 dataset, the DP-CDVAE models, particularly those with the GINE encoder, yield substantially smaller Wasserstein distances for , , and the number of elements compared to other models. For the Carbon-24 dataset, our models also exhibit a smaller Wasserstein distance for compared to the CDVAE model. Ground-state performance Another objective of the structure generator is to generate novel structures that also are close to the ground state. To verify that, the generated structures are relaxed using the DFT calculation where the relaxed structures exhibit balanced internal stresses with external pressures and reside in local energy minima. These relaxed structures are then compared with the generated structures to evaluate their similarity. In this study, we have chosen a set of 100 generated structures from each of CDVAE, CDVAE+Fourier, and DP-CDVAE models for relaxation where CDVAE+Fourier model is CDVAE model with Fourier embedding features of the perturbed coordinates. However, relaxation procedures for multi-element compounds can be computationally intensive. To address this, we have specifically selected materials composed solely of carbon atoms, using the model trained on Carbon-24 dataset. This selection ensures a convergence of the self-consistent field in DFT calculation. Moreover, in the relaxation, we consider the ground state of the relaxed structures at a temperature of 0 K and a pressure of 10 GPa since the carbon structures in the training set are stable at 10 GPa 33 . We here introduce a ground-state performance presented in Table 3 . The StructureMatcher with the same criteria as in the reconstruction performance is used to evaluate the similarity between the generated and relaxed structures. The relaxed structure was used as a based structure to determine if the generated structure can be matched. Four metrics used to determine the similarity are 1) match rate, 2) , 3) and 4) . The and represent the root mean square differences in volume and energy, respectively, between the generated structures and the relaxed structures in the dataset. In Table 3 , the DP-CDVAE model achieves the highest match rate and the lowest and . Although the CDVAE+Fourier model achieves the lowest , the DP-CDVAE model demonstrates the that is comparable to that of the CDVAE+Fourier model.
Discussion The DP-CDVAE models significantly enhance the generation performance, particularly in terms of property statistics, while maintaining comparable COVs to those of CDVAE. Specifically, for Carbon-24 and MP-20 datasets, the density distributions between the generated and ground-truth structures from DP-CDVAE models exhibit substantially smaller Wasserstein distance compared those of the CDVAE model (see Table 2 ). The of the DP-CDVAE model presented in Table 3 is significantly lower than that of the original CDVAE. This is corresponding to smaller Wasserstein distance of shown in Table 2 . The DP-CDVAE model also demonstrates significantly smaller than the original CDVAE. These suggest that our lattice generation closely approximates the relaxed lattice, while also achieving atomic positions that closely resemble the ground-state configuration. This could be an attribute of the DP approach that gradually learns perturbed coordinates, which in turn enhances the quality of sampled coordinates during the reverse process, much like its successful applications in image and molecular structure generation 11 , 21 , 39 . Additionally, the distribution of the number of elements in the unit cells is relatively similar to that of the data in the test set, particularly in the results from the models with GINE encoder. This could be attributed to the capability of GINE to search for graph isomorphism 40 . For the MP-20 dataset, the Wasserstein distances of the values generated by our models are notably lower. This suggests that the crystal structures we generate are more likely to have values within the specific range we are interested in. Hence, by selecting an appropriate training set, we can concentrate on structures with values falling within the synthesizable candidate range. Moreover, is the energy difference between the generated structures and their corresponding relaxed structures. The ground-state energy represents a local minimum that the generated structure is relaxed towards. A value of close to zero indicates that the generated structure is in close proximity to the ground state. In Table 3 , it can be observed that our model achieves the value of 400.7 meV/atom which is about 68.1 meV/atom lower than the of CDVAE. The mode of of our model is 64–128 meV/atom, which is lower than its root-mean-square value (see Fig. S2 in SI). Nevertheless, both the and the mode of exhibit relatively high values. In many cases, the formation energy of synthesized compounds is reported to be above the convex hull less than 36 meV/atom 41 – 43 . To obviate the need for time-consuming DFT relaxation, it is essential for the generated structures to be even closer to the ground state. Therefore, achieving lower values remains a milestone for future work.
The crystal diffusion variational autoencoder (CDVAE) is a machine learning model that leverages score matching to generate realistic crystal structures that preserve crystal symmetry. In this study, we leverage novel diffusion probabilistic (DP) models to denoise atomic coordinates rather than adopting the standard score matching approach in CDVAE. Our proposed DP-CDVAE model can reconstruct and generate crystal structures whose qualities are statistically comparable to those of the original CDVAE. Furthermore, notably, when comparing the carbon structures generated by the DP-CDVAE model with relaxed structures obtained from density functional theory calculations, we find that the DP-CDVAE generated structures are remarkably closer to their respective ground states. The energy differences between these structures and the true ground states are, on average, 68.1 meV/atom lower than those generated by the original CDVAE. This significant improvement in the energy accuracy highlights the effectiveness of the DP-CDVAE model in generating crystal structures that better represent their ground-state configurations. Subject terms
Supplementary Information
Supplementary Information The online version contains supplementary material available at 10.1038/s41598-024-51400-4. Acknowledgements This research project is supported by the Second Century Fund (C2F), Chulalongkorn University. This Research is funded by Thailand Science Research and Innovation Fund Chulalongkorn University and National Research Council of Thailand (NRCT): (NRCT5-RSA63001-04). T.C. acknowledges funding support from the NSRF via the Program Management Unit for Human Resources and Institutional Development, Research and Innovation [grant number B05F650024]. The authors acknowledge high performance computing resources including NVIDIA A100 GPUs from Chula Intelligent and Complex Systems Lab, Faculty of Science, and from the Center for AI in Medicine (CU-AIM), Faculty of Medicine, Chulalongkorn University, Thailand. We acknowledge the supporting computing infrastructure provided by NSTDA, CU, CUAASC, NSRF via PMUB [B05F650021, B37G660013] (Thailand). URL:www.e-science.in.th. The Computational Materials Physics (CMP) Project, SLRI, Thailand, is acknowledged for providing computational resource. Author contributions T.P and N.C. implemented the methodology into the code. T.P., N.C., R.A., and C.A. designed the project and contributed to discussion of the results. S.K. performed the DFT calculations. T.P., N.C., R.A., and T.C. contributed to discussion of the theory and the algorithm. T.P. and T.C. wrote the paper. T.B. supervised all aspects of the project. Data availability The code and datasets generated and/or analysed during the current study are available at https://github.com/trachote/dp-cdvae . Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:01
Sci Rep. 2024 Jan 13; 14:1275
oa_package/f6/42/PMC10787843.tar.gz
PMC10787844
38218925
Introduction With the continuous improvement of the global economy, the prevalence of chronic kidney disease (CKD) continues to elevate which has emerged as a severe public health problem, ≈15% of people suffer from CKD 1 , 2 . CKD is typically characterized by renal interstitial fibrosis and excessive extracellular matrix deposition contributing to the progressive loss of renal function 3 . Unfortunately, existing effective clinical treatments only alleviate the process of renal fibrosis but do not complete the cure for the disease 4 . Therefore, developing a method for anti-renal fibrosis has very important clinical significance 5 . Numerous studies have reported that Smad2/3 positively regulates cell fibrosis-like change and inflammation in renal interstitial fibrosis 6 , 7 . In addition, expression of Smad2/3 in the nucleus can aggregates in the renal tubular cell and accelerate CKD progression 8 , 9 . Therefore, the key to anti-fibrosis is to regulate the expression of Smad2/3 protein. A recent study revealed that carboxyl terminus of Hsp70 interacting protein (CHIP) as an E3 ubiquitin ligase, may directly participate in regulating the stability of Smad2/3 protein and inhibiting tumorigenesis 10 , 11 . This finding suggests that increasing the activity of CHIP in the renal tubular cells to promote the clearance of Smad2/3 may be important for RIF therapy. However, CHIP ligase was susceptible to biochemical and physical instability, which may induce to inhibit Smad2/3 activity weakness and lead to decrease the therapeutic efficacy 12 . Moreover, the low enrichment in the renal for CHIP-mediated ubiquitin degradation significantly limits their research and the future clinical transformation applications for RIF therapy 13 . The key to solving problems is to improve CHIP activity and selectively rich in the injured kidney tissue to promote Smad2/3 ubiquitination and degradation. Extracellular vesicles (EVs) such as microvesicles, are double membrane microparticles (60–1000 nm in size) secreted by cells in a constitutive or inducible manner 14 , 15 . The released EVs naturally function as intercellular messengers by selecting transporting nucleic acids and proteins to distal or nearby recipient cells 16 . A large number of studies have shown the potential of using EVs as mighty and feasible nanocarriers for drug delivery vector in various situations, from tumor therapy to gene regeneration therapy 17 , 18 . Compared with current delivery systems, EVs have a unique advantage in the natural origin, which enables them to escape phagocytosis, extend the half-life of therapeutic agents, and low immunogenicity 19 . In addition, magnetic targeted therapy has become a novel research hotspot in recent years, and magnetic nanoparticles have small size and magnetic guided properties 20 , and targeted migration of tissue damage sites under the action of magnetic fields has broad application prospects 21 . Engineered EVs have shown great potential as a promising therapeutic platform and have attracted considerable attention in tissue regeneration 22 . EVs could be engineered to overexpress various proteins in targeting tissues and active signals pathway for the regulation of fibrosis like cells 23 . Therefore, we use superparamagnetic iron oxide nanoparticles to modify engineering EVs loaded CHIP delivery system to enhance anti-fibrosis therapeutic effect for the effectively treatment of CKD. Here, we provide the valid method for SPION decorated CHIP high-expressing MSC-EVs in CKD renal fibrosis treatment. We found that SPION-EVs-CHIP showed great ability to target injury renal in unilateral ureteral obstruction (UUO) rat. More importantly, SPION-EVs-CHIP significantly reversed collagen deposition by inducing Smad2/3 ubiquitination and degradation of renal tubular cells and inhibiting tubular damage-mediated inflammatory responses as compared to MSC-EVs. Our EV-engineering technology demonstrated that SPION-EVs with CHIP overexpression provide a potential platform for effective renal interstitial fibrosis therapy by inducing Smad2/3 degradation.
Methods Ethics statement for human samples and animal models Written informed consent was obtained from healthy donors, and discarded human umbilical cord tissues were acquired according to the approved protocol of the Institutional Review Board (IRB) at the Affiliated Hospital of Jiangsu University. And conducted in accordance with ethical principles of the World Medical Association Declaration of Helsinki. All animal experimental protocols were approved by the Animal Care and Use Committee of Jiangsu University (Approval number: 2020280). All animal experiments were performed in accordance with the associated relevant guidelines and regulations for working with live vertebrate animals. Animals Sprague-Dawley rat (male, 6–8 weeks old, 180–200 g) were purchased from the Experimental Animal Center of the Jiangsu University. And db/db mice (female, 6–8 weeks old, 18–22 g) were purchased from the Cavens Experimental Animal Co., Ltd (Changzhou). They were provided free access to pellet food and kept water in plastic cages at 20 ± 2 °C and kept on a 12 h light/dark cycle in specific pathogen-free facilities. The experimental endpoints the animals were anesthetized euthanasia with pentobarbital sodium (Sprague-Dawley rat: 50 mg/kg, db/db mice: 30 mg/kg, Sigma-Aldrich) by intraperitoneal injection. Kidneys were fixed with 4% formaldehyde, embedded in paraffin, and sectioned to 4 μm thickness. Cell culture The rat renal tubular cells line NRK-52E was purchased from National Collection of Authenticated Cell Cultures. NRK-52E cells were maintained in DMEM (GIBCO, USA) containing 10% fetal bovine serum, 100 U mL −1 penicillin, and 100 mg mL −1 streptomycin in 5% CO 2 at 37 °C. Mesenchymal stem cells were isolated from human umbilical cord according to a previously described method. In brief, take the umbilical cord from a full-term newborn, wash with PBS and cut off the arteries and veins. Cut it into 2 mm sized tissue blocks and stick them on a 6-well plate with an interval of 5 mm. Add basic DMEM (GIBCO, USA) medium containing 10% fetal bovine serum and 100 U mL −1 penicillin, and 100 mg mL −1 streptomycin in 5% CO 2 at 37 °C. Thereafter, the medium was refreshed every 3 days and continuous culture until the purity of mesenchymal stem cells reached 85%. Extraction and Purification of MSC-EVs-CHIP A lentivirus vector encoding murine CHIP was purchased from genepharma (Suzhou, China). Mesenchymal stem cells were transfected with lentivirus encoding CHIP gene. When the transfection efficiency reached 80% measured by western blotting and flow cytometry analysis. The EVs in medium and FBS were depleted by ultracentrifugation. The medium was replaced and collected after 48 h. MSC-EVs were isolated and purified by ultracentrifugation 33 . In brief, the medium was centrifuged at 800 g for 10 min, 2000g for 10 min, 10 000 g for 30 min, 100,000 g for 70 min at 4 °C, to obtain MSC-EVs. The specific marker expression of CD63, CD9, CD81, and the CHIP expression was measured by western blotting. Biological characterization analysis of MSC-EVs-CHIP Prepared MSC-EVs and MSC-EVs-CHIP samples were pipetted onto a 300 meshes copper and incubated for 3 min. 20 μL of 1% uranyl acetate was pipetted onto the mesh copper and incubated for 1 min, washed and dry for TEM. The images were acquired using transmission electron microscope. MSC-EVs size determination by dynamic light scattering (DLS), in brief the volume of MSC-EVs and MSC-EVs-CHIP samples was increased to 1 mL with PBS and loaded into a quartz cuvette. The size distribution of EVs was measured by dynamic light scattering at 25 °C. Nanometer tracking analyzer detected its zeta potential and nanoparticle size. Renal interstitial fibrosis model construction and treatment RIF was established in Sprague-Dawley rats, anesthetize with inhalation of 2% isoflurane, expose the left ureter through a lateral incision, and ligated with double straps and surgical thread. The rats were randomly assigned into five experimental groups: Control, UUO, MSC-EVs, and MSC-EVs-CHIP, SPION-EVs. Rats were ureteral obstruction for 2 weeks, and intravenously injected with MSC-EVs, MSC-EVs-CHIP and SPION-EVs (2 mg per rat/every three days). After 2 weeks of treatment, the renal and blood serum were collected and analyzed renal function. The biodistribution of MSC-EVs-CHIP and SPION-EVs was performed a PerkinElmer IVIS Lumina II. In brief, MSC-EVs-CHIP and SPION-EVs was labeled with CM-DIR, and the free CM-DIR was washed thrice with cold PBS by ultracentrifugation for 3 times with cold PBS. Then, the CM-DIR-labeled EVs were injected intravenously into rat for 12 h and the CM-DIR signal in various tissues by PerkinElmer IVIS Lumina II. RNA extraction and RT-qPCR Trizol reagent was applied to extract total RNA from cultured cells or tissues in strict accordance with the instructions provided on the kit, followed by the determination of RNA concentration. mRNA was reverse-transcribed to cDNA and subjected to quantitative PCR, which was performed with the BioRad CFX96 Real-Time PCR Detection System (BioRad, USA). mRNA expression was compared using the 2 −ΔΔCt relative quantification method, GAPDH was used as the endogenous control. Western blot analysis Renal tubular cells and kidney tissues were lysed using lysis buffer and the obtained protein lysates were quantitated by BCA assay, then degenerated at 100 °C for 5 min. Proteins were resolved by sodium dodecyl sulfate-polyacrylamide gel electrophoresis, transferred onto polyvinylidene fluoride membranes and blocked with 5% bovine serum albumin. Membranes were immunoblotted with primary rabbit polyclonal antibodies to alpha-smooth muscle actin (α-SMA; 1:500, AB32575), Mothers against decapentaplegic homolog 2 (Smad2/3; 1:1000, CST3102), Fibronectin (FN, 1:1000, CST26836), Collagen I (Collagen I, 1:1000, CST72026) and incubated overnight at 4 °C with constant shaking. Next, the primary antibody-incubated membranes were washed five times with washing buffer and incubated with HRP-coupled secondary antibody for 1 h at room temperature. Protein bands were visualized by western blotting. Band intensity was quantified using the Image J software. The relative expression of the target protein was normalized to the band intensity of β-actin. We have included original western blot chemiluminescent images with corresponding light micrographs showing molecular weight markers for all western blots in Supplementary Fig. 10 . Histologic analysis Renal tissues samples were fixed in 10% formalin solution, embedded in paraffin and sliced into 5μm thick sections. After deparaffinization and rehydration, sections were stained with hematoxylin and eosin (HE), Sirius red and Masson. The area at the junction of the cortex and medulla in the renal section was selected for histological analysis. HE staining was used to assess tubular injury, Sirius red and Masson staining was performed to evaluate the deposition of collagenous fibers in the renal interstitium. The positive area (red) of Sirius red and (blue) Masson staining was quantified using Image J. Immunohistochemistry and Immunofluorescence Paraffin-embedded tissues were heat-fixed, deparaffinized, rehydrated, antigen retrieved, and subsequent process. First, renal tissue sections were heat-fixed for 2 h at 60 °C and deparaffinized in xylene for 30 min, then rehydrated in 100%, 75%, and 50% ethanol respectively for 10 min. Antigen retrieval was performed using 10 × 10 −3 M citrate antigen retrieval solution under high pressure for 30 min. The sections were incubated with 3% H 2 O 2 for 10 min in the dark, blocked with 10% BSA for 1 h and stained with Smad2/3 and Fibronectin antibodies overnight at 4 °C, and washed three times with PBS. 50 μL HRP-labeled goat anti-mouse/rabbit Ig mixture was added to sections for 30 min and washed with PBST thrice. 50 μL DAB reagent was added for color reaction and restained with hematoxylin for 30 s. Finally, the sections were soaked in 50%, 75%, 100% ethanol, and xylene for 10 min respectively to dehydrate. The images of samples were acquired using Leica fluorescence optical microscope. The renal sections after deparaffinized and rehydrated, then stained with Smad2/3, α-SMA, CD66, and CHIP, Slc5a12 antibodies overnight at 4 °C, and washed thrice times with PBS. The sections were stained with DAPI for confocal laser scanning microscopy (GE, USA). NRK-52E cells were washed twice with cold PBS. Then cells were fixed in 4% paraformaldehyde (PFA) for 10 min, blocked with 10% BSA for 1 h at room temperature, and stained with CHIP, Smad2/3, and α-SMA antibodies overnight at 4 °C. The cells were then washed thrice times with cold PBS, and stained with DAPI for confocal laser scanning microscopy (GE, USA). Statistical analysis All the experiments were randomized and blinded. All studies were performed in at least three independent experiments with each experiment including triplicate sets in vitro, or six animals per group in vivo. All data are presented as the mean ± SEM. GraphPad Prism 8.0 software (San Diego, CA, USA) was used for the statistical analysis. Western blotting and immunofluorescence analyses were performed using the Image J software. Data among multiple groups were compared using one-way analysis of variance, followed by a post hoc correction using Tukey’s test. a student’s t-test was used to test the difference between two groups. A value of * p < 0.05 was indicative of a statistically significant difference.
Results Characteristics of MSC derived extracellular vesicles with CHIP high expression Umbilical derived MSCs with high purity were isolated from rat according to previous protocols (Supplementary Fig. 1 ). We next infected MSCs with lentivirus encoding the CHIP gene and selected with puromycin to obtain CHIP high-expressing MSCs, and confocal graph shows that CHIP is mainly expressed in the cytoplasm (Fig. 1a, b ). The morphology of MSCs was not affected by high CHIP expression levels (Fig. 1c ), lipogenesis and osteogenesis experiment slightly induced stem-cell differentiation compared to non-transfected MSCs (Supplementary Fig. 2 ). Then MSCs derived EVs were then isolated and purified according to previously described protocols. Transmission electron microscopy (TEM) imaging shows then cup shaped vesicle morphology of MSC-EVs-CHIP (Fig. 1d ). And the average particle diameter of MSC-EVs-CHIP measured by dynamic light scattering (DLS) is 108 ± 10.6 nm (Fig. 1e ). The zeta potential of MSC-EVs-CHIP was ≈−15 mV (Fig. 1f ). Finally, we confirmed the expression of CHIP in MSCs-EVs by western blotting assay. The results showed that CHIP was also highly expressed in engineered EVs containing EVs-associated proteins, such as CD63, CD9, Alix and Calnexin was expressed negatively (Fig. 1g ), and GFP signals was also found in EVs-CHIP. These results suggested that the genetically engineered MSC-EVs with high CHIP expression were successfully prepared. MSC-EVs-CHIP reduced fibrotic change by inducing Smad2/3 degradation in renal tubular cells Evidence suggests that activation of the Smad2/3 pathway and inflammatory infiltration exacerbate fibrosis in CKD. To test whether MSC-EVs-CHIP induces Smad2/3 degradation and reduce renal interstitial fibrosis, MSC-EVs with GFP-tagged CHIP high expression were incubated with renal tubular cells NRK-52E. As shown by confocal imaging, a high GFP fluorescence signal was observed in NRK-52E (Fig. 2a ). Imaging flow cytometry showed that GFP MSC-EVs-CHIP can be internalized by NE-52E cells (Fig. 2b ). Western blot experiment confirmed high expression of CHIP in NRK-52E (Fig. 2c ). Furthermore, the GFP signal was colocalized with Smad2/3 (Red) in the cells, suggesting that CHIP shifted into the cytoplasm with MSC-EVs (Fig. 2d ). Next, we used fibrosis stimulating factor TGF-β 1 (10 nM) to induce fibrosis-like change in NRK-52E cells and establish in vitro RIF model. In our experiment, it was found that TGF-β 1 stimulates α-SMA overexpress and transform into myofibroblasts cells (Supplementary Fig. 3a ), MSC-EVs-CHIP significantly inhibited Smad2/3 and fibrotic of TGF-β 1 incubated cells, as shown by the confocal microphage and western blot (Fig. 2e, f ). Additionally, Co-immunoprecipitation observed that MSC-EVs-CHIP promoted ubiquitin molecules expression and decreased the expression of Smad2/3 in TGF-β 1 -incubated cells (Fig. 2g ). In order to simulate the mechanical pressure environment in vivo by using stiff gel (30Kpa) to stimulate NRK-52E (Fig. 2h , Supplementary Fig. 3b ), confocal microscopy also showed that the number of Smad2/3 in nucleus was decreased obviously in NK-52E cells treated with MSCEVs-CHIP relative to that in controls (Fig. 2i ), and statistical analysis confirms the results (Fig. 2j ). These results demonstrated that MSC-EVs-CHIP could alleviate TGF-β 1 and stiff gel induced renal tubular cell fibrotic like change by promoting Smad2/3 ubiquitination degradation. MSC-EVs-CHIP migrated to the kidney and ameliorated inflammatory infiltration by reduced Smad2/3 accumulation MSC-EVs-CHIP have been shown to promote Smad2/3 degradation, anti-fibrotic-like change of renal tubular cell, all of which play a critical role in the development and procession of RIF. We further evaluated the therapeutic efficacy of MSC-EVs-CHIP in RIF rat model established by unilateral UUO rat model. First, we assessed the organizational distribution of MSC-EVs-CHIP. CM-DIR labeled MSC-EVs and MSC-EVs-CHIP were injected into rat intravenously for 48 h, then the organ fluorescence was evaluated by PerkinElmer IVIS Lumina II ex vivo imaging. As shown in Figure S4 , MSC-EVs and MSC-EVs-CHIP was enriched in liver, hardly enriched in the kidney of control groups, while accumulated in kidney of UUO treated rat (Supplementary Fig. 4 ). In addition, the fluorescence imaging of the kidney tissue of RIF-rat receiving MSC-EVs-CHIP exhibited significant CM-DIR signals relative to the control (Supplementary Fig. 5a ). Western blotting analysis further showed that MSC-EVs treatment resulted in a high CHIP level in kidney tissue (Supplementary Fig. 5b ). These data suggested that under RIF pathological conditions, MSC-EVs-CHIP have an ability to target the injury kidney. We next investigated whether MSC-EVs-CHIP promote smad2/3 degradation and inhibits inflammation in the kidney of RIF model. Rats were raised for 7 days after unilateral ureteral obstruction, and were then intravenously treated with MSC-EVs or MSC-EVs-CHIP (2 mg per rat/every three days). After one week, we performed immunofluorescence analysis of the renal tissue. In UUO rats, Smad2/3 is highly expressed in the nucleus of renal tubular epithelial cells, and fibrosis-related proteins (Collagen I, α-SMA) are significantly increased (Supplementary Fig. 6 ). While the renal of rat receiving MSC-EVs-CHIP, we observed a significantly reduced expression of Smad2/3 in renal tubular nucleus compared to MSC-EVs treatment, suggesting the downregulation of Smad2/3 in the renal tissue of RIF rat (Fig. 3a ). In addition, the fibrosis related indicators α-SMA and Fibronectin in the kidneys of RIF rats receiving MSC-EVs-CHIP were significantly reduced (Fig. 3b ). We further observed the reduced Smad2/3 accumulation and tubular cells fibrotic like change in the renal of MSC-EVs-CHIP-treated rat by immunofluorescence analysis, suggesting that MSC-EVs-CHIP can significantly rescue RIF pathology by promoting Smad2/3 degradation and recovering normal homeostatic processes (Fig. 3c, d ). Given the critical role of inflammation in the progression of RIF, its effect on inflammation in renal was also evaluated after MSC-EVs-CHIP intervene. Persistent mechanical damage can stimulate the inflammatory response in tubular cells, especially the CD66 positive neutrophil accumulation, as shown in the HE staining and immunofluorescence (Fig. 3e–g ). Encouragingly, the induced inflammation and cytokine (IL-6, IL-1β, TNF-α) release was remarkably inhibited by MSC-EVs-CHIP treatment (Fig. 3h ). These data suggested that CHIP overexpression mediated by MSC-EVs inhibited Smad2/3 and decreased fibrosis index, further remitted the inflammation in RIF-rat. MSC-EVs-CHIP reduced interstitial collagen deposition in UUO model Inject the prepared MSC-EVs-CHIP and MSC-EVs into the tail vein of the UUO model, and collect large mouse notebooks from each group on the 15th day. The biochemical analyzer detected renal function indicators such as creatinine and urea nitrogen, and found that the MSC-EVs-CHIP group significantly downregulated creatinine and urea nitrogen compared to the MSC-EVs group, significantly improving renal function (Fig. 4a, b ). The appearance of kidneys in the UUO group significantly larger and lighter in color, while the kidney tissue significantly returned to control after MSC-EVs-CHIP treatment (Fig. 4c ). Further through the HE staining results of renal tissue pathology showed that the UUO injury group had an increase in tubular vacuolar degeneration, a significant thickening of the glomerular basement membrane, and a tissue damage score of 70%. Compared with the MSC-EVs group, the MSC-EVs-CHIP group restored normal renal tissue structure after intervention, and reduced tubular vacuolar degeneration. The damage score was reduced to 18% (Fig. 4d, e ). The results of Sirius red staining and Masson staining showed significant fibrosis in UUO renal tissue. The change area reached 64%, while the MSC-EVs-CHIP group showed a significant reduction in renal interstitial collagen fibers and a significant decrease in fibrosis degree after treatment, which was better than the MSC-EVs treated group alone (Fig. 4f-h ). Extract kidney tissue proteins from each group for Western blot detection, displaying fibrosis related proteins Fibronectin, Collagen I, and α-SMA were significantly reduced (Fig. 4i ). The above research results indicate that high expression of CHIP enhances MSC-EVs in improving renal function and reducing renal interstitial collagen. In vivo renal-targeting and antifibrotic therapeutic effect of SPION-EVs In order to further enhance the anti-fibrotic effect of MSC-EVs, we used the SPION with surface modified transferrin (Tf) to combine with the transferrin receptor on the surface of MSC EVs CHIP for engineering modification to enhance its ability to target the injured site. Scanning electron microscope observation shows that the size of the nanospheres is about 5 nm, and dynamic light scattering shows that the size of the nanoparticle is about 10 nm (Fig. 5a ). The superparamagnetic modification of transferrin binds to the transferrin receptor on the surface of the MSC-EVs-CHIP membrane after co incubation at 4 °C for 4 h, and SPION-EVs are collected and dissolved in PBS by external magnetic field adsorption (Fig. 5b ). The elliptical vesicle structure combined with multiple SPION was observed by transmission electron microscopy (Fig. 5c ). Analyze the stability and magnetic characteristics of SPION-EVs in aqueous solutions using hysteresis loops and PDI polymer dispersion index (Fig. 5d, e ). The average particle size of SPION-EVs measured by dynamic light scattering (DLS) is 115 ± 11.3 nm, compared to MSC-EVs-CHIP, the average particle diameter increases by about 7 nm, indicating that SPION not obvious change the size of extracellular vesicles (Fig. 5f ). The zeta potential of SPION-EVs was ≈−16 mV (Supplementary Fig. 7 ). Western blot was used to detect the expression of surface positive marker proteins CD9, CD81, Alix and negative marker Calnexin in SPION-EVs, and strong positive expression of CHIP and transferrin receptor TfR (Fig. 5g ). This indicates that we have successfully prepared engineered SPION-EVs. Construct the renal fibrosis treatment model using prepared engineered SPION-EVs under external magnetic field intervention (Fig. 5h ). By using a small animal live imaging system, it was shown that under the action of an external magnetic field, the signal intensity of DIR labeled SPION-EVs was enriched by targeting the liver to the kidneys (Fig. 5i ). The pathological tissue showed a decrease in tubular vacuolar degeneration and a return to normal structure in the SPION-EVs intervention group. Sirius red and Masson staining confirmed that compared to the MSC-EVs -CHIP group, the SPION-EVs group showed a decrease in interstitial collagen fibers and a significant decrease in fibrotic areas (Fig. 5j ). Immunofluorescence showed that SPION-EVs significantly inhibited the nuclear uptake of Smad2/3 in renal tubular cells and decreased α-SMA was superior to the MSC-EVs-CHIP group (Fig. 5k ). The above results indicate that the targeted enrichment of engineered SPION-EVs under magnetic field significantly enhances the anti-fibrotic effect. SPION-EVs attenuated renal interstitial fibrosis in the DKD model To further prove that SPION-EVs is a promising nanoplatform for the delivery of CHIP and treatment of renal fibrosis, we constructed another renal interstitial fibrosis model in diabetic kidney disease (DKD) rat. The DKD injury model was established by feeding high-fat combined with streptozotocin (STZ) for 18 weeks, intervened by tail vein injection of SPION-EVs. Firstly, we assessed the alleviation of SPION-EVs against interstitial fibrosis in DKD rat. As expected, histological analysis of kidney sections at 24 weeks after STZ injection revealed significant tubulointerstitial damage, including tubular atrophy, cast formation, infiltrates of leukocytes, and fibrosis, all of which were markedly ameliorated by SPION-EVs treatment (Fig. 6a, b ). To fully assess the efficacy of SPION-EVs, kidney sections were subjected to Masson and Sirius red staining to determine the fibrosis index and stained with anti-α-SMA/Fibronectin antibody for the detection of fibrosis-like tubules. The results demonstrated that the renal fibrosis was reduced in DKD rat treated with SPION-EVs (Fig. 6c–f ), and reduction of α-SMA, Collagen I, and Fibronectin deposition in renal tissues of SPION-EVs treated DKD rat (Fig. 6g–j ). Moreover, immunofluorescence shows that SPION-EVs can significantly inhibit the nuclear expression of Smad2/3 in renal tubules stimulated by high glucose environment and reduce α-SMA (Fig. 6k ). These data suggested that SPION-EVs could attenuate renal fibrosis and retarding the progress of DKD (Fig. 7 ). Assessment of toxicity and inflammation in SPION-EVs–treated rat model We next examined the concentrations of hepatic enzyme such as alanine aminotransferase and aspartate aminotransferase in the serum of SPION-EVs treated rats showed no obvious differences from those of phosphate-buffered saline (PBS) treated rats (Supplementary Fig. 8a ). Similarly, SPION-EVs and MSC-EVs-CHIP group toxicity on the major organs, including the heart, liver, spleen and lung. The histological analysis showed no evidence of in vivo toxicity of SPION-EVs (Supplementary Fig. 8b ). Consistent with previous studies that engineered cell-derived EVs are biocompatible and not significantly toxic.
Discussion In this study, we reported a method for manufacturing engineered SPION-EVs and evaluated the feasibility, safety, and effectiveness of MSC-EVs with CHIP overexpressed in treating renal interstitial fibrosis in UUO and DKD models. Impressively, provide packed CHIP protein into MSC-derived EVs by genetic engineering with lentivirus transfection (Fig. 1 ), then through nanomaterials SPION decorated MSC-EVs with CHIP overexpressing target and enrich the damaged kidneys, resulting in notable renal protective effects, specifically by promoting Smad2/3 ubiquitination and degradation to reverse the transformation of renal tubular cells into myofibroblast. Our findings not only demonstrate the role of engineered MSC extracellular vesicles in the anti-fibrotic activity, but also as a new renal fibrosis therapeutic agent and target delivery nanoplatform. In a word, engineered SPION-EVs constitute an effective nanotherapeutic for the renal fibrosis treatment. Over the past several years, the clinical studies of RIF therapies have concentrated on antifibrotic, including anti-fibrosis antibodies and YAP inhibitors 24 , 25 . Unfortunately, these treatment strategies have not achieved good results and were not significantly reduce fibrosis in clinical tests. Furthermore, the pathogenesis of RIF is complex and includes inflammation, tubular injury, oxidative stress and metabolic dysfunction 26 . There is increasing evidence that Smad2/3 ubiquitination and degradation in renal interstitial is a promising treatment method for CKD therapy 27 . we developed an SPION decorated engineered MSC-EVs platform by CHIP expression to reduced renal fibrosis and inflammatory infiltration in RIF rats. In vitro, we observed that MSC-EVs-CHIP inhibit Smad2/3 expression in rat renal tubular epithelial cells and relieved cells myofibroblast change induced by TGF-β 1 , and inhibited inflammasome activation. In UUO and DKD rat model, MSC-EVs-CHIP promoted Smad2/3 ubiquitination and degradation, further alleviated tubular cells fibrosis and inflammation. CHIP as an E3 Ubiquitin ligase, that induces protein folding and degradation balance by upregulating ligase expression or activation 28 . And which is an important component of the protein quality control system, preventing and treating diseases related to protein metabolism disorders 29 . CHIP protein could be used as a biological drug for the treatment of RIF. However, due to the distribution of metabolism in vivo , and poor renal target efficacy, the poor stability, lead to the CHIP protein can hardly be used in clinical practice. Moreover, we verified the low enrichment of CHIP protein in kidney tissue by intravenous injection in the RIF rat model. According to our previous works on genetically engineering decorated MSC-EVs for delivery of the key protein molecules, which have packed CHIP protein into MSC-EVs by genetic engineering with lentivirus transfection, then use SPION decorated MSC-EVs-CHIP that resolved the challenges for enrichment and renal-target. Mesenchymal stem cell derived extracellular vesicles as cell-free long-distance delivery systems have displayed apparent advantages, for instance high injury tissue target ability, tissue repairment capacity and low immunogenicity, which have been applicated in clinical therapies 30 . It has reported that MSC-EVs have strong tissues target capacity to realize drug delivery 31 . Our previous research confirmed that MSC-EVs attenuate renal interstitial fibrosis through the kinase ubiquitin system CK1δ/β-TRCP mediated YAP ubiquitination and degradation 24 . MSC-EVs could to some extent reach the site of damaged kidney tissue and exert the effective on tissues fibrosis, however, the fluorescence statistical distribution of MSC-EVs labeled with DIR membrane dyes, we found that MSC-EVs were mainly concentrated in the liver and lungs. This has prompted us to think deeply about the need to enhance the targeted enrichment of extracellular vesicles through engineering transformation. Superparamagnetic iron oxide nanoparticles (SPION) have been widely used in disease therapy for excellent superparamagnetism property. The conjugation of SPIONs with extracellular vesicles has many advantages, including magnetic targeted functionalization, magnetic thermotherapy, and delivery of anti-fibrosis agents 32 . Furthermore, in order to reduce the uptake of MSC-EVs in the liver and lungs after tail vein injection, nanomaterials SPION modified MSC-EVs were used to enhances the penetration and enrichment of MSC-EVs to renal under external magnetic field. In conclusion, we have exploited a therapeutic strategy for renal fibrosis using nanomaterials SPION decorated MSC-EVs-CHIP to increase target the concentration of CHIP in renal tissue. We have constructed a platform for targeted delivery of CHIP by using MSC-EVs and highlighted the potential of SPION-EVs as a promising nanotherapeutic for renal interstitial fibrosis treatment. Our platform may provide potential for clinical application in the future owing to the advantages of engineered MSC-EVs and the novel inducing Smad2/3 protein degradation strategy in RIF treatment.
Renal interstitial fibrosis (RIF) is a fundamental pathological feature of chronic kidney disease (CKD). However, toxicity and poor renal enrichment of fibrosis inhibitors limit their further applications. In this study, a platform for CKD therapy is developed using superparamagnetic iron oxide nanoparticles (SPION) decorated mesenchymal stem cells derived extracellular vesicles with carboxyl terminus of Hsc70-interacting protein (CHIP) high expression (SPION-EVs) to achieve higher renal-targeting antifibrotic therapeutic effect. SPION-EVs selectively accumulate at the injury renal sites under an external magnetic field. Moreover, SPION-EVs deliver CHIP to induce Smad2/3 degradation in renal tubular cells which alleviates Smad2/3 activation-mediated fibrosis-like changes and collagen deposition. The extracellular vesicle engineering technology provides a potential nanoplatform for RIF therapy through CHIP-mediated Smad2/3 degradation. Subject terms
Supplementary information
Supplementary information The online version contains supplementary material available at 10.1038/s41536-024-00348-0. Acknowledgements C.J. and J.Z. contributed equally to this work. The authors thank the members of Qian lab for helpful discussion and paper preparation. This work was supported by the National Natural Science Foundation of China (82172102, 81871496), the Jiangsu Province’s Major Project in Research and Development (BE2021689), the Natural Science Foundation of Jiangsu Province (BK20220527), China Postdoctoral Science Foundation (2023M731376), Zhenjiang Key Laboratory of High Technology Research on Exosomes Foundation and Transformation Application (Grant ss2018003). Author contributions C.J. performed experimental design, tissue procurement, data generation pathological assessments, data analysis and interpretation, and manuscript preparation; J.Z. performed experimental design, data generation, and data analysis; H.S. and L.S. performed tissue procurement, data generation, interpretation, and intellectual contribution. W.X., J.J. provided intellectual contribution and critically appraised the manuscript; H.Q. conceived the study, designed experiments, interpreted data, and prepared the manuscript. Data availability All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:01
NPJ Regen Med. 2024 Jan 13; 9:3
oa_package/0b/12/PMC10787844.tar.gz
PMC10787845
38222223
Introduction Diarrhea is one of the top three infectious diseases responsible for ~8%-10% of total deaths among children below the age of five years; 90% of these deaths are estimated to occur in South Asia and sub-Saharan Africa [ 1 , 2 ]. Rotavirus and diarrheagenic Escherichia coli ( E. coli ) are the two most common pathogens responsible for acute diarrhea in children below the age of five years. Shigella , Cryptosporidium , Campylobacter spp. , and Vibrio cholerae are a few other pathogens frequently associated with diarrhea in children under five [ 3 ], with variable region-specific incidence. The Global Enteric Multicentric Study (GEMS), a large case-control study, determined the etiology of childhood diarrhea in South Asia and sub-Saharan Africa. Rotavirus, Cryptosporidium , E. coli producing heat-stable toxin (ST-ETEC), and Shigella were identified as major attributable organisms for the cause of diarrhea [ 4 ]. Later, stool samples from GEMS were re-analyzed using quantitative polymerase chain reaction (PCR) and revised pathogen-specific burdens were calculated. Pathogen-attributable burden increased significantly, and Shigella was reported as the most common causative agent [ 5 ]. Customarily, Shigella infection is associated with bloody diarrhea, but this re-analysis using PCR techniques documented that most such children have only loose stools without blood. Despite Shigella being reported as the most common cause of acute diarrhea in GEMS re-analysis, there is a paucity of studies evaluating Shigella by molecular methods and comparing the isolation rate with conventional culture techniques. Also, the clinical characteristics and outcomes of children with non-bloody Shigella diarrhea managed as per current WHO diarrhea treatment guidelines are not known. In this study, we determined the frequency of Shigella detected by molecular and conventional methods in under-five children with diarrhea and the clinical profile and outcome of children with Shigella diarrhea over the next three months.
Materials and methods This hospital-based prospective observational study was conducted between November 2017 and April 2019 at the University College of Medical Sciences, Delhi, an urban hospital catering mainly to the low-income population of eastern Delhi and adjoining areas of the state of Uttar Pradesh in India. Approval from the Institutional Ethics Committee for Human Research (IEC-HR) of the institute was obtained before enrolling the study participants (approval number: IEC-HR/2017/32/88). Informed consent was obtained from the parents or caregivers of all the participants enrolled in the study. Participants All consecutive children aged between one month and five years, presenting with acute diarrhea to the outpatient department or pediatric emergency services of the hospital and requiring hospitalization due to any reason (such as dehydration, severe malnutrition, septicemia, etc.), were eligible for inclusion. Acute diarrhea was defined as diarrhea of < seven days duration with or without blood in stools [ 6 ]. Children who received antibiotic therapy seven days prior and those with any other known cause of blood in stools (rectal polyp, bleeding diathesis, inflammatory bowel disease, etc.) were excluded. At the time of enrollment, detailed clinical history, including the treatment prescribed (oral rehydration solution (ORS), zinc, paracetamol, etc.) during the current diarrheal episode, was elicited from the parents or guardian. A detailed clinical examination, including an assessment of hydration status, was performed. Anthropometry was performed on all children after the correction of dehydration. All findings were recorded in a case record form. Collection and processing of stool samples A freshly passed stool sample was collected in a wide-mouthed, clean container. A part of the stool sample was concentrated by the formal ether sedimentation method for wet-mount microscopy for ova, parasites, and cysts. The second part of the sample was enriched in selenite F broth (SFB) and alkaline peptone water (APW) for four to six hours at 37o C and cultured on MacConkey’s medium, xylose lysine deoxycholate (XLD) medium, and bile salt agar (BSA) medium for the isolation of pathogenic enteric bacteria by standard laboratory methods. Antibiotic sensitivity testing of isolates was performed as per Clinical and Laboratory Standards Institute (CLSI) guidelines [ 7 ]. The third part of the sample was stored at -80 °C for PCR. The DNA was extracted from a stool sample using the QIAamp DNA Stool Mini Kit (QIAGEN, Hilden, Germany) and stored at -20 °C. The extracted DNA was subjected to PCR amplification using specific primers for the invasion plasmid antigen H (ipaH) gene at 424 bp for Shigella [ 8 ]. Forward 5’- GCTGGAAAAACTCAGTGCCT-3’ and reverse 5’- CCAGTCCTGTAAATTCATTCT-3’ amplification was performed using a reaction mixture of Taq polymerase, deoxynucleotide triphosphate (dNTP), and buffer (QIAGEN). Thirty-five amplification cycles were performed with the following steps: Initial denaturation at 95 °C for five minutes, cycling denaturation at 95 °C for 30 seconds, annealing at 56 °C for 30 seconds, extension at 72 °C for 40 seconds, and final extension at 72 °C for 10 minutes. The PCR amplicons were loaded onto a 2% agarose gel in a specific order for electrophoresis and run at 150 volts for 20-25 minutes. After electrophoresis, the 0.5 mM ethidium bromide-stained gel was visualized and interpreted for specific amplification patterns on the UV transilluminator. Patient management Hydration status was assessed, classified, and managed with ORS and intravenous fluids as per WHO guidelines [ 9 ]. Oral zinc was administered (10 mg/day for two- to six-month-olds, 20 mg/day for six- to 59-month-olds) for a total duration of 14 days. All children with visible blood in their stools received antimicrobial therapy as per WHO and Indian Academy of Pediatrics (IAP) guidelines [ 9 , 10 ]. Children with bloody diarrhea, aged more than one year, with no risk factors (no malnutrition, no dehydration, preserved appetite) received oral antibiotics (cefixime), while those with malnutrition, severe dehydration, or poor oral acceptance received injectable antibiotics. Enrolled children were evaluated for comorbidities like hypoglycemia, hypothermia, and electrolyte imbalance and managed accordingly. Children with severe acute malnutrition (SAM) were stabilized and managed as per WHO guidelines [ 11 ]. Outcome variables Outcomes were assessed as a percentage of children with a diagnosis of Shigella infection by molecular (PCR) and conventional methods (stool microscopy, culture). Children were discharged once the dehydration was corrected, stool frequency was reduced, and oral acceptance improved. After discharge, follow-up for persistence or recurrence of diarrhea was done on days three and seven, and then fortnightly for three months with simultaneous evaluation for growth faltering. Any recurrence of diarrhea or dysentery over three months after discharge was noted. The need for follow-up was emphasized by personal communication at each visit or by telephonic communication to minimize the attrition rate during follow-up. Sample size In an earlier study [ 12 ], Shigella infection was diagnosed by molecular methods in 11 out of 60 children with acute watery diarrhea (18.33%) and 34 out of 60 children with dysentery (56.66%). Assuming 10% of our enrolled cases to have dysentery (visible blood in stools), the estimated proportion of Shigella positivity by molecular methods was 22.2%. With an estimated proportion of 22.2%, the calculated sample size was 150 children with a relative precision of 30% (absolute precision of 6.66%) at a 95% confidence level. Thus, we planned to enroll 150 children with acute diarrhea. Statistical analysis Data collected were entered in a Microsoft Excel spreadsheet (Microsoft Corp., Redmond, WA) and analyzed using IBM Statistical Package for the Social Sciences (SPSS) software version 25 (IBM Corp., Armonk, NY). Outcome parameters such as frequency of isolation, duration of diarrhea, and recurrence of diarrhea were measured using descriptive statistics. Categorical variables were compared between Shigella -positive and Shigella -negative cases by the chi-square test to identify clinical characteristics and the outcome of Shigella infection. For normally distributed data, continuous variables between the two groups were compared using the Student's t-test, and for data that were not normally distributed, the Mann-Whitney U test was used. Odds ratios and 95% CI were calculated to determine risk factors (e.g., age < one month, socioeconomic status, malnutrition status) and clinical predictors (e.g., blood in stools, vomiting, dehydration) by univariate analysis.
Results Enrollment and baseline characteristics During the study period, 209 children aged between one month and five years with acute diarrhea were hospitalized, out of which 150 were included in the present study. Figure 1 depicts the flow of participants in the study, along with the reasons for the exclusion of cases not enrolled in the study. The median (interquartile range (IQR)) age of enrolled children (72 (48%) males and 78 (52%) females) was 10 (5-23) months; 87 (58%) were infants (age <1 year). Only 94 (63%) were completely immunized for age. The majority (142, 94.7%) of the children’s families belonged to the lower middle, upper lower, or lower socioeconomic class on the Kuppuswamy scale [ 13 ]. The median (IQR) duration of diarrhea at hospitalization was two (one to three) days, and the number of stools passed in the previous 24 hours was 15 (10-20). Out of 150 children, 141 (94%) had watery, mucoid, or loose stools, and nine (6%) had blood in their stools. The most common associated symptom with diarrhea was vomiting (85%), followed by decreased oral acceptance (80%), and fever (40%). Clinical characteristics Undernutrition was prevalent in the enrolled children; 58 (38.7%) were stunted (height-for-age Z-score (HAZ) <−2SD) and 64 (42.7%) were wasted (weight-for-height Z-score (WHZ) <−2SD). Among these, 28 (19%) were severely stunted (HAZ <−3SD), while 40 (26.7%) children had SAM (weight for height Z-score <-3SD or mid-upper arm circumference <11.5 cm for children aged between six months and five years). At admission, dehydration was present in 138 (92%) children; 89 (59%) were severely dehydrated, and 15 (10%) had features suggestive of circulatory shock. All enrolled children received ORS during the hospitalization with a mean (SD) duration of 89.0 (31.3) hours, while IV fluids were administered to 95% of children for a mean (SD) duration of 44.1 (31.6) hours. Antibiotics were used in 101 (67.3%) cases for a mean (SD) of 5.3 (2.8) days. The presence of SAM (37%) and suspected or confirmed sepsis (33%) were the most common indications for the use of antibiotics in children admitted with acute diarrhea, followed by clinically suspected cholera (22%), dysentery (9%), pneumonia (5%), and meningitis (2%). Stool microscopy Shigella sonnei was diagnosed by stool culture in only one child. Escherichia coli was cultured from 67 stool samples, whereas Klebsiella pneumonia , trophozoites of Giardia lamblia and ova of Ascaris lumbricoides , and motile Vibrio cholera in hanging drop were detected in one stool sample each. Shigella was detected in 13 (8.7%) out of 150 stool samples by PCR amplification using the ipaH gene. The only culture-positive case was also detected by PCR. The sensitivity of stool culture was 7.7% against PCR for the diagnosis of Shigella infection. Out of 13 children in whom the Shigella PCR was positive, 11 (84.6%) had non-bloody diarrhea, and only two (15.4%) had dysentery. In 11 children with Shigella PCR positivity and non-bloody diarrhea, the characteristics of the stool did not change to bloody diarrhea until the resolution of the diarrhea episode, but eight of these children had received antibiotics due to associated comorbidities. Among three children with PCR positivity and non-bloody diarrhea managed without antibiotics, two had vomiting and poor oral acceptance at presentation, whereas none of them had fever or abdominal pain. All three children presented with some or severe dehydration, and none of these children developed bloody diarrhea or persistent diarrhea, despite no antibiotic usage. Comparison of the characteristics and outcomes of children who were PCR-positive vs. PCR-negative for Shigella Table 1 compares the clinical and anthropometric characteristics of children with stool PCR positivity for Shigella with those who had a negative test result on PCR. There was no statistically significant difference between the two groups. The mean (SD) duration of diarrhea, as monitored after enrolment, was not significantly different (P=0.94) between PCR-positive participants (2.2 (1.54) days) and PCR-negative participants (2.3 (1.45) days). On univariate analysis, none of the evaluated risk factors or clinical predictors had a significant association with the Shigella infection (Table 2 ). Follow-up Out of 150 enrolled children, 23 were lost to follow-up, whereas 43 had a recurrence of diarrhea over the next three months. Among these, 14 patients required hospitalization, and antibiotics were used in 12 patients. During follow-up, only two patients had dysentery, and one patient developed persistent diarrhea. The recurrence of diarrhea episodes over three months was seen slightly more often among the children who had initial PCR (5, 45.4%) positivity than in PCR-negative participants (38, 37.8%), but this was not statistically significant (P=0.507). Also, there were no statistically significant differences in outcome in terms of growth faltering, recurrence of diarrhea, or hospitalization (Table 3 ).
Discussion In this hospital-based study on under-five children with acute diarrhea, we documented that stool culture's diagnostic yield was very low compared to stool PCR amplification using a specific primer for the ipaH gene for the diagnosis of Shigella infection. The majority of Shigella infections presented with watery diarrhea rather than bloody diarrhea, and a history of blood in stools was a poor marker for the diagnosis of shigellosis. We documented that the molecular diagnostic approach performed better than conventional culture methods for the diagnosis of shigellosis. We could not find any significant clinical predictors associated with Shigella infection. There was no statistically significant difference in the overall outcome between PCR-positive and PCR-negative children at three months of follow-up. The conventional methods of detection of Shigella infection rely on the stool culture, which turns out positive in a small fraction of actual cases of shigellosis due to multiple reasons like low bacterial load, loss of bacterial viability due to changes in ambient temperature and pH during specimen transport and storage, and use of antibiotics before specimen collection, transport, and storage [ 14 ]. Molecular diagnostic methods seem to overcome some of the shortcomings of conventional diagnostic methods, thus improving the diagnosis of Shigella infection. The use of PCR assays based on the amplification of the ipaH gene sequence is one such molecular method for diagnosis in cases of dysentery, as it is carried by all four species of Shigella [ 15 ]. Recent data from other settings also suggest that conventional culture yield is poor in episodes of Shigella diarrhea in young children [ 16 ]. The PCR-derived incidence for Shigella surpassed the original estimates by two-fold in the GEMS re-analysis, thus making it the most attributable pathogen of diarrhea among children under five [ 5 ]. In a re-analysis by PCR of stool samples from the multi-country Malnutrition and Enteric Disease (MAL-ED) Study, Shigella -attributable incidence increased from 4% to 41.3% in children aged between 12-24 months [ 17 ]. Earlier, Dutta et al. from Kolkata, India, reported a Shigella detection rate of 15.3% (46 of 300) by stool PCR as against 7.7% (23 of 300) by stool culture [ 8 ]. The positivity rate in the study by Dutta et al. was higher than that seen in our study, which is likely to be due to the inclusion of a very high proportion (42%; 126 out of 300) of children with dysentery. Similarly, Aggarwal et al. reported that the prevalence of Shigella increased from 8% (17 of 207) to 18.3% (11 of 60) in diarrhea and from 33% (39 of 118) to 56.7% (34 of 60) in dysentery when stool samples were analyzed by culture and PCR, respectively [ 12 ]. In two large population-based surveillance studies from Vietnam and China, an ipaH gene amplification-based PCR assay was used to detect Shigella spp . from the rectal swab specimens of cases presenting with dysentery. Results showed the presence of the ipaH gene in approximately 93%-97% of randomly selected Shigella culture-positive specimens and 46%-58% of randomly selected Shigella culture-negative specimens [ 18 , 19 ]. In a study by Von Seidlein et al. on the analysis of pooled data from studies from six Southeast Asian countries, i.e., Bangladesh, China, Pakistan, Indonesia, Vietnam, and Thailand, Shigella was isolated from 2,927 (5%) of 56,958 diarrhea episodes, and more than half of these were in children under the age of five, and PCR detected the ipaH gene in 33% of samples of culture-negative stool specimens [ 20 ]. In a study from Bangladesh, Islam et al. documented that a PCR assay based on an ipaH probe improved the rate of detection of Shigella in stool samples to 61% from 44% by the conventional culture method [ 21 ]. They also tested 123 strains of E. coli by PCR to identify enteroinvasive E. coli , but none yielded any positive results. Though these studies have included variable age groups, the high detection rates of the ipaH gene in culture-negative stool specimens suggest that earlier estimates of shigellosis burden measured by conventional culture may have underestimated the true disease burden. Blood in stools has been used as a surrogate marker for Shigella diarrhea. The WHO's diarrhea management guidelines recommend antibiotics effective against Shigella only when visible blood is present in stools. In our study, 84.6% (11 of 13) of the children with stool PCR positivity for Shigella presented with non-bloody diarrhea. In the GEMS re-analysis, Shigella spp . was the second highest cause of watery diarrhea, and overall, 40.3% (~527 of 1,310) of cases due to S higella spp. were non-dysenteric [ 5 ]. Likewise, 86.2% of the attributable incidence for Shigella was non-dysenteric in the re-analysis of the MAL-ED cohort study [ 17 ]. The results of a recent meta-analysis showed that the proportion of Shigella infections presenting as dysentery has been decreasing [ 22 ]. This may be related to the diagnosis of such cases by molecular methods or early treatment with antibiotics. However, it is unclear whether Shigella infection without blood in stools needs therapy on similar lines. In our series of cases, antibiotics were administered for other reasons in cases where Shigella was later detected by molecular methods, except in three children in whom the infection resolved without any use of antibiotics. The strength of our study was the prospective nature of enrollment, with a focus on the clinical presentation and outcome of Shigella infection detected with molecular methods. Our study also attempted to evaluate the clinical profile (resolution of diarrhea, change to bloody diarrhea in those with watery diarrhea, and recurrence) during a three-month follow-up period for children with Shigella infection managed as per current diarrhea treatment guidelines, which has not been evaluated in any of the studies comparing detection rates of Shigella between conventional and molecular methods. However, because of antibiotic use for other indications, we were unable to validly comment on the outcome of PCR-positive cases when they are managed without antibiotics. A relatively low Shigella prevalence in our series precluded the identification of any significant clinical marker of Shigella infection. Another limitation of our study was the use of the ipaH gene as a marker for Shigella detection, which may also be found in enteroinvasive E. coli . We enrolled hospitalized patients who primarily had moderate-to-severe diarrhea and concurrent illnesses; therefore, results from our study may not be generalizable to community settings with less severe diarrheal illnesses.
Conclusions The present study concludes that Shigella is an important cause of diarrhea in children under five, often missed by conventional laboratory methods. Blood in stools as a syndromic indicator for Shigella infection needs to be reconsidered as the majority of Shigella diarrhea cases have non-bloody stools. As the current management guidelines for childhood diarrhea recommend antibiotics for only bloody diarrhea, studies evaluating other clinical predictors of Shigella infection and faster and more cost-effective techniques for its molecular diagnosis are required. Large, community-based longitudinal studies are needed to evaluate the outcome of non-dysenteric Shigella infections diagnosed by molecular methods so that management guidelines for such infections can be formulated.
Background and objectives: Shigella is an important cause of diarrhea in children under five, often missed by conventional laboratory methods. Blood in stools has always been a syndromic indicator for Shigella diarrhea, but most cases present with watery diarrhea without blood. This study aimed to determine the frequency of Shigella detected by molecular and conventional methods in children under five. Additionally, we aimed to study the clinical profile and outcome of children with Shigella diarrhea managed as per current diarrhea treatment guidelines. Methods: In this hospital-based prospective observational study, stool samples from 150 children (age range: one month to five years) with acute diarrhea (duration < seven days) were subjected to routine microscopic examination, stool culture, and DNA extraction. The extracted DNA from stored stool samples was subjected to polymerase chain reaction (PCR) amplification using a specific primer for the invasion plasmid antigen H gene sequence (ipaH) gene at 424 bp. Results were interpreted in the context of the percentage of isolation of Shigella by molecular (PCR) and conventional methods (stool microscopy and culture) and the follow-up outcome in terms of recurrence of diarrhea or dysentery and growth faltering over three months after discharge. Results: Shigella infection was diagnosed in stool samples by PCR from 13 (8.7%) children, whereas it was isolated by conventional stool culture in only one (0.7%) child. The sensitivity of culture was only 7.7% against PCR for the diagnosis of Shigella infection, whereas blood in stools had a sensitivity of 15.4%. The majority of Shigella PCR-positive cases (11 out of 13) presented with non-bloody diarrhea. None of the evaluated clinical predictors had a significant association with the Shigella infection. No statistically significant difference was found between PCR-positive and PCR-negative children at the end of follow-up (P>0.05). Conclusion: The majority of children with Shigella infection present with watery diarrhea rather than bloody diarrhea, and a history of blood in stools is a poor marker for the diagnosis of shigellosis. The diagnostic performance of stool culture is also very low compared to stool PCR for the diagnosis of Shigella diarrhea.
CC BY
no
2024-01-15 23:42:01
Cureus.; 15(12):e50546
oa_package/45/9d/PMC10787845.tar.gz
PMC10787846
38222208
Introduction Mucoid degeneration (MD) of the anterior cruciate ligament (ACL) is a rare condition, with a prevalence rate ranging from 0.2% to 1.2% [ 1 ]. The cause of this condition is not well understood and has been the subject of vigorous debate, with only a limited number of reports in the literature detailing its origin [ 2 , 3 ]. It is characterized by the infiltration of a mucoid-like substance within the ACL, which can cause knee pain and restricted motion [ 4 ]. Although it was once considered rare, recent reports suggest that it may be more common than previously thought, implying that it is underdiagnosed or misdiagnosed [ 5 ]. We describe a successful treatment approach for MD of ACL, which involves the application of arthroscopic debridement.
Discussion MD of ACL is an uncommon condition that leads to knee pain and discomfort, primarily impacting individuals in their fourth decade of life [ 6 ]. Kumar et al. [ 7 ] initially coined the term mucoid cystic degeneration to describe this condition affecting the ACL. It is characterized by significant disruption of the ligament's normal appearance and can contribute to knee pain. The predominant symptom associated with MD of ACL is knee pain, primarily located in the posterior aspect [ 8 , 9 ]. This pain is caused by mechanical impingement on the posterior cruciate ligament and the posterior capsule or bone erosions [ 9 ]. Additional clinical manifestations include a mechanical hindrance to extension, varying degrees of swelling, and audible clicking sounds [ 10 ]. MRI serves as a valuable tool for evaluating MD of ACL during preoperative assessment. It shows an ACL that is indistinctly defined, exhibiting an enlarged dimension while retaining a typical orientation, and presenting heightened signal intensity interspersed among visible intact ACL fibers, creating a distinctive "celery stalk" appearance [ 11 ]. On T1-weighted images, MD of ACL typically appears with intermediate signal intensity, while on T2-weighted images, it shows high signal intensity [ 11 , 12 ]. Hodler et al. correlated MRI appearances with histological findings and found that 29 out of 38 ligaments had focal areas of signal increase, suggesting a correlation between the focal MRI signal changes and the presence of degenerative changes in the ligaments [ 13 ]. Arthroscopic observations suggest that MD of ACL involves a hypertrophied, fibrillated ligament with the presence of yellowish mucinous material interspersed among the fibers. This is often accompanied by the absence of the ligamentum mucosum [ 12 ]. Additionally, a lack of smooth synovial lining is typically noted [ 14 ]. Arthroscopic surgery is a discretionary treatment option for MD of ACL. This procedure involves the partial removal of lesions within the ACL, leading to rapid pain relief and improved range of motion, without any persistent symptoms of instability. The reduction in volume and tension within the ACL is often attributed to the notable pain relief [ 15 ]. Debridement of mucinous substance, along with partial resection of ACL, is a recommended and effective therapy that does not cause instability, according to many authors [ 16 ]. While certain authors have underscored the significance of notchplasty as a supplementary step in the procedure, Motmans and Verheyden have challenged this idea by arguing against the necessity of notchplasty. They contend that a comprehensive debridement alone is sufficient to resolve impingement and effectively address the underlying pathology [ 4 ].
Conclusions The clinical and radiological characteristics of MD of ACL can be often inconclusive, potentially leading to a misdiagnosis of a torn ACL. Arthroscopists must be cognizant of this uncommon condition, as they may encounter a mucoid ACL during surgical procedures. This awareness is crucial for an accurate diagnosis and appropriate management of the condition.
Mucoid degeneration (MD) is an uncommon pathological phenomenon that specifically affects the anterior cruciate ligament (ACL). This condition arises from the infiltration of yellowish material within the fibers of the ACL, contributing to the clinical presentation characterized by discomfort and limited mobility. MRI has proven to be the foremost diagnostic modality in effectively distinguishing MD of ACL from other potential pathologies. Preoperative recognition of this condition facilitates straightforward diagnosis, particularly via characteristic findings observed during knee arthroscopy. We present a case of MD of ACL, review prior studies about the condition, and outline its clinical features and symptoms, including those observed in our specific case.
Case presentation The patient was a 57-year-old female who presented to our hospital with a two-year history of right knee pain and limited flexion following a minor injury. A thorough examination was conducted. Notably, there was no apparent swelling or instability observed. The knee exhibited a restricted range of motion, spanning from 0° to 90°, with pain occurring during terminal flexion. Tenderness was localized to the lateral joint line of the knee, and the McMurray test yielded a positive result. Significantly, no clinical signs or symptoms suggestive of instability, ligamentous laxity, or patellofemoral pathology were evident (Figure 1 ). A plain X-ray examination provided valuable insights, revealing subtle degenerative changes, particularly in the medial compartment (Figure 2 ). The patient underwent conservative treatment for several months; however, there was no improvement in the condition. MRI of the knee was performed, which revealed specific findings in the right knee. In the coronal view, a clear thickening of the ACL was evident, characterized by preserved fibers, and a consistent increase in signal intensity. Noteworthy findings on the sagittal view revealed an indistinct ACL presenting a distinctive "celery stalk" appearance (Figure 3 ). Arthroscopy was performed, uncovering femoropatellar arthritis marked by advanced osteocartilaginous lesions and degenerative changes in the lateral and medial meniscus. The continuous fibers, notably the anterolateral bundle, exhibited enlargement and displayed a yellowish degenerative appearance, confirming MD. Additionally, degenerative vertical lesions were evident in the external meniscus. A decision was made to perform debridement of the yellowish lesion of the ACL, and subsequent stability testing through the anterior drawer maneuver yielded satisfactory results (Figure 4 ). At the 10-month follow-up, the patient demonstrated a complete range of knee motion without experiencing pain or instability.
CC BY
no
2024-01-15 23:42:01
Cureus.; 15(12):e50545
oa_package/c8/bb/PMC10787846.tar.gz
PMC10787847
38222263
Approved by: Agustin Ibanez, Latin American Brain Health Institute (BrainLat), Chile
The journal retracts the 18 June 2021 article cited above. Following publication, the publisher uncovered evidence that false identities were used in the peer-review process. The assignment of fake reviewers was confirmed by an investigation, conducted in accordance with Frontiers' policies and the Committee on Publication Ethics (COPE) guidelines. Given the concerns, the editors no longer have confidence in the findings presented in the article. This retraction was approved by the Chief Editors of Frontiers in Aging Neuroscience and the Chief Executive Editor of Frontiers. The authors received a communication regarding the retraction and had a chance to respond. This communication has been recorded by the publisher.
CC BY
no
2024-01-15 23:42:01
Front Aging Neurosci. 2023 Dec 28; 15:1358292
oa_package/c4/9e/PMC10787847.tar.gz
PMC10787848
38222136
Introduction and background The development of more efficient, well-tolerated, and easy combination antiretroviral treatment (cART) has significantly enhanced HIV-positive people's (PLWH) life expectancy, even exceeding that of the general population [ 1 ]. Their extended survival, however, has resulted in a shift in their health profile, with a greater frequency of age-related comorbidities, which frequently manifest 5 to 10 years sooner than in the general population [ 2 ]. Furthermore, certain PLWH are becoming more susceptible to geriatric disorders such as frailty [ 3 ]. The goal of this study is to look into the role of frailty in the holistic treatment of older people with HIV (OPLWH). Frailty, which was initially used by the insurance industry to assess mortality risks in adults over the age of 65 [ 4 ], has recently been broadened to include physical, cognitive, social, emotional, and economic components [ 5 ]. It denotes vulnerability to unfavorable health impacts as a result of several stresses. While frailty was originally studied in community-living persons over the age of 65, it has subsequently been studied across a wide range of demographics and conditions, indicating prevalence rates ranging from 3% to 5% between the ages of 30 and 60 to 25-30% after the age of 80 in the United Kingdom [ 6 ]. Frailty has been proven to be a more accurate predictor of survival in COVID-19 individuals than chronological age or comorbidities [ 7 ]. Frailty, cognitive decline, and other geriatric disorders are becoming more widespread and emerging sooner as PLWH continues to age, with a median age presently in the mid-50s in high-income countries [ 3 ]. Frailty must be identified since it increases the likelihood of acquiring new chronic diseases, such as falling, cognitive impairment, polypharmacy, hospitalization, loss of independence, and mortality. Because aging rates vary, screening for frailty is a critical initial step in the management of OPLWH [ 8 ]. As a result of these developments, the approach to caring for OPLWH has moved from immunovirologic treatment to a more multidisciplinary one that places an emphasis on tailored services [ 9 ]. Specialized programs have been developed to accommodate the particular needs of OPLWH [ 7 ]. While frailty assessments are not necessary for every OPLWH, they are crucial in identifying at-risk individuals since chronological age alone is not always a reliable predictor of age-related disorders in geriatrics [ 8 ].
Conclusions As OPLWH lives longer, healthcare for them is changing. However, there are issues with the growth in frailty and other geriatric problems among OPLWH. Once associated with elderly people, fragility is now a critical factor in determining susceptibility to unfavorable health outcomes. Personalized services and an interdisciplinary approach have been developed for OPLWH. Because frailty is associated with mortality and chronic diseases, it is imperative to identify and treat it early. Research on pharmacotherapeutic alternatives and e-health solutions is still ongoing. To offer complete treatment in this evolving healthcare environment, it is vital to steer clear of ageist healthcare practices and give frailty the attention it demands.
The life expectancy of people living with HIV (PLWH) has greatly increased due to advancements in combination antiretroviral treatment (cART). However, this longer life has also increased the prevalence of age-related comorbidities, such as frailty, which now manifest sooner in this group. Frailty, a term coined by the insurance industry, has been broadened to include physical, cognitive, and emotional elements and has been recognized as a critical predictor of negative health outcomes. With the median age of PLWH now in the mid-50s, treating frailty is critical given its link to chronic diseases, cognitive decline, and even death. Frailty assessment tools, such as the Frailty Phenotype (FP) and the Frailty Index (FI), are used to identify vulnerable people. Understanding the pathophysiology of frailty in PLWH indicates the role of immunological mechanisms. Frailty screening and management in this group have progressed, with specialized clinics and programs concentrating on multidisciplinary care. Potential pharmacotherapeutic solutions, as well as novel e-health programs and sensors, are in the future of frailty treatment, but it is critical to ensure that frailty evaluation is not exploited to perpetuate ageist healthcare practices. This narrative review investigates the changing healthcare environment for older people living with HIV (OPLWH), notably in high-income countries. It emphasizes the significance of identifying and managing frailty as a crucial feature of OPLWH's holistic care and well-being.
Review Frailty: concept and evolution Derived from the Latin word "fragilis" (meaning "breakable"), the term frailty indicates a condition of increased sensitivity. Fried devised the Frailty Phenotype (FP) and the Frailty Index (FI) to emphasize the complexity of the idea in contrast to impairments or comorbidities [ 10 ]. Although the fundamental traits and consequences of frailty are well acknowledged, the absence of a clear operational definition makes diagnosis difficult. There are at least 67 metrics that have been developed; the most often used ones in research are the FP and FI [ 11 ]. The FP diagnoses frailty when three of the five predefined physical criteria - unintentional weight loss, self-reported fatigue, poor hand grip strength, sluggish walking speed, and low physical activity - are met. One or two conditions indicate pre-frailty, whereas none indicate non-frailty. Based on the accumulation of many age-related health problems, the FI computes frailty. Generally speaking, frailty in older adults who live in communities is indicated by a FI greater than 0.25. A prognosis of poor quality and several acute illnesses is indicated by a FI larger than 0.7 [ 12 ]. There are 30 prerequisites in all. Even though the FP is simpler to use, it does require specialized assessments and qualified personnel. The FI, on the other hand, appears more complicated but provides superior risk discrimination and may be determined using electronic health information [ 13 ]. Models predict comparable results when applied to diverse populations, and they also identify subgroups of frail individuals who do not have impairments or comorbidities. Research on the prevalence of frailty varies, with estimates ranging from 2% to 25% [ 9 ]. Other proven methods for diagnosing frailty include the Edmonton Frail Scale (EFS), the Clinical Frailty Scale (CFS), and the Frail questionnaire. Even among those who are not yet diagnosed as frail, a meta-analysis confirmed the association between these markers of frailty and increased rates of death, hospitalization, loss of independence, impairments, falls, fractures, and cognitive decline [ 14 ]. A gradual loss of normal homeostatic mechanisms brought on by biological and environmental stresses that affect cells, tissues, organs, and the overall health of a person ultimately leads to frailty. A chronic inflammatory state is shaped by immunological traits and CMV seropositivity, and immunosenescence is a major factor in this process. Numerous processes and serum markers are associated with this chronic inflammation, which is sometimes referred to as inflammaging [ 15 ]. Hormonal fluctuations, epigenetic alterations, telomere shortening, genetic control, and changes in body composition, such as abdominal fat and muscle loss, can all have an impact on frailty [ 16 ]. The CFS is a measure of independence in older PLWH that shows favorable correlations with the FP and FI [ 17 ]. A multi-domain evaluation called the EFS has earlier been compared to the FP and FI in PLWH [ 18 ]. A simple instrument consisting of five parts is the FRAIL Scale [ 19 ]. Originally intended to predict death in PLWH, the Veterans Aging Cohort Study Index (VACS-I) is currently utilized as a frailty screening tool with construct validity and predictive validity [ 20 ]. These methods aid in identifying persons who, due to their frailty, may require treatments and care. Pathophysiology of frailty in PLWH Frailty in PLWH has molecular underpinnings associated with age-related dysregulation of several physiological systems. Dysregulated systems include those related to the stress response (neuro-immuno-endocrine), metabolism (insulin and mitochondria), and musculoskeletal systems [ 21 ]. When dysregulation escalates to a point where negative outcomes are more likely, frailty is categorized as a syndrome. Other variables are involved, even though the majority of evidence indicates that frailty in OPLWH is comparable to physiologic aging [ 22 ]. Prolonged low-grade HIV replication triggers persistent immunological activation even in suppressed individuals, which results in immunosenescence and inflammation that invariably produce frailty in clinical trials. Additionally, epigenomic dysregulation is at play. Clinical variables such as co-infections, early suppression of HIV replication, CD4/CD8 ratio, and microbiota translocation facilitate the development of frailty. These traits emphasize the role that immunoinflammatory variables play in the development of frailty [ 23 ]. Frailty is a common dysregulated condition of physiological functioning and impaired functional reserve that leads to negative outcomes typical of aging, although it may manifest earlier in PLWH [ 24 ]. Studies employing the FP for frailty diagnosis In the multicenter AIDS Cohort Study (MACS), frailty was measured using an adapted frailty-related phenotype (aFRP) in untreated, seropositive, college-educated, Caucasian males with a mean age of 55. In the same group of males over 65 who tested negative for HIV, frailty was common in 3.4% of cases [ 25 ]. Immuno-virologic traits were associated with frailty, and among PLWH who started cART, frailty was associated with a higher risk of developing AIDS or passing away [ 26 ]. Risk factors for frailty included old age, non-Hispanic black ethnicity, and potentially controllable conditions such as AIDS, smoking, hepatitis C infection, depression, diabetes, and renal disease. Frailty also happens in the absence of concomitant comorbidities [ 25 ]. A 12% frailty prevalence was observed in the AIDS Linked to Intravenous Experience (ALIVE) cohort, which consisted primarily of male African-American injectable drug users (median age 49). Risk factors for this cohort included female sex, advanced HIV disease, lower education, depression, and multimorbidity, as well as older age and HIV infection. Frailty has been associated with all-cause hospitalizations [ 27 ], overall mortality [ 28 ], and chronic illnesses such as lung, heart, and mental disorders. The HIV Infection, Aging, and Immune Function Long-Term Observational Study (HAILO) cohort examined disability and frailty. The prevalence of pre-frailty and frailty was 37% and 6% of the population, respectively, with little overlap between the two conditions. Premature mortality, type II diabetes, and cardiovascular disease have all been related to frailty [ 29 ]. Neurocognitive impairment, obesity, smoking, the first choice of cART (with NNRTI [nucleoside reverse transcriptase inhibitor]-based cART increasing the risk of frailty), and educational level were among the modifiable risk factors. Both moderate alcohol use and physical activity were protective [ 30 ]. Additionally, it was demonstrated that more seropositive women than males had frailty, which is in line with the overall population. 10.0% of uninfected women and 17.3% of seropositive women in the Women's Interagency HIV Study (WIHS) were frail [ 31 ]. Concerns regarding economic and healthcare delivery challenges as the PLWH population ages are raised by other research, like those carried out in South Africa and Spain, which revealed a high prevalence of frailty in PLWH [ 32 ]. Screening of frailty in PLWH When a senior is diagnosed as frail, it signifies more than simply a condition with a dismal prognosis. For instance, one important risk factor for perioperative issues is frailty. By altering recognized preoperative variables for surgical morbidity, pre-habilitation clinics improve outcomes [ 33 ]. An integrated geriatric strategy is becoming more and more recommended for certain older PLWH, especially those who have been classified as weak [ 34 ]. Other surrogates, besides frailty, can help identify PLWH who would benefit from a geriatric assessment. For instance, polypharmacy is more common in PLWH compared to controls. Similarly, decreased functional status as determined by gait speed or the whole Short Physical Performance Battery (SPPB) assessment tool is also more common [ 35 ]. Frailty, functional status, and deficits all interact in PLWH, despite being distinct illnesses [ 36 ]. PLWH frequently experience functional impairments and impairments, especially those who also have concurrent geriatric symptoms [ 36 , 37 ]. Strong PLWH in need of geriatric review may also be indicated by a combination of immunological markers (e.g., a low nadir CD4 count of 200, a plateau CD4 level of 500 on suppressive cART, and a CD4/CD8 ratio of one) [ 38 ]. Weakness is a dynamic state. The majority of non-frail and pre-frail individuals retained their status in a 12-month follow-up analysis involving over 300 treated PLWH, but the majority of frail individuals relapsed to pre-frailty [ 39 ]. Pre-frailty, which affects 30-60% of PLWH, has to be identified since it is associated with unfavorable outcomes. We have previously discussed the elements that contribute to the shift of PLWH in the MACS to frailty. The only factor associated with a reversal of frailty was age [ 40 ]. Guaraldi spent four years studying the MHMC Cohort's frailty transition determinants. The following factors predicted FI at follow-up: baseline FI, female gender, length of HIV infection and cART usage, and history of smoking [ 41 ]. There are currently no guidelines for which PLWH to refer, and it is unclear if geriatric referrals are clinically effective. The public should be evaluated for frailty if they are older than 70. It is reasonable to think about screening PLWH above the age of 50 based on data showing PLWH age-advancement. The responsibilities of geriatricians as knowledgeable consultants or involved team members are being established [ 42 ]. Management of frailty in PLWH In high-income nations, the primary healthcare approach for OPLWH is based on specialized community-based primary care or tertiary care that is offered in infectious disease clinics. Specialized geriatric or aging-HIV clinics, HIV-metabolic clinics, and HIV-rehabilitation programs have been established to address the wider range of social, emotional, and health concerns experienced by OPLWH as their healthcare needs have expanded to include non-HIV-related illnesses [ 43 ]. Numerous of these clinics have adopted care methods designed for senior citizens that are based on geriatric concepts. With an emphasis on modifying the Comprehensive Geriatric Assessment (CGA) to identify the most susceptible OPLWH, these models entail screening for frailty and other geriatric disorders. CGA, a well-researched multidimensional diagnostic procedure, is used to evaluate the functional, psychological, and medical abilities of a subset of elderly people [ 44 ]. After the age of 70, it is advised to do yearly frailty screenings in the general population [ 42 ]. According to recent guidelines, PLWH over 50 should undergo yearly frailty screening using the FRAIL Scale. If the screening findings are positive, the individual should then be sent for a thorough geriatric evaluation [ 45 ]. To test for frailty, HIV clinics that specialize in OPLWH use a variety of measures, such as gait speed, the FRAIL Scale, and the Clinical Frailty Scale [ 45 ]. Many clinics have provided insights into how they manage OPLWH who are frail, using multidisciplinary geriatric evaluations to determine who is most in need [ 34 ]. For instance, the CFS is used by the Chelsea and Westminster Clinic in London to screen older people with HIV (OPLWH). Individuals with a CFS score of 5 or more, which indicates at least mild frailty, are sent to a specialized geriatric HIV clinic [ 46 ]. Clinics may use several measures to screen for frailty. While some use electronic health records to identify the FP, others use the FI. When electronic health data are accessible, the FI may be applied more easily than the FP, which needs specific equipment and training [ 13 , 47 ]. The goals of therapy for OPLWH who have been classified as frail are to prevent and control impairment and comorbidities. Furthermore, geriatric syndrome assessment and treatment are essential elements of this care approach [ 48 ]. To control OPLWH, tactics borrowed from research with frail, uninfected older adults are used. Maintaining the quality of life and possibly averting cognitive decline also depend on acknowledging the increased health risks linked to social isolation. This concept aims to actively involve older people with HIV/AIDS within their social networks and reintegrate HIV care into primary care while facilitating access to community resources [ 49 ]. Fundamental care principles for frail older HIV-negative individuals include implementing strength-training exercise programs, managing sarcopenia, assessing for polypharmacy, treating weight loss and undernutrition with protein-calorie supplements, and evaluating and treating reversible causes of fatigue such as anemia, depression, hypothyroidism, and B12 deficiency. Furthermore, vitamin D levels should be monitored, and it should be administered if necessary [ 50 , 51 ]. Programs for preventing frailty that involve many components have shown promise in reducing the pace of frailty development, boosting cognitive performance, and increasing physical function [ 52 ]. CGA informs care plans that can improve functional abilities, lower the chance of institutionalization, postpone the onset of impairment, cut down on hospital admissions and stays, and increase survival [ 53 ]. A physiotherapist should ideally be available or included in the multidisciplinary team. Exercise programs of varying intensities have enhanced physical function in OPLWH, while short-duration exercise programs have shown beneficial effects on frail older individuals. Improvements in quality-of-life measures in PLWH have also been associated with participation in yoga programs [ 54 ]. While exercise has been shown to be the most effective intervention for frailty, group-based physiotherapy classes have also demonstrated efficacy in protecting against functional decline. On the other hand, addressing the social determinants of frailty is important. This includes, but is not limited to, isolation, loneliness, depression, or any other psychiatric diagnoses. However, this may not always be feasible, as it often depends on the availability of local professional resources and the presence of a community-based support system [ 55 ]. The New Orleans Alcohol Use in HIV (NOAH) study evaluated the body composition, gait speed, and muscle strength of 341 participants living with HIV. Body composition was found to have a modulatory effect on frailty risk among PLWH, which was statistically significant. That is, while obesity was associated with increased risk, greater muscle mass may have had a protective effect, even among individuals who consume alcohol. These findings emphasize the importance of physical activity and weight control in modulating frailty risk [ 56 ]. Furthermore, strict control of the medications PLWH consumes may be beneficial. Polypharmacy may increase the risk of adverse outcomes, and de-prescription is often necessary. For instance, a study demonstrated that frailty was more common in those who used medications with an anticholinergic effect (OR 2.12; 95% CI 0.89-5.0). Therefore, clinicians should be aware of their impact and work to reduce exposure whenever possible [ 57 ]. As mentioned earlier, exercise is a well-established preventive intervention for frailty. However, since many of the OPLWH are already in a frail state, high-intensity training and weight-lifting may be too taxing. Therefore, a study investigated a novel game-based training program, referred to as exergame, and its impact on ameliorating frailty among 10 HIV-infected individuals. The exergame program which was conducted twice weekly for six weeks, incorporated activities such as weight shifting, ankle reaching, and obstacle crossing. After the conclusion of the program, changes in balance, gait, and other parameters were assessed, and improvements were seen in balance and mobility [ 58 ]. E-health systems are being increasingly utilized to provide health information, personalized recommendations, and smartphone reminders to promote healthy habits and help older people preserve and enhance their functional independence. There is much promise in the continuing evaluation of wearable sensors for frailty identification [ 54 ]. The Ecological Momentary Assessment, which gathers real-time patient-reported outcome indicators, can help make it easier to incorporate patient-reported outcomes into innovative healthcare models [ 53 ]. Nonetheless, it is important to maintain vigilance to prevent the possible ageist approach to healthcare delivery from being supported by frailty [ 52 ]. Limitations and challenges While our review provides insight into frailty pathophysiology, screening, detection, and management for PLWH, it also identifies several limitations and research gaps, the most notable of which is the absence of a standardized operational definition for frailty. This presents a challenge not only for consistent diagnosis in the healthcare setting but also for conducting studies with reliable data and results. Therefore, the need to establish a universally accepted definition is paramount. However, more research is needed to explore the most effective and relevant metrics for frailty while considering the diverse impacted populations, including OPLWH. In a comparable context, the variability of estimates of frailty prevalence in the different studies (ranging from 2% to 25%) highlights the importance of understanding any contributing factors and standardizing assessment methods in future studies. On the other hand, it implies that frailty in this population (PLWH) may manifest differently. Furthermore, while our article touches on the integration of geriatric concepts in HIV care, more research is required to assess the feasibility, effectiveness, and scalability of these approaches. Clear guidelines for identifying PLWH who would benefit from frailty assessments and evaluating the clinical effectiveness of geriatric referrals are lacking, and this is an area we aim to address in our ongoing research. Lastly, a more thorough exploration of e-health systems and electronic health records used for estimating frailty, especially those employing metrics such as the Frailty Index, would enhance our understanding of technological advancements in frailty assessment and their real-world applicability. Addressing these gaps will contribute to a more comprehensive and nuanced understanding of frailty, improve diagnostic accuracy and inform targeted interventions for vulnerable populations.
CC BY
no
2024-01-15 23:42:01
Cureus.; 15(12):e50539
oa_package/79/92/PMC10787848.tar.gz
PMC10787849
38222186
Introduction Autism is a disorder distinguished by significant challenges in social interaction and communication coupled with repetitive and stereotypical patterns of behavior and activities. Deficits in social interaction and language development become apparent before the age of three. In children, this condition is referred to as autism spectrum disorder (ASD). ASD is a disorder that affects a child's neurological system, growth, and general development [ 1 ]. A child with ASD often has problems communicating; they may have trouble developing social skills. In the Diagnostic and Statistical Manual of Mental Illnesses-5 (DSM-5), ASD is a neurodevelopmental disorder characterized by persistent deficits in social communication and social interaction across multiple contexts, as well as restricted, repetitive patterns of behavior, interests, or activities. The criteria are grouped into two main domains: social communication and behavior. To be diagnosed with ASD, according to the DSM-5, an individual must demonstrate persistent difficulties in social communication and interaction, along with at least two of the four types of restricted, repetitive behaviors. The severity of ASD is specified based on the level of support required: Level 1 (requiring support), Level 2 (requiring substantial support), or Level 3 (requiring very substantial support) [ 2 ]. It is important to note that the field of psychiatry and psychology may evolve, and updates to diagnostic criteria or new editions of diagnostic manuals may occur. Therefore, it is recommended to consult the latest professional literature or experts in the field for the most up-to-date information on the diagnosis of ASD and other mental health conditions. Different approaches, especially psychosocial therapies, are used to address the core symptoms of ASD. These include specific educational programs in special education, speech, occupational, physical, and behavioral analysis, all of which have a favorable impact on the overall efficacy of treatment. Occupational therapy is frequently utilised in managing ASD, although its efficacy can be limited. Consequently, there is considerable anticipation that psychopharmacology may provide additional assistance to these individuals. The CDC states that ASD is a prevalent illness, typically first identified in early childhood [ 3 ]. The signs and symptoms of attention deficit hyperactivity disorder (ADHD), as listed in the DSM-5, include experiencing difficulties with staying focused, paying attention to details, and listening when spoken to directly. Individuals with ADHD often have trouble finishing homework and organizing projects and are reluctant to engage in demanding activities. They regularly lose critical materials needed for assignments, become easily distracted, and are restless or fidgety. Common behaviors include frequently leaving their seat in situations where remaining seated is expected, being unable to play quietly, and exhibiting restlessness as if driven by an inner motor. Additional symptoms are talking excessively, vocalizing thoughts aloud, and showing a lack of self-control or patience [ 4 ]. The level of impairment in individuals with ASD can vary, affecting perception, cognitive processing, learning, and memory [ 5 ]. Additionally, genetics plays a significant role in autism, as evidenced by the ASD occurrence rate among twins, which ranges from 36% to 95%. These genetic factors particularly influence the repetitive behavioral aspects associated with ASD [ 6 - 8 ]. Clinical assessments using the Indian Scale for Autism Assessment (ISAA) have yielded positive results across the board for symptoms associated with ASD [ 9 - 11 ]. The training intervention was designed to enhance the understanding of managing arousal, preventing annoyance from worsening, and promoting self-regulation skills [ 12 ]. It is essential to ensure that children on the autism spectrum interventions are supported by research evidence. Given the variety of accessible pharmaceutical therapies, compiling comprehensive research on interventions for children with these conditions is imperative [ 13 ]. Physiotherapists are often among the first practitioners to assess children who may be at risk for these disorders [ 14 ]. The study aims to investigate the efficacy of a multimodal physiotherapy approach in addressing the specific challenges associated with ASD, including speech impairment and attention deficit. The case report will focus on an individual with ASD who presents both speech difficulties and attention deficit symptoms. The multimodal physiotherapy intervention will incorporate a combination of physical therapy techniques, sensory integration strategies, and potentially other therapeutic modalities to enhance both motor and cognitive functions. The study seeks to assess improvements in speech abilities, attention span, and overall functional outcomes in individuals with ASD following the implementation of the multimodal physiotherapy intervention. By exploring the potential benefits of this comprehensive approach, the research aims to contribute valuable insights into the development of effective therapeutic strategies for individuals with ASD who experience speech impairments and attention deficits.
Discussion This study presents intriguing findings that warrant careful consideration and interpretation. The observed improvements in speech abilities and attention deficits following the implementation of the multimodal physiotherapy intervention suggest the potential efficacy of this comprehensive approach. The integration of physical therapy techniques and sensory strategies appears to have a positive impact on both motor and cognitive functions, addressing the unique challenges faced by individuals with ASD who experience concurrent speech impairments and attention deficits. The individualized nature of the treatment plan is a notable strength, emphasizing the importance of tailoring interventions to the specific needs of individuals with ASD. In the broader context of interventions for ASD, this study contributes to the growing body of literature exploring novel therapeutic approaches. By incorporating physical therapy and sensory strategies into a multimodal intervention, the research underscores the potential benefits of simultaneously addressing both motor and cognitive aspects. This aligns with the evolving understanding of the interconnected nature of sensory-motor functions and cognitive processes in individuals with ASD. The findings may inform clinicians and therapists working with this population, offering insights into developing more comprehensive and tailored interventions. Despite the promising results, it is essential to approach the study's implications cautiously and recognize the need for continued exploration of multimodal approaches within more extensive and diverse samples, allowing for a more robust understanding of their effectiveness across the spectrum of ASD presentations. The physiotherapy interventions presented in this protocol are a comprehensive and multidisciplinary approach to addressing attention deficit and speech impairment in a three-year-old child with developmental anomalies. Each intervention is tailored to target specific aspects of the child's condition, aiming to improve their overall communication and cognitive abilities. Auditory integration training (AIT) involves exposure to specially filtered and modulated music to enhance auditory processing, which may improve sensory integration and attention in individuals with developmental disorders [ 15 ]. Sensory integration therapy (SIT) focuses on improving the brain's processing of sensory information, particularly for children with sensory sensitivities often seen in ASD [ 16 ]. Holding therapy, based on the belief that autism is linked to a lack of maternal bonding, has its efficacy and safety debated. Facilitated communication involves a skilled facilitator assisting a child with communication challenges by using an output device for spelling words or phrases [ 17 ]. Music therapy is recognized for its potential to enhance interaction and communication skills in individuals with ASD [ 18 ]. The Picture Exchange Communication System (PECS) is an established method for children with ASD to communicate, offering a structured approach for non-verbal or minimally verbal individuals to express needs and desires using pictures or symbols [ 19 ]. Parent-mediated communication is a crucial component of the child's support system, equipping parents to effectively understand and respond to their child's communication. However, the effectiveness of each intervention can vary from one individual to another, and ongoing evaluation and research are needed to determine the long-term impact and effectiveness of these interventions. ADHD is common in children, and the implementation of therapy in mental health, complemented by interdisciplinary collaboration, is crucial [ 20 ]. The pediatric population benefits from the expertise of physiotherapists who specialize in treating children. These professionals possess a deep understanding of the interplay between early childhood development and the body's systems and functions, as well as typical child growth patterns. Physiotherapists use many advanced devices and tools, as well as generic skills, applying their additional training in development and growth to treat infants through teenagers. Physiotherapists use an extensive span of treatment approaches in the pediatrics field. In each case, careful assessment determines the treatment approach and protocol. However, it is crucial to acknowledge the study's limitations, such as the single-case design, which may restrict the generalizability of the findings. Further research with larger sample sizes and controlled designs is warranted to confirm and extend the current results.
Conclusions In conclusion, this case report provides valuable insights into a holistic intervention strategy for individuals with ASD facing concurrent challenges of speech impairments and attention deficits. The observed improvements in speech abilities and attention following the multimodal physiotherapy intervention suggest the potential effectiveness of integrating physical therapy techniques and sensory strategies. The individualized nature of the treatment plan highlights the importance of tailoring interventions to the specific needs of each individual with ASD. While the findings are promising, it is crucial to recognize the study's limitations, including its single-case design, which may limit generalizability. Further research with larger and more diverse samples and controlled designs is essential to validate and extend these preliminary results. The study contributes to the broader understanding of therapeutic approaches for ASD by emphasizing the interconnected nature of sensory-motor functions and cognitive processes. The multimodal physiotherapy approach offers a nuanced perspective for clinicians and therapists, suggesting potential benefits in concurrently addressing both motor and cognitive aspects. As we move forward, ongoing research should continue to explore and refine multimodal interventions, considering their applicability across the diverse spectrum of ASD presentations and informing the development of more comprehensive and tailored therapeutic strategies for individuals with ASD experiencing speech impairments and attention deficits.
Autism is a disorder distinguished by significant challenges in social interaction and communication coupled with repetitive and stereotypical patterns of behavior and activities. Deficits in social interaction and language development become apparent before age three. In children, this condition is referred to as autism spectrum disorder (ASD). Attention Deficit Hyperactivity Disorder (ADHD) is a neurodevelopmental disorder characterized by persistent patterns of inattention, hyperactivity, and impulsivity. Individuals with ADHD may struggle with sustaining attention, organizing tasks, and completing assignments. They may exhibit hyperactive behaviors such as fidgeting, difficulty staying seated, and impulsive actions like interrupting others. ADHD can significantly impact daily functioning and is often diagnosed in childhood, with symptoms potentially persisting into adulthood. The disorder has three main subtypes: predominantly inattentive, predominantly hyperactive-impulsive, and combined presentation. Treatment typically involves a combination of behavioral interventions, psychoeducation, and, in some cases, medication, aiming to provide support and strategies for individuals to manage their symptoms effectively in various aspects of life. Cognitive impairment in ASD varies, meaning it could be at the sensory perception level of cognitive processing, learning, and memory. The goal of the training intervention was to control physiological arousal, enhance awareness, keep annoyance from getting worse, and encourage self-regulation abilities. In this case report, we discuss the approach of multimodal physiotherapy for autism with speech impairment and attention deficit. Furthermore, physiotherapy needs to find a position in the new mental health care paradigm in order to contribute to mental health care.
Case presentation Patient information We present a case involving a three-year-old male child characterized by a spectrum of developmental anomalies. Predominant concerns encompass intellectual disability, hyperactivity, speech dysfluency, and attentional deficits. The patient was delivered at term following an uneventful gestational period of 9 months and 5 days via a lower segment cesarean section (LSCS). Postnatally, delayed initiation of the birth cry necessitated a brief admission to the neonatal intensive care unit (NICU) for observation. Challenges in breastfeeding ensued due to attentional constraints, resulting in a reliance on artificial feeding. Additionally, the patient was identified as having a calcium deficiency. At birth, he weighed 2.4 kilograms and presented with congenital anomalies, which were first observed by his mother at 2.5 years of age. Parental reports at this age indicated deficits in attention, speech fluency, social integration, and communication skills. His vocalizations remained mostly indiscernible babbling until 36 months, when he produced his inaugural linguistically coherent word. He exhibited challenges in apprehending and executing tasks and directives. Motor milestones were accomplished within normative time frames. Ambulation with support was initiated at nine months, and independent ambulation was achieved at 10 months. Equilibrium mastery was attained at 14 months. Regarding ADHD, the patient exhibited a diminished attentional span and proclivity for distractibility. Restlessness and loud vocalizations during play were evident, along with difficulty in maintaining a stationary seated position. A sensitive, temperamental disposition was noted. Modest progress was ascertained by both therapeutic interventionists and parents, with circumscribed strides in attentional endurance and communication acumen. The patient remains reliant on external feeding and continues to necessitate diaper use. Proficiency in toilet training remains unattended. Clinical examination The child exhibits integration with an examiner but shows less interaction with the surrounding environment. His overall activity level was average, but he exhibited a notable deficit in orientation to time, place, and person. His speech was unimpaired, and his hearing was within normal limits. However, his attention span was observed to be poor, which contributed to difficulties in maintaining focus and engagement. The anthropometric measurements are detailed in Table 1 . Investigation The patient underwent a Brainstem Evoked Response Audiometry (BERA) examination, which revealed normal auditory pathway function, indicating no detectable abnormalities in the auditory system (Table 2 ). Physiotherapy intervention The patient underwent physiotherapy treatment for four weeks (Table 3 ). Throughout the treatment, both the patient and his family participated in counseling sessions and followed an exercise regimen. Figure 1 shows a child undergoing holding therapy under the supervision of a therapist, and Figure 2 illustrates the Picture Exchange Communication System (PECS) therapy. Follow-up and outcome measures After four weeks of treatment, outcome measures were assessed. Table 4 summarizes the findings of the scales after the completion of treatment.
CC BY
no
2024-01-15 23:42:01
Cureus.; 15(12):e50547
oa_package/f8/cf/PMC10787849.tar.gz
PMC10787871
38221988
Introduction Road accidents are a significant public health concern worldwide, with an estimated 1.35 million deaths caused by road traffic accidents each year. 1 Developing countries, such as India, are disproportionately affected, with over 150,000 fatalities reported annually. 1 Road safety is a major concern in India, with a large number of accidents and fatalities reported each year. According to the Ministry of Road Transport and Highways, there were 449,002 road accidents in India in 2019, resulting in 151,113 deaths and 451,361 injuries. 2 Road accidents pose a threat to health, leading to approximately 1.35 million deaths globally each year. Countries such as India are particularly affected, experiencing over 150,000 fatalities annually. This alarming situation emphasizes the importance of comprehending the factors that contribute to these accidents in order to develop prevention measures. Our research is motivated by the urgency to improve road safety in high-risk areas, like India, where the number of accidents and fatalities remains distressingly high. Our research delves into the modeling of accident severity, a statistical technique in the field of road safety. Traditional statistical models like logit and probit have been used since the 1990s to predict traffic accident severity, but they have limitations, especially when their underlying assumptions are violated. 3 , 4 On the other hand, Artificial Intelligence (AI) models offer adaptability and can handle complex nonlinear relationships without being constrained by such assumptions. We focus on the Random Forest (RF) algorithm, an advanced machine learning model with distinct advantages over other algorithms for predicting accident severity. 5 Through tuning parameters and preprocessing techniques, we aim to improve the performance of RF even further, making our academic contribution significant in the pursuit of more accurate and reliable predictions for accident severity. The modelling process involves analyzing data on past accidents and identifying the factors that contributed to their occurrence and severity. 6 These factors can include road conditions, weather, driver behaviour, and vehicle type, among others. The goal of accident severity modelling is to identify the most important factors contributing to accidents and to develop evidence-based strategies to improve road safety and reduce the number and severity of accidents. 7 – 9 Statistical models have been widely used for predicting traffic accidents’ severity. 3 , 4 Artificial intelligence models, in contrast, do not make any assumptions and are more adaptable. They can handle intricate nonlinear relationships and generally offer higher predictive accuracy than statistical approaches. 5 RF, in particular, has been successfully applied in various contexts, and its performance can be further improved by tuning key parameters and careful data preprocessing. 4 , 5 , 10 , 11 The performance of the RF algorithm is significantly influenced by the selection of hyperparameters. 12 To optimize its performance, identifying the optimal parameter values is crucial. The main objective of our study is to develop a predictive model for the severity of traffic accidents on Indian highways using RF models due to their accuracy and interpretability. The findings of our study will be used to develop a predictive model for accident severity that can inform road safety policies and interventions. This model can be used to identify high-risk areas and prioritize resources for accident prevention and mitigation. Study Areas The study areas selected are the National Highways two stretches as mentioned below 1. Pune-Sholapur Section of NH-9 in km.144/400 to Km. 249/000 in the State of Maharashtra ( Figure 1 ). 2. Six-Laning of Barwa-Adda-Panagarh Section of NH-2 from km 398.240 to km 521.120 including Panagarh Bypass in the States of Jharkhand and West Bengal ( Figure 2 ). The study areas for this research project were selected based on specific criteria. Firstly, the researchers had prior experience of working on one of the stretches, which is the Pune-Sholapur Section of NH-9 in km.144/400 to Km. 249/000 in the State of Maharashtra. This experience could have provided insights and knowledge that could be useful in conducting the study. Additionally, data was also provided by the same concessionaire as of the previous stretch on request for another stretch, which is the Six-Laning of Barwa-Adda-Panagarh Section of NH-2 from km 398.240 to km 521.120 including Panagarh Bypass in the States of and West Bengal. This data could have been relevant to the research objectives and could have assisted in achieving the desired outcomes.
Methods The proposed methodology for this research involves the following steps for implementing a RF model machine learning technique for accident severity prediction. Data Preparation: The first step in implementing a RF model for accident severity prediction is to collect and prepare data. Raw data on road accidents for the selected stretches of the highway can be obtained from secondary sources such as the Ministry of Road Transport and Highways (MoRTH) and the National Highways Authority of India (NHAI). 2 Data wrangling and mining techniques can be used to clean and preprocess the data. Feature Selection: Once the data is prepared, selecting appropriate features for the model becomes crucial. Feature selection plays a vital role in reducing the dimensionality of the data and enhancing the model’s accuracy. There are several techniques available for feature selection, such as statistical tests, correlation analysis, and principal component analysis (PCA). 13 Model Training: In the next step, an RF model can be trained on the preprocessed data. The model can be developed using a machine learning-based framework, as described in Breiman’s work on RF. 14 The RF algorithm involves bagging and random feature selection techniques to create multiple decision trees that are aggregated to form a stronger learner. 15 RF Algorithm Formulation: The RF algorithm can be represented as: where X are the input features, B is the number of trees, and T b(X) is the prediction of the b-th individual decision tree. Parameter Tuning: To improve the performance of a RF model, it is important to fine-tune its parameters. The three key parameters that significantly impact the tuning performance of the RF model are the total number of trees (n_estimators), the number of features used for each node segmentation (max_feature), and the maximum depth of a tree (max_depth). 16 In the construction of the RF model for predicting accident severity, Gini impurity is employed as a criterion to evaluate the significance of different explanatory variables. Gini impurity, a measure utilized within the framework of decision trees (the base learners in a RF), is crucial for the optimal selection of features at each node split. It offers a quantitative metric to discern the effectiveness of a variable in segregating the target classes. Mechanism of Gini Impurity: In the context of binary classification, the Gini impurity for a node is calculated as: where, pk ​ is the proportion of samples classified to class k at that node, and the summation operates over all classes. A lower Gini impurity score suggests a higher purity of the node, indicating an enhanced classification. Gini Importance in RF: In the developed RF model, the Gini impurity plays a dual role: Node Splitting : It aids in the identification of the most significant variable at each node by evaluating the potential reduction in impurity for each split, and Feature Importance : Post model training, the average decrease in impurity caused by each feature across all trees is computed, known as Gini importance. This metric offers insights into the relative significance of different features for the prediction task. Model Evaluation: After training the RF model and optimizing its parameters, it is important to evaluate the model’s performance. Various evaluation metrics can be used, including accuracy, precision, recall, F1 score, and Area Under the Curve - Receiver Operating Characteristics (AUC-ROC) curve. 17 Model Implementation: Once the model has been trained and evaluated, it can be deployed for accident severity prediction. The methodology can be designed using Python for building the model and forecasting the severity of road traffic accidents on Indian highways. Source data Data on road accidents from selected stretches of highways was obtained from the Concessionaires of the National Highways Authority of India (NHAI) for two projects: Pune-Solapur and Bengal (BAEL) Section. For the Pune-Solapur Section of NH-9, which is located between km. 144/400 and km. 249/000 in the state of Maharashtra, accident dates from 2013 to 2018 were used. For the Six-Laning of Barwa-Adda-Panagarh Section of NH-2, which includes Panagarh Bypass and is located in the States of Jharkhand and West Bengal Stretch, accident dates from 2015 to 2019 were used for the stretch between km 398.240 and km 521.120. The raw data was subject to exploratory data analysis, as detailed in the following section. Data Preparation In this stage, data gathering and exploration is performed using secondary source data. The dataset consists of 3257 observations out of which the 1855 observations are of Bengal (BAEL) Section and 1402 observations are of Pune- Solapur and 32 variables, including the target variable “accident severity.” The 32 attributes and their corresponding mappings are presented in Table 1 . Data Modelling The RF classification algorithm has been employed in this study to forecast the severity of road traffic accidents in India. This section details the procedure for implementing the model, performance evaluation, and discuss the results obtained. The RF algorithm is written using python programming language. The target variable for the random RF is selected as the’Accident Severity’ which has classes as Fatal, Grevious Injury, Minor Injury and No Injury and indexed as 1-Fatal, 2-Grevious Injury, 3-Minor Injury, 4-No Injury. The dataset is partitioned into training and testing sets with a ratio of 80% and 20%, respectively. The hyperparameters’n_estimators’ and’max_depth’ are specified, and a grid search is conducted with cross-validation (cv=5) to identify the optimal hyperparameters. The best parameters and scores are obtained. The best estimator is fit on the training data. Predictions are made on the test data and the accuracy of the model is obtained. The algorithm and programme for Accident Severity Modelling using RF are written in the Python programming language, and the code is made available to the public for further development. The source code can be accessed via the software availability statement. Accuracy analysis on test data: Three metrics were employed to evaluate the effectiveness of the algorithms: accuracy, precision, and recall. These metrics are defined as follows: Accuracy: The formula for a metric that measures the proportion of correctly predicted observations to the total number of observations is represented as: Precision is a metric that indicates the ratio of correctly predicted positive observations to the total number of predicted positive observations, and is calculated using the formula: Recall is a metric that reflects the ratio of correctly predicted positive observations to the total number of actual positive observations, and is determined using the formula:
Result and Discussion Model Performance The classification model used three hyperparameters -’max_depth’: 10,’max_features’:’sqrt’, and’n_estimators’: 100, and the results generated a confusion matrix for the training set. The matrix indicated the number of correctly and incorrectly classified instances for each class. The classification report provided precision, recall, and f1-score for each class, along with support. The model showed high precision and recall for class 1 but low precision and recall for classes 2, 3, and 4, with an overall accuracy of 67% and a weighted average f1-score of 0.64 on the training set. The macro average f1-score, which assigns equal weight to each class, was 0.53. The optimal parameters for a RF classifier model were determined through a grid search, with a max depth of 2, n estimators of 5000, and a random state of 0. The model was then applied to the test data, and the predictions were saved in an Excel file called “predicted output3.xls” for further analysis. The accuracy of the model on the test data was determined to be 0.4147, or approximately 41.47%, indicating that it accurately predicted the severity of traffic accidents in about 41.47% of test cases. Predicted outputs Comparative analysis of observed and predicted accident severity index against dates The actual accident severity indices are represented by the observed values, while the predicted values are generated by the RF model using the input features. The following is a summary ( Figure 3 ) of the comparison between the observed and predicted values: On dates such as 25-02-2017, 17-04-2017, and 22-04-2017, the RF model accurately predicts the accident severity index. In a number of instances, the model predicts a lower accident severity index value than the observed value. 18-02-2017, 23-02-2017, and 27-03-2017, for example. Occasionally, the model overestimates the accident severity index by predicting a higher value than the observed value, as on 24-05-2017 and 20-10-2017. In general, the model frequently predicts a severity index of 2 for accidents, even when the observed values are distinct. This may indicate a bias in the model, possibly as a result of an imbalance in the training dataset, in which severity index 2 occurs more frequently than other categories. Comparative analysis of observed and predicted accident severity index against time Figure 4 displays the date, day of the week, and time of the accident, as well as the observed and predicted accident severity indices. The plotted for the 165 rows of predicted data doesn’t fit in A4 sheet hence the data is published and the link is provided in the Tableau graphs visuals availbility [i]. The dataset contains accident data from February 18, 2017 to December 31, 2017, as determined by Tableau analysis of the plot generated from the provided Excel table. The observed accident severity index ranges from 1 to 4, where 1 corresponds to the least severe accident and 4 to the most severe accident. The observed severity index for the vast majority of accidents in the dataset is 3, followed by 4. 2 indicates a less severe accident, while 4 indicates a more severe accident. The majority of accidents within the dataset have a predicted severity index of 2, followed by an index of 1. The analysis of the scatter plot reveals that the predicted severity index is typically lower than the observed severity index. This suggests that the model used to predict the severity of accidents is not always accurate and could be improved. Comparative analysis of observed and predicted accident severity index against Location and Chainages- RHS The Tableau plot ( Figure 5 ) presents a detailed visual analysis of accident data on the right-hand side of the road. The data is organized by date and day of the week, displaying the accident location, observed accident severity index, and predicted accident severity index for each incident. The plot effectively illustrates the spatial distribution of accidents and their severity over time, enabling the identification of patterns and trends. The Tableau plot doesn’t fit in A4 sheet hence the data is published and the link is provided in the Tableau graphs visuals availability [ii]. It is evident from the analysis that the majority of accidents have an observed severity index of 2 or 3, indicating a moderate severity. However, the predicted accident severity index largely remains at 2, indicating that the predictions may be somewhat conservative and do not fully capture the observed severity range. In addition, there appears to be no correlation between the day of the week and the frequency or severity of accidents across the different days of the week. This may suggest that external factors, such as traffic patterns or weather conditions, have a greater impact on the occurrence and severity of accidents than the day of the week. Comparative analysis of observed and predicted accident severity index against Location and Chainages- LHS The graph displays ( Figure 6 ) the date, day of the week, and accident location on Left Hand Side (LHS) of the road, as well as the observed and predicted accident severity indices. The plotted of predicted data doesn’t fit in A4 sheet hence the data is published and the link is provided in the Tableau graphs visuals availability [iii]. The scatterplot reveals that the majority of accidents on the left side of the road had a severity index of 2 or 3, with only a few instances of severity index 1 and 4. This indicates that the majority of collisions on the left side of the road were of moderate severity. In the majority of cases, the predicted accident severity index was 2, with only a few instances of values 3 and 4. This suggests that the predictive model may be biased towards predicting less severe accidents. There was no discernible pattern or trend between the day of the week and the occurrence of accidents. Accidents appeared to occur every day, indicating that the day of the week may not be a significant predictor of accident severity on the left side of the road. The accident locations, as measured by Accident Location A Chainage km, were scattered along the roadway at various distances. This suggests that there may not be a particular accident hotspot or concentration on the left-hand side of the road. Data Recording and availability The recording of road accident data in India must comply with the MoRTH & IRC guidelines, utilizing the Road Accident Recording and Reporting Formats. Despite this, there exists a need for a more advanced data recording system to effectively model road safety. The digital monitoring of road accidents can increase the frequency of data collection and minimize the absence of crucial information. Often, the lack of a system or individual to document the accident leads to the absence of important road accident data. This missing data can be regained through the use of machine learning, thus enhancing the accuracy of road safety modeling.
Result and Discussion Model Performance The classification model used three hyperparameters -’max_depth’: 10,’max_features’:’sqrt’, and’n_estimators’: 100, and the results generated a confusion matrix for the training set. The matrix indicated the number of correctly and incorrectly classified instances for each class. The classification report provided precision, recall, and f1-score for each class, along with support. The model showed high precision and recall for class 1 but low precision and recall for classes 2, 3, and 4, with an overall accuracy of 67% and a weighted average f1-score of 0.64 on the training set. The macro average f1-score, which assigns equal weight to each class, was 0.53. The optimal parameters for a RF classifier model were determined through a grid search, with a max depth of 2, n estimators of 5000, and a random state of 0. The model was then applied to the test data, and the predictions were saved in an Excel file called “predicted output3.xls” for further analysis. The accuracy of the model on the test data was determined to be 0.4147, or approximately 41.47%, indicating that it accurately predicted the severity of traffic accidents in about 41.47% of test cases. Predicted outputs Comparative analysis of observed and predicted accident severity index against dates The actual accident severity indices are represented by the observed values, while the predicted values are generated by the RF model using the input features. The following is a summary ( Figure 3 ) of the comparison between the observed and predicted values: On dates such as 25-02-2017, 17-04-2017, and 22-04-2017, the RF model accurately predicts the accident severity index. In a number of instances, the model predicts a lower accident severity index value than the observed value. 18-02-2017, 23-02-2017, and 27-03-2017, for example. Occasionally, the model overestimates the accident severity index by predicting a higher value than the observed value, as on 24-05-2017 and 20-10-2017. In general, the model frequently predicts a severity index of 2 for accidents, even when the observed values are distinct. This may indicate a bias in the model, possibly as a result of an imbalance in the training dataset, in which severity index 2 occurs more frequently than other categories. Comparative analysis of observed and predicted accident severity index against time Figure 4 displays the date, day of the week, and time of the accident, as well as the observed and predicted accident severity indices. The plotted for the 165 rows of predicted data doesn’t fit in A4 sheet hence the data is published and the link is provided in the Tableau graphs visuals availbility [i]. The dataset contains accident data from February 18, 2017 to December 31, 2017, as determined by Tableau analysis of the plot generated from the provided Excel table. The observed accident severity index ranges from 1 to 4, where 1 corresponds to the least severe accident and 4 to the most severe accident. The observed severity index for the vast majority of accidents in the dataset is 3, followed by 4. 2 indicates a less severe accident, while 4 indicates a more severe accident. The majority of accidents within the dataset have a predicted severity index of 2, followed by an index of 1. The analysis of the scatter plot reveals that the predicted severity index is typically lower than the observed severity index. This suggests that the model used to predict the severity of accidents is not always accurate and could be improved. Comparative analysis of observed and predicted accident severity index against Location and Chainages- RHS The Tableau plot ( Figure 5 ) presents a detailed visual analysis of accident data on the right-hand side of the road. The data is organized by date and day of the week, displaying the accident location, observed accident severity index, and predicted accident severity index for each incident. The plot effectively illustrates the spatial distribution of accidents and their severity over time, enabling the identification of patterns and trends. The Tableau plot doesn’t fit in A4 sheet hence the data is published and the link is provided in the Tableau graphs visuals availability [ii]. It is evident from the analysis that the majority of accidents have an observed severity index of 2 or 3, indicating a moderate severity. However, the predicted accident severity index largely remains at 2, indicating that the predictions may be somewhat conservative and do not fully capture the observed severity range. In addition, there appears to be no correlation between the day of the week and the frequency or severity of accidents across the different days of the week. This may suggest that external factors, such as traffic patterns or weather conditions, have a greater impact on the occurrence and severity of accidents than the day of the week. Comparative analysis of observed and predicted accident severity index against Location and Chainages- LHS The graph displays ( Figure 6 ) the date, day of the week, and accident location on Left Hand Side (LHS) of the road, as well as the observed and predicted accident severity indices. The plotted of predicted data doesn’t fit in A4 sheet hence the data is published and the link is provided in the Tableau graphs visuals availability [iii]. The scatterplot reveals that the majority of accidents on the left side of the road had a severity index of 2 or 3, with only a few instances of severity index 1 and 4. This indicates that the majority of collisions on the left side of the road were of moderate severity. In the majority of cases, the predicted accident severity index was 2, with only a few instances of values 3 and 4. This suggests that the predictive model may be biased towards predicting less severe accidents. There was no discernible pattern or trend between the day of the week and the occurrence of accidents. Accidents appeared to occur every day, indicating that the day of the week may not be a significant predictor of accident severity on the left side of the road. The accident locations, as measured by Accident Location A Chainage km, were scattered along the roadway at various distances. This suggests that there may not be a particular accident hotspot or concentration on the left-hand side of the road. Data Recording and availability The recording of road accident data in India must comply with the MoRTH & IRC guidelines, utilizing the Road Accident Recording and Reporting Formats. Despite this, there exists a need for a more advanced data recording system to effectively model road safety. The digital monitoring of road accidents can increase the frequency of data collection and minimize the absence of crucial information. Often, the lack of a system or individual to document the accident leads to the absence of important road accident data. This missing data can be regained through the use of machine learning, thus enhancing the accuracy of road safety modeling.
Conclusion The RF classifier model predicted the severity of traffic accidents with an overall accuracy of 67% on the training set and approximately 41.47% on the test set. Indicating possible bias or imbalance in the training dataset, the model tended to predict a lower severity index than the observed values. There were no discernible relationships between the day of the week and the occurrence or severity of accidents. The performance of the model can be enhanced by correcting the dataset imbalance and refining the model’s hyperparameters. The observed and predicted accident severity indices were compared against a number of variables, including dates, times, and locations on both sides of the road. In some instances, the model accurately predicted the accident severity index, but it frequently underestimated accident severity. No discernible patterns or trends were observed in terms of accident location, indicating that external factors may have a greater influence on the occurrence and severity of accidents. To improve road safety modelling, it is essential to adopt a more sophisticated data recording system consistent with MoRTH and IRC recommendations. Digital monitoring of road accidents can increase the frequency of data collection and reduce the loss of vital information. Integrating machine learning techniques can contribute to more effective interventions and decision-making in the field of traffic accident prevention and mitigation. In the field of accident severity modeling our research stands out for its contributions. We focus on leveraging Artificial Intelligence (AI) models, which excel at capturing the relationships that traditional statistical methods often overlook. Our, in depth study of the Random Forest (RF) algorithm, combined with careful parameter adjustment and data preprocessing highlights its potential in this area. This research addresses not concerns but also specifically tackles India’s road safety challenges providing insights applicable worldwide as well as tailored solutions for the region. A key aspect of our approach is our unwavering dedication to improving accuracy positioning our work as a standard for precise and reliable accident severity predictions. Overall this study makes a contribution to literature, in this field.
No competing interests were disclosed. Background: Road accidents claim around 1.35 million lives annually, with countries like India facing a significant impact. In 2019, India reported 449,002 road accidents, causing 151,113 deaths and 451,361 injuries. Accident severity modeling helps understand contributing factors and develop preventive strategies. AI models, such as random forest, offer adaptability and higher predictive accuracy compared to traditional statistical models. This study aims to develop a predictive model for traffic accident severity on Indian highways using the random forest algorithm. Methods: A multi-step methodology was employed, involving data collection and preparation, feature selection, training a random forest model, tuning parameters, and evaluating the model using accuracy and F1 score. Data sources included MoRTH and NHAI. Results: The classification model had hyperparameters ‘max depth’: 10, ‘max features’: ‘sqrt’, and ‘n estimators’: 100. The model achieved an overall accuracy of 67% and a weighted average F1-score of 0.64 on the training set, with a macro average F1-score of 0.53. Using grid search, a random forest Classifier was fitted with optimal parameters, resulting in 41.47% accuracy on test data. Conclusions: The random forest classifier model predicted traffic accident severity with 67% accuracy on the training set and 41.47% on the test set, suggesting possible bias or imbalance in the dataset. No clear patterns were found between the day of the week and accident occurrence or severity. Performance can be improved by addressing dataset imbalance and refining model hyperparameters. The model often underestimated accident severity, highlighting the influence of external factors. Adopting a sophisticated data recording system in line with MoRTH and IRC guidelines and integrating machine learning techniques can enhance road safety modeling, decision-making, and accident prevention efforts. Amendments from Version 1 We express our gratitude to the reviewers for their constructive comments that greatly enhanced the manuscript's quality. In response to Reviewer-1's insightful observation on the term "accident", we have clarified the context and regional usage of the term in the paper, while also acknowledging the global standards. As per suggestions, we've incorporated the Random Forest algorithm formulation in the methods section and discussed the significance of the Gini impurity test in understanding the importance of explanatory variables. Addressing Reviewer-2's feedback, we've emphasized the motivation behind our work and its academic contribution in the Introduction section, ensuring a comprehensive understanding of our study's significance. We revisited and corrected specific literature citations, particularly references 7, 14, and 15. The choice of factors, their contribution, and the transferability of the developed model are now prominently highlighted in a dedicated subsection. Further, to address the concerns about the novelty, we've incorporated a detailed paragraph in the conclusion, elucidating the innovative aspects and unique contributions of our predictive model in comparison to existing literature. We believe these revisions have amplified the clarity, depth, and significance of our research, ensuring alignment with the journal's standards and objectives.
Future Scope The study presented provides a good starting point for future research in the field of road safety modeling and accident prevention for Indian highways. However, with the limitations of the present study there opens potential areas for future research as mentioned below which will be taken up in continuation. Dataset improvement: The study identified the possibility of dataset bias and imbalance affecting model performance. Future research will focus on improving the quality and quantity of data, reducing bias and improving model performance. This will involve exploring alternative data sources, enhancing data collection methods, and addressing data quality issues. Model improvement: The study used the RF algorithm to develop a predictive model for traffic accident severity. In future research, other machine learning algorithms or ensemble models to improve model performance will be explored. Additionally, refining hyperparameters and addressing dataset imbalance will be done to improve model accuracy. External factors analysis: The study highlighted the influence of external factors on accident severity prediction. Future research can focus on exploring the impact of external factors such as weather conditions, road infrastructure, and driver behavior on accident severity. This can enhance the accuracy of predictive models and inform decision-making in accident prevention efforts. Real-time monitoring: The study highlighted the need for a sophisticated data recording system in line with MoRTH and IRC guidelines. Future research can focus on developing a real-time monitoring system that can capture road safety data in real-time and provide insights for accident prevention efforts.
Acknowledgement We are grateful to National Highways Authority of India and IL&FS Engineering and Construction Company for making the raw accident data available. Data availability Zenodo. Data for Accident Severity Prediction Modelling for Indian Highways Case Study, https://doi.org/10.5281/zenodo.7773156 . 18 This project contains the following underlying data: • Accdataset_hk_PS_BAEL_Combined.csv (The dataset consists of 3257 observations out of which the 1855 observations are of Bengal (BAEL) Section and 1402 observations are of Pune- Solapur.) • predicted_output_1.xlsx (This is level-2 processed data derived from raw accident data using prediction modeling. The data has been indexed from 1 to 4 for further analysis, and there are a total of 165 rows in the predicted output observations. Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0). Software availability • Github: https://github.com/humera-k/RF_Accident_Severity • https://zenodo.org/badge/latestdoi/616376786 Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0). Tableau graphs visuals availability 1. Accidental_Analysis_1 | Tableau Public (Comparative analysis of observed and predicted accident severity index against time) 2. Accidental_Analysis_1 (Comparative analysis of observed and predicted accident severity index against Location and Chainages-Right hand Side (RHS)) 3. Accidental_Analysis_1 (Comparative analysis of observed and predicted accident severity index against Location and Chainages-Left Hand Side (LHS))
CC BY
no
2024-01-15 23:43:45
F1000Res. 2023 Oct 20; 12:494
oa_package/95/3e/PMC10787871.tar.gz
PMC10787873
37815684
Introduction Breast cancer is the most common malignancy affecting women worldwide [ 1 ]. In the United States, approximately 5% of patients with breast cancer have de novo metastatic disease at diagnosis, and at least 20% of those initially diagnosed with early-stage breast cancer subsequently develop metastatic disease [ 2 ]. Despite advances in therapies, the prognosis of patients with metastatic breast cancer (MBC) remains poor, with a median time to progression on first-line therapy of 9.7 months for triple-negative breast cancer (TNBC) and 25.3 months for HR + , HER2 negative breast cancer and a median overall survival (OS) of 23 months for TNBC and 63.9 months for HR + , HER2 negative breast cancer [ 3 – 5 ]. Patients with metastatic disease develop cumulative toxicity from multiple lines of chemotherapy, as well as chemotherapy resistance, which limits efficacy. In order to improve survival while maintaining quality of life, it is important to identify new treatment regimens for patients with MBC. New chemotherapy combinations may improve the duration of response (DOR) and also may provide a superior chemotherapy backbone for the addition of targeted agents in future studies. Eribulin mesylate, a nontaxane microtubule dynamics inhibitor, is a structurally simplified, synthetic analog of the natural product Halichondrin B, isolated from the marine sponge Halichondria okadai [ 6 , 7 ]. Eribulin suppresses polymerization and sequesters tubulin into nonfunctional aggregates [ 6 , 8 – 10 ] and has other cytotoxic effects, including vascular remodeling, reversal of the epithelial–mesenchymal transition, induction of the differentiation, and suppression of migration and invasion [ 11 , 12 ]. Eribulin is administered as a 2- to 5-min IV infusion without the need for premedications, so it is easier to administer than other chemotherapy agents [ 13 ]. Eribulin was approved as monotherapy for the treatment of taxane and anthracycline-resistant MBC based on results from the EMBRACE trial, which reported a 2.5-month improvement in median OS with eribulin compared to treatment of physicians choice (TPC, 13.1 months vs 10.6 months; HR 0.81, 95% CI 0.66–0.99 p = 0.041) in women with heavily pre-treated MBC [ 14 ]. Response rate (RR) was significantly longer with eribulin, with a non-significant numerical difference in PFS. In another study, eribulin was compared to capecitabine as an earlier line therapy (up to two lines of chemotherapy) and showed no difference in OS or PFS, but a pooled subset analysis suggested improved OS with eribulin in patients with HER2-negative and TNBC [ 15 , 16 ]. In both studies, treatment-related adverse events (AEs) with eribulin included neuropathy and neutropenia, requiring higher rates of growth factor administration [ 17 – 19 ]. Combination chemotherapy with docetaxel and cyclophosphamide (TC) has become a standard treatment option for early-stage lower risk breast cancer based on data from several studies showing improved or similar outcome compared to anthracycline-based regimens [ 20 – 22 ] Compared to doxorubicin and cyclophosphamide (AC), treatment with TC improved OS in patients with up to three positive axillary nodes [ 20 ]. Based on encouraging efficacy with TC, we hypothesized that eribulin combined with cyclophosphamide (EC) would be effective in taxane-resistant disease with tolerable toxicity. The aim of this study was to determine the maximum tolerated dose (MTD) of EC, followed by a dose expansion study to estimate the clinical benefit rate (CBR) of EC in patients with advanced breast cancer (ABC). We also performed correlative studies to assess the correlation of circulating tumor cells (CTCs) with response and survival.
Methods Patients Male or female patients ≥ 18 years who had histologically confirmed locally advanced, unresectable or metastatic carcinoma of the breast of all subtypes were eligible to enroll with no limitation on prior lines of therapy. Eligibility included an Eastern Cooperative Oncology Group performance status (ECOG PS) of 0–2, measurable disease, adequate organ and bone marrow functions (neutrophils > 1.0 × 10 9 /L, platelets > 100 × 10 9 /L, hemoglobin > 9 g/dL, total bilirubin < 1.5 × upper limit of normal (ULN), AST and ALT ≤ 3 × ULN or ≤ 5 × ULN in patients with known liver metastasis, creatinine ≤ 1.5 × ULN or ≥ 60 mL/min for patients with creatinine levels > 1.5 × institutional ULN), ≤ grade 1 peripheral neuropathy, and a life expectancy of at least 3 months. Patients with stable treated brain metastases were also eligible to enroll. Exclusion criteria included known active central nervous system (CNS) metastases and/or carcinomatous meningitis, a corrected QT interval (cQT) > 480 ms, or significant cardiovascular disease within the past 6 months. The study was approved by the University of California San Francisco (UCSF) Comprehensive Cancer Center Protocol Review Committee on Human Research, and written informed consent was obtained from all patients prior to trial enrollment. The trial was registered at ClinicalTrials.gov (NCT01554371). Study design Patients were treated using a 3 + 3 dose confirmation strategy for eribulin with dose expansion at the MTD. Cyclophosphamide was given at a fıxed dose of 600 mg/m 2 on day 1 of a 21-day cycle; eribulin was given day 1 and 8 every of a 21 day cycle, and was escalated from 1.1 mg/m 2 at dose level 0 (DL0) to 1.4 mg/m 2 at dose level 1 (DL1); both drugs were given intravenously. Dose expansion occurred at DL1. The primary objective of dose escalation was to determine the MTD of EC. The highest dose level at which no more than one of six subjects experienced a dose limiting toxicity (DLT) defined the MTD. DLTs were defined as grade 3 or 4 clinically evident non-hematologic toxicity; grade 4 neutropenia, thrombocytopenia lasting > 7 days, febrile neutropenia or any clinically significant toxicity grade 2 or higher that required more than 14 days to resolve occurring within the first 21 days of combination therapy. The primary objective of the dose expansion was CBR at three months; we selected CBR at 3 months rather than 6 months, as these patients were heavily pre-treated so we anticipated shorter responses to therapy in the later line setting. Secondary objectives were RR, DOR, PFS, safety, and correlation of CTCs with clinical benefit. Drug doses were modified for treatment-related toxicity. These toxicities were neutropenia (absolute neutrophil count [ANC] < 1000/μL), thrombocytopenia, rash, GI toxicity, liver abnormalities and neuropathy. If toxicity occurred in a patient, dose reductions were managed as follows. Dose levels -1 and -2 for eribulin were 1.1 mg/m 2 and 0.7 mg/m 2 , for cyclophosphamide one dose reduction was allowed to 500 mg/m 2 . Dose reductions were sequential, with the first dose reduction for cyclophosphamide, followed by eribulin to dose level -1 then -2 for persistent toxicity. If toxicity persisted despite these dose reductions and/or if the participant experienced a cycle delay of three or more weeks, study treatment was discontinued (Supplement 1). Concomitant medication Patients received prophylactic antiemetics and premedications according to standard institutional guidelines. Colony stimulating growth factor use was allowed at the discretion of the treating physician. Palliative radiotherapy was permitted to control bone pain as long as the irradiated area was limited in extent. Other investigational agents and potent inhibitors or inducers of CYP3A4 were not permitted. Assessments Baseline evaluations included medical history, a physical examination, Eastern Cooperative Oncology Group (ECOG) Performance Status (PS) tumor imaging with computed tomography (CT), bone scan, laboratory tests (hematology, blood chemistry), a serum pregnancy test for females of child-bearing potential, and an electrocardiogram with QTc measurement. In the dose confirmation cohort (phase Ib), the response to EC was evaluated every 6 weeks until disease progression according to the investigator, based on objective tumor assessments using RECIST version 1.1 criteria. In the dose-expansion cohort (phase II), response to EC was evaluated after study start, then every 9 weeks until end of study therapy. Safety/tolerability Safety evaluations at baseline and subsequent visits included AEs, clinical laboratory tests, physical examination, and vital signs. AEs were assessed and AE severity was graded in accordance with the National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE version 4.0). Both agents were held for grade 3 or febrile neutropenia (ANC < 1000/μL). Filgrastim or pegylated-filgrastim myeloid growth factor support was encouraged for ANC < 1500/μL and was allowed at the discretion of the treating physician in order to maintain adequate blood counts. Filgrastim was given for ANC < 1000/μL at any time, or as prophylaxis in patients at risk for neutropenia. Neuropathy was assessed using the 10-point Modified Total Neuropathy Score at the start of each cycle and at study termination. Correlative studies Whole blood samples were obtained in fixative-containing tubes (CellSave tubes, Veridex (currently Menarini)) and processed in the laboratory of Dr. John Park at University of California, San Francisco for CTC identification and enumeration using the CellSearch system. Samples with 5 CTCs per 7.5 mLs of blood were considered CTC-positive. Statistical analysis The study followed a standard dose-confirmation schema (phase Ib portion) with three patients per cohort (3 + 3 design) for a total of six patients. A two-stage design was performed in the dose-expansion (phase II portion) with a possible total of 40 patients. An overall RR of 25% was considered clinically meaningful. Using a two-stage Simon’s minimax design, the null and alternative hypothesis were H0: p0 < 10% versus Ha: p1 > 25% for the proportion of patients with complete or partial response (PR) by RECIST criteria. Based on the type I error of 5% and the type II error rate of 20%, p0 = 10% and p1 = 25%. Secondary efficacy variables were analyzed using Kaplan–Meier methods, with a corresponding median and a 95% CI.
Results Patient characteristics Patients with histologically confirmed metastatic or ABC with any number of prior lines of therapy were eligible to enroll in this study, and 44 patients enrolled in total. Baseline patient characteristics are summarized in Table 1 . The median age was 56 years (range 33–82 years). 31 patients (70.4%) had HR + /HER2- disease, 12 patients (27.3%) had TNBC, and 1 patient (2.3%) had HR + /HER2 + disease. Patients had a median of 1 prior line of hormone therapy (range 0–6) and 2 prior lines of chemotherapy (range 0–7). Most patients (97.7%) had visceral disease. The most common metastatic sites of disease were bone, lymph nodes, liver, and lung (Table 1 ). Antitumor activity The median duration of treatment was 14.7 weeks (1.8–53.3 weeks). The CBR was 79.5% (35/44; 7 PR, 28 SD). The median PFS was 16.4 weeks (95%CI: 13.8–21.1 weeks). (Figure 1 a). The median DOR was 16.4 weeks (13.8–21.1 weeks). Clinical response to EC therapy is summarized in Table 2 . Individual patient characteristics of those who had a PFS 24 weeks on EC are summarized in Table 3 . The CBR at 3 months in patients with HR + /HER2- disease was 83.9% (n = 31), and 12.9% of patients had a PR. The median PFS was 18.1 weeks. In the 12 patients with TNBC, the CBR was 66.7% with a PR rate of 25%. The median PFS was 10.8 weeks. As expected, PFS was longer in those with HR + disease versus those with TNBC (18.1 vs 10.8 weeks; p = 0.067) (Fig. 1 b). The CBR among patients who had received (neo)adjuvant treatment with an anthracycline and/or a taxane (A/T) was 81.8% (18/22), similar to that of the overall study population. There was no difference in PFS among patients who had received prior A/T (n = 24) versus those who had treatment without A/T (n = 20) (21.1 vs.15.1 weeks respectively, p = 0.4251). There was no difference in PFS among patients who received 0–2 prior lines of chemotherapy versus who received lines of chemotherapy (18.1 vs. 14.7 weeks respectively, p = 0.6736). In patients with HR + disease, patients who had received 0–2 prior lines of chemotherapy (n = 17) had a median PFS of 21.1 weeks whereas patients who had received lines of chemotherapy (n = 15) had a median PFS of 14.3 weeks. In patients with TNBC, patients who had received 0–2 prior lines of chemotherapy (n = 9) had a median PFS of 9.8 weeks whereas patients who had received lines of chemotherapy (n = 3) had a median PFS of 19.1 weeks. Drug exposure and safety No DLTs were identified in the dose confirmation (phase Ib) portion of the study. Three patients were treated at DL0 (eribulin 1.1 mg/m 2 ) and 3 were treated at DL1 (eribulin 1.4 mg/m 2 ), the MTD. Thirty-seven patients (84.1%) completed at least three cycles of treatment and 21 (47.7%) received cycles of treatment. The median number of cycles delivered was 5.8 (1.1–17.8); the median exposure was 3.6 weeks and nine patients received treatment for 6 months or longer (20.5%). Seventeen patients (38.6%) had 1 cyclophosphamide dose reduction and 12 patients (27.3%) had 1 eribulin dose reduction. Adverse events (AEs) are summarized in Table 4 . Twenty-eight patients (63.6%) experienced a grade 3/4 AE, the most common of which were neutropenia (47.7%, n = 21), fatigue (4.5%, n = 2), dyspnea (4.5%, n = 2), and anemia (2.3%, n = 1). Febrile neutropenia was reported for three patients (6.8%). Most patients (77.3%) received myeloid growth factors. Treatment related AEs led to dose adjustment (interruption/delay or reduction): 26 patients (59.1%) had a dose interruption/delay and 17 patients (38.6%) underwent dose reduction due to a treatment-related AE. Dose reductions due to neutropenia included a decrease of cyclophosphamide to 500 mg/m 2 (n = 17) and of eribulin to 1.1 mg/m 2 (n = 12). Five patients discontinued treatment due to fatigue (n = 3) or neutropenia (n = 2). Biomarkers CTCs in blood were enumerated at baseline and during treatment to explore the correlation between CTC levels and response to EC. Of the 44 evaluable patients, 26 had baseline CTC data. Of these, 14 were CTC-positive (53.8%). There was no significant difference in the mean CTCs/7.5 mLs of blood at baseline between subtypes (t-test p = 0.6435). There was no significant association between CTC status at baseline and treatment response at first scan (Fisher p = 0.5901). 18 of the 26 patients had paired CTC data at baseline and follow-up, 2 of whom were CTC positive at baseline and turned CTC-negative on treatment. The change in CTC status was not significantly associated with response to treatment (Fisher p = 0.3355). The median PFS was significantly shorter in patients who were CTC-positive at baseline compared to those who were CTC-negative (13.1 vs. 30.6 weeks, p = 0.011). Eighteen patients had on-treatment CTC data available (either during treatment or at the end of study). Of these, nine were CTC-positive (50%). The median PFS was significantly shorter in patients who were CTC-positive during treatment compared to those who were CTC-negative (13.1 weeks vs. 30.6 weeks, p = 0.035).
Discussion This trial was conducted to assess the safety, efficacy, and tolerability of EC for the treatment of patients with metastatic or ABC. CBR and PFS were 79.5% and 16.4 weeks respectively, comparing favorably to historic data of single agent eribulin for ABC (PFS 14.8 weeks) [ 14 ]. Our data demonstrate that EC has activity in extensively pretreated patients, as 68.2% of patients who enrolled had been treated with three to seven prior lines of chemotherapy. Of note, responses were observed in patients who had received prior anthracycline and taxane therapies. Patients with both HR + and TNBC were enrolled. In particular, metastatic TNBC is an aggressive breast cancer subtype associated with poor clinical outcomes highlighting the importance of identifying novel treatment approaches. Previous phase III studies demonstrated statistically significant improvements in OS in patients with metastatic TNBC treated with eribulin versus treatment of physician choice in both subgroup and pooled analyses. Specifically, in a pooled analysis of the EMBRACE study and Study 301, eribulin significantly improved OS compared with TPC in patients with TNBC (HR: 0.74, p = 0.006) [ 16 , 23 ]. In our study, 12 patients with metastatic TNBC received EC, with a CBR of 66.7% and a PR rate of 25%. The median PFS was 10.8 weeks with EC treatment. Among patients with metastatic TNBC, the heavily pre-treated patients had longer responses to EC than the less heavily-pretreated patients, although numbers are small; future larger studies can evaluate this further. The adverse events of eribulin in this study is consistent with what has been reported in previous studies [ 14 , 19 ]. In this study, the most frequently reported treatment related AEs were fatigue (68.2%), neutropenia (59.1%), nausea (56.8%), constipation (50%), peripheral neuropathy (47.7%), dyspnea (40.9%), headache (36.4%), and anorexia (36.4%), which reflect the known toxicity profiles of eribulin and cyclophosphamide. The most common grade 3/4 AE was neutropenia (47.7%; 6.8% febrile neutropenia). The incidence of neuropathy was 47.7%, but no patient experienced grade 3/4 neuropathy. Similarly, prior clinical trials also report high incidence of neutropenia and neuropathy with the use of eribulin. In the EMBRACE study, 52% of participants experienced neutropenia (grade 3: 8%, grade 4: 1%) and 35% of participants experienced peripheral neuropathy (grade 3: 8%, grade 4: 0.4%) [ 14 ]. In real-world studies of patients treated with eribulin grade 3/4 neutropenia occurred in 12% of patients and grade 3/4 neuropathy occurred in 2.6% of patients [ 24 ]. Overall, given that the patient population was heavily pre-treated in our study, it was reassuring that the toxicity profile of this combination chemotherapy regimen was similar to those previously reported in single agent studies. In this study, we administered eribulin using a 21-day cycle at a dose of 1.4 mg/m 2 on days 1 and 8 combined with cyclophosphamide at a fıxed dose of 600 mg/m 2 on day 1 of each cycle. Previous studies demonstrated that eribulin was more tolerable when administered on a 21-day schedule compared to a 28-day schedule [ 25 ]. Alternative schedules of eribulin administration have been investigated to provide better tolerance in patients who experienced myelosuppression. A modified biweekly regimen which provides additional time for bone marrow recovery may potentially improve safety compared with the 21-day dosing regimen [ 26 ]. In a prospective phase 2 trial, biweekly eribulin (1.4 mg/m 2 on days 1 and 15 of a 28-day cycle) was tolerable and had comparable antitumor activity in patients who were intolerant of the standard eribulin schedule [ 26 ]. Dose reductions due to neutropenia required patients to decrease to 500 mg/m 2 cyclophosphamide (n = 17) and to 1.1 mg/m 2 in eribulin (n = 12), consistent with expected hematologic toxicity in this heavily pre-treated population. Only five patients discontinued treatment due to AEs. Previous studies have suggested that the presence of CTCs in patients with MBC is associated with a worse prognosis [ 27 , 28 ] and can predict treatment response and progression [ 29 ]. Based on these prior work, we performed an exploratory CTC analysis to address whether CTC levels correlate with response to EC. Consistent with previous studies [ 27 , 28 ], the CTC positivity rate was 53.8% (14 of 26 patients) at baseline. Median PFS was significantly shorter in patients who were CTC-positive at baseline or during treatment compared to those who were CTC-negative. This study has several notable strengths. First, few studies have evaluated combination chemotherapy in patients with MBC who have received multiple lines of prior chemotherapy [ 14 , 30 ]. Inclusion of this patient population in our study suggests that the EC regimen may be more generalizable to real-world treatment scenarios. Second, this trial included patients with multiple breast cancer subtypes: most patients had HR + /HER2- breast cancer, with a smaller number of patients with TNBC, and one patient with HR + /HER2 + disease. The RR to EC was higher in patients with HR + /HER2- disease compared to patients with TNBC, as expected, but our study was not powered to fully detect differences between subtypes. Third, the CTC data provides interesting correlative data, and we found that the median PFS was significantly shorter in patients who were CTC-positive at baseline compared to those who were CTC-negative, providing rationale to continue to study the prognostic and predictive value of CTCs in ABC. This study also has several limitations. First, the primary endpoint of CBR provides important clinical information about response to EC, but this trial is not designed to provide data about OS. Second, since patients were heavily pre-treated, treatment history was fairly heterogenous, which clearly impacts both response to therapy and toxicity. This was a single-arm study so it is not possible to determine the efficacy of this regimen compared to others, and further studies are needed to clarify the efficacy of this regimen. Lastly, our study was conducted before regulatory approval of pembrolizumab plus chemotherapy for PD-L1 + metastatic TNBC, sacituzumab govitecan for metastatic TNBC and heavily pre-treated HR + /HER2- MBC, and trastuzumab deruxtecan for HER2-low MBC. However, as patients being treated with eribulin today will be even more heavily pre-treated than the patients in this trial, the activity of this combination regimen may have implications for current therapeutic options. In conclusion, the results of this trial demonstrate that EC has antitumoral activity in heavily pretreated patients with locally ABC or MBC. Importantly, EC demonstrated a manageable tolerability profile. These results support the additional clinical development of EC as a novel treatment combination for the treatment of ABC.
Purpose We hypothesized that eribulin combined with cyclophosphamide (EC) would be an effective combination with tolerable toxicity for the treatment of advanced breast cancer (ABC). Methods Patients with histologically confirmed metastatic or unresectable ABC with any number of prior lines of therapy were eligible to enroll. In the dose escalation cohort, dose level 0 was defined as eribulin 1.1 mg/m 2 and cyclophosphamide 600 mg/m 2 , and dose level 1 was defined as eribulin 1.4 mg/m 2 and cyclophosphamide 600 mg/m 2 . Eribulin was given on days 1 and 8 and cyclophosphamide on day 1 of a 21-day cycle. In the dose expansion cohort, enrollment was expanded at dose level 1. The primary objective was clinical benefit rate (CBR), and secondary objectives were response rate (RR), duration of response (DOR), progression-free survival (PFS), and safety. Results No dose-limiting toxicities were identified in the dose escalation cohort (n = 6). In the dose expansion cohort, an additional 38 patients were enrolled for a total of 44 patients, including 31 patients (70.4%) with hormone receptor-positive (HR +)/HER2- disease, 12 patients (27.3%) with triple-negative breast cancer (TNBC), and 1 patient (2.3%) with HR + /HER2 + disease. Patients had a median age of 56 years (range 33–82 years), 1 prior line of hormone therapy (range 0–6), and 2 prior lines of chemotherapy (range 0–7). CBR was 79.5% (35/44; 7 partial response, 28 stable disease) and the median DOR was 16.4 weeks (range 13.8–21.1 weeks). Median PFS was 16.4 weeks (95% CI: 13.8–21.1 weeks). The most common grade 3/4 adverse event was neutropenia (47.7%, n = 21). Fourteen of 26 patients (53.8%) with circulating tumor cell (CTC) data were CTC-positive ( 5 CTC/7.5 mL) at baseline. Median PFS was shorter in patients who were CTC-positive vs. negative (13.1 vs 30.6 weeks, p = 0.011). Conclusion In heavily pretreated patients with ABC, treatment with EC resulted in an encouraging CBR of 79.5% and PFS of 16.4 weeks, which compares favorably to single-agent eribulin. Dose reduction and delays were primarily due to neutropenia. The contribution of cyclophosphamide to eribulin remains unclear but warrants further evaluation. NCT01554371. Supplementary Information The online version contains supplementary material available at 10.1007/s10549-023-07073-0. Keywords
Supplementary Information Below is the link to the electronic supplementary material.
Funding This work was supported in part by funding to the UCSF Regents from Eisai, Inc. Data availability Enquiries about data availability should be directed to the authors. Declarations Conflict of interest Amy Jo Chien: Research funding to institution: Merck, Puma, Amgen, Seagen. Michelle E. Melisko: Research funding to institution: Astra Zeneca, Novartis, KCRN Research, Puma, Seattle Genetics. For spouse: Speaker bureau/honoraria: AstraZeneca, Genentech, Gilead. Stock Ownership: Merrimack. John Park: Consulting/honoraria: AstraZenica, Genentech, Diichi, Gilead. Equity/stock ownership: Merrimack Pharma. Hope S. Rugo: Research support for clinical trials through the University of California: Pfizer, Merck, Novartis, Lilly, Roche, Daiichi, Seattle Genetics, Macrogenics, Sermonix, Boehringer Ingelheim, AstraZeneca, Astellas and Gilead. Honoraria from: Puma, Samsung, Mylan, Chugai, Blueprint, and NAPO. The remaining authors, declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
CC BY
no
2024-01-15 23:42:01
Breast Cancer Res Treat. 2024 Oct 10; 203(2):197-204
oa_package/13/b8/PMC10787873.tar.gz
PMC10787874
37204642
Introduction Many studies support the notion that when parents are consistent in their responses to their child’s misbehavior, children show less externalizing behavior (Barry et al., 2009 ; Gardner, 1989 ; Gryczkowski et al., 2010 ; Halgunseth et al., 2013 ; Lengua & Kovacs, 2005 ). However, these studies have focused on global self-reports, which mostly confound consistency of parental reactions across multiple episodes of misbehavior (e.g., punishing misbehavior in one episode, and condoning it in another) with consistency within a single episode of misbehavior (e.g., threatening with punishment, but leaving it in the end). Additionally, observational studies have mostly focused on within-episode consistency (Del Vecchio & O’Leary, 2006 ; Gardner, 1989 ). Therefore, it is not known whether within- and across-episode consistency play similar roles in the early development of externalizing problems. In this study, we use a daily diary approach to examine parental consistency both within and across episodes, offering a unique possibility to differentiate these two types of consistency. We examine associations between these aspects of consistency, and how they are associated with child externalizing behavior, both concurrently and longitudinally. Differentiating Within- from Across Disruptive Behavior Episode Consistency Empirically, studies using questionnaire measures of inconsistent discipline have indeed found it to be associated with more externalizing behavior in children (Barry et al., 2009 ; Gryczkowski et al., 2010 ; Halgunseth et al., 2013 ; Lengua & Kovacs, 2005 ). However, research to date has not separated within-episode consistency from across-episode consistency. For instance, the frequently used Alabama Parenting Questionnaire (Frick et al., 1999 ) includes items in the inconsistency scale asking parents whether they threaten to punish, but then do not do so in the end, or whether they let their child out of punishment early, which are examples of inconsistent responding within a single episode of misbehavior. Additionally, items are included asking parents whether their punishment depends on their mood, which concerns consistency across episodes of misbehavior. It is important to note that these types of inconsistency do not necessarily co-occur. Parents may be relatively inconsistent in their responses within a single episode, perhaps not feeling competent enough to follow through with an initial course of action and giving up along the way (Deković et al., 2010 ), yet they may be consistent in this response style across multiple episodes of misbehavior. Alternatively, parents may be consistent within an episode of misbehavior by providing a negative consequence and sticking to it, but respond in different ways to new episodes, for instance switching to ignoring the misbehavior or trying to redirect attention by making a joke. They may switch approaches because they are in a different mood state themselves (Rueger et al., 2011 ), or because they felt their approach was not effective in a previous instance. Several theoretical frameworks would predict that both types of consistency would be associated with increased externalizing problems. Attachment theory postulates that children form secure attachments to their caregivers if these are consistently responsive to their needs. With consistency, children learn that they can trust their caregiver to provide them with a secure base and a safe haven (Ainsworth et al., 2015 ). With inconsistency in caregiving, in contrast, children learn that their environment is unpredictable and insecure, increasing the risk of attachment problems and problem behavior (Madigan et al., 2016 ). Supporting this notion, unpredictable behavior from parents has been shown to impact the stress response in both very young children (Noroña-Zhou et al., 2020 ), as well as older children (Manczak et al., 2018 ). A dysregulated stress system may in turn result in problems with self-regulation, resulting in heightened disruptive behavior (Wesarg et al., 2020 ). Additionally, unpredictability likely hampers children’s ability to develop a sense of self-efficacy, because it prevents them from developing a sense of control over situations (Bandura, 1978 ; Lippold et al., 2016 ). Low self-efficacy, in turn, makes it more difficult to regulate anger and frustration, resulting in increased levels of disruptive behavior. Social learning theories emphasize operant conditioning principles which predict that within-episode inconsistency would be associated with increased problematic behavior because failing to follow-through with negative consequences reinforces the child for showing the misbehavior (Patterson, 1982 ). Additionally, consistent negative consequences across different disruptive behavior episodes, or at least a lack of reward, would quickly result in extinction of problematic behavior, whereas inconsistent discipline – with intermittent patterns of positive or negative reinforcement – makes it difficult for children to learn that their behavior is not acceptable. Both social learning and attachment frameworks would thus predict that within- as well as across-episode consistency would play a role in the maintenance of disruptive behavior. Although most questionnaire studies have confounded within- and across-episode consistency, some studies have specifically investigated within-episode consistency. Using observations of parent-child interactions, parents with children who were high on externalizing behavior were less likely to follow-through on an initial demand than parents of children who were low on externalizing behavior (Gardner, 1989 ). Additionally, mothers of aggressive toddlers have been observed to be more likely to react with both overreactivity and laxness to instances of child aggression than mothers of non-aggressive toddlers (Del Vecchio & O’Leary, 2006 ). However, with observations it is more difficult to assess how this within-episode consistency relates to across episode consistency. As observations are already so time-consuming, observing (enough) episodes of misbehavior to examine across episode consistency would be especially difficult. Using Daily Diaries to Assess Parental Consistency Daily diary assessments allow for overcoming the abovementioned limitations of single-time surveys or observations. Single-time self-reports of parental consistency may be biased, as correctly judging how consistent one is in their reactions might be even more difficult than broadly gauging whether one frequently reacts in a certain way (Lippold et al., 2016 ). A more valid way to assess parental consistency may therefore be to repeatedly ask parents about their actual behavior, and analyze the consistency across their responses. A small scale study examined parents’ daily fluctuations in their overreactive and lax discipline, with lax discipline indicating inconsistency by not following through. Although overreactivity and laxness were positively correlated at the between-person level, indicating that mothers who were generally more overreactive were also more lax, there was a negative association at the within-person level, indicating that when mothers were overreactive in a certain instance, they were less likely to be lax at that time, and vice versa (Passini et al., 2013 ). These results indicate that mothers can be inconsistent across episodes with regards to their within-episode consistency. However, associations with child externalizing behavior were not examined here. Another diary study did examine associations with externalizing behavior and found that for mothers with 5- 8-year old children maternal consistency across one week was associated with less child externalizing behavior (Villarreal et al., 2021 ). Consistency was operationalized as the within-person fluctuations in destructive conflict characteristics, which included both maternal punitive behavior as well as child and mother negativity, making it unclear whether this was an association with inconsistency in maternal behavior specifically. Yet another study examined daily reports of parents harsh and warm reactions to child misbehavior and found that consistency in the level of warmth was associated with less child ADHD symptoms, whereas consistency in the level of harshness was not (Li & Lansford, 2018 ). As these studies operationalized inconsistency as fluctuations in mean levels of parenting behaviors, we do not know whether parents exhibited different types of responses in the same instance of misbehavior. For instance, some parents may be more likely to only punish the child by taking away a privilege, whereas other parents are more likely to yell at the child, take away a privilege and also comfort the child, with the latter type of response perhaps especially confusing to the child. A previous observational study has indeed found that some parents were more likely to use both positive and negative discipline strategies within the same episode than others, but did not examine associations between this inconsistency and externalizing behavior (van Zeijl et al., 2007 ). The Present Study In the present study, we use daily diary data to differentiate consistent responding within a single episode of misbehavior from consistent responding across multiple episodes of misbehavior and examine how they are each associated with the severity of disruptive behavior in children. We assess within-episode consistency as the mean number of different reactions to a specific episode of misbehavior, distinguishing positive attention, positive consequence, negative consequences, negative attention, and ignoring. Across-episode consistency is assessed as the overall dispersion of mothers’ reactions across all possible categories, taking into account the total number of episodes of misbehavior across a week. Parental reactions that were concentrated in fewer reaction categories were indicative of more consistency. We examine associations between the two types of consistency in two independent samples: a community sample ( N = 134) of mothers of 1.5 to 3.5 year old children who completed a daily diary for 7 days, and an at-risk sample with heightened disruptive behavior ( N = 149) of 3 to 8 year old children who filled out a daily diary for 14 days. Including these two samples has several benefits. First, we are able to examine whether associations between the two types of consistency conceptually replicate across multiple samples. Second, we can investigate whether both types of consistency are equally associated with disruptive behavior during a developmental stage when disruptive behavior starts to emerge and is relatively more normative, as during a developmental stage when disruptive behavior for most children has started to decline (Tremblay, 2010 ) as a result of increases in children’s verbal-skills and overall self-regulation (Kuhn et al., 2016 ). This allows us to investigate whether the two types of consistency play a similar role in the early emergence as in the maintenance of more persistent problem behavior across development. Contemporary accounts of social learning theories would predict that across-episode consistency may be less relevant for problem behavior that persists into preschool age, as repeated coercive cycling in parent-child dyads is thought to result in increasingly rigid, mutually negative interactions over time (Granic & Patterson, 2006 ). We also examine the added value of computing these types of consistency from daily diary data over a single questionnaire assessment. To this end, we assessed whether it is a better predictor of children’s externalizing behavior one year later than a measure of consistency derived from a general questionnaire as administered in a single – less time-consuming baseline assessment – asking parents to estimate how often in the past month they showed the reactions that we also included in the diary study. Although this ‘general consistency’ measure confounds within- and across-episode consistency, it may still be a better measure of consistency than some of the current measures that are used. Rather than asking parents to report on how consistent they are, we merely asked parents to indicate how often they reacted a certain way, and compute consistency by calculating the dispersion of parents’ responses across the different reaction categories. This approach makes it less likely that this association is for instance explained by parents scoring themselves as inconsistent due to a more negative self-view (Smit et al., 2021 ). This likely plays a role in more traditional questionnaire measures, as most parents will realize that threatening with punishment and then not following through is not an effective parenting strategy. Our approach will allow us to examine whether taking multiple days of measurements to assess consistency is really necessary, or whether we have enough information when we just ask parents how often they react a certain way overall. Additionally, associations between the general consistency and within- and across-episode consistency can be examined to provide an indication of the validity of these measures.
Method Sample We included two samples, to allow for conceptual replication: Sample 1 is a community sample of 134 mothers of 1.5–3.5 year old children ( M = 30 months, 44.3% girls), who reported on their child’s temper tantrums (frequency and severity) and their responses to these tantrums – in general across the past month and daily for 7 days. Mothers were predominantly, but not exclusively, highly educated (79% higher vocational or university education), and were not selected for experiencing any particular difficulties with their child. Seven percent indicated that they raised their child without a partner. No information regarding ethnicity was collected for this sample. Between February 2016 and June 2017, undergraduate students recruited mothers with children between one and five years old for a research practical. They recruited mothers through online parenting fora and Facebook, and face-to-face outside in Amsterdam. Mothers who participated were also asked to forward the invitation for the study to other mothers. Mothers were informed about the study and gave informed consent in the online study environment. They filled out the general questionnaire regarding: children’s temper tantrums and their own reactions, their personality and sense of parenting competence ( N = 884). Mothers who indicated that their child was between 1.5 and 3.5 years old were asked if they would like to participate in an additional daily diary study, and N = 382 indicated that they would like to receive more information. They were contacted by telephone, with N = 220 eventually participating. For this study, we only selected participants if they had participated in at least 4 days of the study ( N = 185), and who had reported reactions for at least two tantrums, resulting in a final sample of N = 134. Participants completed an average of 6.76 days ( SD = 0.62, range = 4–7 days). For this sample, mothers additionally reported on their child’s externalizing behavior one year later ( n = 86). Participants who dropped out of the study did not differ significantly from those who participated one year later with regards to age of the mother ( T (131) = 1.66, p = 0.062) or child ( T (132) = -1.65, p = 0.050), the child’s sex ( χ 2(1) = 0.29, p = 0.589) or mothers’ educational level ( χ 2(4) = 0.74, p = 0.947). Additionally, there were no significant differences in children’s tantrum severity ( T (132) = 0.90, p = 185) and mothers’ within-, across-, or general consistency at T1 ( T (132) = -0.10, p = 0.922; T (132) = -0.09, p = 0.930; T (117) = -0.21, p = 0.833, respectively). Mothers who filled out the general questionnaire had a chance of winning a gift certificate of 50 euros. Mothers who participated at least four days of the diary study received a small gift (a small book for their child – 2 euros) by mail. The study was approved by the ethical review board of the Department of Child Development and Education at the University of Amsterdam (#2015-CDE-6367). Sample 2 consists of 149 parents (94% mothers) of 3–8 year old children ( M = 5.88; 46% girls) oversampled for disruptive behavior – 17% had received parenting support for disruptive child behavior prior to the study; seven percent still received support during the study. Parents reported on how they generally responded to their child’s disruptive behavior as well as their daily responses for 14 days. Parents were predominantly, but not exclusively, highly educated (78% higher vocational or university education). Ten percent indicated that they raised their child without a partner. Culturally, 93% identified as Dutch, of which 19% identified as bicultural (mainly other European cultures or Moroccan). Others self-identified as Moroccan, other European, Asian, Surinamese, or Turkish. This roughly represents the Dutch population where around 25% of families has at least parts of their roots outside the Netherlands, most often in Turkey, Morocco and Surinam (Centraal Bureau voor de Statistiek [CBS], 2020 ). Parents were recruited between March 2020 and June 2021, through social media, primary schools across the Netherlands, and databases from the University of Amsterdam of parents who consented to be contacted for research projects. Children with disruptive behavior problems were oversampled by advertising the study as targeting parents of children with mild to moderate levels of disruptive behavior. Parents who signed up were contacted by phone to explain the study procedures. Parents who agreed to participate signed informed consent, completed a baseline assessment (i.e., demographics and trait measures) with a link to daily online daily questionnaire ( N = 156). For this study, we only selected participants if they had participated for at least 8 days of the study, and who had reported reactions to at least two disruptive behaviors, resulting in a final sample of N = 149. Participants completed an average of 13.22 days ( SD = 1.34, range = 8–14 days). Parents received €50 for completing the study. Study procedures were approved by the Ethical Review Board of the department of Child Development and Education of the University of Amsterdam (2019-CDE-11055). Measures Parental Consistency In Sample 1, parents reported on how they responded to their child’s tantrums both in general over the past month – before they started the diary study, and for each tantrum that took place during the diary study (for a maximum of seven tantrums a day). Parents rated their responses from a list of 11 behaviors. We made a functional classification based on social learning principles (Patterson, 1982 ), differentiating punishment and reward from lack of punishment or reward, and positive and negative attention: negative consequence (2 items: ‘I sent my child to their room/corner/time-out’, ‘I punished my child’), positive consequence (‘I negotiated with my child’, ‘I gave in to my child’), withholding attention (‘I didn’t, I let my child cool off’, ‘I ignored my child’), negative attention (‘I became angry with my child’, ‘I grabbed my child’, ‘I spoke sternly to my child’), positive attention (‘I distracted my child’, ‘I comforted my child’). In the questionnaire about responses to tantrums in general, participants indicated how often they tended to respond that way (1 = never ; 2 = almost never ; 3 = < half the time ; 4 = about half the time ; 5 = > half the time ; 6 = almost always ; 7 = always ), and we computed a mean score per category. In the daily diaries parents indicated whether or not they responded that way in that particular instance (0 = no ; 1 = yes ), allowing for multiple responses. Parents received a score of 1 in a category when answered yes to at least one of the responses in that category. For our measure of across-episode consistency from the daily diary data and the general consistency measure from the baseline questionnaire, we calculated the Index of Qualitative Variation (IQV), which is a measure of variation for nominal variables – where a standard deviation cannot be computed due to qualitative rather than quantitative differences between categories, using the following formula (Frankfort-Nachmias & Leon-Guerrero, 2018 ): For each response category we first computed what proportions of the total number of responses they were for each individual. The squared proportions of each of the categories are summed and then subtracted from 1 and multiplied by K, which is the number of categories (5 in our study). This is then divided by the number of categories minus 1. The resulting value can range from 0 to 1.00, with higher scores indicating greater inconsistency. Therefore, we subtracted this value from 1, so that higher scores indicated greater consistency. From the diary data, we additionally computed a measure of within-episode consistency. For each tantrum, we summed the total number of responses in the different categories (potential range 1–5), and then computed a mean score across all tantrums that were reported during the study. The observed range was 1–3 with higher values indicating less consistency. For ease of interpretation we recoded this variable so that higher values indicated greater consistency by subtracting the values from 3. The final variable thus ranged from 0–2. The intraclass correlation coefficient (ICC) for within-episode consistency was 0.20. This value is similar to ICCs that have previously been reported for parenting variables in diary studies, such as psychological control and autonomy support (Mabbe et al., 2018 ), with somewhat higher levels of around 0.33 also reported for psychological control (Aunola et al., 2013 ). In Sample 2, parents were asked how they responded to their child’s disruptive behavior in general at the start of the study, and how they responded to their child’s most challenging disruptive behavior that particular day (if any) in the diary study. Parents rated their responses from a list of 13 behaviors, which we grouped into the same five categories as for Sample 1: negative consequence (4 items: ‘I sent my child to their room for at least an hour’, ‘I gave my child a short time-out, away from others’, ‘I took something nice away from my child (e.g., toys or screen time)’, I gave my child extra chores (e.g., set the table)’), positive consequence (2 items: ‘I gave my child his/her way’, ‘I gave in to my child’), withholding attention (2 items: ‘I did nothing’, ‘I talked about it with my child afterwards’), negative attention (3 items: ‘I yelled/swore’, ‘I said things I didn’t mean’, ‘I threatened with punishment, but did not punish’), positive attention (2 items: ‘I begged my child to stop’, ‘I used humor to distract my child’). In the general questionnaire, participants indicated how often they tended to respond that way on a 5-point Likert type scale (1 = less than once a week, 2 = once a week ; 3 = few times a week ; 4 = once a day ; 5 = several times a day ), and we computed a mean score per category. In the daily diaries study parents indicated whether or not they responded that way in that particular instance (0 = no , 1 = yes ), allowing for multiple responses. Parents received a score of 1 in a category when answered yes to at least one of the responses in that category. When parents reported that their child had not shown any disruptive behavior that day, the response category was coded as missing. Like in Sample 1, we computed the IQV as a measure of across-episode consistency from the daily diary data and a measure of general consistency from the baseline questionnaire, and the mean number of selected categories of responses per episode as our measure of within-episode consistency. For within-episode consistency, the ICC was 0.18. Child Externalizing Behavior In Sample 1, we calculated a measure of severity of the child’s tantrum behavior from the daily diary reports, by summing for each tantrum the total number of aggressive (hitting, kicking, biting, throwing an object, pushing/pulling, spitting, grabbing) and self-injurious behaviors (banging head, holding breath, freezing). A previous study on this sample found that a profile with elevated levels on these behaviors was predictive of both internalizing and externalizing problems above and beyond tantrum frequency and duration (Van den Akker et al., 2022 ). The ICC for tantrum severity was 0.24. One year later (T2), parents in Sample 1 filled out 24 items of the Externalizing Problem Behavior Scale (the attention problem and aggressive behavior problem subscales, e.g., “My child does not seem to feel guilty after misbehavior”) of the Dutch version of the Child Behavior Checklist (1, 5–5) (Achenbach & Rescorla, 2000 ). Parents were instructed to indicate for the past 2 months how characteristic the item was of their child's behavior, with each item rated as 0 ( not true ), 1 ( sometimes/somewhat true ), or 2 ( often/very true ). Cronbach's alpha for the present sample was 0.89. For Sample 2, rather than indicating how many disruptive behaviors children had displayed, parents rated children’s overall level of disruptive child behavior at T1 each day (i.e., “how disruptive was your child’s behavior today?”) on a 1 − 10 scale. A mean score across the 14 days was computed. The ICC was 0.32. Analysis Plan Hypotheses and analyses were registered on the Open science Framework ( https://osf.io/tecr4/?view_only=0c6f1e3d6b2f46e49c5599c4c168be3c ). We first winsorized outliers (outside 1.5* IQR) to the nearest value if there was a gap in data between that range and the outlier. In Sample 1, for the within-episode consistency measures as derived from the daily diaries we identified two outliers, and for the across-episode consistency derived from the questionnaire asking about tantrums in general, we identified one outlier. For the severity of daily disruptive behavior we identified six outliers, and for externalizing behavior we identified two outliers. In Sample 2, for the within-episode consistency measure as derived from the daily diaries we identified one outlier, and for the across-episode consistency we identified four outlier. For the severity of daily disruptive behavior we identified four outliers. To answer our first research question- whether our measures of within- and across-episode consistency measure different but related aspects of consistency – we computed correlations. Next, we performed regression analyses to predict the severity of daily disruptive behavior from the within- and across-episode consistency measures to see whether they were uniquely associated. In a next step, we examined whether associations were significant above and beyond mean levels of the daily parental reactions. These analyses control for child sex and age and parental educational level and are performed on both Samples 1 and 2. As 11 parents in Sample 2 received parenting support for their child’s behavior, we also controlled for received support in Sample 2. Finally, we performed regression analysis in SPSS (version 28) to examine – in Sample 1 – whether within- and across- episode consistency as derived from the daily diary reports longitudinally predict child externalizing behavior problems one year later (T2), over and above a measure of consistency derived from estimates of parental behavior across the past month, controlling for the severity of temper tantrum behavior as reported in the diary study at T1.
Results On average, children in Sample 1 had an average 5.90 tantrums during the 7-day period ( SD = 4.21, range 2–20), and for children in Sample 2 the mean level of disruptive behavior was rated 3.32 on the 10 point scale across the 14 days ( SD = 1.21, range 1.14–6.54). Descriptives and intercorrelations for Samples 1 and 2 are provided in Table 1 . In both samples, within- and across episode consistently were significantly associated. Associations were strong, but not so strong as to indicate that they would actually be measuring the same thing. In Sample 1, only across-episode consistency was negatively associated with disruptive behavior severity; in Sample 2, both within- and across-episode consistency were negatively associated with disruptive behavior severity. Within- and Across-episode Consistency and Severity of Child Disruptive Behavior To examine whether within- and across-episode consistency were uniquely associated with the severity of daily disruptive behavior, we performed regression analyses. In Sample 1, the first step, controlling for age and sex of the child and educational level of the parent was not significant ( F (3,130) = 0.47, p = 0.707, R 2 = 0.01). Adding across-episode and within-episode consistency resulted in a significant improvement of the model (Δ F (2,128) = 6.95, p = 0.001, Δ R 2 = 0.10): when parents were more consistent across disruptive behavior episodes, children displayed less severe disruptive behavior, whereas within-episode consistency was not significantly associated with severity of daily disruptive behavior (Table 2 ). Results of Sample 2 conceptually replicated the findings of Sample 1. The first step, controlling for sex of the child and educational level of the parent was not significant ( F (3,144) = 0.32, p = 0.808, R 2 = 0.01). Adding across-episode and within-episode consistency resulted in a significant improvement of the model (Δ F (2,142) = 9.66, p < 0.001, Δ R 2 = 0.12): only across-episode consistency, not within-episode consistency, was significantly associated with severity of daily disruptive behavior (Table 3 ). In a next set of regression analyses, we examined whether within- and across-episode consistency predicted the severity of daily disruptive behavior, above and beyond mean levels of the different response categories. In Sample 1, adding the mean levels of the proportions of the five parental responses across the seven days did not result in a significant improvement over the model including only age and sex of the child and educational level of the parent (Δ F (5,125) = 0.47, p = 0.801, Δ R 2 = 0.02), indicating that how much parents displayed a certain type of reaction was not predictive of the child’s disruptive behavior. Adding within- and across-episode consistency to the model did result in a significant improvement (Δ F (2,123) = 8.63, p < 0.001, Δ R 2 = 0.12). Across-episode consistency was predictive of daily disruptive behavior severity, whereas within-episode consistency was not. For model coefficients, see Table 2 . Different from Sample 1, in Sample 2, adding the mean levels of the proportions of the five parental responses did result in a significant improvement over the model including only sex and educational level of the parent (Δ F (5,139) = 6.21, p < 0.001, Δ R 2 = 0.18). Providing negative consequences and giving negative attention to disruptive behavior, were each associated with more severe daily disruptive behavior. Here, adding within- and across-episode consistency to the model did not result in a significant improvement (Δ F (2,137) = 0.55, p = 0.581, Δ R 2 = 0.01), indicating that the association between consistency and child disruptive behavior was explained by the individual negative responses. For model coefficients, see Table 3 . Prediction of Externalizing Problems One Year Later In Sample 1, we examined whether the consistency measures predicted externalizing behavior one year later, controlling for the severity of daily disruptive behavior at T1 and for parent-reported consistency as derived from a one-time questionnaire about general responses to tantrums. The first step was significant ( F (4,73) = 5.01, p = 0.001, Δ R 2 = 0.22): more severe daily disruptive behavior was predictive of more externalizing behavior one year later. Above and beyond this effect, less consistency as derived from parents’ reports of how frequently they generally displayed certain responses to their child’s tantrums (i.e. general consistency as computed from the baseline measure), was predictive of more externalizing problems. Importantly however, adding within- and across-episode consistency to the model did not result in a significant improvement (Δ F (2,71) = 0.001, p = 0.999, Δ R 2 = 0.00). These results indicate that consistency in parental responses as derived from their reports of how often in the last month they displayed certain reactions, was longitudinally predictive of externalizing problems, whereas within- and across-episode consistency as derived from the daily diary measures were not. For model coefficients, see Table 4 .
Discussion Aim of this study was to investigate how within- and across-episode parental consistency in responding to misbehavior are associated to externalizing problem behavior in children, using a daily diary approach. Within- and across-episode consistency were moderately strongly correlated with each other, but only across-episode consistency was associated with the severity of daily disruptive behavior. In Sample 1, this association was significant above and beyond the content of the parental reactions, whereas in Sample 2, the association was explained by the fact that parents who were more consistent across episodes were less likely to provide negative consequences or negative attention for the disruptive behavior. When we compared the longitudinal predictive value of the measures of consistency derived from the daily diaries to a measure of consistency derived from how often parents indicate they usually react, we found that the measures from the daily diary did not predict externalizing behavior problems one year later, whereas parental consistency as derived from the general questionnaire did. Within- and Across-episode Consistency In both samples, we found that the two types of consistency were significantly associated with each other, but the associations were not so strong that they indicated that they reflected the same underlying construct. This indicates that some parents were relatively higher on across-episode consistency whereas others were relatively higher on within-episode consistency. These results support the idea that it is relevant to separate the two types of consistency. The correlations between the two types of consistency were quite similar across the two different samples, as were the associations between the two types of consistency as computed from the daily diary data and the ‘trait’ measure of consistency that was computed based on how parents indicated that they generally responded to disruptive behavior in the baseline measure. These associations provide some validation of these measures. Interestingly, despite moderately strong correlations between within- and across episode consistency, when associations between the two types of consistency and the severity of daily disruptive behavior were examined, across-episode consistency was significantly associated with the severity of daily disruptive behavior in both samples, whereas within-episode consistency was not. That within-episode consistency was not associated with disruptive behavior severity is not in line with observational findings that lower within-episode consistency differentiated mother-child dyads with conduct-problems from those without (Gardner, 1989 ), and mothers of aggressive toddlers from those without (Del Vecchio & O’Leary, 2006 ). These findings may indicate that with regards to within-episode consistency, the rewarding nature of the final response – when a parent eventually gives in or does not follow through on their initial demand – is more important in explaining this effect of within-episode consistency rather than the mere variation of types of responses as was assessed by our measure. Parents who reward the child for misbehavior are likely to first provide negative attention for instance, scolding the child, and only give in after a sequence of different types of reactions (Gardner, 1989 ). Alternatively it may mean that across-episode consistency is actually more strongly associated with disruptive behavior. As previous studies examining these rewarding interaction sequences have not controlled for across-episode consistency, more research is necessary to examine whether this association also disappears when across-episode consistency is taken into account. We add to previous findings that inconsistency in responses across episodes of misbehavior may be specifically associated with more severe disruptive behavior, regardless of the variation in types of responses within single episodes. In Sample 1, parents who were less consistent not only varied more within- or across-episode in how they responded to children’s tantrum, but also more frequently used each of the responses, both positive (e.g., positive consequences such as ‘giving in’) and negative (e.g., negative consequences such as ‘punishing’). Importantly, it was the variation between responses rather than the frequency of the individual responses that was associated with child disruptive behavior. This might indicate that parental consistency in responding is more important for lowering child disruptive behavior than how parents respond specifically. Alternatively, it might mean that when children show more disruptive behavior, parents are more likely to try out different ways of responding in an attempt to deal with it. Other studies have found that behavioral or emotional variation is associated with more maladjustment in young children as observed at a more micro time-scale, across real-time interaction. For instance, variability in affective displays has been related to more externalizing problems in mother-toddler dyads (Lunkenheimer et al., 2011 ), as has behavioral variability (Lunkenheimer et al., 2020 ). Our findings support the idea that, in a non-clinical sample, predictable parental responses are most important in reducing disruptive behavior. Unpredictable behavior from parents has been shown to impact the stress response in infants, with a blunted cortisol response to a painful stressors for infants of mothers who’s behavior was less predictable (Noroña-Zhou et al., 2020 ), and variability in the affective quality of mother-child interactions and even in the timing of leisure activities, has been associated with an increased production of proinflammatory cytokines, an index of stress-reactivity, for youth (Manczak et al., 2018 ). More research is necessary to understand whether these processes play a role in explaining the association between across-episode consistency in parenting behavior and child disruptive behavior. In Sample 2, a sample with older children who were at-risk for problem behavior, across-episode consistency was no longer associated with disruptive behavior above and beyond the individual reactions, whereas the frequency of providing negative consequences and negative attention were associated with more severe disruptive behavior. It seems that, whereas in Sample 1 it did not matter so much what parents did to reduce child disruptive behavior, as long as they did it consistently across episodes, in this sample negative responding was specifically associated with disruptive child behavior. Perhaps for families with older children with elevated levels of disruptive behavior as in Sample 2, parent and child have more strongly established patterns of negative responding to each other (Granic & Patterson, 2006 ). In support of this idea, in studies of school aged children, affective variability has been associated with less rather than more behavioral problems (Granic et al., 2007 ; Hollenstein et al., 2004 ). Settling into a rigid, negative interaction style is a process that takes place in the interaction between parent and child over several years. Heightened variability in other areas may still have negative effects in older children and adolescents. For instance, higher variability in experienced stressors has been associated with worse emotional adjustment in adolescents (Zheng et al., 2022 ), and higher variability in daily activities is associated with lower psychological well-being in young adults (Lee et al., 2018 ). Daily Diary Measures In this study, daily disruptive behavior was associated with externalizing problems one year later, and within- and across-episode consistency were associated with our measure of general consistency. Thus, it appears that the daily diary measures were tapping some of the micro-level processes giving rise to increases in problems at a developmental timescale (Granic & Patterson, 2006 ). However, it also appears that these associations between parenting and child behavior did not cross-over from one level to the other, as the general parental consistency measure was predictive of externalizing problems one year later, whereas the daily measures of consistency (within- and across-episode consistency) were not. Additionally, general consistency was in turn not associated with daily disruptive behavior severity, whereas the daily measure of across-episode consistency was. It thus seems that the parenting and child behavior measures that were measured on a more similar timescale were more likely to be associated with each other. A previous study had similar findings in this regard, with daily measures of parenting variability associated with global parenting measures, but only global measures associated with a measure of the child’s ADHD symptoms (Li & Lansford, 2018 ). Although there, ADHD symptoms became significantly associated with variability in parental warmth after controlling for parental ADHD symptoms and several types of stress, and daily symptom expression was not assessed. More research is necessary to understand how inconsistency in daily parent-child interactions may eventually increase externalizing problems over months and years. Although daily diary measures are especially helpful in differentiating within- from across-episode consistency, the measure of how often parents indicated to react to their children’s disruptive behavior a certain way over the past month – the general consistency measure – was more predictive longitudinally of externalizing behavior than the daily diary measures. Although this measure again confounds within- and across-episode consistency, it may still be a better measure of consistency than some of the other measures of general consistency. As we asked directly about very specific reactions, our measure is likely a more valid measure of actual consistency in responding (Morsbach & Prinz, 2006 ). At the same time, we would like to note that the validity of this measure deserves further scrutiny. Strengths and Limitations This study has several strengths. First, we included two samples of daily diary data that allowed us to differentiate within- from across-episode consistency and examine how our results would replicate across samples. Second, the analyses were registered on the OSF before conducting them. In addition to these strengths, some limitations are also worth mentioning. First, we did not differentiate different types of disruptive behavior episodes. Perhaps some episodes were more similar to each other than others, with a child yelling after not getting what it wanted in separate instances more similar than hitting a sibling in frustration about losing a game. Parents might react differently to different types of misbehavior, but consistently so within the types of misbehavior. Relatedly, in the functionally based categorization of behaviors in this study, certain responses were collapsed into categories as they were highly similar in their function – with these categorizations preregistered. Categories of positive attention, and ‘getting what you want’ were differentiated as they are likely different enough to be inconsistent, as are receiving negative attention or being punished for instance. An even higher level of abstraction could also be chosen, where anything ‘positive’ is contrasted with anything ‘negative’. At present, it is not known how different responses must be to contribute to inconsistency. Relatedly, the categorization of the parental reactions was based on a social learning theory perspective. However, inconsistencies in other aspects of the response might also be relevant. For instance, for several of the reactions that parents could choose from, it would be possible to be quite calm or quite frustrated while doing so, and these differences in affective quality and intensity might also contribute to inconsistency. More research is necessary to investigate whether inconsistency computed from other aspects of parental responses shows similar associations as the inconsistency measures derived from the categorization we made here. Second, both samples consisted of families with mostly highly-educated parents, raising questions about how generalizable these findings are to populations with different educational backgrounds. Additionally, whereas for Sample 2 it was clear that it was representative of ethnicities in the Netherlands, for Sample 1 information about ethnic diversity of the sample was not collected, making it impossible to draw any conclusions about this. Third, although similar measures were available for both samples included in this study, these studies were not designed to be the same, and as a result varied in multiple design aspects, making it impossible to draw any conclusions about why the results may have differed between them. Fourth, there are other aspects of consistency that we have not included in this study. For instance, consistency between different caregivers’ reactions might also play a role adjustment problems (Dwairy, 2010 ). Conclusion Results of this study show that it is meaningful to separate parental consistency within- from consistency across-episodes of misbehavior as they are correlated, but not strongly so. Furthermore, the aspects of consistency may be differentially important for the severity of child disruptive behavior as it is displayed in daily life, and there is some indication that across-episode consistency might be more important than actual responses, at least in a general population sample of toddlers. However, the actual responses were more important in our sample of early elementary school aged children from an at-risk population. Findings thus suggest that different risk factors (across-episode consistency or negative responding specifically) for disruptive behavior might apply to different subpopulations.
Conclusion Results of this study show that it is meaningful to separate parental consistency within- from consistency across-episodes of misbehavior as they are correlated, but not strongly so. Furthermore, the aspects of consistency may be differentially important for the severity of child disruptive behavior as it is displayed in daily life, and there is some indication that across-episode consistency might be more important than actual responses, at least in a general population sample of toddlers. However, the actual responses were more important in our sample of early elementary school aged children from an at-risk population. Findings thus suggest that different risk factors (across-episode consistency or negative responding specifically) for disruptive behavior might apply to different subpopulations.
Consistent discipline is thought to reduce early child externalizing behavior. It is unclear, however, whether consistency is important mainly within episodes of misbehavior (e.g., threatening with discipline but then giving in) or across episodes of misbehavior (e.g., disciplining each instance of misbehavior). Using a daily diary approach, we examine whether these two types of consistency are associated with disruptive child behavior, concurrently and prospectively. We included two samples (Sample 1: N = 134, M agechild = 30 months, 44% girls; Sample 2: N = 149, M agechild = 5.88 years; 46% girls, at-risk sample) with daily reports of child disruptive behavior and parental responses (Sample 1 = 7 days; Sample 2 = 14 days). Sample 1 parents additionally reported on their reactions over the past month and their child’s externalizing behavior one year later. Within-episode consistency was assessed by the average number of parental reactions per episode; across-episode consistency by the Index of Qualitative Variation; and general consistency by parents’ report of how they had responded to child disruptive behavior in the past month. In both samples correlations between within- and across-episode consistency were significant, but not so strong that they were not differentiated. Again in both samples, regression analyses provided evidence for unique predictive value of across-episode, not within-episode, consistency for daily disruptive behavior. Parental general consistency was longitudinally associated with fewer externalizing problems, whereas within- and across-episode consistency were not. It appears meaningful to differentiate within- from across-episode consistency to better understand the relevance of different aspects of consistency. Keywords
Appendix I: Questions as included in the daily diaries Parental Reactions Sample 1 For each tantrum during that day in daily diary, question: Sample 2 For the most difficult to manage disruptive behavior reported that day, question: For answer options, see Table 5 . Answer categories for each option in both samples were: check box, coded unchecked = 0, checked = 1. Multiple checks allowed. Child Externalizing Behavior Sample 1 For each tantrum that occurred that day, question: For answer options, see Table 6 . Answer categories for each option in both samples were: check box, coded unchecked = 0, checked = 1. Multiple checks allowed. Sample 2 Question: ‘ How disruptive was your child’s behavior today?’. Answer categories were: slider 1 − 10. Author Contribution All authors contributed to the study conception and design. Material preparation and data collection were performed by Alithe Van den akker, Patty Leijten, and Peter Hoffenaar, data analysis was performed by Alithe Van den akker. The first draft of the manuscript was written by Alithe Van den akker and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. Funding Study 2 was funded by the Dutch Organization for Health Research and Development (ZonMw) under grant number 636320007 to Patty Leijten. The funder had no role in the design of this protocol, the collection of data, the data analysis, or the interpretation or publication of the study results. Data Availability Data are available from the first author upon request. Compliance with Ethical Standards Conflict of Interest The authors declare that they have no conflict of interests. Ethical Approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.
CC BY
no
2024-01-15 23:42:01
Res Child Adolesc Psychopathol. 2024 May 19; 52(1):79-92
oa_package/7e/9a/PMC10787874.tar.gz
PMC10787875
38217693
Introduction Multiparametric MRI (mpMRI) of the prostate is nowadays considered the key tool for the diagnosis of clinically significant prostate cancer (csPCa) [ 1 ]. The quality of MRI has progressively increased in the last years and standardised multiparametric sequences acquisition and reporting have been established, with the development of the Prostate Imaging – Reporting and Data System (PI-RADS) scoring system, first published in 2012 by the European Society of Urogenital Radiology (ESUR) [ 2 ]. The updated PI-RADS versions 2 and 2.1 are nowadays broadly used in clinical practice, overcoming some ambiguities and limitations related to the overall scoring system of the former version [ 3 , 4 ]. Nevertheless, PI-RADS has not being designed for staging purposes; on this regard, the ESUR developed another score system based on capsular alterations detectable at MRI and related to the likelihood of an extracapsular extension (ECE) of the lesion [ 5 ]. During treatment planning, the identification of ECE as a strategy for local staging is crucial in order to achieve an adequate balance between cancer control and preservation of potency and continence, ultimately obtaining the best surgical, oncological and functional results. The main objective of this study was to identify clinical, pathological and radiological parameters associated to ECE in a single institution cohort of patients undergoing mpMRI prior to radical prostatectomy (RP), as by whole-mount prostate sections for definitive histological assessment. Furthermore, we evaluated the usefulness and the inter-observer variability of the overall LIKERT (subjective operator assessment for the likelihood of a csPCa, from 1 to 5), ECE-LIKERT (subjective operator assessment for the likelihood of ECE, from 1 to 5), PI-RADS v2 and ESUR-ECE, in order to evaluate their performance in predicting the ECE risk in the same cohort of patients.
Patients and methods Study design and population This is a retrospective analysis of patients who had undergone mpMRI before RP at Fundació Puigvert – Barcelona (ES), between April 2013 (date of the implementation of mpMRI driven pathway) and December 2017. The mpMRI was requested at urology consultant discretion, either before or after the diagnostic biopsy, as no specific recommendations for requesting an mpMRI were still put in place during the period in observation nor in the international guidelines neither in our internal protocol. Overall, 126 patients met the selection criteria (mpMRI undertaken within 6 months before surgery, either before or after the diagnostic biopsy; availability of full set of data in observation) and their medical records were collected and reviewed. The patients’ data were managed according to our institutional review board protocol, in full compliance with both the principles of the latest version of the Declaration of Helsinki and of the Spanish adaptation of the General Data Protection Regulation (organic law 3/2018, December the 5th 2018). mpMRI protocol A 3-Tesla mpMRI examination with a pelvic phased-array surface coil was performed for all patients. The mpMRI protocol-included T2-weighted (T2W) sequences in three planes, Diffusion Weighted Imaging (DWI) sequences with high b -values (> 1200 s/mm 2 ) and apparent diffusion coefficient (ADC) map, and Dynamic Contrast-Enhanced (DCE) sequences with a bolus of gadolinium contrast medium injection. Imaging acquisition protocol was rigorously compliant with the PIRADS v1 guidelines, in force during the period in observation [ 2 ]. It is Important to note that these guidelines also included recommendations for waiting a period of 4–6 weeks between a biopsy and the eventual MRI to minimise the effect of eventual haemorrhage, and in case of substantial persisting artifacts to repeat the MRI within further 4 weeks or so. These recommendations were duly followed, and interestingly remained substantially unvaried along the updated v2 and v2.1 versions [ 3 , 4 ]. All images were retrospectively and independently reassessed by two expert radiologists (L.G. and J.H.) with at least 4 years of experience in prostate mpMRI, both blinded to clinical and histological data; PI-RADS v2 was used to score the MRI explorations, as the updated v2.1 was published posteriorly to the radiological revision of the imaging tests. The maximal index lesion size (ILS), length of capsular involvement (LCI) by tumour, number of lesions and location were recorded whenever visible from the dominant sequence involved. The radiologists also used an overall-LIKERT and ECE-LIKERT scores, as by subjective impression of the likelihood of significant malignancy and extracapsular extension (1 = very unlikely; 5 = very likely), according to the recommendation of the PREDICT (Prostate Diagnostic Imaging Consensus Meeting) panel [ 6 ]. Furthermore, the likelihood of ECE was evaluated using the ESUR MRI scoring guidelines of extra-prostatic disease [ 5 ]. Reference standard Whole-mount histological sections from the RP specimens were used as the reference standard. The specimens were fixed in 10% buffered formalin, and sectioned into horizontal sections of 3–4 mm. All tissues were paraffin-embedded, and 3–4 microns sections were obtained and stained with hematoxylin–eosin; then, the sections of the tissue were assessed by a single expert uropathologist (F.A.), blinded to mpMRI data. The uropathologist recorded cancer location, size, volume, and Gleason grade group according to the International Society of Urological Pathology (ISUP) consensus conference of 2014 [ 7 ]. Outcomes and statistical analysis The primary outcome consisted in identifying the parameters associated to the ECE at the RP specimen. Descriptive data were expressed as median and interquartile range (IQR, 25–75 quartile). Analysis between groups was performed using Student’s t test (Mann–Whitney U test in variables without normal distribution) for continuous variables, and Chi-square (Fisher’s exact test with observed frequencies < 5) for categorical variables. Quantitative (continuous) variables were transformed to binary (categorical) variables before inclusion in logistic models using the best predictive cut-off point obtained with Receiver Operating Characteristics (ROC) curve analysis. Univariate and multivariate logistic regression models were performed including ECE as a dependent variable. Preoperative clinical, pathological and mpMRI variables with p value < 0.2 at the univariate analysis were included as independent variables using the backward stepwise logistic regression analysis. Predictors from the final model were used to calculate the likelihood of ECE according to the following equation: Exp( β )/[1 + Exp( β )], where β = [− 3.00 + X *(predictor A ) + Y *(predictor B )] for two predictors. The quality of the final model was assessed using the Hosmer–Lemeshow goodness-of-fit-test. Inter-observer agreement between the two radiologists regarding the MRI features/scores was assessed using the intraclass correlation coefficient (ICC) with the 95% confidence interval, applying a two-way ICC with a random rater assumption. The agreement (match) between a PCa lesion detected at mpMRI and an equivalent lesion at whole-mount histological sections from the RP specimens was assessed using Cohen Kappa coefficient, whose results were categorised in standardised ranges < 0.4 poor agreement; 0.4–0.6 moderate agreement; 0.61–0.8 substantial agreement; 0.81–1 as excellent. A p -value < 0.05 was considered statistically significant for all cases. Statistical analysis was performed using R studio (V2.5) package.
Results The clinical, bioptic and MRI features are summarised in Table 1 . The median age at prostate biopsy, PSA and PSA-density were 66.6 years, 7.2 ng/ml and 0.2 ng/ml 2 , respectively. Overall, 41 patients (32.5%) had T2 or T3 clinical stage. Intraprostatic perineural invasion (IPNI) and ISUP group grade > 3 were observed in 26 (20.6%) and 29 (23%) patients, respectively. The median index lesion size (ILS) and length of capsular involvement (LCI) were 12 and 9 mm (mm), respectively. MRI readings based on PI-RADS v2, ESUR and LIKERT scores are available in the Supplementary material. Pathology data and ECE analysis are summarised in Table 2 and Supplementary material, respectively. The median ILS of RP specimen was 16 mm. Accordingly, MRI underestimated the ILS by a 25% in comparison to the true specimen size. Overall, ECE was found in 35 (27.8%) patients; definitive ISUP grade > 3 was observed in 40 (31.7%) patients. The following variables were significantly associated to ECE: length of biopsy core (LBC; median: 7 vs 4 mm, p < 0.001), intraprostatic perineural invasion (48.6% vs 9.9%, p < 0.001), ILS (median: 20 vs 10 mm, p < 0.001), LCI (median: 17 vs 7 mm, p < 0.001), PSA density (0.24 vs 0.14, p = 0.001), and biopsy ISUP grade > 3 (40% vs 16.4%, p = 0.003). At ROC Curve analysis, the best cut-off points for LBC, LCI and ILS were identified as 5.5, 9.5 and 11 mm, respectively. Overall, both LCI and IPNI showed statistical significance ( p < 0.001) at multivariate logistic regression model (Table 3 ). The probability to detect ECE with the generated model was 81.4% by including the two variables (Supplementary material). The model was calibrated with an overall p-value of 0.985 by using the Hosmer–Lemeshow test ( R 2 = 66%), and the predictive accuracy was evaluated through ROC curve analysis, with an area under the curve (AUC) of 0.83 [95% CI (0.76–0.90)], p < 0.001 (Fig. 1 ). Univariate and multivariate logistic regression analyses were also conducted to identify predictors in the subgroup of patients with seminal vesicle invasion (SVI), but they were not included in the final model because of the low number of events ( n = 8). Data analysis and comparison of ROC curve between final model and imaging scores models (PI-RADS v2, LIKERT, ESUR) are available in the Supplementary material. Concordance between the two radiologists was evaluated for ILS, PI-RADS v2, LIKERT and ESUR score, with an overall ICC > 0.6 ( p < 0.001) in all parameters examined (Supplementary material). Correlation for ILS was 0.66 and 0.62 for L.G. and J.H., respectively ( p < 0.001).
Discussion The prognosis of PCa is highly related to tumour stage, and ECE is the most common adverse feature found in histopathology after RP [ 8 ]. Inadequate selection of patients with ECE to nerve-sparing RP increases the risk of positive surgical margins with subsequent need for either adjuvant or salvage radiotherapy. In order to improve preoperative risk assessment, numerous nomograms have been developed in the last years, mostly based on the association of multiple clinical risk factors (PSA, biopsy ISUP grade, clinical stage, etc.) [ 9 – 12 ]. The Partin tables were among the first tools developed for the prediction of pathological stage, and have been widely used for several years for surgical planning [ 10 ]. Other predictive tools were subsequently developed using different variables (e.g. percentage of positive biopsy cores), but none of them reached a comparable popularity as for the Partin tables [ 9 , 12 ]. Nevertheless, they could not distinguish unilateral from bilateral ECE [ 13 ]. With the advent of mpMRI, further nomograms have been developed in an attempt to improve the accuracy to predict the ECE. Feng et al. first reported that mpMRI could improve the performance of Partin tables and MSKCC nomogram regarding ECE prediction [ 14 ]. Giganti et al. developed a nomogram exploiting clinical and MRI parameters with strong accuracy for ECE prediction [ 15 ], subsequently confirmed by an external validation conducted by Alves et al. [ 16 ]. One of the most important features of their model was the excellent concordance between MRI-tumour volume and the ADC map of tumour lesion. More recently, Gandaglia et al. developed a model to predict ECE, SVI and stage upgrading in patients diagnosed with MRI-targeted and concomitant systematic biopsies [ 17 ], achieving an AUC of 73% (ECE), 81% (SVI) and 73% (upgrading) at internal validation. Nevertheless, a meta-analysis of de Rooij et al. reported high specificity but low sensitivity for mpMRI accuracy in local staging; overall, staging based on MRI alone lacks sensitivity in detecting ECE, especially in case of focal, minimal or microscopic extension because of limitation in spatial resolution [ 18 , 19 ]. Moreover, the degree of underestimation increases with smaller radiologic tumour size and lower PI-RADS scores [ 20 ]. The MRI and histology biopsy features have been variably reported in literature in the recent past as predictors of ECE [ 21 ]. Baco et al. found that MRI-tumour LCI well correlated with ECE, with an AUC that outperformed the Partin tables; they also found higher accuracy for microscopic-ECE detection with a 20 mm threshold [ 22 ]. Similarly, Kongnyuy et al. identified MRI-LCI as a promising predictor of ECE, positive pathological lymph nodes and biochemical recurrence. The 12.5 mm cut-off showed the highest sensitivity (77%) and specificity (59%) in predicting ECE. The AUC was comparable to that of the Partin tables, outperforming them with LCI and PSA combination [ 23 ]. Moreover, in a recent meta-analysis of Li et al., the LCI showed high diagnostic performance in predicting ECE, with a pooled sensitivity and specificity of 0.79 and 0.77, respectively. When subgroup analysis was performed comparing different threshold values, lower LCI cut-off values yielded slightly better sensitivity and comparable specificity, without substantial differences between sub-groups [ 24 ]. In addition to LCI, IPNI is another acknowledged parameter often associated with ECE, possibly because in the 85% of cases the ECE goes through neurovascular bundles by dissection of the intraprostatic perineural spaces for tissue planes of least resistance [ 25 ] Perineural invasion is defined as the tumour invasion into the perineural sheath, and during the years, it has been associated with tumour progression and prognosis of several malignancies, including prostate cancer [ 26 ]. In a recent study on upper urinary tract urothelial carcinomas, Lin et al. found that PNI-positive patients had unfavourable pathological features, including high pathological stage, high tumour grade and lymphovascular invasion, leading to worse progression-free survival (PFS) [ 26 ]. Algaba et al. found that IPNI is correlated to cancer volume and higher percentage of extraprostatic cancer [ 27 ]. Same finding was reported more recently by Leyh-Bannurah et al., being IPNI the only histological parameter found significantly associated to ECE in their nomogram [ 28 ] Nevertheless, IPNI has not yet been reported among the most powerful histological features even in the latest version of the EAU guidelines, so that our finding might prompt its inclusion among the reporting recommendations for the prostatic biopsy. In our study, we identified both the LCI and the IPNI as predictors of ECE: the LCI cut-off that best correlated to ECE was 9.5 mm, which was a similar finding reported also by Li et al. [ 24 ]. Interestingly, in our cohort, the ILS at MRI did not show the same degree of association to the ECE as by the LCI; furthermore, MRI underestimated pathological tumour size by 25%. Overall, these data may have several implications: (1) Prostate mpMRI alone, including singular features and scoring systems, do not adequately predict ECE, except LCI—especially in combination to a histology biopsy feature, as IPNI in our series. (2) A change of PI-RADS score 5 definition should be prompted in the future PI-RADS updated version: score 5 is attributed to ILS ≥ 15 mm, but this threshold was chosen by the PI-RADS steering committee on the basis of old studies of the’90, when csPCa and ECE were found to be correlated to a tumour volume ≥ 0.5 cm 3 (15 mm in major axis) at whole-mount RP specimen [ 29 , 30 ]. If differences of PI-RADS scores 4 to 5 are to be based on the risk of ECE, LCI should be the preferred MRI feature and with a lower cut-off, as the 15 mm cut-off at MRI might underestimate for a quarter the actual tumour size at specimen. (3) We found excellent or substantial inter-readers agreement for relevant MRI variables, thus strengthening the importance of high-quality imaging acquisition and readers’ skills for their adequate assessment. This latter matter has been popularised with the introduction of a dedicate score (PIQUAL) about the quality of the MRI sequences and the ability to make decision on the basis of it [ 31 ]. Main limitations of the study includes (1) the retrospective design of our analysis, even though significant efforts have been done in reviewing MRI images by two radiologists blinded to final histology; (2) the number of cases is limited, especially because MRI implementation in the clinical practice has substantially increased in more recent years; (3) the MRI images were reassessed according to the PI-RADS v2, as the version 2.1 became available on a later stage to that phase of our study. Nevertheless, it is very unlikely that the minor changes in score reporting of the latest version would have had an impact on the outcome of our study, as shown in a recent publication comparing the v2.0 vs v2.1 diagnostic performance with no difference in concordance rates between targeted biopsy and radical prostatectomy (doi: 10.2214/AJR.23.29964).
Conclusions The mpMRI confirms limitations in predicting the ECE at the RP specimen, even when involving tailored scoring systems. On the other side, MRI-LCI is the most robust factor associated to ECE; in our series, we found a strong predictive accuracy when combined with the IPNI presence. This outcome may prompt a change in the definition of PI-RADS score 5, by reducing IL size cut-off to 10–12mm, or by replacing IL size reporting with LCI (cut-off 9–10mm).
Objectives To identify the predictive factors of prostate cancer extracapsular extension (ECE) in an institutional cohort of patients who underwent multiparametric MRI of the prostate prior to radical prostatectomy (RP). Patients and methods Overall, 126 patients met the selection criteria, and their medical records were retrospectively collected and analysed; 2 experienced radiologists reviewed the imaging studies. Logistic regression analysis was conducted to identify the variables associated to ECE at whole-mount histology of RP specimens; according to the statistically significant variables associated, a predictive model was developed and calibrated with the Hosmer–Lomeshow test. Results The predictive ability to detect ECE with the generated model was 81.4% by including the length of capsular involvement (LCI) and intraprostatic perineural invasion (IPNI). The predictive accuracy of the model at the ROC curve analysis showed an area under the curve (AUC) of 0.83 [95% CI (0.76–0.90)], p < 0.001. Concordance between radiologists was substantial in all parameters examined ( p < 0.001). Limitations include the retrospective design, limited number of cases, and MRI images reassessment according to PI-RADS v2.0. Conclusion The LCI is the most robust MRI factor associated to ECE; in our series, we found a strong predictive accuracy when combined in a model with the IPNI presence. This outcome may prompt a change in the definition of PI-RADS score 5. Supplementary Information The online version contains supplementary material available at 10.1007/s00345-023-04720-5. Keywords Open access funding provided by Università degli Studi di Sassari within the CRUI-CARE Agreement.
Supplementary Information Below is the link to the electronic supplementary material.
Acknowledgements We are thankful to Cristina Esquinas Lopez, Professor of Statistics at at Vall d’Hebron Research Institute, Universitat Autonoma de Barcelona (Spain) for undertaking the statistical analysis on the manuscript. [email protected] Funding Open access funding provided by Università degli Studi di Sassari within the CRUI-CARE Agreement. None. Declarations Conflict of interest The authors have nothing to disclose.
CC BY
no
2024-01-15 23:42:01
World J Urol. 2024 Jan 13; 42(1):37
oa_package/5e/6f/PMC10787875.tar.gz
PMC10787876
37833450
Introduction Invasive lobular carcinoma (ILC) is the second most common histological subtype of breast cancer after invasive ductal carcinoma (IDC), representing 10–15% of all cases [ 1 ]. In the metastatic setting, ILC differs in its pattern of metastatic sites, often involving the bone, and gastrointestinal tract [ 2 – 4 ]. Additionally, several studies demonstrate worse overall survival (OS) in metastatic ILC compared to IDC, even when evaluating patients with similar receptor subtypes [ 4 – 6 ]. While investigators have evaluated surgical outcomes by histologic subtype in early-stage disease, there are scant data evaluating the use of primary-site surgery in the metastatic setting in those with ILC versus IDC. The current recommended therapy for metastatic breast cancer is systemic therapy, with local therapy reserved for palliation of symptoms [ 7 ]. While retrospective studies and institutional series have found associations between primary tumor resection and longer survival in those with metastatic breast cancer [ 8 ], most randomized control trials have not demonstrated such a survival advantage [ 9 – 11 ]. A previous study found that ILC patients with bone-only metastases had longer OS than those with visceral metastases when given a combination of chemotherapy and surgery, but it is unclear whether this reflects the more indolent course of osseous metastases, or an ILC specific effect of treatment [ 4 , 12 ]. As such, whether histologic subtype should factor into patient selection for primary-site surgery is unknown. Prior analyses have suggested that if surgery of the primary tumor is associated with improved survival, this may be more likely in those with hormone receptor (HR) positive disease, or bone-only metastases [ 10 , 13 ]. Given that ILC is largely HR-positive, HER2-negative, and has a propensity for bone metastases, we wondered if primary-site surgery use differed in patients with metastatic ILC compared to IDC. We used the National Cancer Database (NCDB) to evaluate differences in practice patterns and management of patients with metastatic ILC compared to metastatic IDC. Specifically, we investigated the following questions: whether rates of primary-site surgery differ by histologic subtype and whether selection factors associated with undergoing primary-site surgery differ by histologic subtype. As secondary endpoints, we evaluated the use of chemotherapy and radiotherapy relative to surgery by histologic subtype, and the association between primary-site surgery and OS in ILC and IDC cohorts in both unmatched and propensity score matched multivariable models.
Methods Data source and study cohort The NCDB is a national comprehensive clinical surveillance resource representing over 70% of all newly diagnosed cancer cases in the United States and includes patient demographics, clinical information, and survival outcomes [ 14 , 15 ]. Participants User Files from 2010 to 2016 were used. Due to the de-identified nature of the public-access user files, the study was exempted from institutional review board approval. Since most ILC tumors are HR-positive and HER2-negative, we limited analysis to tumors with this receptor subtype. Tumors that were estrogen receptor (ER) and/or progesterone receptor (PR) positive were considered HR-positive. Histology codes were used to identify cohorts, with the ILC cohort comprising those with codes for ILC or mixed ILC/IDC (histology codes 8520, 8522, and 8524 if invasive behavior), and the IDC cohort comprising codes for IDC or invasive mammary carcinoma not otherwise specified (histology codes 8500, 8501, 8502, 8503, and 8523 if invasive behavior). We excluded patients with stage I-III disease, histologic subtypes other than IDC or ILC, individuals who died within 6 months of their diagnosis, and those missing critical clinical information including disease stage, HR-status, HER2-status, or treatment type. Clinical measures Charlson-Deyo Co-Morbidity Index (CDCI) was recorded as a measure of severity of co-morbid conditions. Age at diagnosis was subdivided into under 50 years and over 50 years to estimate pre- and post-menopausal status, respectively. Metastatic disease sites were categorized as bone-only, visceral-only, bone and visceral, or unknown [ 4 ]. We utilized data on treatment facility type, insurance status, and median income in univariate and multivariate analyses. Statistical methods We compared clinicopathologic and demographic features between the ILC and IDC cohorts using chi-square tests for categorical variables and t-tests for continuous variables. We investigated factors associated with receiving surgery for the primary tumor, radiotherapy, chemotherapy, and timing of chemotherapy relative to surgery by histologic subtype. For univariate analyses, we used Kaplan-Meier plots and log-rank tests to assess associations between receipt of surgery and OS by histologic subtype, and by timing of chemotherapy and receipt of radiotherapy. We also evaluated treatment facility type, insurance type, and median income quartiles by surgery and histologic subtype. For multivariate analysis, we used Cox proportional hazards models to account for confounders with OS. The multivariable model included age, CDCI (0/1+), metastatic site (bone-only versus all other), and receptor subtype (ER-positive, PR-positive, HER2-negative versus ER-positive, PR-negative, HER2-negative). Finally, we performed propensity score matching including age, tumor grade, receptor subtype, site of metastatic disease, CDCI (0/1+), and treating facility variables to account for likelihood of having primary-site surgery to determine the association between primary-site surgery and OS. Within each histology category among those who had survival data available patients who had surgery (ILC n = 1,444; IDC n = 4,924) were matched to patients who did not have surgery (ILC n = 3,553; IDC n = 10,894) using the greedy nearest neighbor matching algorithm. Matching was restricted to observations that had propensity scores in the extended common support region (ILC 0.05–0.71; IDC 0.06–0.66), which extends the common support region by 0.25 times a pooled estimate of the common standard deviation of the logit of the propensity score. The PSMATCH procedure in SAS version 9.4 was used to perform matching. To account for the matched nature of the sample, log-rank tests and Cox models were stratified on the matched pairs. Hypothesis tests were two-sided, and the significance threshold was set to 0.05. Statistical analyses were performed using Stata 16 and SAS version 9.4.
Results Study cohort There were 100,147 patients with stage IV breast cancer, of whom 25,294 had HR-positive, HER2-negative invasive lobular or ductal histology, and met study criteria (Fig. 1 ). Of these patients, 19,171 (75.8%) had IDC and 6,123 (24.2%) had ILC. Within the ILC cohort, 4,484 (73.2%) had pure ILC, with the remaining having mixed ILC/IDC. Median follow-up time of the ILC cohort was 27.2 months (IQ range 14.7–41.5 months), which was similar to the median follow-up time of 26.8 months (IQ range 14.6–42.6 months) for the IDC cohort (Table 1 ). Patients with ILC were slightly older than those with IDC (mean age 64 years versus 61 years, p < 0.001) and differed significantly by race ( p < 0.001), with a higher proportion of White patients (79.4% versus 74.2%, p < 0.0001) (Table 1 ). Additionally, ILC and IDC patients differed by tumor grade ( p < 0.001), with ILC patients having a higher proportion of grade 1 tumors (22.1% versus 9.48%, p < 0.0001) and more bone-only metastases than those with IDC (60.8% versus 45.2%, p < 0.001). While overall most patients were treated at community cancer centers, slightly more ILC patients than IDC patients were treated at academic centers (ILC 36.0% versus IDC 33.5%, p = 0.001) (Table 1 ). ILC patients were significantly more likely to have public insurance (ILC 54.7% versus IDC 51.6%) and less likely to be uninsured (ILC 3.82% versus IDC 5.15%, p < 0.0001). Patients with ILC had higher median income, with 39.7% in the highest quartile compared to 35.7% of IDC patients in the highest median income quartile ( p < 0.0001). Primary-site surgery by histologic subtype In the overall study population, 7,158 (28.3%) underwent primary-site surgery. Although the absolute difference is small, primary-site surgery was performed less often in those with metastatic ILC than those with metastatic IDC (n = 1,644 [26.8%] versus n = 5,514 [28.8%] respectively, p = 0.004) (Table 2 ). This difference was more pronounced when restricting the analysis to patients with pure ILC, where only 23.8% underwent primary-site surgery. Additionally, the slightly lower rate of primary-site surgery among patients with ILC compared to IDC was observed among those with bone-only or unknown site of metastases, but not among those with visceral metastases (Fig. 2 ). Among those who had primary-site surgery in both histologic cohorts, the site of metastatic disease was significantly more likely to be bone-only compared to other sites (ILC 63.5%; IDC 48.5%, p < 0.001) (Table 3 ). For both cohorts, patients with private insurance were significantly more likely to receive primary-site surgery (ILC 46.4%; IDC 49.4%, p < 0.0001) compared to patients with public insurance (Table 3 ). Both ILC and IDC patients who received surgery were equally likely to have been treated at an academic treating facility (ILC n = 415 [26.3%]; IDC n = 1,318 [26.2%]) compared to a community setting (ILC n = 1,164 [73.7%]; IDC n = 3,714 [73.8%]). Among those who underwent primary-site surgery (n = 7,158), the differences by histology reflected those seen in the overall study population. Those with ILC were older (ILC mean age 61.7 years versus IDC 58.1 years, p < 0.001), had more T3 tumors (ILC 30.8% versus IDC 15.5%, p < 0.001), had more N3 nodal status (ILC 35.3% versus IDC 20.5%, p < 0.001), and had more grade 2 disease (ILC 61.4% versus IDC 45.9%, p < 0.001) (Table 1 ). There was no significant difference in mastectomy rate between the two cohorts (Table 2 ). Positive surgical margins were significantly more common in those with ILC who underwent lumpectomy compared to IDC (15.7% versus 11.2%, p = 0.025) (Table 1 ). The factors associated with undergoing primary-site surgery were similar in the ILC and IDC cohorts. In both groups, primary-site surgery was less common in older patients, and more common in those with larger tumors (except T4) and higher N category (Table 4 ). The odds of undergoing primary-site surgery were highest for ILC patients with pathologic stage T3 disease versus T1 (OR 2.65, 95% CI 1.17–3.51, p = 0.002; Table 4 ) whereas the odds of surgery for patients with IDC were highest in pathologic stage T2 disease versus T1 (OR 1.71, 95% CI 1.30–2.25, p < 0.001). In both groups, those with T4 disease had significantly lower odds of primary-site surgery compared to T1 (Table 4 ). Over time, the odds of undergoing primary-site surgery decreased. Specifically, the odds of surgery decreased by 16% per each additional year of diagnosis (OR 0.84 and p < 0.001 for both groups; 95% CI 0.82–0.87 in ILC group; 95% CI 0.83–0.85 in IDC group, Table 4 ). Radiotherapy by histologic subtype The use of radiotherapy overall was lower for patients with ILC than IDC (29.1% versus 37.9%, p < 0.001) (Table 2 ). In IDC patients who had surgery, 51.5% also had radiation, while in ILC patients who had surgery, 42.5% also had radiation (Table 3 ). Among those who received radiotherapy, there was no difference in the rate of radiation to local versus distant sites by histologic subtype ( p = 0.55). Use and timing of chemotherapy More IDC patients received chemotherapy (41.3% ILC versus 47.4% IDC, p < 0.001), while those with ILC were more likely to receive endocrine therapy (83.5% ILC versus 78.6% IDC, p < 0.001). For those who had primary-site surgery, the sequence of chemotherapy and surgery differed by histologic subtype; while 40.5% of patients with IDC had chemotherapy prior to surgery, only 29.0% of patients with ILC had chemotherapy prior to surgery ( p < 0.001). Survival analyses in unmatched cohorts Overall, patients with ILC had slightly but significantly shorter OS than those with IDC (median 38 months ILC versus 40 months IDC, p = 0.006). In both cohorts, primary-site surgery was associated with significantly improved OS (Table 1 ). In the ILC cohort, undergoing primary-site surgery was associated with 35% lower risk of death compared to those who did not undergo surgery (HR 0.65, 95% CI 0.57–0.68, p < 0.001) (Fig. 3 ). This association persisted when controlling for age, CDCI (0/1+), metastatic site, and receptor subtype (HR 0.64, 95% CI 0.58–0.70, p < 0.001). The timing of surgery (before or after systemic chemotherapy) was not significantly associated with OS among those with ILC in unadjusted analysis nor after controlling for age, CDCI (0/1+), metastatic site, and receptor subtype. Similarly, patients in the IDC cohort who underwent primary-site surgery had an associated 40% lower risk of death compared to those without surgery (HR 0.60, 95% CI 0.57–0.63, p < 0.001) (Fig. 3 ). This remained true after controlling for age, CDCI (0/1+), metastatic site, and receptor subtype (HR 0.61, 95% CI 0.58–0.64, p < 0.001). Unlike for those with ILC, timing of surgery was significantly associated with OS. Patients with IDC who had chemotherapy before surgery had 24% less risk of death compared to those who had surgery prior to chemotherapy (HR 0.76, 95% CI 0.70–0.84, p < 0.001). This association persisted after controlling for age, CDCI (0/1+), site of metastasis, and receptor subtype (HR 0.83, 95% CI 0.76–0.92, p < 0.001). The type of surgery (mastectomy versus lumpectomy) was not significantly associated with different OS in either group. However, we found a significant statistical interaction between having surgery for the primary tumor and the site of radiation. Primary-site surgery was associated with a greater reduction in risk of death among those who had local radiation compared to those who had distant radiation (HR 0.35, 95% CI 0.3–0.4 versus HR 0.67, 95% CI 0.61–0.74 respectively, test of interaction p < 0.001). This interaction between primary-site surgery and site of radiation was similar in both the ILC and IDC cohorts separately. Survival analyses in propensity score matched cohorts The propensity score model included age, tumor grade, receptor subtype, metastatic site, CDCI (0/1+), and treating facility variables, with 3,089 ILC patients (991 with surgery, 2,098 without surgery) and 11,216 IDC patients (3,429 with surgery, 7,787 without surgery) in each cohort with available data for matching. In both the ILC and IDC matched samples there was still a significant association between having surgery for the primary tumor and improved OS (ILC HR 0.71, 95% CI 0.59–0.85, p < 0.001; IDC HR 0.67, 95% CI 0.61–0.74, p < 0.001).
Discussion Recent randomized trial data suggest that the role of primary-site surgery in the management of patients with metastatic breast cancer is limited to local control in select cases, with no evidence of impact on OS [ 9 – 11 ]. However, optimal selection criteria for primary-site surgery are unknown, with decisions being made on an individualized basis in clinical practice [ 13 ]. Given the known differences in surgical management in the early stage setting and disease patterns in the metastatic setting between ILC and IDC, we explored whether the use of primary-site surgery differs in HR-positive HER2-negative metastatic lobular versus ductal breast cancer. In this cohort of 25,294 patients from the NCDB, we found that a high proportion of patients overall underwent primary-site surgery in the setting of metastatic breast cancer (28.3%). It is important to note that these data represent patients diagnosed between 2010 and 2016, prior to the publication of randomized trials demonstrating the lack of OS benefit from primary-site surgery [ 9 – 11 ]. Indeed, we found that the odds of undergoing primary-site surgery significantly decreased over time. While primary-site surgery was more common in those with bone-only metastases, and those with ILC were more likely to have such a disease pattern, the overall usage of primary-site surgery in the ILC cohort was slightly but significantly lower than in the IDC cohort. Although a slightly smaller proportion of patients with ILC had primary-site surgery, the majority of factors associated with receiving surgery did not differ between the lobular and ductal groups; in both groups, primary-site surgery was more common among younger patients, those with T2 or T3 tumors, more nodal disease, and private insurance. Interestingly, while patients with ILC had larger tumors than those with IDC, there was no difference in the rate of mastectomy by histologic subtype. This differs from the early-stage setting, where lobular histology is associated with higher mastectomy rates. Similar to the early-stage setting, however, those with metastatic ILC who had primary-site surgery experienced significantly higher positive margin rates after lumpectomy than those with metastatic IDC. This suggests that the local control benefit of primary-site surgery might be attenuated in those with ILC, who may require more extensive surgery to achieve negative margins. We did find an association between local radiotherapy and improved OS in this cohort; whether this association reflects a relationship between improved local control and survival outcomes versus improved outcomes in those selected to have radiation is unknown. Of note, patients with ILC were significantly less likely to receive radiation than those with IDC, which is consistent with other studies [ 16 , 17 ]. Interestingly, we found significantly lower odds of primary-site surgery in patients with T4 tumors in both lobular and ductal groups. Since palliation is the most accepted purpose of primary-site surgery in the stage IV setting, we would have expected higher rates of surgery in those with T4 tumors. Alternatively, these tumors may have been deemed unresectable; one of the challenges of analyzing this retrospective dataset is the inability to discern the reasons for performing primary-site surgery. This limitation likely impacts the strong association between primary-site surgery and improved OS that we found in both groups. For example, in both the ILC and IDC cohorts, patients who had private insurance were more likely to have surgery compared to patients who had public insurance. The improved outcomes associated with primary-site surgery may reflect improved access to care as opposed to a biologic effect of surgery. This is consistent with prior data; retrospective series tend to show a survival advantage associated with primary-site surgery [ 4 , 12 , 13 ], whereas recent randomized trial data do not [ 9 , 11 ]. Such discrepancies suggest selection bias in which patients undergo primary-site surgery. We are likely unable to account for the many factors that influence why surgery would be used in some patients versus others despite using propensity score matched models. Of more interest, perhaps, is the finding that the use of pre-operative systemic therapy was associated with improved OS in the IDC cohort, but not in the ILC cohort. We suspect that pre-operative systemic therapy in the IDC cohort may have helped to select patients who would have more durable response to therapy, and therefore have improved OS. In contrast, response to therapy in those with ILC may be more difficult to ascertain, or less likely to be associated with outcomes. For systemic therapy, those with ILC were significantly more likely to receive endocrine therapy than those with IDC, despite all studied cases being HR-positive. Likewise, those with IDC were more likely to receive chemotherapy. This treatment pattern has been observed in previous literature and may point to the notion that early-stage ILC has reduced sensitivity to chemotherapies, or perceived as such, and is therefore utilized less frequently [ 18 , 19 ]. However, more recent studies show that in the metastatic setting, response to eribulin and CDK4/6 inhibitors may be similar between ILC and IDC [ 20 , 21 ]. These findings highlight the need to identify lobular specific therapies for those with metastatic disease. As a secondary endpoint, we looked at OS by histology. Similar to our findings, worse OS in those with metastatic ILC has been shown in other studies as well [ 2 , 3 , 22 ]. While the underlying reason for this difference is unclear, it suggests that ILC is indeed biologically different than IDC, given differential outcomes despite restricting the study population to those with HR-positive, HER2-negative tumor types, and ILC tumors being of lower grade than IDC tumors. One potential explanation could be that those with metastatic ILC may have an overall higher burden of disease than is typically detected on standard imaging modalities [ 23 ]. To our knowledge, this is the largest reported study evaluating primary-site surgery by histologic subtype in metastatic breast cancer. However, this study is subject to several limitations, including selection bias, lack of detailed systemic therapy information, radiation field data, and the absence of local recurrence events as an endpoint. However, the findings reflect real-world management patterns which appear to differ by histologic subtype. While ILC has long been regarded as a less aggressive tumor type, our findings from this large NCDB study are consistent with others showing worse outcomes in ILC than IDC. The differences between the IDC and ILC groups in this study were relatively small, however, it is interesting to note that histology appears to be influencing management. The use of primary-site surgery was slightly lower, and the use of both radiotherapy and chemotherapy were much lower in those with metastatic ILC compared to metastatic IDC. It is unclear what is driving the lower usage of chemotherapy and radiotherapy in ILC cases; this may reflect an underlying bias that lobular tumors are more indolent and slow growing. Coupled with shorter OS in the ILC cohort, these findings reinforce the need for further study to determine histologic subtype-specific management options. In regard to surgical management, the significantly larger tumor size and higher positive margin rates in the ILC cohort suggest that if primary-site surgery is to be utilized, one should consider a larger excision and likely incorporate radiotherapy to maximize potential benefit of locoregional intervention. Further work is needed to improve management outcomes for those with metastatic ILC.
Purpose Primary site surgery for metastatic breast cancer improves local control but does not impact overall survival. Whether histologic subtype influences patient selection for surgery is unknown. Given differences in surgical management between early-stage lobular versus ductal disease, we evaluated the impact of histology on primary site surgery in patients with metastatic breast cancer. Methods The National Cancer Database (NCDB, 2010–2016) was queried for patients with stage IV HR-positive, HER2-negative invasive lobular carcinoma (ILC) and invasive ductal carcinoma (IDC). We compared clinicopathologic features, primary site surgery rates, and outcomes by histologic subtype. Multivariable Cox proportional hazard models with and without propensity score matching were used for overall survival (OS) analyses. Results In 25,294 patients, primary site surgery was slightly but significantly less common in the 6,123 patients with ILC compared to the 19,171 patients with IDC (26.9% versus 28.8%, p = 0.004). Those with ILC were less likely to receive chemotherapy (41.3% versus 47.4%, p < 0.0001) or radiotherapy (29.1% versus 37.9%, p < 0.0001), and had shorter OS. While mastectomy rates were similar, those with ILC who underwent lumpectomy had significantly higher positive margin rates (ILC 15.7% versus IDC 11.2%, p = 0.025). In both groups, the odds of undergoing surgery decreased over time, and were higher in younger patients with T2/T3 tumors and higher nodal burden. Conclusion Lobular histology is associated with less primary site surgery, higher positive margin rates, less radiotherapy and chemotherapy, and shorter OS compared to those with HR-positive HER2-negative IDC. These findings support the need for ILC-specific data and treatment approaches in the setting of metastatic disease. Keywords
List of abbreviations Charlson-Deyo Co-Morbidity Index Estrogen receptor Human epidermal growth factor receptor 2 Hormone receptor Invasive ductal carcinoma Invasive lobular carcinoma National Cancer Database Overall survival Progesterone receptor Author contributions Supervision, study concept and design, and interpretation of data were provided by Rita A Mukhtar. The first draft of the manuscript was written by Harriet Rothschild and all authors commented on previous versions of the manuscript. Amy Shui performed statistical analysis and data interpretation. All authors read and approved the final manuscript. Funding Author HTR was supported by the National Center for Advancing Translational Sciences, National Institutes of Health (NIH) (Grant number TL1 TR 001871). The contents of this manuscript are solely the responsibility of the authors and do not necessarily represent the official views of the NIH. Author AMS is part of the Biostatistics Core that is generously supported by the University of California, San Francisco Department of Surgery. Author RAM is supported by National Cancer Institute Award K08CA256047. For the remaining authors no sources of funding were declared. Data Availability The datasets analyzed during the current study are available in the National Cancer Database, https://www.facs.org/quality-programs/cancer-programs/national-cancer-database/ . Declarations Ethics approval Due to the de-identified nature of the public-access user files in the National Cancer Database, the study was deemed exempt from institutional review board approval. Competing interests The authors have no relevant financial or non-financial interests to disclose.
CC BY
no
2024-01-15 23:42:01
Breast Cancer Res Treat. 2024 Oct 13; 203(2):245-256
oa_package/92/6e/PMC10787876.tar.gz
PMC10787878
0
Introduction Terpenoids are a large group of natural compounds constructed from isoprene units (C5). Many different terpenoids are used in the food, beverage, cosmetic, and pharmaceutical industries (Tetali 2019 ). The current industrial supply of terpenoids relies mainly on chemical synthesis or extraction from plants. The chemical synthesis of terpenoids harboring complex structures has been actively researched (Phil 2018 ). However, the multi-stereo center in the complex terpenoids requires diastereoselective synthesis, resulting in multi-step total synthesis with high costs and low yield (Santosh et al. 2020 ; Le et al. 2022 ). Therefore, the chemical synthesis of terpenoids harboring complex structures is expensive and requires fine-controlled and complex processes. Plant extraction has been also used to supply terpenoids; however, it is limited by the presence of various compounds harboring similar physicochemical properties in plants and the difficulty of separating these compounds (Rodger et al. 2011 ). In addition, natural plant supplies are limited owing to their finite amount (Nur and Henrik 2017 ). Therefore, the engineering of microorganisms to produce terpenoids has emerged as a promising approach. Research on terpenoid production using microorganisms has rapidly increased in the twenty-first century. Saccharomyces cerevisiae and Escherichia coli have been used to produce plant terpenoids by heterologous expression of their genes because they have advantages for production, such as the presence of a native terpenoid synthetic pathway and facile genetic manipulation (Navale et al. 2021 ; Sun et al. 2020 ). Saccharomyces cerevisiae offers several advantages for terpenoid production, including a high sugar catabolic rate, a relatively fast growth rate, and its GRAS status (Claudia et al. 2017 ). In addition, saccharomyces cerevisiae has an inherent terpenoid production pathway, the mevalonate (MVA) pathway, in which isopentenyl diphosphate (IPP) or dimethylallyl diphosphate (DMAPP), essential intermediates of sterols and ubiquinones, are synthesized from acetyl-CoA (Guo et al. 2020 ). Metabolic reactions in the MVA pathway proceed in the cytosol, which, in turn, are used for terpenoid production. Recently, terpenoid production using organelles has attracted attention owing to their ability to produce acetyl-CoA. Peroxisomes are considered potential acetyl-CoA pools because β-oxidation of fatty acids proceeds in them (Hammer and Avalos 2017 ). In fact, a terpenoid, β-amyrin, was produced at high level (57.8 mg/g dry cell weight (DCW)) by introducing two enzymes of the MVA pathway into peroxisomes and four downstream enzymes into the cytosol or peroxisomes (Du et al. 2022 ). The concentration of acetyl-CoA in the mitochondria is as much as 20–30 times higher than that in the cytosol because of the high enzymatic activity of the mitochondrial pyruvate dehydrogenase complex (Weinert et al. 2014 ). Thus, mitochondria are considered suitable organelles for the introduction of the MVA pathway for terpenoid production (Duran et al. 2020 ). For example, patchoulol, a terpenoid, was produced at a high level (19.24 mg/L) by introducing all the enzymes of the MVA pathway into the mitochondria and a fusion protein of the downstream enzymes into the cytosol (Tao et al. 2022 ). In another example, amorphadiene was produced at 400 mg/L by introducing several enzymes of the MVA pathway into the mitochondria and two downstream enzymes into the cytosol (Yuan and Ching 2016 ). Although introducing the MVA pathway into the mitochondria is attractive for producing terpenoids in yeast, there may be a limit because the mitochondria comprise a small percentage of the entire cell (Uchida et al. 2011 ). We hypothesized that terpenoid production will increase by introducing the MVA pathway into the mitochondria when mitochondrial volume increases. In the present study, we constructed yeast strains to analyze squalene production or β-carotene production and investigated the relationship between mitochondria-mediated terpenoid (squalene or β-carotene) production and mitochondrial volume.
Materials and methods Genes for yeast strain construction The genes used for the construction of the yeast strains were prepared as described in the Supplementary Information. The plasmids, primers, and DNA fragments used are listed in Tables S1 , S2 , and S3 , respectively. Escherichia coli NovaBlue (Merck Millipore, Darmstadt, Germany) was used for plasmid preparation using the Orthodox method. Yeast strains Saccharomyces cerevisiae BY4741 was used as the host strain. The strains used in this study are listed in Table 1 . The methods used for strain construction are described in the Supplementary Information. Mitochondrial volume analysis Mutants with defective mitochondrial morphology were selected from the Yeast Knockout Collection (Open Biosystems, Huntsville, AL, USA) database. The total mitochondrial volume per cell was quantified using a previously described method (Viana et al. 2015 ). The mutant strains were transformed with pGK426-MLS-GFP and cultured in 5 mL of synthetic complete media without uracil (2% glucose, 1.46 g/L Yeast Synthetic Drop-out Medium Supplements; Merck, Darmstadt, Germany) overnight at 30 °C with shaking at 200 rpm. The strains were transferred to YPD media (2% glucose, 2% peptone, and 1% yeast extract) at 0.1–0.4 OD 600 and cultured further for 4 h under the same conditions. When the OD 600 reached 0.5–1.0, the cultures were diluted 8–16-fold with YPD, transferred to a 96-well glass plate coated with concanavalin-A, and incubated at 30 °C for 20 min. The supernatant was removed and fresh YPD was added to each well. MLS-GFP signals in mutants were detected using an Olympus FV1000 confocal microscope (Olympus, Tokyo, Japan). Z-stack images of MLS-GFP were constructed and mitochondrial volumes were calculated using ImageJ and MitoGraph software. The mitochondrial volume in one cell was calculated as the average of at least 10 cells per strain. Analyses for IPP/DMAPP, squalene, and β-carotene contents For squalene production, strains, pre-incubated in 5 mL of YPD at 30 °C overnight, were inoculated into 30 mL of YPD at 0.01 OD 600 and incubated at 30 °C for 24 h in 250-mL baffled flask with shaking at 200 rpm. For β-carotene production, YPD supplemented with 200 μg/mL hygromycin (Nacalai Tesque, Japan) was used, and sampling was performed 48 h after inoculation. DCW was determined using the equation, DCW/OD 600 = 0.561 derived beforehand. Intracellular IPP/DMAPP were extracted as previously described (Luo et al. 2020 ). Briefly, 2 mL of each culture was centrifuged at 5,000 × g for 2 min, and the supernatant was removed. Cells were lysed in an acetonitrile/methanol/water (40:40:20) solution. The solution was chilled at –20 °C for 20 min, followed by centrifugation at 16,000 × g for 10 min. The supernatant was collected, dried, and resuspended in 50 μL of water. IPP/DMAPP were analyzed using liquid chromatography-tandem mass spectrometry (LC–MS/MS) (Agilent Nexera 1260 series high-performance liquid chromatography system and 6460 Triple Quad LC/MS; Agilent Technologies, Santa Clara, CA, USA). Separation, detection, and quantification of IPP /DMAPP were performed as previously described (Kato et al. 2012 ); however, IPP and DMAPP could not be separated from each other. Thus, in this study, the contents are presented as the sum of IPP and DMAPP (referred to as IPP/DMAPP). Intracellular squalene was extracted as previously described (Zhu et al. 2021 ). Briefly, 600 μL of each culture was centrifuged at 5,000 × g for 2 min and the supernatant was removed. 600 μL of ethyl acetate and approximately 200 μL of grass beads (diameter of 0.5 mm) were added to the cells, which were then lysed by shaking at 1,500 rpm for 10 min using a grinder (shake Master NEO BMS, Japan) with a pre-cooled sample holder at 4 °C. The cell lysate was centrifuged at 21,880 × g for 10 min and the upper organic solvent was collected. A GCMS-QP2010 (Shimadzu, Japan) equipped with a DB-5MS column (15 m × 0.25 mm, 0.25 μm film thickness; Agilent Technologies) was used to analyze squalene. The oven temperature was maintained at 100 °C for 2 min, gradually increased to 250 °C at a rate of 20 °C/min, maintained for 2 min, then increased to 325 °C at a rate of 50 °C/min and maintained for 2 min. Intracellular β-carotene was extracted as previously described (Ma et al. 2022 ). Briefly, 100 μL of each culture was centrifuged at 5,000 × g for 2 min and the supernatant was removed. A total of 900 μL of dimethyl sulfoxide was added to the cells, and β-carotene was extracted after incubation at 50 °C for 1 h. After the incubation, 450 μL of methanol was added and the mixtures were centrifuged at 21,880 × g for 5 min. The supernatants were collected. A high-performance liquid chromatography (HPLC) (Shimadzu, Japan) equipped with a BDS Hypersil C18 column (4.6 × 150 mm 2 , 5 μm particle size; Thermo Fisher Scientific, Waltham, MA, USA) and a SPD-20A UV–VIS detector (Shimadzu) was used to analyze β-carotene. The oven temperature was maintained at 25 °C. Methanol and acetonitrile were used as mobile phases at a flow rate of 0.28 mL/min and 0.52 mL/min. β-carotene was detected by absorbance at 450 nm. The representative chromatograms of IPP/DMAPP, squalene, and β-carotene were shown in Figs, S1 , S2 , and S3 . Measurement of intracellular oxidation levels The reactive oxygen species (ROS) level was analyzed using ROS Assay Kit -Photo-oxidation Resistant DCFH-DA- (Dojindo, kumamoto, Japan). Briefly, strains, pre-incubated in 5 mL of YPD at 30 °C overnight, were inoculated into 5 mL of fresh YPD at 1.0 OD 600 and incubated at 30 °C for 4 h in test tubes with shaking at 200 rpm. Then, 1 mL of each culture was centrifuged at 5,000 × g for 2 min, and the supernatant was removed. The cells were treated according to the manufacturer’s protocol and the fluorescence was measured with λ ex = 500 nm and λ em = 540 nm using a plate-reader (INFINITE M NANO + ; Tercan, Männedorf, Switzerland). Spot assay using dithiothreitol (DTT) The effect of DTT addition to media on growth was analyzed as previously described (Bode et al. 2013 ). Briefly, strains, pre-incubated in 5 mL of YPD at 30 °C overnight, were inoculated at 1.0 OD 600 into 5 mL of fresh YPD and incubated at 30 °C for 4 h in a test tube with shaking at 200 rpm. OD 600 was adjusted to 1, and tenfold serial dilutions were spotted at 3μL onto YPD, YPD + DTT 0.5 mM, or YPD + DTT 5 mM plates. The plates were incubated at 30 °C for 24 h.
Results The strain expressing the MVA pathway in mitochondria Proteins can be delivered to the mitochondria by fusion with a mitochondrial localization signal (MLS); however, their enzymatic activities are sometimes interrupted in the mitochondria if the MLS remains fused to them (Duran et al. 2020 ). To introduce the entire MVA pathway into the mitochondria and activate it, we constructed a BY4741 background strain, SSY1, expressing all the enzymes of the MVA pathway fused with the MLS of CoxIV (Fig. 1 a), because it is cleaved away in mitochondria. The MLS of CoxIV consists of 26 amino acids; the first 25 amino acids and last glutamine are required for translocation and cleavage, respectively. To ascertain the action of MLS, we constructed a strain, expressing a GFP fused with MLS, and examined the localization of GFP as compared with MitoTracker. As shown in Fig. 1 b, the signal of the MLS-fused GFP coincided with that of MitoTracker, confirming that the MLS of CoxIV worked effectively in translocation to the mitochondria. Thus, we hypothesized that all enzymes of the MVA pathway fused with MLS were also translocated to the mitochondria. Next, we examined whether the MVA pathway introduced into the mitochondria functioned correctly by measuring the amount of IPP/DMAPP in SSY1 compared to that in BY4741. The IPP/DMAPP content in SSY1 was 164 nmol/g DCW, which was 15-fold higher than that in BY4741, 11 nmol/g DCW (Fig. 1 c). This result indicates that the MVA pathway introduced into the mitochondria functioned and was beneficial for terpenoid production, as previously reported (Lv et al. 2016 ). The total mitochondrial volumes in mutants with defective mitochondrial morphology Mitochondrial volume has been reported to be decreased in a deletion mutant of MGM1 , a mitochondrial morphology-related gene (Bernhardt et al. 2015 ). We hypothesized that mutants with defective mitochondrial morphology could be used to examine the effect of total mitochondrial volume on terpenoid production in yeast expressing the MVA pathway in their mitochondria. We selected 13 genes, FZO1 , MGM1 , UGO1 , DNM1 , FIS1 , MDV1 , CAF4 , MMM1 , MDM10 , MDM12 , MDM31 , MDM32 , and MDM33 (Ohsumi and Shimoda 2007 ). GFP fused with the MLS of CoxIV was expressed in BY4741 and the deletion mutants, and mitochondrial volumes were calculated from the fluorescent signals of GFP using the confocal microscopic Z-stack method (Fig. 2 ). As expected, we found differences in mitochondrial volume among them. The only strain with a higher volume than that in BY4741 was Δ mdm32 (2.2 m 3 /cell), which was 1.8-fold higher than that in BY4741 (1.2 μm 3 /cell). The strains with volumes lower than those in BY4741 were Δ mdm10 , Δ mdm12 , Δ fzo1 , Δ mmm1, Δ mdm31, Δ mdm33, Δ mgm1 , and Δ ugo1 . The strain with the lowest volume was Δ ugo1 (0.4 μm 3 /cell), which was threefold lower than that in BY4741. The effect of mitochondrial volume on mitochondria-mediated terpenoid production We hypothesized that an increase in mitochondrial volume may augment the terpenoid production in yeast expressing the MVA pathway in their mitochondria. To examine the effect of mitochondrial volume on terpenoid production, we used mutants with defective mitochondrial morphology, Δ mdm32 , Δ fzo1 , Δ mgm1 , and Δ ugo1 . Four mutants were constructed from SSY1, and the relationship between mitochondrial volume and IPP/DMAPP or squalene content was examined. In SSY2 (Δ mdm32 ) harboring a large mitochondrial volume, both the contents of IPP/DMAPP (25 nmol/g DCW) and squalene (707 nmol/ g DCW) increased 1.3- and 2.8-fold higher than that in SSY1 (19 nmol/g DCW for IPP/DMAPP, 256 nmol/ g DCW for squalene) (Fig. 3 a and b). In contrast, in SSY3 (Δ fzo1 ), SSY4 (Δ mgm1 ), and SSY5 (Δ ugo1 ) harboring a low of mitochondrial volume, the contents of IPP/DMAPP and squalene decreased compared with that in SSY1, with the exception of squalene in SSY3. The contents of IPP/DMAPP in SSY3, 4, and 5 were 4.65, 1.29, and 1.32 nmol/g DCW, which were 4.2-, 15.1-, and 14.7-fold less than those in SSY1. The concentrations of squalene in SSY3, 4, and 5 were 233, 14, and 25 nmol/ g DCW, which were 1.1-, 18.2-, and 10.1-fold less than those in SSY1. Furthermore, we found that the contents of both IPP/DMAPP and squalene were clearly correlated with the mitochondrial volume (Fig. 3 c and d). In addition, we investigated the effect of the mitochondrial volume on β-carotene production to evaluate the effectiveness of our approach on other terpenoid production. In SSY7 (Δ mdm32 ) harboring a large mitochondrial volume, the contents of IPP/DMAPP (15 nmol/g DCW) and β-carotene (1609 nmol/ g DCW) were 1.2- and 1.4-fold higher than those in SSY6 (12.4 nmol/g DCW for IPP/DMAPP, 1132 nmol/ g DCW for β-carotene) (Fig. 4 a and b). In contrast, the contents of IPP/DMAPP in SSY8 (Δ fzo1 ), 9 (Δ mgm1 ), and 10 (Δ ugo1 ) were 3.0-, 2.7-, and 12.4-fold lower than those in SSY6 at 4.1, 4.6, and 1.0 nmol/g DCW, respectively. The concentrations of β-carotene in SSY8, 9, and 10 were 1030, 491, and 678 nmol/ g DCW, 1.1-, 2.3-, and 1.7-fold lower than those in SSY6, respectively. Furthermore, the contents of both IPP/DMAPP and β-carotene were correlated with the mitochondrial volume (Fig. 4 c and d). These data indicate that the increase in mitochondrial volume augmented terpenoid production in yeast expressing the MVA pathway in the mitochondria.
Discussion In this study, we demonstrated the beneficial effect of the MVA pathway introduced into the mitochondria for terpenoid production by measuring IPP/DMAPP, an essential precursor of terpenoids. Furthermore, using mutants with defective mitochondrial morphology, we found that the contents of IPP/DMAPP and terpenoids (squalene or β-carotene) were positively correlated with the mitochondrial volume and showed differences in the mitochondrial volume. These results not only support the advantages of introducing the MVA pathway into mitochondria for terpenoid production but also suggest that increasing the mitochondrial volume might enhance the production. Both IPP/DMAPP and terpenoids contents were positively correlated with the mitochondrial volume. However, notably, there was a difference in the degree of increase between IPP/DMAPP and terpenoids in SSY2 (Δ mdm32 ) and SSY7 (Δ mdm32 ) (1.3-fold for IPP/DMAPP and 2.8-fold for squalene in squalene production, 1.2-fold for IPP/DMAPP and 1.4-fold for in β-carotene production). IDI1 was not introduced into the mitochondria of any of the strains used in this study. Thus, DMAPP was not synthesized in the mitochondria. The growth defect compared to the wild-type strain is believed to be caused by the accumulation of the pyrophosphate compounds, mevalonate-5-PP, IPP, and DMAPP in the mitochondria (Zhu et al. 2021 ), which was also observed in this study (Fig. 1 C). Pyrophosphate compounds can react with adenosine monophosphate-amino acids through aminoacyl tRNA synthetases and convert them to toxic ATP analogs, which inhibit mitochondrial adenine nucleotide translocase and F1-ATPase (Mönkkönen et al. 2006 , Mookerjee-Basu et al. 2010 ). However, we first speculated that the defective mitochondrial morphology or the introduction of MVA pathway enzymes into the mitochondria might cause mitochondrial dysfunction and induce oxidative stress. Therefore, we measured the ROS levels and growth of BY4741, mdm32 deletion strain, SSY1, and SSY2 (Δ mdm32 ) and found that the ROS levels in SSY2 (Δ mdm32 ) tended to be slightly higher compared to those in BY4741 (Fig. S4 ). Moreover, the addition of DTT as a reducing agent to media improved the growth of the strain, which was exposed to oxidative stress (Bode et al. 2013 ). Thus, a DTT assay was performed using the SSY2 (Δ mdm32 ) strain, which revealed that DDT addition to YPD media did not improve the growth of SSY2 (Δ mdm32 ) (Fig. S5 ). Therefore, the slight increase in ROS levels in SSY2 (Δ mdm32 ) was not the cause of the growth defect of SSY2 (Δ mdm32 ), supporting the theory that growth defects are attributed to the accumulation of the pyrophosphate compounds mentioned in the previous study (Zhu et al. 2021 ). Thus, it is plausible that IPP/DMAPP mainly contains IPP accumulated in the mitochondria. After production in the mitochondrial matrix, IPP is exported to the cytosol, where squalene production proceeds (Zhu et al. 2021 ). During export, IPP must pass through the inner and outer mitochondrial membranes. Several transporters selectively pass through small molecules in the inner membranes (Duran et al 2020 ). Porin is present in the outer membrane and allows small molecules (< 5 kDa) to pass freely between the cytosol and intermembrane space. Thus, passage through the inner membrane is thought to be critical for IPP export from the mitochondria to the cytosol. Mdm32 is a protein present in the mitochondrial inner membrane, and loss of Mdm32 leads to defects in mitochondrial morphology (Okamoto and Shaw 2005 ). Although the mechanism underlying the IPP export remains to be elucidated, structural changes in the inner mitochondrial membrane may enhance IPP export or diffusion. Regarding the Δ fzo1 strains (SSY3 (Δ fzo1 ) and SSY8 (Δ fzo1 )), although both IPP and terpenoid decreased in the Δ fzo1 strains compared to the normal strains (SSY1 and SSY6), we also observed a difference in the degree of decrease between IPP and squalene content: 76% for IPP and 9% for squalene (Fig. 3 a and c) and that between IPP and β-carotene content: 68% for IPP and 9% for β-carotene (Fig. 4 a and c). These differences seem to arise from the balance between IPP synthesis in the mitochondria and IPP export. Due to the low mitochondrial volume, a sufficient amount of enzymes in the MVA pathway could not be transported to the mitochondria; thus, the rate of IPP synthesis decreased. The IPP produced via the mitochondria in the Δ fzo1 strains seem to be exported smoothly to the cytosol, because of its low synthetic activity, and is subsequently metabolized to squalene or β-carotene. Thus, although the terpenoid level in the Δ fzo1 strains was slightly lower than that in the normal strains, the difference in terpenoid levels between the normal strains and the Δ fzo1 strains was not as drastic as that in IPP. In contrast, the IPP/DMAPP levels in the Δ fzo1 strains decreased drastically compared with those in the normal strains (Figs. 3 a and 4 a). The Fzo1 protein resides in the outer mitochondrial membrane, and the loss of Fzo1 may not affect IPP export or diffusion due to the presence of porins in the outer membrane, suggesting that there may be no difference in the export ability of mitochondria. There may be a limit to export, and the IPP in the normal strains probably accumulates in the mitochondrial matrix. In the present study, SSY1 to 10 showed growth defects compared to the wild-type strain, BY4741. The defects were severe in the Δ mdm32 strains (SSY2 and SSY7). The deletion of FZO1 , MGM1 , UGO1 , or MDM32 leads growth defects by dysfunction of mitochondria (Okamoto and Shaw 2005 ). In the normal and Δ mdm32 strains, the IPP/DMAPP levels were high (Figs. 3 a and 4 a), indicating that pyrophosphate compounds may accumulate in the mitochondria. Therefore, the severe growth defects in the Δ mdm32 strains is probably due to both mitochondrial dysfunction and pyrophosphate compound accumulation in the mitochondria. As the normal strains had no mitochondrial morphological defects. Thus, the growth defect in the normal strains may only be due to the accumulation of pyrophosphate compounds in the mitochondria. In contrast, the growth defects in Δ fzo1 strains, Δ mgm1 strains (SSY4 and SSY9), and Δ ugo1 strains (SSY5 and SSY10) may be due to mitochondrial dysfunction because their IPP levels, that is, their activities in the MVA pathway introduced in the mitochondria, were considerably low. Further approaches to expand the volume of mitochondria without growth defects by mitochondrial dysfunction and enhance the export of IPP, inhibiting pyrophosphate compound accumulation, would be required to increase terpenoid production using our proposed mitochondria-based strategy.
Abstract Terpenoids are widely used in the food, beverage, cosmetics, and pharmaceutical industries. Microorganisms have been extensively studied for terpenoid production. In yeast, the introduction of the mevalonate (MVA) pathway in organelles in addition to the augmentation of its own MVA pathway have been challenging. Introduction of the MVA pathway into mitochondria is considered a promising approach for terpenoid production because acetyl-CoA, the starting molecule of the MVA pathway, is abundant in mitochondria. However, mitochondria comprise only a small percentage of the entire cell. Therefore, we hypothesized that increasing the total mitochondrial volume per cell would increase terpenoid production. First, we ascertained that the amounts of isopentenyl diphosphate (IPP) and dimethylallyl diphosphate (DMAPP), the final molecules of the MVA pathway, were 15-fold higher of the strain expressing the MVA pathway in mitochondria than in the wild-type yeast strain. Second, we found that different deletion mutants induced different mitochondrial volumes by measuring the mitochondrial volume in various deletion mutants affecting mitochondrial morphology; for example,Δ mdm32 increased mitochondrial volume, and Δ fzo1 decreased it. Finally, the effects of mitochondrial volume on amounts of IPP/DMAPP and terpenoids (squalene or β-carotene) were investigated using mutants harboring large or small mitochondria expressing the MVA pathway in mitochondria. Amounts of IPP/DMAPP and terpenoids (squalene or β-carotene) increased when the mitochondrial volume expanded. Introducing the MVA pathway into mitochondria for terpenoid production in yeast may become more attractive by enlarging the mitochondrial volume. Key points • IPP/DMAPP content increased in the strain expressing the MVA pathway in mitochondria • IPP/DMAPP and terpenoid contents are positively correlated with mitochondrial volume • Enlarging the mitochondria may improve mitochondria-mediated terpenoid production Supplementary Information The online version contains supplementary material available at 10.1007/s00253-023-12922-5. Keywords Open Access funding provided by Kobe University.
Supplementary Information Below is the link to the electronic supplementary material.
Acknowledgements The authors are grateful for the support from NEDO Projects No. P16009 (Development of production techniques for highly functional biomaterials using plant and other organism smart cells). Author contribution SY designed and performed the experiments, interpreted the results, and helped write the manuscript. TB was involved in planning and performing the experiments. TK performed an interpretation of the results and helped write the manuscript. AK reviewed the research plan and manuscript. TH was the supervisor in research planning, experiments, interpretation of the results, and writing of the manuscript. All authors read and approved the final manuscript. Funding Open Access funding provided by Kobe University. Data availability The data generated and/or analyzed during the current study are available from the corresponding author upon reasonable request. Declarations Ethical approval This article does not contain any studies with human participants or animals performed by any of the authors. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:01
Appl Microbiol Biotechnol. 2024 Jan 13; 108(1):1-9
oa_package/7d/53/PMC10787878.tar.gz
PMC10787880
0
Introduction The asymmetric reduction of C = C double bonds is widely applied in the pharmaceutical, fine chemical, and agrochemical industries, as up to two stereogenic centers can be created in those processes. Unfortunately, the chemo-catalyzed reduction of olefins using hydrogen gas and transition metal catalysts with expensive chiral ligands suffers from poor environmental sustainability, narrow substrate range, or low enantioselectivity (Hollmann et al. 2021 ). The biocatalyzed reduction provides an alternative green synthetic route with a broad substrate range and excellent enantioselectivity. Ene-reductases (ERs), NAD(P)H-dependent flavoproteins from the old yellow enzyme (OYEs) family, are powerful biocatalysts for the asymmetric reduction of activated alkenes, generating the products up to two new stereogenic centers with excellent enantioselectivities (Hall and Bommarius 2011 ; Toogood and Scrutton 2018 ). Since the first OYE was isolated from yeast, many OYEs have been identified from bacteria, yeasts, fungi, plants, and algae (Parmeggiani et al. 2022 ). The OYEs could be used in the reduction of C = C double bonds bearing an electron-withdrawing group, such as nitro, aldehyde, ketone, carboxylic acid, ester, imide, or nitrile moiety. The enzymes of the OYE family were proposed to be divided into five subclasses based on the phylogenetic and biochemical analysis, called class I (classical OYEs), class II, class III (thermophilic-like OYEs), class IV, and class IV (fungal OYEs). Most of the known OYEs belong to the classical OYE subgroup, which contains OYE1-like OYEs from fungi, OYE3 from Saccharomyces sp . , PETNR-like OYEs from bacteria, and OPR-like OYEs from plants (Peters et al. 2019 ). These enzymes have been used in the synthesis of various industrial interested building blocks, such as ( R )-citronellal (Zheng et al. 2018 ), amino acid derivatives (Winkler et al. 2012 ), and ( R )-3-Hydroxy-2-methylpropanoate (Stueckler et al. 2010 ). Thermophilic enzymes have great potential for industrial use as they have high resistance to temperature and organic solvents. Since the identification of the first thermophilic-like OYE YqjM from Bacillus subtilis (Kitzing et al. 2005 ), another 10 thermophilic-like OYEs belonging to class III have been isolated and characterized, including XenA from Pseudomonas putida (Griese et al. 2006 ), Ts OYE from Thermus scotoductus SA-01 (Opperman et al. 2008 ), TOYE from Thermoanaerobacter pseudethanolicus (Adalbjornsson et al. 2010 ), Gk OYE from Geobacillus kaustophilus (Schittmayer et al. 2011 ), Chr -OYE3 from Chryseobacterium sp. CA49 (Xu et al. 2014 ), Geo ER from Geobacillus sp. 30 (Tsuji et al. 2014 ), OYERo2 from Rhodococcus opacus 1CP (Riedel et al. 2015 ), F OYE-1 from Ferrovum sp. JA12 (Scholtissek et al. 2017 ), Ca OYE from Chloroflexus aggregans (Robescu et al. 2021 ), and Pf ER2 from Pseudomonas fluorescens (Shi et al. 2021 ). Moreover, four thermal-tolerant OYEs were also identified from other subclasses, including Bc OYE (class IV) from Bacillus coagulans (Zhou et al. 2014 ), YqiG (class IV) from B. subtilis (Sheng et al. 2016 ), Ct OYE (class I) from Chroococcidiopsis thermalis , and Gs OYE (class I) from Galdieria sulphuraria (Robescu et al. 2020 ). All the thermophilic-like OYEs showed activity optima around 40–55 °C exhibited high thermostability and high resistance toward organic solvents, so they have huge potential for industrial application. Several thermophilic-like OYE structures have been revealed, including YqjM (PDB: 1Z41) (Kitzing et al. 2005 ), XenA (PDB: 2H8X) (Griese et al. 2006 ), CrS (PDB: 3HF3) (Opperman et al. 2010 ), TOYE (PDB: 3KRU) (Adalbjornsson et al. 2010 ), Gk OYE (PDB: 3GR7) (Schittmayer et al. 2011 ), Rm ER (PDB: 5OCS) (Opperman 2017 ), and Ca OYE (PDB: 7O0T) (Robescu et al. 2021 ), which showed that OYE homologs are highly conserved. The general catalytic mechanism of OYE-mediated reduction is well understood. The electron-withdrawing group moiety forms an H-bond interaction with two donor residues, usually histidine and asparagine, and then the addition of hydride from the reduced flavin to the Cβ of the activated alkene, followed by proton transfer to the Cα of the substrate from a tyrosine or cysteine residue. Although many OYEs have been identified over the last years, the demand for new OYEs keeps growing to overcome application limitations, such as low catalytic activity and poor stability in harsh reaction conditions. Fungi appear to be valuable sources of OYEs, but most of the fungal OYEs are unexplored. Aspergillus sp . produces many secondary metabolites, and it has been proposed that ene-reductases participate in the synthetic pathways. For example, the Aspergillus fumigatus old yellow enzyme EasA reduces chanoclavine-I aldehyde to dihydrochanoclavine aldehyde in the ergot alkaloid pathways (Cheng et al. 2010 ). More recently, four OYEs were identified from Aspergillus niger and Botryotinia fuckeliana (Robescu et al. 2022a , 2022b ). Herein, we mined the OYEs in the filamentous fungus Aspergillus flavus to expand the ene-reductase toolbox. A new thermophilic-like OYE ( Af OYE1) was identified and carefully characterized, which exhibited broad substrate scope and excellent enantioselectivity.
Materials and methods Reagents Alkene substrates and racemic reduction products were purchased from MilliporeSigma (MA, USA), Aladdin (Shanghai, China), or TCI (Shanghai, China). Glucose dehydrogenase from Bacillus was purchased from Aladdin (G139687, Shanghai, China), and NADH disodium salt hydrate was purchased from TCI (Shanghai, China). All protein commercial crystallization kits were purchased from Hampton Research (CA, USA), MiTeGen LLC (NY, USA), Qiagen (Hilden, Germany), or Molecular Dimensions (OH, USA). All other reagents and solvents were commercially available and were used without further purification. General analysis methods 1 H NMR spectra were acquired on a Bruker-400 spectrometer in CDCl 3 , and all signals were reported in parts per million (ppm) downfield relative to tetramethylsilane. Optical rotations were measured with a Perkin Elmer 341 polarimeter. Gas chromatographic analyses were performed on a Thermo Scientific TRACE 1300 gas chromatograph equipped with a CHIRASIL-DEX CB column (Agilent Technologies, USA), and using a flame ionization detector, nitrogen was used as the carrier gas at 5 mL/min, the split ratio was 1:20 (v/v), and the column temperature was programmed as being kept at 50 °C for 1 min and then upgraded to 150 °C at the rate of 3 °C/min. The protein purities were analyzed using SDS-PAGE using 4–12% polyacrylamide gradient gel (SurePAGE, Genscript, Nanjing, China) running in 50 mM MOPS, 50 mM Tris base, 0.1% SDS, and 1 mM EDTA, at pH 7.7, and stained with Coomassie Blue. Protein concentrations were measured using the Bradford assay (Coomassie Brilliant Blue G-250) and bovine serum albumin as the protein standard. Mining OYE from Aspergillus flavus and plasmid construction The thermophilic-like old yellow enzyme YqjM was submitted to a BLASTp program against the A. flavus strain NRRL3357 proteins in UniProtKB/Swiss-Prot database, which returned an NADPH dehydrogenase afvA (UniProt: B8N8Q9, GenBank: XP_041144250.1) with 41.7% identities and an NADP-dependent oxidoreductase lnbE (UniProt: B8NWW6) with 34% identities. Here, the NADPH dehydrogenase afvA was selected, and its gene sequence with an N-terminal His6-tag followed by a tobacco etch virus (TEV) protease cleavage site was codon-optimized by OPTIMIZER ( http://genomes.urv.es/OPTIMIZER/ , See Supplementary Information for nucleotide sequence) (Puigbò et al. 2007 ) for the expression in E. coli and synthesized by GENEWIZ (Nanjing, China). The synthesized gene was inserted into the pET28b vector between the Nco I and Xho I enzyme sites, creating the plasmid pET28b- afvA , and the sequence was confirmed via sequencing (Shenggong, Shanghai, China). Protein production and purification The plasmid pET28b- afvA was transferred into E. coli BL21 (DE3) for protein production. LB medium containing kanamycin (50 μg/mL) was inoculated with E. coli BL21 (pET28b- afvA ), and the culture was incubated at 37 °C and 220 rpm. Then, isopropyl-β-D-thio-galactoside (IPTG) was added into the culture at the final concentration of 1 mM when the OD 600 reached about 0.5. The culture was further cultured at 16 °C for 18 h for protein expression, and then the cells were harvested by centrifugation (5000 g, 4 °C) for 6 min. The cell pellets were re-suspended in HEPES buffer (25 mM, pH 7.5, 10 M NaOH solution was used to adjust pH) containing 300 mM NaCl, 20 mM imidazole, and 1 mM tris(2-carboxyethyl)phosphine (TCEP), and then the cells were lysed by low-temperature ultra-high‐pressure homogenizer (JN-mini, JNBio, China) at 1100 bar and 4 °C. The lysis was centrifuged at 18,000 g and 4 °C for 1 h to remove the cell debris. The clear lysate was filtered through a surfactant-free cellulose acetate membrane with a 0.45 μm pore size and loaded on the His TrapTM HP column, which was pre-equilibrated with HEPES buffer (25 mM, pH 7.5) containing 300 mM NaCl, 20 mM imidazole, and 1 mM TCEP. Proteins were eluted by HEPES buffer (25 mM, pH 7.5) containing 300 mM NaCl and 1 mM TCEP with a linear imidazole concentration gradient from 20 to 500 mM. The peak fractions were collected and followed by His6-TEV protease cleavage and dialysis against HEPES buffer (25 mM, pH 7.5) containing 300 mM NaCl and 1 mM TCEP overnight. The dialysate was loaded onto a His TrapTM HP column again, which was pre-equilibrated with HEPES buffer (25 mM, pH 7.5) containing 300 mM NaCl and 1 mM TCEP. Flow through and wash fractions that contain the desired purified protein were collected and checked by SDS-PAGE. The fractions containing desired purified protein were pooled and concentrated by Amicon® Ultra-15 Centrifugal Filter Unit (10 K, Merck Millipore) and used as the catalyst. To prepare protein samples for crystallization, the purified protein was further purified using gel filtration (HiLoad 16/60 Superdex pg 200, GE Healthcare) with HEPES buffer (25 mM, pH 7.5) containing 100 mM NaCl and 1 mM TCEP. The peak fractions were concentrated and used for crystallization. General procedures of AfOYE1-catalyzed reductions The purified Af OYE1 was used in all the measurements. All the reaction mixtures contained 5 mL potassium phosphate buffer (100 mM, pH 7.0), 0.1 μM purified Af OYE1, 5 U glucose dehydrogenase from Bacillus , 50 mM glucose, 20 mM NADH, and 8 mM substrate, and then the reactions were conducted at 45 °C for 10 min in triplicates. Then, the mixtures were extracted with ethyl acetate (1 mL), and then the yields were determined by GC with an external standard calibration curve. One unit of enzyme activity was defined as the amount of enzyme producing 1 μmol product per minute under the assay condition. The irreversible thermal inactivation ( T 1/2 ) of Af OYE1 was analyzed by heating the purified Af OYE1 in potassium phosphate buffer (pH 7.0) at different temperatures (30–70 °C) for 5 min, chilled on ice. The residual activities on 2-cyclopentenone were then measured under standard assay conditions. To determine the optimal reaction conditions, 8 mM 2-cyclopentenone was used as a substrate, and the reactions were initiated by the addition of purified Af OYE1. The reaction mixtures in 5 mL potassium phosphate buffer (100 mM, pH 7.0), were conducted at 25–60 °C for 10 min, or in 5 mL potassium phosphate buffer (100 mM, pH 5.5–8.0) were conducted at 45 °C for 10 min. Then, the mixtures were extracted with ethyl acetate (1 mL), and the yield of cyclopentanone was determined by GC with an external standard calibration curve. Preparative biotransformation was carried out in 50 mL potassium phosphate buffer (100 mM, pH 7.0), 0.5 μM purified Af OYE1, 20 U glucose dehydrogenase from Bacillus , 200 mM glucose, 20 mM NADH, and 50 mM substrate, and then the reactions were conducted at 45 °C for 10 h. Then, the mixtures were extracted with ethyl acetate (3 × 50 mL), the organic phase was combined and dried by anhydrous NaSO 4 , the solvents were removed under certain reduced pressure, and then the final product was purified with column chromatography and subjected to NMR and optical rotation analysis. The kinetic parameters K m and k cat of Af OYE1 for 2-cyclopentenone were determined by measuring the production of cyclopentanone. The reaction mixture containing varying concentrations of 2-cyclopentenone in 5 mL potassium phosphate buffer (100 mM, pH 7.0) was incubated under the optimal reaction conditions (pH 7.0, and 45 °C) at 220 rpm for 2.0 min. The mixtures were extracted with ethyl acetate (1 mL), and then the yield of cyclopentanone was determined by GC. The parameters were calculated using the Prism program (GraphPad, San Diego, CA, USA). Determination of the flavin cofactor The flavin content of Af OYE1 was determined spectrophotometrically from absorption scans (350–750 nm) using a SpectraMax® M2e (Molecular Devices, USA) microplate reader. Free flavin was released from Af OYE1 by the protein denaturation through incubating in a boiling water bath for 10 min; then, the samples were centrifuged at 15,000 g for 10 min. The supernatant, Af OYE1 solution, and standard flavin mononucleotide (FMN) solution were analyzed by microplate readers to determine flavin content. Crystallization Initial crystallization screening of Af OYE1 (20 mg/mL) was carried out at 18 °C with high-throughput sparse matrix crystallization trails using a Gryphon crystallization robot (Art Robbins Instruments). MRC 96-well two-drop standard plates were adopted with the sitting-drop vapor diffusion method by mixing 0.5 μL protein and 0.5 μL reservoir solution in the initial screening. A 24-well plate was used with the hanging-drop vapor diffusion method by mixing 1 μL protein and 1 μL reservoir solution in manual optimization steps. Well-diffracted crystals were achieved in the condition containing 0.1 M Bis–Tris propane, pH 6.5, 15% (w/v) PEG 3350, and 0.2 M sodium nitrate, after seeding with smaller crystals initially obtained in the same condition. Crystals were cryoprotected in mother liquor supplemented with 25% (v/v) ethylene glycerol and snap-frozen in liquid nitrogen for data collection. Data collection and structural determination The X-ray diffraction data were collected at 100 K Crystallography Beamline 18U1 (BL18U1) using a Pilatus 3 S 6 M detector at Shanghai Synchrotron Radiation Facility (SSRF, Shanghai, China) and at the wavelength of 0.979 Å. Datasets were processed using XDS (Kabsch 2010 ). The structure was solved by molecular replacement using a modified monomer model of ene-reductase (PDB code: 5OCS, sharing 42% identity with Af OYE1) (Opperman 2017 ) with all amino acids mutating to alanine in coot as a search model. An Af OYE1 dimer was positioned in the asymmetric unit using PHENIX Phaser-MR (McCoy et al. 2007 ). All refinement and model-building procedures were carried out by PHENIX.refine (Adams et al. 2010 ) and COOT (Emsley and Cowtan 2004 ). FMN was automatically imported from the Coot dictionary. The final parameters obtained for the best models reached a Rfactor/Rfree of 0.144/0.167 at 1.55 Å (Table S1 ). All the structure figures were prepared by PyMOL. Phylogenetic analysis The sequence alignment of 83 known OYEs (Table S2 ) and Af OYE1 was done by ClustalW, and then phylogenetic analysis was conducted using the Maximum Likelihood statistical method based on the Jones-Taylar-Thornton (JTT) model in MEGAX. The initial tree for the heuristic search was obtained by applying NJ/BioNJ algorithms to a matrix of pairwise distance that was estimated by a p -distance and then selecting the topology with superior log likelihood value (− 28,266.79). The visualizing phylogenetic tree was produced by an online tree viewer iTOL Version 6.5.8, and the assignment of the subclasses was based on a previous evolutionary study. Construction of point mutations The mutants were constructed by site-directed mutagenesis using primers listed in Table S3 and confirmed by sequencing (Shenggong, Shanghai, China). The correct plasmid was transformed into E. coli BL21 for protein production.
Results Mining OYEs from A. flavus and sequence analysis To identify new OYEs from A. flavus , the thermophilic-like OYE YqjM from Bacillus subtilis was used as a reference to search the A. flavus proteins in UniProtKB/Swiss-Prot database. The returned results showed that A. flavus contains two OYEs, including a putative NADPH dehydrogenase afvA (UniProt: B8N8Q9) and an NADP-dependent oxidoreductase lnbE (UniProt: B8NWW6). The afvA and lnbE have 41.7% and 34% identities with YqjM, respectively. Here, the afvA was selected since it has higher identities with the known YqjM, and then it was overexpressed in E. coli for further analysis. Functional analysis showed that afvA has the ability to reduce 2-cyclopentenone into 2-cyclopentanone. The Af OYE1 was previously named afvA, which was annotated as an NADPH dehydrogenase. afvA was proposed to catalyze the hydroxylation of siderin and 7-demethyl siderin to the corresponding hydroxylsiderin and hydroxydemethylsiderin (Uka et al. 2020 ). However, we found that afvA could not hydroxylate the siderin analogs, such as 6-hydroxy-4-methylcoumarin, 6-methylcoumarin, and 7-methylcoumarin, which indicated that afvA might not be able to catalyze the hydroxylation. Here, we annotated afvA as Af OYE1 based on the functional analysis. The phylogenetic analysis of Af OYE1 and 83 known OYEs was conducted, and the results showed that Af OYE1 belongs to the group of Class III OYEs (previous thermophilic-like OYEs group) (Fig. 1 ) (Peters et al. 2019 ). The sequence alignments showed that Af OYE1 conserved the two fingerprint motifs of the thermophilic-like OYE homologs and the catalytic residues of OYEs, including His211, His214, and Tyr216 (Fig. S1 ) (Kitzing et al. 2005 ). Af OYE1 has the highest similarities to the OYEs from Botryotinia fuckeliana ( Bf OYE4) and Aspergillus niger ( An OYE8), with 53.6% and 51.7% identities, respectively (Robescu et al. 2022a ). The phylogenetic analysis and sequence properties also indicated that Af OYE1 is an ene-reductase other than a hydroxylase. Expression and characterization of the recombinant The heterologous expression of Af OYE1 with an N-terminal His6-tag was conducted in the host E. coli BL21 (DE3) at 16 °C. SDS-PAGE analysis showed that Af OYE1 was produced with a molecular mass of around 43 kDa ( Fig. S2 ), which was in accordance with the predicted molecular mass derived from the amino acid sequence. The UV–VIS absorption spectra of the purified Af OYE1, released flavin, and standard FMN suggested that Af OYE1 contained an FMN molecule as its prosthetic group and the FMN was non-covalently bound to the protein (Fig. 2 ). The supernatant after boiling of Af OYE1 and centrifugation turned into bright yellow, and both the released flavin and standard FMN had identical spectra, which had maxima absorbance at about 370 and 450 nm. The flavin contained Af OYE1 exhibited maxima absorbance at 370 and 460 nm. Evaluation of the thermostability of AfOYE1 and optimal reaction conditions The Af OYE1 thermal stability was evaluated by detecting the midpoint for thermal inactivation ( T 1/2 ). The irreversible thermal deactivation curve showed that the T 1/2 of Af OYE1 was about 60 °C (Fig. 3 A). Moreover, the catalytic activity of Af OYE1 decreased slowly after 5 min heat treatment at the temperature below 55 °C, and then a drastic decline was observed at the temperature above 55 °C. To further characterize Af OYE1, the effects of temperature and pH on its catalytic activity with 2-cyclopentenone were investigated carefully. Af OYE1 exhibited high catalytic activities at temperatures 40–50 °C and observed maximum activity (0.85 U/mg) at 45 °C (Fig. 3 B). The data showed that Af OYE1 still conserved 70% of the maximal activity at 60 °C (0.6 U/mg) and over 60% of the maximal activity when the temperature was down to 20 °C (0.52 U/mg). Moreover, Af OYE1 had high catalytic activity in a wide range of pH (pH 5.5–8.0) (Fig. 3 C). The highest catalytic activity (1.06 U/mg) of Af OYE1 was observed at pH 7.0, and it displayed over 85% of the maximal activity at pH 5.5 (0.91 U/mg) and pH 8.0 (0.94 U/mg). Evaluation of the catalytic activity of AfOYE1 toward activated alkenes and kinetic analysis The Af OYE1-catalyzed bioreduction system was further applied to different kinds of activated α, β-unsaturated alkenes. The results showed that cyclic enones ( 1 , 2 , 3 , and 4 ), acrylamide ( 5 ), nitroalkenes ( 6 and 7 ), and α, β-unsaturated aldehydes ( 8 ) could be reduced by Af OYE1 (Fig. 4 ). However, cyclic imide ( 9 ) and α, β-unsaturated esters ( 12 and 13 ) could not be accepted as substrates for Af OYE1. Af OYE1 displayed the highest activity toward 2-cyclohexenone with TTN of 2050. For the Af OYE1-catalyzed bioreduction of cyclic enones, it had higher activity toward cyclohexenones than cyclopentenones ( 3 vs. 1 , 4 vs. 2 ). Similar to other OYEs in the thermophilic-like group, Af OYE1 was shown to have a restricted substrate spectrum (Amato and Stewart 2015 ). The results showed that the substituent on the cyclic moiety had an important effect on the catalytic activity. Af OYE1 could catalyze the reduction of α-methyl substituted cyclopentenone ( 2 ) and cyclohexenone ( 4 ). However, it could not convert the β-methyl substituted cyclopentenone ( 10 ) and cyclohexenone ( 11 ) into the corresponding product. Moreover, the Af OYE1-catalyzed reduction had excellent enantioselectivities toward the prochiral alkenes. The bioreduction of 2-methylcyclopentenone ( 2 ) and 2-methylcyclohexenone ( 4 ) yielded the corresponding ( S )-2-methylcyclopentanone ([α] 25 D = + 16.5° (c 0.4, CHCl 3 ) (( S ), [α] 25 D = + 114.9° (c 0.52, CHCl 3 ) (Shimoda et al. 2004 )) and ( S )-2-methylcyclohexanone with > 99% ee. The kinetic parameters K m and k cat of Af OYE1 were investigated with varying concentrations of 2-cyclopentenone (Fig. S3 ). The K m value determined by non-linear regression was 0.061 ± 0.005 mM −1 , which is significantly smaller than those of most of the known OYEs (Robescu et al. 2022b , 2020 ; Xu et al. 2014 ). The apparent k cat of the Af OYE1 on 2-cyclopentenone was 34.12 ± 0.74 min −1 , resulting in catalytic efficiencies of 9.32 mM −1 s −1 . Crystal structure of AfOYE1 The structure of Af OYE1 (PDB: 8J59) was determined to have a maximum resolution of 1.55 Å in the P1211 space group with an elongated dimer per asymmetric unit (Table S2 ). The overall structure of Af OYE1 showed an expected eight-stranded (α, β)-barrel fold of triosephosphate isomerase (TIM), which is highly conserved in the thermophilic-like OYEs (Fig. 5 A). On the top of the β-barrel core of each monomer encircles one FMN cofactor, which points toward the C-terminal side and is buried within the active site cavity. Similar to other thermophilic-like OYEs, the PDBePISA analysis ( https://www.ebi.ac.uk/msd-srv/prot_int/cgi-bin/piserver ) of the Af OYE1 structure showed that the inverse dimeric architecture with an interacting surface of 492 Å 2 is contributed by α1-helix, the peculiar long N-terminal loops, and the C-terminal helix of each protomer, which contains 7 hydrogen bonds and 6 salt bridges among the 41 interfacing residues (Fig. 5 B and Table S4 ) (Krissinel and Henrick 2007 ). Meanwhile, Af OYE1 has a slightly negative surface but a highly positive inner surface at the entrance of the active site, which endows its ability to bind the negatively charged FMN cofactor (Fig. S4 ). The DoGSiteScorer analysis ( https://proteins.plus ) of the structure showed that the catalytic cavity of Af OYE1 is quite deep, and the FMN cofactor is heavily buried at the bottom of the catalytic hole with a depth of 24.23 Å (Table S4 ). As shown by the difference electron density map F o - F c , the si -face of the FMN faced the solvent, while the re -side was in contact with the protein backbone tightly via the extensive hydrogen bonding and hydrophobic interactions with the side chain and main chain elements (Fig. 5 C and Fig. S5 ) (Robescu et al. 2022b ). Meanwhile, the histidine pair (His211/His214) together with the highly conserved proton donor Tyr216 interacts with a buffer-sourced chloride anion and is positioned on the si -face of the isoalloxazine ring of FMN. It indicated that His/His together with Tyr 216 participated in the binding and correct orientation of substrate (Fig. 5 C) (Robescu et al. 2022b , 2020 , 2021 ). Until now, two OYE structures from filamentous fungi, including OYE4 from B. fuckeliana ( Bf OYE4, PDB: 7BLF) (Robescu et al. 2022a ) and OYE from A. niger ( An OYE8, PDB: 7QFX) (Robescu et al. 2022b ), have been reported. The superposition structures of Af OYE1 with Bf OYE4 and An OYE8, with an r.m.s.d. of 0.694 Å and 0.588 Å, respectively, showed that they have similar monomer folding, except the α6-helix for the residues (from Phe271 to Glu280), the loop region (from Ser316 to Gly325), and the cap subdomain region (from Ala132 to Thr182) (Fig. S6 ). The protruding α6-helix and the long N-terminal loops of Af OYE1 work dimerization function in multi-asymmetry units (Fig. S7 ). However, the protruding α6-helix shows random loops in other OYEs (Fig. S6 ). The loop region from Ser316 to Gly325 runs on the enzyme surface and reaches the active site entrance, which contributes to the size and feature of the catalytic cavity (Fig. S8 A). The residues Ile318 and Ile320 in the loop have a distance to the FMN flavin group of 5.1 Å and 4.1 Å, respectively. Meanwhile, the residue Ile320 points its side chain toward the top of the catalytic cavity, which is stabilized by a hydrogen bond with Arg371 at the C-terminal α-helix (from Arg371 to Gln375) (Fig. S8 B). The cap region of Af OYE1 with large unstructured turns is located on the top of the entrance of the active site (from Ala132 to Thr182), while the Bf OYE4 shows α-helices and the Gs OYE (PDB: 6S0G) shows β-hairpins (Robescu et al. 2020 ) (Fig. S8 C), which may involve interacting with NADH/NADPH and adjusting the selectivity toward FMN cofactor (Adalbjornsson et al. 2010 ; Knaus et al. 2016 ; Pompeu et al. 2012 ; Pudney et al. 2007 ). The C-terminal α helix of each monomer (from Thr395 to Phe401) points to the active sites of the adjacent monomer, and the “finger” residue Trp399 interacts with the respective FMN flavin group at a distance of 3.8 Å (Fig. S8 B). This may play an important role in enzyme assembling, TIM barrel catalytic domain surface stabling, and two protomers dimerization (Fig. S8 B). Analysis of the catalytic residues of AfOYE1 Both the sequence alignment and structure analysis indicated that Af OYE1 highly conserved the catalytic residues, including His211, His214, and Tyr216. Here, the function of those residues was investigated. Both His211 and His214 were mutated into leucine, creating the variants H211L and H214L. The catalytic activity analysis showed that these two variants still conserved about 30% of catalytic activity of the wild type Af OYE1 (Fig. 6 ). Previously, the histidine pair has been considered as the binding motifs of OYE homologs, such as His164 and His167 in YqjM (Shi et al. 2020 ), His172 and His175 in Crs (Opperman et al. 2010 ), and His182 and Asn185 in “thermophilic-like” OYE homologs (Riedel et al. 2015 ). The variants H211L and H214L had only 30% catalytic activity of the wild type, which indicated that the histidine residues in the catalytic center were important for substrate binding. Furthermore, the tyrosine residue in the catalytic center of OYE was proposed as a proton donor of OYE, which contributed to the transfer of proton to substrate at the α-position (Kitzing et al. 2005 ). Therefore, Tyr216 was mutated to phenylalanine (Y216F) to investigate whether Tyr216 serves as the proton donor in Af OYE1. The results showed that the variant Y216F conserved 30% catalytic activity of the wild type (Fig. 6 ), which indicated that Tyr216 did act as a proton donor in Af OYE1. However, alternative proton donors might exist, because the variant Y216F was still able to reduce 2-cyclopentenone.
Discussion The Af OYE1 was previously named afvA, which was annotated as an NADPH dehydrogenase. The proposed function of afvA was to hydroxylate siderin and 7-demethyl siderin in the biosynthesis of aflavarin (Cary et al. 2015 ). Here, the recombinant Af OYE1 was shown to have the ability to reduce the activated alkenes. However, it could not catalyze the hydroxylation of the siderin analogs, such as 6-hydroxy-4-methylcoumarin, 6-methylcoumarin, and 7-methylcoumarin. These data indicated that Af OYE1 might not play as a hydroxylase in the biosynthesis of aflavarin. Similar to other OYEs in the thermophilic-like group, the recombinant Af OYE1 was shown to have a restricted substrate spectrum (Amato and Stewart 2015 ). It has the catalytic activity toward cyclic enones, acrylamide, nitroalkenes, and α, β-unsaturated aldehydes, but the cyclic imide and α, β-unsaturated esters could not be accepted as substrate. Until now, over twenty OYE homologs from Class III OYEs have been identified, including Bf OYE4 from filamentous fungus B. fuckeliana and An OYE8 from filamentous fungus A. niger . Although Bf OYE4 and An OYE8 were phylogenetically clustered in the group of Class III OYEs, the thermostability of these two filamentous fungi source OYEs is similar to those of non-thermostable OYEs (Robescu et al. 2022b ). This newly identified Af OYE1 from filamentous fungus A. flavus had high catalytic activity at 40–50 °C and with the optima temperature at 45 °C. Moreover, the thermal stability analysis showed that the midpoint for thermal inactivation ( T 1/2 ) of Af OYE1 was 60 °C. All the data indicated that Af OYE1 is a thermostable OYE. Therefore, the first thermostable OYE is identified from A. flavus , which provides a new optional thermostable OYE in asymmetric reduction of activated alkenes. Af OYE1 catalyzed the bioreduction of 2-methylcyclopentenone and 2-methylcyclohexenone with excellent ( S )-enantioselectivity, producing the corresponding ( S )-2-methylcyclopentanone and ( S )-2-methylcyclohexanone with > 99% ee. Until now, most of the known OYEs are ( R )-stereoselectivity, and only 4 OYEs, including KYE1 from Kluyveromyces lactis (Yanto et al. 2011 ), YersER from Yersinia bercovieri (Yanto et al. 2011 ), Ct OYE from Chroococcidiopsis thermalis (Yanto et al. 2011 ), and Gs OYE from Galdieria sulphuraria (Robescu et al. 2020 ), were ( S )-stereoselectivity in the reduction of 2-methylcyclohexenone to 2-methylcyclohexanone. However, Af OYE1 only shared 23–33% of identities with those four OYEs. The identification of a new thermostable Af OYE1 would provide a robust enzyme for ( S )-enantioselective bioreductions. The novel structure of the loop of Ser316 to Gly325 narrows the entrance of the catalytic pocket. This 10aa-long loop of Af OYE1 spans the entrance of the catalytic pocket and bumps close to the entrance of the catalytic pocket, resulting in a narrow entrance of the catalytic center (Fig. S9 A). However, the corresponding loop in Ca OYE (Ser265 to Try280), An OYE8 (Ser308 to Phe323), Bf OYE4 (Ser324 to Tyr339), YqjM (His 255 to Pro262), Ts OYE (Ser260 to Val277), and Gk OYE (Ser250 to Gln265) was away from the entrance of the catalytic pocket (Fig. S9 A), resulting in an open catalytic pocket (Robescu et al. 2022b , 2021 ). Meanwhile, similar to the residue Arg325 of Ca OYE and Phe368 of Rm ER, the “finger” residue Trp399 of Af OYE1, which is located in the C-terminal α helix of each monomer and points to the FMN flavin group, and together with the cap region (residues Ala132 to Thr182) may interact with NADH/NADPH and adjust the selectivity of FMN cofactor (Opperman 2017 ). The functional analysis of the conserved catalytic residues His211, His214, and Tyr216 of Af OYE1 showed that all three residues were important to its catalytic activity. However, the mutagenesis of those three residues did not eliminate their catalytic activity. The histidine pair has been considered as the binding motifs of OYE homologs, such as His164 and His167 in YqjM (Shi et al. 2020 ), His172 and His175 in Crs (Opperman et al. 2010 ), and His182 and Asn185 in “thermophilic-like” OYE homologs (Riedel et al. 2015 ). The Af OYE1 variants H211L and H214L conserved 30% catalytic activity of the wild type, which showed that those two histidine residues played an important role in the substrate binding. Meanwhile, the variants H211L and H214L still exhibited activity, which indicated that other residues in the catalytic center of Af OYE1 might enable the enzyme to bind the substrate when the histidine was replaced. The replacement of Tyr216 by phenylalanine resulted in a dramatic decrease in catalytic activity, indicating Tyr216 acts as a proton donor in the Af OYE1 (Kitzing et al. 2005 ). Thus, a common catalytic mechanism can be anticipated for Af OYE1 and other OYEs. On the other hand, the variant Y216F could still reduce 2-cyclopentenone into 2-cyclopentanone, which suggested that the proton transfer process could be compensated. Previously, it has also been shown that the replacement of Tyr177 in CrS resulted in the variant only harboring 20% of the catalytic efficiency (Opperman et al. 2010 ). In summary, a new thermophilic-like OYE Af OYE1 was identified from A. flavus . It could reduce a spectrum of activated alkenes, had an optimal temperature of 45 °C, and exhibited high reduction activity in a wide pH range (pH 5.5–8.0). Unlike most of the known OYEs, the Af OYE1-catalyzed asymmetric reduction had ( S )-selective with excellent enantioselectivity, which expands the toolbox of thermostable ( S )-selective OYEs. The crystal structure of Af OYE1 revealed that it had a special catalytic cavity, which may be relevant to its enantioselectivity. Our results indicate that fungi are good sources for the identification of new OYEs. We believe the newly identified Af OYE1 will give an alternative for asymmetric alkene hydrogenation in industrial processes.
Abstract Old yellow enzymes (OYEs) have been proven as powerful biocatalysts for the asymmetric reduction of activated alkenes. Fungi appear to be valuable sources of OYEs, but most of the fungal OYEs are unexplored. To expand the OYEs toolbox, a new thermophilic-like OYE ( Af OYE1) was identified from Aspergillus flavus strain NRRL3357. The thermal stability analysis showed that the T 1/2 of Af OYE1 was 60 °C, and it had the optimal temperature at 45 °C. Moreover, Af OYE1 exhibited high reduction activity in a wide pH range (pH 5.5–8.0). Af OYE1 could accept cyclic enones, acrylamide, nitroalkenes, and α, β-unsaturated aldehydes as substrates and had excellent enantioselectivity toward prochiral alkenes (> 99% ee). Interestingly, an unexpected ( S )-stereoselectivity bioreduction toward 2-methylcyclohexenone was observed. The further crystal structure of Af OYE1 revealed that the “cap” region from Ala132 to Thr182, the loop of Ser316 to Gly325, α short helix of Arg371 to Gln375, and the C-terminal “finger” structure endow the catalytic cavity of Af OYE1 quite deep and narrow, and flavin mononucleotide (FMN) heavily buried at the bottom of the active site tunnel. Furthermore, the catalytic mechanism of Af OYE1 was also investigated, and the results confirmed that the residues His211, His214, and Tyr216 compose its catalytic triad. This newly identified thermophilic-like OYE would thus be valuable for asymmetric alkene hydrogenation in industrial processes. Key points A new thermophilic-like OYE AfOYE1 was identified from Aspergillus flavus, and the T 1/2 of AfOYE1 was 60 °C AfOYE1 catalyzed the reduction of 2-methylcyclohexenone with (S)-stereoselectivity The crystal structure of AfOYE1 was revealedv Supplementary Information The online version contains supplementary material available at 10.1007/s00253-023-12963-w. Keywords
Supplementary Information Below is the link to the electronic supplementary material.
Acknowledgements We thank SSRF 18U1 beamline scientists for their help with data collection and the NMR center in the College of Science at Henan Agricultural University for taking NMR spectra. We acknowledge English editorial assistance by Maria Ajmal (Henan Agricultural University, China). Author contribution NL, YW, and YM carried out the experimental work and drafted the manuscript; YL, SZ, SW, and PM analyzed the data and revised the manuscript; NL, YH, and HL designed the experiments and revised the manuscript. All authors have read and approved the final manuscript. Funding This study was funded by the National Natural Science Foundation of China (Nos. 32171472 and 31900876), the Key Scientific Research Projects for Higher Education of Henan Province (No. 22A180013), the Key Scientific and Technological Project of Henan Province (No. 232102311151), and High-level Talent Scientific Research Startup Fund Program of Henan University of Technology (2019BS020). Data availability All data generated or analyzed during this study are included in this published article (and its supplementary information files). Declarations Ethics approval This article does not contain any studies with human participants or animals performed by any of the authors. Conflict of interest The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:01
Appl Microbiol Biotechnol. 2024 Jan 13; 108(1):1-11
oa_package/54/2b/PMC10787880.tar.gz
PMC10787881
0
Introduction The model yeast species Saccharomyces cerevisiae is used in many fundamental and applied research applications, such as production of many industrially relevant compounds (Krivoruchko and Nielsen 2015 ) and as biosensors to various substances, for example, heavy metals or estrogens (Martin-Yken 2020 ). The compounds produced in yeast include D-lactic acid (DLA) (Baek et al. 2016 ), which is used for production of stereocomplex type poly-lactic acid, a promising biodegradable polymer (de Albuquerque et al. 2021 ). Genetically engineered yeast strains produce up to 112 g/L (1.24 M) of DLA in neutralizing conditions or over 53.2 g/L (0.59 M) of this substance without neutralizing agents (Ishida et al. 2006 , 2011 ; Baek et al. 2016 ; Yamada et al. 2017 ; Mitsui et al. 2020 ). Efficient sensing systems for DLA would also be relevant in medicine, as D-lactic acidosis is a rare but serious neurologic condition specific to individuals with short bowel syndrome (Kowlgi and Chhabra 2015 ; Petersen 2005 ). However, a systematic evaluation of yeast cell response is scarce not only for DLA but even for its incomparably more common enantiomer L-lactic acid (LLA). The response to LLA has been studied in several yeast strains on the levels of viability or growth rate reduction, as well as transcription. Specifically, it was found that lactic acid (LA) in industrially relevant concentrations of 90 or even 280 mM LA (presumably LLA) had a rather limited effect on the metabolism of S. cerevisiae in thermostat cultures but did affect the energy status of the cell by provoking a reduction in the ATP content (Thomsson and Larsson 2006 ). Another study estimated the minimal inhibitory concentration of LA (presumably LLA) as 2.5% w/v (278 mM), while 0.2% began to stress the cells (Narendranath et al. 2001 ). There are also two studies on the transcriptional effect of LLA. One of them dealt with LLA in comparison with acetic and hydrochloric acids with DNA microarrays in shake flask cultures. The authors found that these organic acids triggered relatively similar gene expression perturbations and affected cell wall and metal metabolism; the latter was intermediated by the Atp1p transcription factor (Kawahata et al. 2006 ). Another study, using chemostat cultures, also found iron metabolism remodeling; it was very pronounced at pH 5 and 500 mM LLA and much less severe at pH 3 and 900 mM LLA (Abbott et al. 2008 ). To the best of our knowledge, there is no published data on the effect of DLA on yeast transcriptome or proteome. In this study, we evaluated the transcriptional response of S. cerevisiae to varying concentrations of DLA (from 0.05 mM to 45 mM) and 45 mM of LLA in order to check if this model species possessed any promoters that would quantitatively respond to DLA but not LLA and could thus be promising for designing a yeast-based stereo-specific biosensor to lactic acid. Such transcriptome-based approach has already been successfully applied to find a 1-butanol sensing promoter in S. cerevisiae (Shi et al. 2017 ). In addition, we aimed at enriching the data on transcriptional response to DLA in comparison to LLA. We found that the concentrations of DLA of 0.05 or 0.5 mM did not trigger any changes in gene expression compared to the control samples. The genes activated in response to 5 mM DLA were enriched in those controlling cell wall organization, while the genes upregulated upon 45 mM DLA treatment included several genes functioning in lactate metabolism and iron uptake. Finally, the genes responding to LLA contained many genes known to respond to this and other weak acids.
Materials and methods Yeast cultivation The yeast strain used for this work was BY4742 ( MATα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0 ; Baker Brachmann et al. 1998 ). For the experimental exposures, overnight suspension cultures were inoculated from an independent BY4742 colony on solid YPD medium (1% yeast extract, 2% peptone, 2% D-glucose, and 2% agar) and grown in synthetic medium containing 0.67% (w/v) yeast nitrogen base, 2% (w/v) glucose, 150 mM NaCl, 20 mg/L L-histidine HCl, 100 mg/L L-leucine, 30 mg/L L-lysine HCl, and 20 mg/L uracil at 28 °C with orbital shaking (120 rpm). The addition of NaCl had the purpose of mimicking the conditions in the blood plasma for further use in sensor applications. The next day, approximately 1.5 optical density units of each culture were collected by centrifugation (1700 × g for 5 min at room temperature) and resuspended in 3 ml of the same media (control) or media with lactic acid. OD600 was recorded at the beginning of the exposure and in 3–4 h. After this time, the experimental cultures were collected by centrifugation and frozen at -80 °C for RNA extraction. The relative growth rate was calculated as the difference between logarithmic (base 2) final OD600 and initial OD600 divided over incubation time in hours. This experiment was performed in total seven times with independent suspension cultures with 0.05, 0.5, 5, or 45 mM DLA, or 45 mM LLA and a control medium; three of these replicates were used for the RNA sequencing. Moreover, similar exposures were performed with eight independent suspension cultures with 45 mM DLA, 45 mM sodium D-lactate (DLS), and a control medium; five of these replicates were used for qPCR testing. RNA extraction, sequencing, and quantitative PCR (qPCR) RNA extraction, library construction, and sequencing were performed by the CeGaT company (Tübingen, Germany). RNA isolation was performed with the RNeasy kit (Qiagen, Hilden, Germany) according to manual (RNeasy Mini Handbook) with slight modifications. Cells were homogenized by mechanical disruption. After the addition of RLT buffer (Qiagen, Hilden, Germany) and glass beads, the samples were vortexed three times for 3 min. After each vortexing step, the samples were cooled on ice. Then, the lysate was centrifuged for 2 min at maximum speed, and the supernatant was transferred to new microcentrifuge tubes, combined with the same volume of 70% ethanol and mixed by pipetting. Sequencing libraries were prepared with the TruSeq Stranded mRNA kit (Illumina Inc., CA, USA) and sequenced at 2 × 100 bp with a NovaSeq 6000 (Illumina Inc., USA). Demultiplexing of the sequencing reads was performed with Illumina bcl2fastq v2.20, and adapters were trimmed with Skewer v 0.2.2 (Jiang et al. 2014 ). For each sample, between 2.8 and 7.4 Gb were sequenced. RNA extraction for qPCR-based gene expression analysis was performed with the RNASwift method according to the original protocol (Nwokeoji et al. 2016 ) using GeneJET spin columns (Thermo Scientific, Waltham, MA, USA) and buffers provided with the RNeasy kit (Qiagen, Hilden, Germany) at the last step. RNA purification was performed according to the recommendation of the buffer manufacturer. Then, RNA was treated with RapidOut DNA removal kit (Thermo Scientific, Waltham, MA, USA) to remove residual genomic DNA. Then, RNA concentration was measured with the Nano-300 (Allsheng, Hangzhou, China) micro-spectrophotometer, and approximately 60–70 ng of DNA-free RNA was used for reverse transcription, which was performed with RevertAid reverse transcriptase and the corresponding buffer (Thermo Scientific, Waltham, MA, USA), RiboLock RNase inhibitor (Thermo Scientific, Waltham, MA, USA), dNTPs and Oligo(dT)18 primers (Thermo Scientific, Waltham, MA, USA) according to the recommendation of the enzyme manufacturer. Then, 1 μL of the resulting cDNA of the resulting solution was used for 10-μL qPCR. The amplification was performed using a StepOne Plus instrument (Thermo Scientific, Massachusetts, USA) with the 5X qPCRmix-HS SYBR Hi-Rox (Evrogen, Moscow, Russia) and primers (5 pmol each) specific for the following genes: ACT1 and CDC19 used as reference genes; AQR1 ; DLD3 ; FIT2 ; and YPS3 . Primer sequences (Supplemental Table S1 ) for ACT1 and CDC19 were taken from the work by Cankorur-Cetinkaya et al. ( 2012 ); the other primer pairs were designed with NCBI Primer Blast (Ye et al. 2012 ). Amplification efficiency was tested for each primer pair with serial dilutions of a control cDNA sample and lied in the range of 88–100% (Supplemental Table S1 ). Data analysis and availability Quality control of raw data was performed with FastQC v0.11.9 and summarized with MultiQC v1.13 (Ewels et al. 2016 ). The R64-1–1 release of S. cerevisiae strain S288C genome (Engel et al. 2014 ) was downloaded from Ensembl (Cunningham et al. 2022 ) release 108 and used as a reference. The reads were aligned to the genome with hisat2 (Kim et al. 2019 ) v2.2.1, sorted with samtools (Li et al. 2009 ) v1.9 and quantified with featureCounts (Liao et al. 2014 ) from subread v2.0.4. The resulting count table was further processed with the DESeq2 (Love et al. 2014 ) v1.34.0 for R (R Core Team 2022 ) v4.1.2 to compare expression levels. The figures were prepared using the ggplot2 (Wickham 2016 ) v3.4.2, enhancedVolcano (Blighe et al. 2023 ) v1.12.0 and DEGReport (Pantano et al. 2023 ) v1.30.3 packages for R. The data from the Abbott et al. ( 2008 ) manuscript, which were used for comparison, were downloaded from the NCBI GEO database (accession number GSE10066) with the script generated by the GEO2R service (Edgar et al. 2002 ), which utilizes the GEOquery (Davis and Meltzer 2007 ) 2.62.2, limma (Ritchie et al. 2015 ) 3.50.3, and DESeq2 packages for R. Gene ontology (GO) term and publication enrichment analyses were performed with YeastMine (Balakrishnan et al. 2012 ; Cherry et al. 2012 ) using the database of 1 Apr 2023. All the code used is available at GitHub (Drozdova 2023 ). The raw and processed RNA sequencing data are also available from the NCBI GEO repository under the accession number GSE231937.
Results Overview of transcriptional response to lactic acid enantiomers The performed analysis revealed differentially expressed genes (hereafter DEGs; absolute log2 fold change > 1 and adjusted p -value < 0.05) only in the case of the two highest DLA concentrations (5 mM and 45 mM), as well as in the case of 45 mM LLA (Fig. 1 a–c). The concentrations of 0.5 mM DLA and below did not produce any significant transcriptional response (Fig. 1 d, e). Overall, the presence/absence of DEGs correlated with the growth inhibition: whenever growth was inhibited, we recorded differential expression (Supplemental Fig. S1 ; Supplemental Table S2 ). Finally, there were 10 genes that were differentially expressed between the maximal concentration of DLA and the same concentration of LLA (see below in the section “Transcriptional response to LLA and search for DLA-specific genes”). Furthermore, we functionally characterized DEG lists with gene ontology terms using YeastMine (Table 1 ). We found that the genes upregulated in response to 45 mM were enriched with those participating in lactate biosynthesis and metabolism (these genes will be characterized in detail below) and siderophore transport. The genes upregulated in response to a lower concentration of DLA (5 mM) were connected to the cell wall, while those downregulated in these conditions contained three genes regulating leucine biosynthesis. In the case of 45 mM LLA, we only found one very general enriched GO term for downregulated genes, generation of precursor metabolites, and energy. Transcriptional response to DLA In the case of the highest concentration of DLA, we found significant enrichment of two GO terms connected to lactate (Table 1 ). The DEGs annotated with the terms “lactate metabolic process” (GO:0006089) and “lactate biosynthetic process” (GO:0019249) largely overlapped and contained DLD1 (YDL174C), DLD3 (YEL071W), SNO4 (YMR322C), and HSP32 (YPL280W). The former two genes indeed encode D-lactate dehydrogenases, mitochondrial Dld1 and cytoplasmic Dld3 (Pallotta 2012 ); according to the literature, Dld3 can also oxidize D-2-hydroxyglutarate to α-ketoglutarate (Becker-Kettern et al. 2016 ). The latter two are specific small chaperones (Gong et al. 2009 ). Unfortunately, none of these four genes was both DLA-specific and quantitatively responding to DLA (Fig. 2 a). In the case of 5 mM DLA, the main groups of upregulated genes were those associated with cell wall biogenesis (Table 1 ; Supplemental Table S4 ). This effect is not probably specific for DLA, as a similar effect was found for different organic acids (Kawahata et al. 2006 ). These genes also reacted to 45 mM DLA, even though less strongly (Supplemental Fig. S2 ). Overall, the transcriptional responses to 5 mM DLA and 45 mM DLA correlated quite well (Fig. 2 b, c). Transcriptional response to LLA and search for DLA-specific genes We found fewer DEGs in response to LLA in comparison to DLA; moreover, there were no overrepresented GO terms for upregulated genes and only a rather vague term “generation of precursor metabolites and energy” in the case of downregulated genes. However, the lists of genes differentially expressed in response to LLA were associated with many (over 20) publications enriched in some of these genes (Supplemental Table S4 ). Impressively, four of the six manuscripts enriched in LLA-upregulated genes dealt with the Haa1 transcription factor, which was indeed shown to mediate the adaptation of yeast cells to lactic and other weak acids (Fernandes et al. 2005 ; Mira et al. 2010 , 2011 ; Sugiyama et al. 2014 ). Generally, the changes triggered by LLA correlated well to the changes observed in response to the same concentration of DLA (Fig. 3 a, b). In order to reveal if there were any genes that quantitatively responded to DLA and did not respond to LLA, we performed a clustering analysis of 214 genes that were differentially expressed in at least one condition. Within the obtained six clusters, none had the desired pattern for a DLA sensor, i.e., monotonous increase or decrease in response to DLA and absence or very slight response to LLA (Fig. 3 c). It is worth mentioning that the first cluster featured genes which were seemingly affected more by the high concentrations of DLA than by LLA, but in fact, the changes were very subtle (Supplemental Fig. S3 ). In addition, we analyzed the genes differentially expressed between 45 mM DLA and 45 mM LLA (Fig. 1 f). There were ten such genes, YHL028W ( WSC4 ), YLR054C ( OSW2 ), YLR121C ( YPS3 ), YGR189C ( CRH1 ), YGR146C ( ECL1 ), YGL255W ( ZRT1 ), YHR209W ( CRG1 ), YLR205C ( HMX1 ), YKR091W ( SRL3 ), and YCR005C ( CIT2 ). Some of these genes ( WSC4 , YPS3 , CHR1 , and OSW2 ) regulate cell wall assembly. Wsc4, Yps3, and Crh1 have functions in maintaining cell wall integrity (Verna et al. 1997 ; Krysan et al. 2005 ; Cabib et al. 2007 ). Osw2 is a protein of unknown function, which is putatively involved in spore wall assembly (Coluccio et al. 2004 ). Several genes ( ZRT1 , ECL1 , and HMX1 ) are implicated in metal transport. Zrt1 is a zinc transporter (Zhao and Eide 1996 ). Ecl1 is a protein of unknown function upregulated by overexpression of the other iron deprivation-responding transcription factor, Aft2 (Rutherford et al. 2003 ). HMX1 encodes a heme oxygenase, and expression of this gene is regulated by the iron deprivation-responding transcription factor Aft1 (Protchenko and Philpott 2003 ). Finally, the link between some of the genes and LA stress was unclear. Crg1 is a small molecule methyltransferase regulating lipid homeostasis in response to a drug cantharidin (Lissina et al. 2011 ), Cit2 is a citrate synthase (Kim et al. 1986 ), and Srl3 (or Whi7) participates in cell cycle regulation (Gomar-Alba et al. 2017 ). Of these genes, five reacted to both 5 mM and 45 mM DLA but not to 45 mM LLA (Fig. 3 d). These could be candidates for qualitative sensors for DLA but required further exploration. Neutralization compensates for the effect of the DLA on growth rate and transcription of selected genes During all previous analyses, we found several groups of genes that reacted to one (45 mM) or two (5 and 45 mM) concentrations of DLA, and we also found that these treatments slowed down yeast growth (Supplemental Fig. S1 ). In order to check if the slow growth was caused by the lower pH values, we performed the same treatment but with 45 mM sodium D-lactate (DLS) obtained with addition of the same amount of NaOH (the concentration of additional NaCl in the media was adjusted to sum up to 150 mM) and indeed observed compensation of the growth defect (Fig. 4 a). Thus, the observed slow growth in yeast treated with 45 mM DLA is explained by pH shift and not by the influence of the D-lactate ion. Moreover, we checked if the expression of several genes that we found to be DLA-responsive changed in response to DLS. For this analysis, we chose the AQR1 , DLD3 , and FIT2 genes, which were activated to 45 mM DLA but did not respond to 5 mM DLA, as well as the YPS3 gene that responded to both concentrations of DLA. These genes belong to different functional groups. AQR1 (YNL065W) encodes a membrane protein from the major facilitator superfamily, which provides the cells with resistance to short-chain monocarboxylic acids (Tenreiro et al. 2002 ). FI T2 (YOR382W) is a cell wall mannoprotein involved in the siderophore transport (Protchenko et al. 2001 ), and we chose it as a representative of a larger group of iron uptake-related proteins (Table 1 ). DLD3 (YEL071W) codes for a protein with D-lactate dehydrogenase activity in vitro, but there is evidence that in vivo it contributes to D-lactate synthesis (Chelstowska et al. 1999 ; Becker-Kettern et al. 2016 ). Finally, YPS3 (YLR121C) encodes an aspartic protease required for cell wall integrity (Olsen et al. 1999 ), and we chose it as a representative of the cell wall integrity-related genes and also because it had strong changes in expression in response to both DLA concentrations but not to LLA (Fig. 3 d) and thus could act as qualitative DLA sensor. However, we found that the relative expression levels of all of these genes were very similar in the control samples and in those treated with 45 mM DLS. Taken together, our data suggest that the expression changes in response to DLA were mostly triggered by the change in the pH value.
Discussion In this study, we explored the transcriptional response of S. cerevisiae to LA enantiomers. One of the goals of this work was to find genes that would specifically and quantitatively respond to DLA but not to LLA. We did not find such genes. In general, we found that the response to high concentration of DLA and LLA was quite similar and even more pronounced to DLA than to LLA. It is possible that the reason for this difference is that LLA is metabolized faster. There was no response to the concentrations of DLA of 0.5 mM and below. It is possible that higher DLA concentrations (> 50 mM) or longer exposures (about a day) could have made the effect more pronounced and reveal differentially expressed genes. However, higher concentrations would be outside the range of DLA concentrations typically found in biological fluids. While normal levels of LLA in mammal blood plasma are about 1–2 mM, the levels of DLA are about two orders of magnitude lower, 0.01–0.07 mM (Ewaschuk et al. 2005 ). In general, the levels of < 0.2 mM are considered normal (Zhang et al. 2003 ). The most well-known condition leading to D-lactic acidosis in human is short bowel syndrome, during which plasma DLA levels may reach millimolar concentrations (Zhang et al. 2003 ; Yilmaz et al. 2018 ). Similar symptoms in ruminants appear upon elevated carbohydrate level in their diet (Lorenz and Gentile 2014 ) and may lead to DLA plasma levels as high as about 25 mM (Ewaschuk et al. 2005 ). Thus, an ideal biosensor for DLA should be sensitive at 0.1–5 mM concentration range. Similarly, the need to use longer exposures of sensor yeast would also hinder its usage in biosensor applications. So, we find it unlikely that a quantitative yeast sensor to DLA may be constructed based on the native yeast transcriptional networks, but the possibility of integrating a heterologous cassette remains open. Such a cassette has been described for Pseudomonas , but its sensitivity starts from about 20 mM DLA (Singh et al. 2019 ), which is also far from ideal for monitoring DLA levels in biological fluids. Importantly, we present the first dataset on yeast transcriptional response to DLA. Generally, we found that if the particular concentration did not inhibit growth, we did not see a transcriptional response either. This result is very similar to the study, in which the authors compared the transcriptome-wide responses to different alcohols in search for 1-butanol sensor and found that the samples treated with ethanol, which did not cause significant growth inhibition, clustered with the control samples, while samples treated with 1-butanol and 1-propanol, which were much more toxic, clustered separately (Shi et al. 2017 ). Intriguingly, the response to 5 mM and 45 mM was seemingly different if judging by enriched GO terms (5 mM DLA caused overexpression of cell wall-related genes, while 45 mM led to increased expression of lactate metabolism and siderophore transport genes), but the transcriptional profiles in response to 5 mM and 45 mM DLA were highly correlated (Fig. 2 c), and more than half of the DEGs were shared between the two comparisons (Fig. 1 g, h). In general, the groups of cell-wall related genes and genes of iron/siderophore uptake have already been described to respond to weak acid stress (Kawahata et al. 2006 ; Abbott et al. 2007 , 2008 ), so our findings fully corroborate the previously published data but at the same time suggest that this response might be pH-dependent rather than specific for the lactic acid. However, comparison of four weak acids, benzoate, sorbate, acetate, and propionate, showed that the transcriptional responses were largely specific (Abbott et al. 2007 ), proving that the response to pH was not the only reason for gene expression changes. We have checked this hypothesis for the FIT2 gene, which codes for cell wall mannoprotein involved in the siderophore transport (Protchenko et al. 2001 ) as a representative of the iron/siderophore uptake functional group and YPS3 , the gene encoding an aspartic protease required for cell wall integrity (Olsen et al. 1999 ), as a representative of the cell wall genes group. Both genes did not respond to D-lactate treatment if pH was compensated (Fig. 4 b). Moreover, in this experiment we also measured the expression levels of the AQR1 gene. It encodes a membrane protein from the major facilitator superfamily, which provides the cells with resistance to short-chain monocarboxylic acids (Tenreiro et al. 2002 ). We found that AQR1 also only responded to DLA at low pH. Interestingly, the authors of the original manuscript noted that the expression of this gene was not stimulated by weak acid stress (Tenreiro et al. 2002 ). We found that it was upregulated in response to 45 mM DLA, while Abbott et al. ( 2007 ) found it as a common gene downregulated in response to the four weak organic acids they tested. Aqr1 has not been shown to have a role in lactic acid transport, but it acts as a lactic acid exporter and has been shown to be important for yeast co-cultivation with lactic acid bacteria (Velasco et al. 2004 ; Kapetanakis et al. 2021 ). While the transcriptional response to DLA has not been explored before, there are studies on transcriptional changes in response to LLA (Kawahata et al. 2006 ; Abbott et al. 2008 ). We have compared the changes in the genes differentially expressed in response to 45 mM LLA according to our data and each of the two studies and found substantial positive correlation (Supplemental Fig. S4 ), even though all experimental designs were different. In the case of the work by Kawahata et al. ( 2006 ), experimental exposures were performed in shake flask cultures with 0.3% LLA (about 33 mM), and the S288C strain (parental to BY472) was used. Two experimental designs were used. First, acid shock was performed by pre-growing the cultures to the optical density at 660 nm (OD660) of 1.0 and exposing them to LLA for 30 min. The second design, acid adaptation, involved diluting overnight cultures to the OD660 = 0.1 in the media with LLA and growing until OD660 reached 1. The number of overlapping DEGs was quite low in both cases (five genes), but the changes in the transcription of these genes were mostly similar to our results (Supplemental Fig. S4 a, b). In the study by Abbott et al. ( 2008 ), chemostat cultures of the CEN.PK 113-7D strain (not closely related to S288C and BY4742) were subjected to quite high concentrations of LLA, namely, 500 mM LLA at pH 3 and 900 mM LLA at pH 5. The lists of DEGs shared in our results and these data were larger (over 30 genes in each case), and the correlation of our 45 mM-LLA exposure was much higher with 500 mM LLA than with 900 mM LLA. We were particularly interested in the D-lactate metabolism genes DLD1 and DLD3 . Originally, both of these genes were shown to code for a mitochondrial and cytoplasmic D-lactate dehydrogenases, respectively (Lodi and Ferrero 1993 ; Chelstowska et al. 1999 ). According to our results (Fig. 2 a; Supplemental Table S3 ), DLD1 was upregulated in response to 5 mM DLA (fold change = 1.92 and adjusted p = 0.04), as well as in response to 45 mM DLA (fold change = 2.76 and adjusted p = 0.0001) and had a similar trend in response to 45 mM LLA, even though the difference did not reach statistical significance (fold change = 1.85 and adjusted p = 0.06). Interestingly, Abbott et al. ( 2008 ) also found upregulation of DLD1 to the highest LLA concentration used in their experimental design, 900 mM (fold change = 4 and adjusted p = 0.0002). This non-stereo-specific regulation could be interesting to explore further. While the Dld1 enzyme is the major D-lactate dehydrogenase, Dld3 is a minor D-lactate dehydrogenase and mostly acts as a transhydrogenase coupling D-2-hydroxyglutarate degradation to DLA synthesis (Lodi and Ferrero 1993 ; Chestowska et al. 1999; Becker-Kettern et al. 2016 ). In our experiment, DLD3 was only upregulated in response to 45 mM DLA but not 5 mM DLA or lower concentrations and did not react to DLS, also corroborating the idea of its very minor role as a D-lactate dehygrogenase. It is possible that this protein acts at high DLA concentrations to prevent cell damage by low pH. In general, our data enrich our understanding of the yeast transcriptome-wide response to LLA and provide the first description of the response to DLA. We found that even though the response to different stereoisomers of lactic acid had quite significant similarities to the response to other weak acids tested previously and largely dependent on pH, there are large differences between DLA and LLA responses, which probably reflect the difference in their role in yeast biology. The role of pH in the DLA response highlights the importance of controlling and optimizing lactic acid production in yeast under neutralizing and non-neutralizing conditions separately, which is also corroborated by a recent work on a recombinant yeast strain with improved lactic acid tolerance and lactic acid yield under non-neutralizing conditions (Yamada et al. 2021 ).
Abstract The model yeast, Saccharomyces cerevisiae , is a popular object for both fundamental and applied research, including the development of biosensors and industrial production of pharmaceutical compounds. However, despite multiple studies exploring S. cerevisiae transcriptional response to various substances, this response is unknown for some substances produced in yeast, such as D-lactic acid (DLA). Here, we explore the transcriptional response of the BY4742 strain to a wide range of DLA concentrations (from 0.05 to 45 mM), and compare it to the response to 45 mM L-lactic acid (LLA). We recorded a response to 5 and 45 mM DLA (125 and 113 differentially expressed genes (DEGs), respectively; > 50% shared) and a less pronounced response to 45 mM LLA (63 DEGs; > 30% shared with at least one DLA treatment). Our data did not reveal natural yeast promoters quantitatively sensing DLA but provide the first description of the transcriptome-wide response to DLA and enrich our understanding of the LLA response. Some DLA-activated genes were indeed related to lactate metabolism, as well as iron uptake and cell wall structure. Additional analyses showed that at least some of these genes were activated only by acidic form of DLA but not its salt, revealing the role of pH. The list of LLA-responsive genes was similar to those published previously and also included iron uptake and cell wall genes, as well as genes responding to other weak acids. These data might be instrumental for optimization of lactate production in yeast and yeast co-cultivation with lactic acid bacteria. Key points • We present the first dataset on yeast transcriptional response to DLA. • Differential gene expression was correlated with yeast growth inhibition. • The transcriptome response to DLA was richer in comparison to LLA. Supplementary Information The online version contains supplementary material available at 10.1007/s00253-023-12863-z. Keywords
Supplementary Information Below is the link to the electronic supplementary material.
Author contribution PD, AG, MT, and EB designed the research. PD, AS, AV, and EI performed the experiments. PD, AG, AV, and EZ analyzed the data and prepared text and figures. MT and EB supervised the project. All authors read and approved the final manuscript. Funding The work was supported by the Russian Science Foundation (interdisciplinary project 20–64-47011 performed in association with the Petrozavodsk State University, Petrozavodsk, Russia; link to information about the project: https://rscf.ru/en/project/20-64-47011/ ). Data availability The datasets generated during and analyzed during the current study are available in the GitHub repository, https://github.com/drozdovapb/S_cerevisiae_lactate_transcriptome and the NCBI GEO repository under the accession number GSE231937. Declarations Ethics approval This article does not contain any studies with human participants or animals performed by any of the authors. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:01
Appl Microbiol Biotechnol. 2024 Jan 13; 108(1):1-12
oa_package/b8/86/PMC10787881.tar.gz
PMC10787882
37861889
Introduction A substantial proportion of cancers are associated with pathogenic variants (PV) in hereditary cancer genes [ 1 , 2 ], including an estimated 10% of breast, 10% of colon, and 20% of ovarian cancers. Approximately, 1 in 300 to 500 people in the population will carry a PV in either BRCA1 or BRCA2 ( BRCA ; the genes most commonly associated with hereditary breast and ovarian cancer (HBOC)) [ 3 – 6 ], and 1 in 370 individuals will carry a pathogenic variant (PV) in one of the Lynch syndrome genes (the genes most commonly associated with hereditary colorectal cancer) [ 7 ]. Early identification of a PV in a cancer susceptibility gene provides individuals with the opportunity for enhanced surveillance and risk-reducing interventions which can significantly reduce the morbidity and mortality of these cancers [ 8 – 13 ]. Identification of carrier status also provides the opportunity for cascade testing in relatives [ 14 – 16 ]. Genetic testing for cancer susceptibility genes is currently offered predominantly to individuals who are considered high-risk for either BRCA or Lynch syndrome based on clinical and family history criteria. However, only around half of the carriers of PVs in BRCA and Lynch syndrome genes will meet clinical and family history criteria for genetic testing [ 5 , 17 , 18 ]. Additionally, current approaches to testing for hereditary cancer genes are associated with inequities in referral and access to testing [ 19 , 20 ]. An alternative approach is population-based hereditary cancer genetic testing which may provide a more clinically effective strategy for early identification of high-risk individuals [ 5 , 21 ]. To date, population-based testing has been primarily studied in the context of BRCA testing in the Ashkenazi Jewish population [ 22 , 23 ], and data are limited in the general population, especially among under-represented racial and ethnic groups. The objective of this study was to determine the yield of hereditary cancer gene PVs among unselected diverse women attending breast imaging centers, as a potential strategy for more complete identification of high-risk individuals who could benefit from enhanced surveillance and/or risk reduction interventions.
Materials and methods Study population This retrospective cohort study included unselected female patients who were offered and underwent genetic testing at the time of breast imaging at three imaging centers (Memorial MRI and Diagnostics, Texas) from November 2020 through March 2022. All patients arriving at the imaging centers were given a written flier with an invitation to undergo genetic testing for a panel of hereditary cancer genes. Patients were also offered the option of an online genetic information session with a board-certified genetic counselor prior to testing. The lead clinician investigator (DM) served as the ordering clinician for the testing at all three centers. A limited number of providers using the imaging centers opted out of having their patients participate. Only patients (including those with a previous history of breast cancer) undergoing routine breast imaging, either by mammogram or ultrasound, were included. Patients undergoing imaging for newly diagnosed breast cancer were excluded. Clinical, demographic, and family cancer history information were ascertained through test requisition forms and family cancer history questionnaires completed by the patients. Patient questionnaires included clinical questions needed for Tyrer–Cuzick breast cancer risk assessment. Race/ethnicity (ancestry) was self-reported. National Comprehensive Cancer Network (NCCN) guidelines for Genetic/Familial High-Risk Assessment: Breast, Ovarian, and Pancreatic Cancer [ 24 ] and Lynch Syndrome (LS) were reviewed to determine if genetic testing would have been guideline indicated for a particular patient [ 25 ]. For the patients unaffected with breast cancer who completed the clinical portion of their questionnaire for breast cancer risk assessment, a Tyrer–Cuzick breast cancer risk 5 years and lifetime risk score was calculated [ 26 ]. This score was only reported back to patients if their genetic testing (described below) was negative/uninformative for PVs associated with increased breast cancer risk. It was not available if enough information about personal/family history was not provided to compute a Tyrer–Cuzick score or the patient had a personal history of breast cancer. If the patients’ clinical and/or family history met NCCN guidelines for hereditary cancer testing, they were given the option to request insurance coverage for testing or self-pay. Patients who did not meet guidelines were offered the option to self-pay and financial assistance options were available to eligible patients, based on their income and family size. Genetic information sessions performed by board-certified genetic counselors (Natera, Inc.) were available to patients on a pre- and post- test basis. All patients with a PV in a breast cancer-associated predisposition gene were offered in person risk counseling by the lead clinician investigator (DM). This study was granted a waiver of consent process under 45 CFR 46.116(d), a waiver of the requirement for documentation of informed consent according to 45 CFR 46.117(c)(2), and a waiver from the HIPAA Authorization Requirement according to 45 CFR 46.164.512(i) (Salus IRB, ID# 21204—01A). Hereditary cancer testing Next-generation sequencing (NGS)-based hereditary cancer risk assessment was carried out utilizing a multiplex gene panel (40 or 53 genes) testing (EmpowerTM, Natera, Inc. in collaboration with Baylor Genetics). The targeted regions of the genes associated with hereditary cancer syndromes are enriched using a capture-based method and sequenced by next-generation sequencing (NGS) using the Illumina platform. The variants detected in exons and within 20 bp of the exon/intron boundary are reported, unless otherwise specified. Read depth analysis is used to detect copy number variation (CNV) for genes. Positive sequencing results from certain genes or regions with highly homologous sequences in the genome are confirmed by gene-specific long-range PCR and Sanger sequencing. Multiplex ligation-dependent probe amplification (MLPA), PCR-based methods, and/or array comparative genomic hybridization (aCGH) are used to confirm copy number changes involving the genes in the test. All patients underwent testing for at least 25 clinically actionable genes (Table 1 ). Genes were considered clinically actionable based on the presence of established NCCN and/or peer-reviewed consensus management recommendations for enhanced surveillance, or risk-reducing interventions, and family cascade testing if a PV was detected. Clinically actionable genes were categorized as “high risk” or “moderate risk” based on their reported relative risk for cancer, relative risk of > 4, and relative risk of 2–4, respectively (Table 1 ). Testing included both HBOC and Lynch syndrome (LS) genes (NCCN HBOC, NCCN Colorectal Cancer (CRC)). Variants were classified consistent with guidelines from the American College of Medical Genetics and Genomics and the Association for Molecular Pathology as previously described [ 26 – 28 ]. Only likely pathogenic and pathogenic variants (PVs) were considered in the analysis: benign and likely benign variants, and variants of unknown significance were not considered. For the purposes of the current analysis, variants were not considered clinically actionable (i.e., had no potential impact on patient care) and were not included if they: Were in genes only associated with autosomal recessive disease association (e.g., monoallelic MUTYH carriers which may conflate estimates of pathogenic variant prevalence) or Were low penetrance variants in clinically actionable genes, such as CHEK2 c.470 T > C [p.Ile157Thr]. Analysis The sociodemographic characteristics and personal and family history of cancer of the study population were explored. Patients’ characteristics were stratified based on the presence and type of PV. The prevalence of P/LP variants was calculated. For patients with a P/LP variant, the proportion who did and did not meet NCCN guidelines for genetic testing was evaluated. We also evaluated the proportion of patients with PV who would have qualified for additional screening based on the empiric risk model (i.e., Tyrer–Cuzick score > = 20%).
Results Study population A total of 1,943 women undergoing breast imaging elected to have hereditary cancer genetic testing during the study period (Table 2 ). Median age was 66 years (range 18–89 years). Self-reported race and ethnicity were Asian 5.0% ( N = 85); Black 20% ( N = 339); White 38% ( N = 650); and Hispanic 32% ( N = 534). A personal history of cancer was documented for 7.5% ( N = 146) (Fig. 1 ), a family history of cancer in 42.3% ( N = 822), and no personal or family history of cancer in 50.2% ( N = 975). A personal history of breast or ovarian-related cancers was recorded in 4% ( N = 80), endometrial cancer in 0.8% ( N = 15), and colorectal cancer in 0.5% ( N = 10). Overall, 18.2% (354/1943) of patients met current NCCN guidelines for hereditary breast and ovarian cancer (HBOC) gene testing, 3.7% (71/1943) met NCCN guidelines for Lynch syndrome genetic testing, and 1.0% (19/1943) met both HBOC and Lynch syndrome guidelines for testing (Table 3 ). Prevalence of pathogenic variants Among 1943 patients who received genetic testing, 39 (2%) were identified as carriers of a PV in an autosomal dominant clinically actionable HBOC-related or LS gene (Table 3 ). Of the 39 PVs identified, 84.6% ( N = 33) were in HBOC-related genes which corresponds to a prevalence of 1.7% (33/1943) in the total cohort. The remaining 15.4% ( N = 6) PVs were in LS genes which corresponds to a prevalence of 0.3% (6/1943) in the total cohort. The most common PVs were in CHEK2 (10/39; 25.6%); PALB2 (8/39; 20.5%); BRCA2 (6/39; 15.4%); and PMS2 (5/39; 12.8%) (Fig. 2 ). Of the 34 PVs where race/ethnicity were known, 47% were detected among non-White patients (Table 2 ). The PV prevalence (%) was distributed across the respective ancestral groups as follows: Black 3/339 (0.89%); White 16/650 (2.4%); Asian 3/85 (3.5%), and Hispanic 9/534 (1.7%). Patients with a PV were over 50 years of age at the time of their testing in 82.1% (32/39) of cases. This is similar to the 1587/1943 (81.7%) of patients aged 51 or older who were tested in the total cohort and consistent with National screening guidelines for women at average risk [ 27 ] (Table 2 ). Guideline eligibility for genetic testing or enhanced surveillance breast cancer Only 38.5% (15/39) of PV carriers met either NCCN guidelines for HBOC testing or LS testing prior to genetic testing. Of these, 25.6% (10/39) met HBOC criteria for genetic testing with PVs in HBOC-related genes and 7.7% (3/39) had a PV in an LS gene, and 5.1% (2/39) met criteria for LS testing with PV in HBOC-related genes. Notably, 5 out of 6 patients with a BRCA2 PV and 5 out of 8 patients with a PALB2 PV did not meet criteria for HBOC testing. The frequencies of clinically actionable PVs identified in high- or moderate-risk genes in patients who either met or did not meet NCCN guidelines for inclusion of the genes are shown in Fig. 2 . Data allowing calculation of the patient’s Tyrer–Cuzick score were available for 64.1% (25/39) of patients with a PV in an autosomal dominant clinically actionable HBOC (Table 3 ). Of these, only 8% (2/25) had a Tyrer–Cuzick score ≥ 20 which would have triggered health insurance coverage for increased surveillance for breast cancer, absent in the PV finding (Table 3 ).
Discussion We found that 2% (39/1943) of women undergoing hereditary cancer gene testing as part of their clinical care at the time of breast imaging had PVs in hereditary cancer genes that had implications for cancer surveillance and clinical management. Importantly, the population undergoing testing was racially and ethnically diverse compared to previous studies [ 20 , 28 ], predominantly over the age of 50 and without a personal history of cancer (92.5%). This was similar or more diverse compared to CDC population data on race and ethnicity for females aged 55 to 74 years who were resident in Texas in 2021 (4.8% Asian, 12.5% Black, 80.7% White, and 28.1% Hispanic) [ 29 ]. Among women who had a PV in either a HBOC or LS gene, only 38.5% met NCCN guidelines for testing of either of these conditions. Of those who had a PV in an HBOC gene and had data needed for a Tyrer–Cuzick score, only 8% had an estimated lifetime score of ≥ 20%. Therefore, while testing as part of clinical care at the time of breast imaging identified some women who were eligible for hereditary cancer gene testing based on NCCN guidelines, it predominantly identified women with PVs who neither met NCCN guidelines for testing nor the Tyrer–Cuzick threshold for increased surveillance for breast cancer [ 30 ]. Consequently, without this opportunity for genetic testing, women with PVs may not have accessed risk-appropriate enhanced surveillance. Identification of individuals with PVs in HBOC and LS genes provides opportunities for cancer prevention and earlier detection [ 13 , 31 ]. For people with PVs in HBOC genes, there are recommendations and options for earlier mammography, screening breast MRI, chemoprophylaxis, and risk-reducing surgeries, such as mastectomy and bilateral salpingo-oophorectomy [ 8 , 10 , 11 , 13 ]. Knowledge of specific PVs can also impact decisions about surgery and adjuvant therapies [ 32 – 34 ]. The median age of the population in the current cohort was 66 years and previous research has suggested that HBOC gene testing in younger women would be the most cost-effective approach to testing [ 35 ]. Nonetheless, a study of the remaining lifetime risk for the subset of women who were older than 65 years in the population-based CARRIERS project [ 36 ] indicated that BRCA1 , BRCA2 , and PALB2 PVs were associated with enough breast cancer risk to warrant high-risk screening [ 37 ]. For people who have PVs in LS genes, there is also compelling evidence of the benefit of colonoscopy in reducing mortality from CRC [ 9 , 38 ]. Additionally, in women with PVs in LS genes, the lifetime risk of endometrial cancer is similar to that of CRC, which can largely be prevented by hysterectomy [ 39 ]. For women with PVs who chose to share this information with family members, it can provide more accurate risk assessment and cascade testing [ 14 – 16 ]. The yield of PVs is typically higher (~ 10%) [ 13 ] when we use the guideline criteria to screen the eligibility for genetic testing. However, this study demonstrated a meaningful yield (~ 2%) of clinically actionable PVs among participants who did not meet any guideline. Amplifying this point, 5 out of 6 BRCA2 carriers identified by this unselected approach did not meet any guideline and there is ample evidence of reduction in cancer-specific and all-cause mortality by standard of care gene-specific clinical management [ 8 , 40 , 41 ]. Thus, the universal testing approach and increasingly cost-effective genetic testing are likely to have a high impact despite a modest yield. Perhaps one of the most striking findings in this clinical cohort was the diversity of the population that accessed the testing. Later stages of cancer at the time of diagnosis and lower cancer survival rates are clearly documented in Black people in the USA [ 42 ]. There are data indicating that people of color or Hispanic ancestry are less likely to be referred for genetic counseling or testing and are less likely to take up genetic testing in the absence of a personal history of cancer [ 20 , 42 , 43 ]. As demonstrated in our study, increasing access and convenience of testing may help overcome barriers to more equitable testing [ 43 , 44 ]. Limitations of this study include that women under 40 years of age without documented increased risk are unlikely to have routine mammography and therefore, may not be represented in this cohort. Thus, BRCA carriers may be under-represented. Nonetheless, several BRCA2 carriers, the majority of which did not meet any testing guidelines, were identified representing a critical opportunity for screening and prevention. Another limitation of the current cohort is that the number of patients who declined hereditary cancer testing or who may have had genetic testing prior to this study was unknown. Given the uncertainty about the total number of women who received the invitation to have genetic testing, the potential benefit of our strategy with regard to the yield of actionable PVs in the imaging center population could be over- or underestimated. Finally, though the racial and ethnic composition of the study participants were exceptionally diverse, we do not know if there were significant differences in the uptake among the respective groups [ 45 ]. Nonetheless, (as cited above) the population who underwent testing is diverse and generally representative of the ethnic makeup of the state of Texas. Further, we do not have qualitative or quantitative data on how the patients made their decision to participate and receive testing as it was beyond the scope of our study. However, we believe that this will be important for future research in the context of population health implementation. Finally, the presence of a PV or elevated empiric risk (e.g., > 20%) does not guarantee insurance coverage, and access has been problematic across different healthcare systems and among the underinsured. Nonetheless, we believe that this report provides additional evidence supporting access to risk-appropriate care. Without granular insurance data, we note that all of the individuals in the study at the least had access to the imaging centers. Additionally, while genetic information sessions both before and after testing were available to all patients who underwent testing, formal pre-test genetic counseling was not required. Requiring pre-test genetic counseling may itself present a barrier to testing [ 46 ], so there may be a trade-off between the benchmark of full-genetic counseling and the use of abbreviated genetic information sessions to improve access to potentially life-saving information, especially in population health settings. Additional research is needed to establish what constitutes optimal pre- and post-test counseling and informed consent for patients receiving genetic testing in non-genetic/population health settings. However, in the meantime, the results of this and similar studies suggest that testing in a diverse imaging center population can extend the reach of genetic cancer risk assessment, has a clinically meaningful and actionable yield of cancer-associated PVs, and can help address inequity in access to testing and risk-appropriate screening and prevention.
Purpose Up to 10% of all breast cancers (BC) are attributed to inherited pathogenic variants (PV) in BC susceptibility genes; however, most carriers of PVs remain unidentified. Here, we sought to determine the yield of hereditary cancer gene PVs among diverse women attending breast imaging centers, who could benefit from enhanced surveillance and/or risk reduction interventions. Methods This cross-sectional retrospective cohort study included consecutive women, unselected for personal or family cancer history, who were offered genetic testing for hereditary cancer genes at the time of breast imaging at three centers (November 2020–March 2022). Results Among 1943 patients (median age: 66 years), self-reported race/ethnicity was White (34.5%), Hispanic (27.7%), African American (17.9%), Asian (4.5%), Ashkenazi Jewish (0.6%), Other (3.5%), and missing (13.0%). Thirty-nine patients (2%) were identified as carriers of a PV in an autosomal dominant clinically actionable hereditary breast and ovarian cancer (HBOC)-related or Lynch syndrome gene, most frequently, BRCA2 (6/39; 15.4%), PALB2 (8/39; 20.5%), CHEK2 (10/39; 25.6%), and PMS2 (5/39; 12.8%). Of the 34 PVs with known race/ethnicity, 47% were detected among non-White patients. Overall, 354/1,943 (18.2%) of patients met NCCN guidelines for HBOC gene testing and only 15/39 (38.5%) patients with an autosomal dominant clinically actionable PV met guidelines. Conclusion This population health approach extended the reach of genetic cancer risk assessment in a diverse population and highlighted the limits of a guideline-based approach. This may help address inequity in access to risk-appropriate screening and cancer prevention. Keywords
Acknowledgements We would like to acknowledge Sofia Hurtado, BS and Mayra Rodas, BS from Natera, Inc. for data acquisition and Urmi Sengupta, PhD, from Natera, Inc., for assistance with the development of the manuscript. Author contributions LW, VS, and JW have contributed to the study concept and design. Data acquisition, analysis, and interpretation were performed by LW, DM, MM, KH, MY, and NK. LW, DM, VS, AR, MM, KH, YS, MY, NK, and JW were involved in the administrative process. LW, VS, and JW supervised the study. Original drafting of the manuscript was performed by LW, VS, and MY. Critical revision and editing of the manuscript were performed by LW, DM, VS, AR, MM, KH, YS, MY, and JW. All authors read and approved the final version of the manuscript for submission. Funding Not applicable. Data availability This is descriptive summary data on genetic testing outcome. The specific variant level data are contributed to ClinVar by Baylor Genetics, Houston, Texas. Declarations Competing interests VS, MM, KH, YS, MY, RG, and WS are employees of Natera, Inc. with stocks or options to own stocks. LW, DM, and NK declare no conflict of interest. Ethical approval This study was granted a waiver of consent process under 45 CFR 46.116(d), a waiver of the requirement for documentation of informed consent according to 45 CFR 46.117(c)(2), and a waiver from the HIPAA Authorization Requirement according to 45 CFR 46.164.512(i) (Salus IRB, ID# 21204—01A).
CC BY
no
2024-01-15 23:42:01
Breast Cancer Res Treat. 2024 Oct 20; 203(2):365-372
oa_package/86/b6/PMC10787882.tar.gz
PMC10787883
38217724
Introduction Robotic surgery has revolutionized the treatment of several diseases given the improved clinical outcomes, such as less blood loss, shorter length of hospital stay and fewer complications compared with laparoscopy or open approach. The main limitation has been the high cost due to purchase and maintenance of the da VinciTM robotic surgical system (Intuitive Surgical, Sunnyvale, CA, USA) that represented the only alternative of the global market so far. After the Intuitive’s patent expiry, novel robotic platforms became available, including the Revo-I, the Senhance, the Versius, Avatera, Hinotori, and Hugo RAS. Some of these new systems share features differing from those of the da Vinci, both in terms of console, robotic units, and arm; thus, the introduction of novel platforms poses specific challenges to be addressed by the whole operating team; information and training about indications, setup, and outcomes are mandatory. The Versius CMR is one of those new systems that has been approved in the UK in 2018. It displays an open surgical console with hand controllers and a head-up display (HUD). Versius is a multi-modular system with independent bedside unit (BSU), one dedicated to visualization with an endoscopic camera. The HUD provides the surgeon with a three-dimensional, high-definition visualization. The system carries the potentials of versatility, due to the open console design—improving communication in the OR—and adaptability with up to 4 BSU located around the operating table with multiple allowed configurations. The hand controllers are ergonomically designed and, unlike the Da Vinci and other new systems, accommodate all functions—including camera movement and energy delivery—without pedal control [ 1 – 3 ]. After strong evidence of feasibility developed in the pre-clinical setting [ 4 ], clinical outcomes from the official “Versius Robotic Surgical System Registry” are yet to come; a low rate of conversion to alternative technique, serious adverse events, and 90-day mortality is anticipated [ 5 ]. However, even if recognized viable from wide series, the description of each individual Versius procedures with specific challenges is still crucial. The aim of the article is to report a comprehensive analysis of the Versius system for pelvic surgery, by describing indications, setup, and early outcomes in a multicentric study.
Methods This is a retrospective observational study involving two Institutions, ASST Santi Paolo and Carlo, Milan, and at the Apuane Hospital, Massa, Italy. The Versius CMR was installed by September, 2022, at the former institution, and by 2021 in the latter. The console surgeons involved were urologists (BR and DM), gynecologists (GG, MF), and general surgeons (AP). All except two were prior Da Vinci users; the latter (AP an MF) had an extensive laparoscopic background. All interventions performed in the pelvic area with the Versius were included in the analysis. Data about indications, intra-, and post-operative course were prospectively collected. Training The training involved the whole OR Team (console surgeon, bedside assistant, scrub nurse, circulating nurse). For surgeons, the Versius training package consisted of a 3.5-day program following 10 h of online didactic training; it includes dry box exercises and wet lab sessions simulated in an operating room using cadaveric models. Interventions Urologic pelvic procedures, gynecological procedures, pelvic procedures from general surgeons were considered. Colectomy and hernia repair were excluded. Interventions could be either performed for oncological or non-oncological purposes; reconstructive interventions with the use of devices (i.e., mesh) were included as well. Data collection and analysis The following variables were collected: Demographics: age, gender; Surgical indications Intra-operative data: OR setup, number of ports, number of robotic arms, console time, estimated blood loss (EBL), conversion to alternative approaches, complications invoking a change in surgical strategy; use of additional devices for repair (i.e., mesh); malfunctioning of the system requiring a re-start Post-operative data: complications classified by Clavien– Dindo, transfusion rate, length of stay (LOS), 30- and 90-day re-admission Data were inserted in a dedicated data base (Excel, Microsoft) and analyzed by an external reviewer (ST) with formal habilitation as a coordinator with the Versius and the Da Vinci systems. A descriptive analysis of all variables was performed. Console Time, EBL, and LOS were provided as mean value and range.
Results Overall, a total of 171 interventions were performed with the Versius CMR at the ASST Santi Paolo and Carlo (120) and at the Apuane Hospital (51). Forty-two of them involved pelvic procedures. A full list of interventions stratified by center and specialty is reported in Table 1 . Twenty-two interventions had an oncological indication (localized prostate cancer) whereas the remaining ones had a non-oncological or functional purpose. The mostly performed procedures were radical prostatectomy (22) and annexectomy (9). The OR setup for urological, gynecological and general surgical procedure (rectopexy and sigmoidectomy) is depicted in Fig. 1 a–d, respectively. No intra-operative complication invoking a change in surgical strategy occurred nor conversion to open or laparoscopic surgery. Two Clavien II and one IIIb complications were evident (pelvic hematoma requiring transfusion, a urinary tract infection treated with antibiotic therapy and a bowel obstruction due to port-site herniation). Table 2 reports console time, EBL, complications, and LOS stratified by procedure. Malfunctioning/alarms requiring the whole system re-activation occurred in 2 different cases. An adjustment in trocar placement according to patients’ height was required in 2 patients undergoing RALP, in which the trocar was moved caudally. In two cases, a pelvic organ prolapse (POP) was repaired concomitant with other gynecological procedures: an hysterectomy, bilateral salpingo-oophorectomy, and lateral suspension with a titanium-coated polypropylene mesh, according to Dubuisson technique. A single case, a 53-year-old woman, underwent a ventral rectopexy for full-thickness rectal external prolapse. Positive margin (PM) rate can be evaluated for radical prostatectomies. The outcomes are variable, one series showed 83.3% of PM rate on 18 cases (11 pT3, published data) [ 6 ] and in the other one (4 pT2c cases), no PM were found. At a follow-up ranging from 9 to 11 months, a single case of biochemical recurrence was noticed. One re-admission at 30 days (herniation) was recorded, and none at 90 days.
Discussion The current series represents the first one completely focused on pelvic surgery with the new Versius surgical system. Sparse reports evaluating single indications with Versius [ 7 – 9 ] are available yet; similarly, some series reported on the use of Versius for abdominal surgery, the majority including preliminary procedures—hernia repair, cholecystectomy—to become familiar with the system [ 10 , 11 ]. In the present series, we intentionally addressed the pelvic area, to evaluate if robotic surgery with a novel platform may address the challenges posed by such interventions, often made up by complex tasks (dissection and reconstruction) in small spaces with limited accessibility and range of movements. A comprehensive analysis of pelvic approach with Versius has been already reported in the pre-clinical setting: Vasdess et al. [ 11 ] described the possible setup for prostate surgery on cadaveric models, exploring multiple options of trocar placement with either 3- and 4-arm configurations. To note, authors evaluated also the feasibility of radical cystectomy (even if in vivo cystectomy has not yet performed so far with the Versius). Our series confirms the feasibility of Versius pelvic surgery in different settings without major complication; actually, we did not face the need for conversion to other approaches. In preliminary series, conversion has been reported in up to 4–6% of Versius cases [ 12 , 13 ] together with a certain variable degree of adverse events [ 13 – 15 ]. Herein, a single Clavien IIIb complication was reported (bowel herniation through abdominal wall requiring surgical repair); the occurrence is seemingly unrelated to the use of a novel robotic system. As far as surgical margin status is concerned, radical prostatectomy is the only oncological intervention we included. In this setting, PM rate from a single institution series appears of importance (83.3%); the small sample size and a high rate of pT3 could have accounted for the occurrence [ 6 ]. Moreover, PM is counterbalanced by low BCR, (a single case showing a raise in post-operative PSA out of 22 RALP), provided long-term PSA is still awaited. Unlike other series dealing with the Versius, some complex cases have been performed as well in the present article: this is the case of surgery for endometriosis and of a case of post-radiation sigmoidectomy. Some specifics to pelvic surgery may apply to the Versius system. Surgical instruments are shorter than those of the Da Vinci (30 cm): the issue should be considered and an estimation of the distance to the target area (and to other areas to be reached, i.e., pelvic nodal dissection) should be pre-planned. Port placement can be, therefore, moved caudally than the conventional Da Vinci configuration, especially considering patient’s height. In two RALP cases, we had to move one port more caudally to make the instrument reach the target area (urethral stump). The BMI of the patient is another issue that needs to be taken into account: in the current series, extreme BMIs have been successfully managed with a prior accurate planning of port and BSU placement. This was the case of gynecological surgery, which encompassed a range of BMI from 16 to 43 (non-published data). One of the major concerns of multi-modular robotic system is the likelihood of external clashing between arms, a point somehow counterbalancing the versatility invoked by new systems with independent units [ 16 ]. Overall, whereas the Da Vinci Xi allows for a linear trocar placement with a standardized docking, such a step with the Versius should be accurately planned to minimize clashing or limitation in instruments motion [ 16 ]. Another feature typical of the Versius system is the docking of the robotic arm to the instrument and not to the trocar, as occurs with the Da Vinci or Hugo RAS; if the use of a non-dedicated trocar may represent an advantage, on the other side collisions between the robotic arm and the skin of the patient are recognized by the system and given as an alarm. The issue should be taken into account as well during the setup, especially in pelvic surgery in which a Trendelenburg of the patient is required and angles with the trocar could be relevant. Beyond technical details, some general considerations may arise from the current experience. Versius can be used by surgeons from different disciplines and it heavily comprises diverse surgical specialties; the same occurs with the Da Vinci, even if major users have initially been urologists [ 1 ]. It is representative that in a center owning three robotic systems such as the ASST Santi Paolo and Carlo (Da Vinci, Versius CMR and Hugo RAS), the majority of Versius cases have been accomplished by gynecologists and general surgeons. In line with these findings, laparoscopists seem to be those surgeons mostly advantaging from the introduction of Versius. The instruments are designed to mimic the articulation of the human arm and the wristed joints with seven degrees of freedom overcome the difficulties of laparoscopic surgery. The use of laparoscopic trocars—not robotic dedicated—raises opportunity for a hybrid procedure; thus, laparoscopists appreciate Versius features designed to simplify laparoscopic surgery. Severe malfunctioning of the Versius has been reported in few cases and has been fixed with the re-start of the system (2 cases). If compared to the Da Vinci, it should be recognized that the latter has reached its fourth generation, whereas the Versius system is at its very first one; technological improvement and optimization are awaited. Noticeably, similar considerations may apply also to other novel robotic platforms that entered the market within the last two years. As far as the Versius is concerned, technological updates are already developing: an improvement in arm clash recovery has been recently released—i.e., maintenance of arm engagement from the surgeon in case of clash—and other enhancements are yet to come, such as improvement of bipolar devices and the implementation of an energy sealer device [ 12 ]. The article is not devoid of limitations. First, the small sample size and short follow-up preclude any conclusion about mid- and long-term outcomes. The different kind of surgeries included make outcome assumptions weak; however, the report of long-term clinical follow-up was beyond the purpose of the paper, that merely aimed to address the feasibility of Versius surgery in the pelvic area. Second, no matched comparison with the Da Vinci nor a subjective comparison is herein provided. Some console surgeons (BR, DM, GG) and assistants (MCS, FT, MS) are currently operating on multiple platforms, but a subjective perception about systems is not herein provided. However, an objective comparison of the first RALP cases with the Versius and Hugo RAS has been described by our group throughout the metrics developed by the ERUS working group for radical prostatectomy [ 17 ]. Further comparative analysis of robotic systems is expected in the very next future, to highlight differences and peculiarity of each platform trying to draw definite outcomes.
Conclusions From the present series, pelvic surgery with the Versius system is feasible without severe intra- or peri-operative complications; long-term oncological and functional outcomes are yet to be defined. All steps of pelvic surgery are reproducible with the Versius, provided a proper surgical setup and trocar placement are pursued. Versius can be easily adopted by surgeons of different disciplines and background within the same institution; a further multi-specialty implementation is expected, and a cost analysis is awaited to highlight its future role into healthcare systems.
Introduction Versius CMR is a novel robotic system characterized by an open surgical console and independent bedside units. The system has potentials of flexibility and versatility, and has been used in urological, gynecological, and general surgical procedure. The aim is to depict a comprehensive analysis of the Versius system for pelvic surgery. Methods This is a study involving two Institutions, ASST Santi Paolo and Carlo, Milan, and Apuane Hospital, Massa, Italy. All interventions performed in the pelvic area with the Versius were included. Data about indications, intra-, and post-operative course were prospectively collected and analyzed. Results A total of 171 interventions were performed with the Versius. Forty-two of them involved pelvic procedures. Twenty-two had an oncological indication (localized prostate cancer), the remaining had a non-oncological or functional purpose. The mostly performed pelvic procedure was radical prostatectomy (22) followed by annexectomy (9). No intra-operative complication nor conversion to other approaches occurred. A Clavien II complication and one Clavien IIIb were reported. Malfunctioning/alarms requiring a power cycle of the system occurred in 2 different cases. An adjustment in trocar placement according to patients’ height was required in 2 patients undergoing prostatectomy, in which the trocar was moved caudally. In two cases, a pelvic prolapse was repaired concomitant with other gynecological procedures. Conclusions Pelvic surgery with the Versius is feasible without major complications; either dissection and reconstructive steps could be accomplished, provided a proper OR setup and trocar placement are pursued. Versius can be easily adopted by surgeons of different disciplines and backgrounds; a further multi-specialty implementation is presumed and long-term oncological and functional outcomes are awaited. Keywords Open access funding provided by Università degli Studi di Milano within the CRUI-CARE Agreement.
Author contributions MCS: protocol and project development, manuscript writing. MDM: project development, data management. JM: data collection, data analysis. MF: project development, data management. APC: project development, data management, manuscript editing. LM: data management. CC: data management. AM: data collection, data analysis. TC: data collection, data analysis. EP: data collection, data analysis. MS: data management. FT: data management. ST: data analysis, manuscript editing. SA: data collection. LS: data collection. MA: data collection. AM: project development. PPB: project development, supervision. SM: supervision. BR: project development, supervision. GG: project development, data management, manuscript editing. Funding Open access funding provided by Università degli Studi di Milano within the CRUI-CARE Agreement. Data availability Data will be shared upon appropriate request. Declarations Conflict of interest Authors have no potential conflict of interest to disclose. The research involves human participants All patients signed an informed consent to the procedure they underwent; all interventions were planned and performed according to the current international guidelines and treatment recommendation.
CC BY
no
2024-01-15 23:42:01
World J Urol. 2024 Jan 13; 42(1):31
oa_package/36/d9/PMC10787883.tar.gz
PMC10787884
37812362
Introduction Breast cancer in men (male breast cancer; MBC) is a rare entity accounting for 0.5–1% of all breast cancer diagnoses and an estimated lifetime risk of 1:1000 [ 1 ]. Due to its rarity and lack of prospective dedicated MBC trials, the treatment guidelines are mainly based on extrapolation from randomized evidence from trials including mainly women with breast cancer (female breast cancer; FBC) [ 2 ]. The utilization of neoadjuvant chemotherapy (NAC) has steadily increased over the years due to its potential to de-escalate breast and axillary surgery as well as to provide an opportunity for response-based tailored adjuvant therapy [ 3 ]. There is solid evidence supporting the use of NAC in FBC, while the utilization and effectiveness of NAC in MBC is less studied [ 3 , 4 ]. In fact, two retrospective studies have shown conflicting results regarding potential sex disparities on the effectiveness of NAC [ 5 , 6 ], whereas one of them also showed a lower utilization of NAC in MBC compared to FBC [ 5 ]. Considering the limited and conflicting evidence on the role of NAC in MBC, the aims of the present nationwide, register-based, retrospective cohort study were to investigate the utilization of NAC in men and women with early breast cancer, and to compare the effectiveness of NAC between men and women in terms of pathologic complete response (pCR).
Patients and methods Study design, data sources, participants and data collection For this nationwide, register-based retrospective cohort study , all patients with stage I–III invasive breast cancer diagnosed in Sweden between January 1, 2008 and December 31, 2019 were identified through the National Quality Registry for breast cancer (Nationellt Kvalitetsregister för bröstcancer; NKBC). NKBC has a high coverage (99.8%) and data completeness ensuring the validity of data and the generalizability of the study results [ 7 ]. Using the ten-digit personal identity number, data from NKBC was linked to other national databases of interest to build the research database BCBaSe 3.0 ( https://cancercentrum.se/samverkan/regional-cancer-centres/research-and-innovation/register-based-research-databases/) . The study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Regional Ethics Committee, Stockholm (Approval number: 2019–02610). All patients with stage I-III breast cancer in NKBC were included except those with lacking information on estrogen-receptor (ER) status, progesterone-receptor (PgR) status, HER2-status or information on preoperative or postoperative TNM stage. Patients who did not undergo surgery were also excluded. The number of patients treated with NAC but not underwent surgery were 91 females (1.4% among females with NAC) and 1 male (4.0% among males with NAC). The decision of no surgery has been considered unrelated to disease progression since all patients had a registered adjuvant therapy and none of the patients was registered with metastatic disease during the first three months from diagnosis (Fig. 1 ). Patient demographics (age, educational level, household income, health care region at diagnosis), tumor characteristics (tumor size, histological grade, clinical stage, pathological stage, morphological type and surrogate molecular subtypes based on immunohistochemistry (IHC) status) and treatment characteristics (data on breast and axillary surgery, chemotherapy, radiotherapy and endocrine therapy) were collected. Outcomes and definitions IHC-subtyping was used to classify tumors into three surrogate subtypes, namely luminal (ER or PgR-status ≥ 10%, HER2-negative), HER2-positive (any ER and PgR status, HER2-positive), and triple negative (ER and PgR status < 10% and HER2-negative) breast cancer. For the research question on NAC utilization, patients were classified as treated with NAC if there was a treatment strategy including NAC, irrespective of chemotherapeutic agent used. For the research question on pCR, we defined pCR as the absence of invasive breast cancer in the surgical specimens from breast and axillary lymph nodes. Statistical analysis Data were summarized using descriptive statistics including numbers and percentages for categorical variables and median with range for continuous variables. Comparisons of patient characteristics between males and females were made by Pearson’s Chi-squared test, Fisher ́s exact test, or Mann–Whitney test as appropriate. For the first research question on NAC utilization, a multivariable logistic regression analysis was performed to analyze the association between sex and the likelihood of receiving NAC while adjusting for prespecified patient- and tumor characteristics including age at diagnosis, educational level, household income, health care region at diagnosis, clinical T and N stage, morphological type, histological grade, and IHC-based subtype. For the second research question on pCR, a multivariable logistic regression analysis was conducted to evaluate the impact of sex on the odds of pCR, adjusted for prespecified parameters including age at diagnosis, clinical T and N stage, histological grade, and IHC-subtype. All p values reported were two-sided, and p values of < 0.05 were considered statistically significant. All statistical analyses were performed using the SPSS statistical package (IBM Corp. Released 2021. IBM SPSS Statistics for Windows, Version 28.0. Armonk, NY: IBM Corp).
Results Characteristics of study cohort In total, 114,290 patients with breast cancer were registered in NKBC between 2008 and 2020. Applying the inclusion and exclusion criteria, 82,888 patients (82,401 women and 428 men) with stage I-III breast cancer who underwent surgery and had adequate information for IHC-subtyping were identified. This cohort comprised the NAC utilization cohort. When the cohort was restricted only to patients who received NAC, 6487 breast cancer patients (6463 women and 24 men) were available for analyses related to the effectiveness of NAC. A flowchart diagram of patients’ selection process is shown in Fig. 1 . Table 1 summarizes patient, tumor and treatment characteristics in the NAC utilization cohort, by sex. Men with breast cancer were older, had a lower educational level, more advanced anatomical stage at diagnosis (both T and N stage), higher histological grade, fewer lobular carcinomas (1.5% vs. 13.8% in women) and a different IHC-subtype distribution (luminal 86.9% vs. 77.3%; HER2-positive 12.5% vs. 13.4%, triple negative 0.6% vs. 9.2%, respectively). Treatment patterns, including breast and axillary surgery as well as adjuvant therapeutic approaches followed the statistically significant differences in anatomical staging and IHC-subtype. Regarding NAC effectiveness cohort, 6463 women (7.8%) and 24 men (4.9%) received neoadjuvant chemotherapy, respectively. When comparing baseline characteristics of women and men treated with NAC, statistically significant differences regarding educational level, IHC-subtype, use of adjuvant endocrine therapy, as well as type of breast surgery were seen (Table 2 ). The utilization of NAC seems to be steadily increased over time for both males and females in Sweden as shown in Table 2 . Factors associated with NAC utilization patterns Using multivariate logistic regression model, using the complete case analysis method, no statistically significant difference in NAC utilization between women and men was observed (Odds Ratio (OR): 1.135; 95% Confidence Interval (CI): 0.606–2.128). The total number of patients included in this model was 78,760. Factors associated with higher likelihood of NAC were: young age, high educational level, high household income, treatment in certain healthcare regions Stockholm/Gotland, South, or Southeast), high clinical T and N stage, high histological grade, HER2-positive and triple negative IHC-subtype and ductal histology (Table 3 ). Factors associated with pCR after NAC No statistically significant difference in pCR rates were observed between women and men in the multivariate logistic regression analysis, using the complete case analysis method (OR: 1.141; 95% CI 0.141–9.238). The total number of patients included in this model was 6215. Factors associated with higher pCR rates were young age, high histologic grade, and HER2-positive IHC-subtype (Table 4 ).
Discussion Using nationwide, register-based data from Sweden, we found no evidence of sex disparities regarding utilization of NAC in breast cancer patients when analyses were adjusted for patient- and tumor characteristics. In addition, the effectiveness of NAC in terms of pCR seems to be similar between men and women with breast cancer, supporting the current recommendations on treating men with breast cancer with the same principles as women with regard to NAC indications. Due to the rarity of MBC and subsequently the lack of prospective trials, potential differences in utilization and effectiveness of NAC between men and women have been investigated only through register-based studies. Regarding utilization of NAC in men with breast cancer, Cao et al. analyzed data from the United States National Cancer Database (NCDB) between 2004 and 2016, and found that men with node positive (N +) disease were less likely to be treated with NAC when compared to women. Interestingly, Cao et al. found an underutilization of oncological treatment in men with breast cancer in general, a pattern not seen in the present study cohort. Breast cancer treatment practices may vary between countries, and may also have changed over time. Our study cohort included patients diagnosed during more recent years, when sex disparities in breast cancer treatment strategies have been acknowledged [ 8 ], thus leading to efforts to mitigate these disparities. Differences between study cohorts with regard to age (a higher proportion of older adults in our cohort) and stage (only patients with N + disease in NCDB cohort) could also explain the partly conflicting study results. Also, with regard to NAC effectiveness in men compared to women with breast cancer, the current literature shows somewhat conflicting results. Cao et al. found similar pCR rate between men and women treated with neoadjuvant chemotherapy, as well as a comparable overall survival. Leone et al. found the odds for pCR in women compared to men to be nearly twice as high when studying patients diagnosed 2010 to 2016 from the same database as Cao. Interestingly, the difference in pCR rates observed in the latter study was mostly driven by differences in pCR within the luminal HER2-negative and luminal HER2-positive subgroups. Our results are in accordance with Cao et al., but differ from Leone et al. Although the number of included patients in certain subgroups in our study cohort was not large enough to enable subgroup analyses, we included this potential confounding factor into the multivariate model when the impact of sex on NAC effectiveness was analyzed. The lack of difference in pCR rates between men and women with breast cancer, in spite of the higher percentage of luminal breast cancer in men, could possibly be explained by different distribution of Luminal A/B tumors between men and women, i.e. a higher proportion of more high risk Luminal B tumors in men [ 9 , 10 ]. Luminal B breast cancer is associated with higher pCR rates than Luminal A [ 11 ], possibly balancing the chance of pCR between the two cohorts. Another potential explanation of similar NAC effectiveness between men and women despite the dominance of luminal tumors in men could be a higher presence of an immunological-enriched tumor microenvironment in male luminal tumors [ 12 ], a condition that has been associated with improved pCR rates in all breast cancer subtypes [ 13 ]. The comparison of patient- and tumor-related characteristics between men and women with breast cancer confirmed some well-established differences between the sexes; older age at diagnosis, more advanced N-stage at diagnosis, the dominance of luminal subtype and the rarity of TNBC and lobular histology in men with breast cancer. Considering NAC utilization in general, some study results deserve attention. Our study results confirm a tumor-driven approach regarding NAC utilization with a higher use in more advanced and biologically more aggressive disease, which is in line with the current evidence and clinical practice. On the other hand, our study results imply some socioeconomic and geographic inequalities with lower odds to receive NAC among patients with lower income, lower education and among those from specific regions. Although similar inequalities have been reported previously [ 14 , 15 ], such disparities are not acceptable and a deeper understanding is necessary in an effort to eliminate healthcare-related inequalities. An interesting finding in terms of factors associated with higher pCR rates was the association between young age and a higher pCR rate. This finding is in accordance with a pooled analysis from eight randomized trials, where younger patients had higher odds of pCR, thus supporting the notion that tumor biology in younger patients might be more aggressive also within subtypes and therefore more susceptible to chemotherapy [ 16 ]. The study has several limitations that need to be addressed. First, data on planned treatment was used rather than actual treatment given, as NKBC data on planned treatment have a higher validity than given treatment for the studied time period. As a result, however, a risk of misclassification between NAC and primary surgery for some patients does exist. Second, the duration of planned NAC is lacking. However, the Swedish guidelines have steadily and throughout the years, recommended the use of six cycles (q3w) of chemotherapy, similar to the recommendation for adjuvant chemotherapy. In our study cohort, we lacked information about dose-dense chemotherapeutic regimens. However, this strategy has been rather uncommon in Sweden during the study period and there is no reason to believe that there would be any sex disparity in using dose dense regimens that could impact the prognosis. The limited sample size for men with breast cancer in some specific subtypes is also a limitation as it precludes from relevant subgroup analysis. To mitigate this source of bias, we tried to adjust the multivariate analyses using parameters of potential interest as breast cancer subtype. One could argue that the exclusion of patients who did not undergo surgery can result in immortal-time bias if disease progression during NAC is the reason for no surgery. However, within the group of patients treated with NAC, the proportion of patients who did not undergo surgery was extremely low in both sexes and was considered unrelated to disease progression, thus eliminating the risk for immortal-time bias. Finally, the nature of collected data for the present study does not allow any information about the role of patient or clinician in treatment decision regarding the type and sequence of treatment strategy between sexes. Acknowledging the relative limited number of men with breast cancer included in the analyses, the current study did not find any sex disparities either in the NAC utilization or effectiveness supporting the current recommendations on treating men with breast cancer similar to women with regard to indications for NAC. The observed socioeconomical and geographical disparities in NAC utilization deserves a deeper understanding before designing strategies to eliminate these inequalities towards an equitable access to breast cancer care.
Purpose Evidence supporting the use of neoadjuvant chemotherapy (NAC) in early breast cancer is based on studies mainly including women, whereas the utilization and effectiveness of NAC in men is less studied. The present study aimed to investigate the utilization and effectiveness of NAC in men and women with early breast cancer. Methods Eligible patients were identified through the Swedish National Breast Cancer Quality Register, that includes all newly diagnosed breast cancer cases in Sweden from 2008 and onwards. For the treatment utilization analysis, all patients with stage I–III between 2008 and 2020 were included (n = 82,888), whereas for the effectiveness analysis the cohort was restricted to patients receiving NAC (n = 6487). For both analyses, multivariate logistic regression models were applied to investigate potential sex disparities in NAC utilization and effectiveness, adjusted for patient- and tumor characteristics. Results In the NAC utilization analysis, 487 men and 82,401 women with stage I–III were included. No statistically significant difference between sexes in terms of NAC utilization was observed (adjusted Odds Ratio (adjOR): 1.135; 95% Confidence Interval (CI) 0.606–2.128) with an overall utilization rate of 4.9% in men compared to 7.8% in women. Among the 24 men and 6463 women who received NAC, the pathologic complete response (pCR) rates were 16.7% and 21.2%, respectively (adjOR: 1.141; 95% CI 0.141–9.238). Conclusion The present study did not find any sex disparities in NAC utilization or effectiveness in terms of pCR. This supports the current recommendations of treating men with breast cancer with the same indications for NAC as women. Keywords Open access funding provided by Uppsala University.
Author contributions AV conceived the original idea about the study and study’s design. Material preparation, data collection and analysis were performed by AV. The first draft of the manuscript was written by AS and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. Funding Open access funding provided by Uppsala University. This study did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Data availability Data are available from register holders (Statistics Sweden, Swedish National Board of Health and Welfare, the Regional Cancer Center Stockholm Gotland) for researchers with relevant ethical approvals and who meet the criteria for access to confidential data. The data are not publicly available due to restrictions by Swedish and European law, in order to protect patient privacy. Declarations Conflict of interest AS, MS and AV have no competing interests. IF has received institutional research grants from MSD unrelated to the current work.
CC BY
no
2024-01-15 23:42:01
Breast Cancer Res Treat. 2024 Oct 9; 203(2):235-243
oa_package/e1/ed/PMC10787884.tar.gz
PMC10787885
38217834
Introduction Robotic-assisted surgery (RAS) has gained widespread diffusion over the last years, demonstrating the ability to overcome the technical limitations of conventional laparoscopy. Enhancements provided by robotic assistance include three-dimensional view and magnification, increased dexterity with 7-degrees of freedom of robotic instruments, tremor filtering, and improved surgeons’ ergonomics [ 1 , 2 ]. Some major drawbacks must be considered before using RAS in children: anesthesia, placement of trocars, and technical difficulties related to small space [ 3 ]. Nevertheless, RAS has been described as safe and feasible option for a wide range of surgical indications in children, including urological, oncological, and gastrointestinal pathologies [ 4 – 8 ]. Several reports have investigated safety and feasibility of RAS in pediatric population, compared with different approaches (open or laparoscopic) [ 9 – 12 ]. To the current state, the field of pediatric gynecology remains the least explored, with only few pediatric reports of application of RAS for gynecological indications [ 13 – 16 ]. This descriptive, retrospective study aimed to report a multicenter experience regarding the application of RAS for gynecological indications in pediatric patients.
Materials and methods All children and adolescents up to 18 years of age, operated using RAS for gynecological indications in 4 different institutions over a 3-year period, were included. The exclusion criteria were patients over 18 years old as well as all gynecological surgical conditions not treated with RAS. The surgical centers were contacted via mail and those accepting to participate to the study, were required to fill a study form, with requested information, for each enrolled patient. Patient baseline, including age at time of surgery, weight, possible comorbidities, clinical presentation, pre-operative diagnosis, and side of pathology, were reported in the first section. Details of operative technique, such as type of procedure, number of robotic and/or accessory ports, use of sealing device, method for specimen extraction, were reported in the second section. Operative results, including robot docking time, total operative time, length of stay (LOS), requirement time of pain medication, complication rate, conversion rate, pathology, and follow-up results, were analyzed in the third section. All data were elaborated using the statistical software Microsoft Excel, Windows vers.11. Descriptive statistics were used to present findings, and quantitative variables were expressed as median (range) to report the data. The study received appropriate Institute Review Board (IRB) approval.
Results Patient baseline Twenty-three girls, with median age at surgery of 12.3 years (range 0.6–17.8) and median weight of 47.2 kg (range 9–73), received RAS for gynecological indication in the study period and were included. Associated comorbidities were reported in 5/23 (21.7%). Most patients (16/23, 69.5%) were symptomatic at time of diagnosis, with non-specific abdominal/pelvic pain being the most frequent presentation. Pre-operative work-up included abdominal ultrasonography (US), pelvic computed tomography (CT) and/or magnetic resonance imaging (MRI), and voiding cystourethrogram (VCUG) in selected cases. Serum tumor markers, such as beta human chorionic gonadotropin (β-HCG), alfa-fetoprotein (α-FP), Cancer Antigen 125 (Ca125), lactic dehydrogenase (LDH), and human epididymis secretory protein 4 (HE-4), were performed in all patients with adnexal mass. Pre-operative diagnosis was ovarian cyst ( n = 4), ovarian “complex” mass ( n = 12), fallopian tube lesion ( n = 3), uterine cyst ( n = 1), gonadal dysgenesis ( n = 1), pelvic paravaginal mass ( n = 1), cloaca malformation ( n = 1), and high-confluence urogenital sinus (UGS) ( n = 1). One patient (4.3%) presented concomitant ovarian “complex” mass and paratubal cyst. Installation and operative technique All RAS procedures were carried out using the da Vinci Xi Surgical System (Intuitive Surgical, Sunnyvale, CA, USA). The patient was placed supine on the operative table and appropriate age-sized Foley catheter was inserted using sterile precautions pre-operatively. Three robotic arms, one 12-mm with 12–8 mm reducer, for the 3D, 0-degree, robotic optic, and two 8-mm ports to accommodate the robotic instruments, were placed on the umbilical line in all procedures. A fourth 5-mm accessory port was also placed. The robot was finally docked over the patient’s feet. Robotic vessel sealer was adopted in all procedures. Indocyanine green (ICG) near-infrared fluorescence (NIRF) was adopted in ovarian mass to check the resection margins and guide intra-operative decision making and in paratubal lesion to check the vascular permeability of the fallopian tube following the removal of the lesion (Fig. 1 ). A 10-mm bag-retrieval, introduced through the umbilical port, was adopted for specimen extraction. Video 1 reproduces the technique of robotic-assisted resection of ovarian mass using ICG-NIRF. Operative results The RAS procedures included: ovarian cystectomy ( n = 10), salpingo-oophorectomy ( n = 6), bilateral gonadectomy ( n = 1), salpingectomy ( n = 1), paratubal cyst excision ( n = 1), Gartner cyst excision ( n = 1), paravaginal ganglioneuroma resection ( n = 1), fistula closure in UGS ( n = 1), and vaginoplasty using ileal flap in cloaca malformation ( n = 1). Median operative time was 144.9 min (range 64–360), and median docking time was 17.3 min (range 7–50). Conversion to open or laparoscopy was not necessary in any case. Median LOS was 2.1 days (range 1–7), and median analgesic requirement was 2.2 days (range 1–6). One patient (4.3%) needed redo-surgery for recurrent Gartner cyst (Clavien 3b). The histopathology confirmed diagnosis of ovarian serous cystadenoma ( n = 2), ovarian functional follicular cyst ( n = 2), mature cystic teratoma ( n = 6), immature teratoma ( n = 4) (Fig. 2 ), ovotestis ( n = 2), streak gonads in Turner syndrome SRY + ( n = 1), paratubal cystadenoma ( n = 2), Gartner cyst ( n = 1), and ganglioneuroblastoma ( n = 1). The median length of follow-up was 2.2 years (range 0.5–4.5). No patients required adjuvant chemotherapy following surgery and none reported recurrence of tumoral pathology. All results are summarized in Table 1 .
Discussion Despite the growing number of indications in pediatric urology, the application of RAS remains still limited in other fields of pediatric surgery. Analyzing the pediatric literature, very few reports on RAS application in gynecology are available, with limited case series or single-case observations [ 13 – 15 ]. Our study collected the number of 4 pediatric surgery units with high volume robotic activity and 24 patients, operated over a 3-year period, were enrolled. Despite the small number of patients included, our preliminary results were promising and showed that RAS may be fully applicable even to gynecological indications in pediatric patients. Improved dexterity, coordination, and visualization were provided by robot assistance. The absence of intra- and post-operative complications also confirmed the safety and feasibility of this approach in children. Moreover, our study added new elements to the current knowledge. First, the previous reports have described benign ovarian pathology as main indication to RAS [ 13 – 15 ]. Our study introduced further undescribed indications, such as tubal, uterine, and vaginal malformations, and demonstrated the feasibility of complex reconstructive procedures of internal genitalia such as vaginoplasty using ileal flap using robotic approach. Based on our experience, we believe that the key-points for an optimal management of such pathology using RAS are correct trocar placement, use of sealing device and ICG-NIRF technology, use of endobag for specimen extraction, and teamwork. The most critical step is the placement of trocars, especially in children [ 17 ]. Improper placement of the trocars would limit robotic manipulation in the abdominal cavity and increase the chances of instrument conflicts due to the small surface area of the abdominal wall of children and the relatively small space in the abdominal cavity. As described by Xie et al. [ 14 ], we always adopted three robotic arms and a fourth accessory laparoscopic port for the bedside surgeon. We placed the robotic ports on the umbilical line to keep proper distance from the pelvic area and have enough working space for manipulation of giant masses or insertion of retrieval bag. Our standard operation order was to insert the trocar for the scope first and then insert the trocars for the robotic arms. Furthermore, if use of specimen retrieval bag is planned, our suggestion is to place 12-mm umbilical robotic port with 12–8 mm reducer to use the 8-mm robotic scope. At time of specimen extraction, the optic is moved to one working arm and the 10-mm retrieval bag is inserted into the abdominal cavity through the umbilical robotic port and the specimen extraction is finally done under direct vision. Use of robotic vessel sealer is very helpful to perform a bloodless dissection of anatomic structures or resection of giant tumors. In some indications, such as ovarian tumors, use of ICG-NIRF was very helpful to visualize the resection margins of the mass and help guide intra-operative decision between salpingo-oophorectomy and ovarian-sparing surgery. This technology required an intra-operative administration of ICG (0.5 mg/kg) via intravenous route and in a matter of 60 s, fluorescence appeared in the target organs, allowing to identify the resection margins and the vascularization of the mass [ 18 – 20 ]. Recently, ICG-NIRF was also adopted during removal of paratubal lesion, to check the vascular permeability of the fallopian tube following the resection of the lesion. Use of endobag is needed for extraction of resected tumors with high suspicion of malignancy. We suggest adopting large endobags (volume up to 1000 mL) and extract the specimen by enlarging the umbilical incision and, whenever possible, aspirating the liquid content of cystic masses before extraction, to avoid additional large Pfannestiel incision. Finally, the teamwork is essential to perform a smooth operation, shorten the learning curve for docking, and ultimately reduce total operative time and anesthetic times. Limitations of the presented study are the small number series, the limited follow-up period, and the multi-institutional participation, that made the data hardly comparable. The number of accumulated procedures in children is difficult to compare with that in adults. Thus, it was necessary collect the data of different pediatric surgery units to collect more conspicuous evidence.
Conclusion This preliminary experience showed that RAS is safe and feasible for surgical treatment of pediatric gynecological pathology, although no conclusive data are available to confirm its superiority over traditional laparoscopy. Case–control and comparative prospective studies will help to delineate better the advantages of this new technology as well as its optimal use in pediatrics. The primary focus for future studies should therefore be on quality management, optimization of patient outcomes for the largest number of patients, and surgical and team training.
Robotic-assisted surgery (RAS) is increasingly adopted in the pediatric population. This retrospective multicenter study aimed to report application of RAS for gynecological indications in pediatric patients. The medical records of all girls with gynecological pathology, operated in 4 different institutions over a 3-year period, were retrospectively collected. Robot docking time, total operative time, length of stay (LOS), requirement time of pain medication, complication rate, conversion rate, and pathology were analyzed. Twenty-three girls, with median age of 12.3 years (range 0.6–17.8) and median weight of 47.2 kg (range 9–73), received the following RAS procedures: ovarian cystectomy for ovarian cyst/mass ( n = 10), salpingo-oophorectomy for ovarian complex mass ( n = 6), bilateral gonadectomy for Turner syndrome SRY + ( n = 1), salpingectomy for fallopian tube lesion ( n = 1), paratubal cyst excision ( n = 1), Gartner cyst excision ( n = 1), paravaginal ganglioneuroma resection ( n = 1), fistula closure in urogenital sinus ( n = 1), and vaginoplasty using ileal flap in cloaca malformation ( n = 1). Median operative time was 144.9 min (range 64–360), and median docking time was 17.3 min (range 7–50). Conversion to open or laparoscopy was not necessary in any case. Median LOS was 2.1 days (range 1–7), and median analgesic requirement was 2.2 days (range 1–6). One patient (4.3%) needed redo-surgery for recurrent Gartner cyst (Clavien 3b). This preliminary experience showed that RAS is safe and feasible for surgical treatment of gynecological pathology in pediatric patients, although no conclusive data are available to confirm its superiority over traditional laparoscopy. Randomized, prospective, comparative studies are needed to identify the gold standard approach for such indication. Supplementary Information The online version contains supplementary material available at 10.1007/s11701-023-01767-9. Keywords Open access funding provided by Università degli Studi di Napoli Federico II within the CRUI-CARE Agreement.
Author contributions All authors contributed to the study conception and design. All authors performed material preparation, data collection, and data analysis. M.E. and C.E. wrote the first draft of the manuscript. G.E. and C.D.M. prepared Fig. 1 . A.C. and M.E. prepared video 1 . All authors read and approved the final manuscript. Funding Open access funding provided by Università degli Studi di Napoli Federico II within the CRUI-CARE Agreement. The authors declare that no funds, grants, or other supports were received during the preparation of this manuscript. Data availability All data will be available on request. Declarations Conflict of interest The authors have no relevant financial or non-financial interests to disclose. Ethics approval This is a retrospective study. No ethical approval is required. Consent to participate Written informed consent was obtained from all individual participants included in the study and their parents.
CC BY
no
2024-01-15 23:42:01
J Robot Surg. 2024 Jan 13; 18(1):20
oa_package/d2/98/PMC10787885.tar.gz
PMC10787886
37249705
Introduction In the DSM-5, a “with Limited Prosocial Emotions” specifier was added to the Conduct Disorder diagnostic criteria to differentiate those children and adolescents who also display a callous absence of empathy and remorse, shallow or narrow affect, and deficient concern regarding performance in activities (American Psychiatric Association, 2013 ). This addition was based on research extending the affective dimension of the adult psychopathy personality construct downward to advance developmental theories (Frick, 2022 ; Frick et al., 2014 ). Greater knowledge on how these traits manifest earlier in development, when they may be more malleable, can inform clinical prevention and intervention efforts. In research settings, this specifier is commonly referred to as callous-unemotional (CU) traits and, like psychopathy, these traits are associated with several unique behavioral, emotional, and cognitive correlates relative to other externalizing and antisocial constructs (Frick et al., 2014 ). However, empirical work on CU traits has relied on panel (or cross-sectional) designs that primarily investigate within-person change. Such panel designs do not sufficiently assess within-person fluctuations and have comparatively low ecological validity–two important limitations of this line of research that can be circumvented through intensive longitudinal methodologies. Correlates of Callous-unemotional Traits Studies have pinpointed several distinguishing correlates of CU traits. Individuals with conduct problems and normative CU traits are often characterized by a defiant or difficult temperament with heightened levels of emotion dysregulation, negative affect, and impulsive aggression to provocation (Frick & Viding, 2009 ). In contrast, those with elevated conduct problems and CU traits do not tend to show the same types of emotional reactivity and are instead discerned by a fearless disposition reflected as emotional hypoactivity, reduced sensitivity to punishment, and often greater aggression severity (Fanti et al., 2016 ; McMahon et al., 2010 ). Despite the name, CU traits do not always confer risk for blunted emotional responsivity. For example, although differentiated on fear processing, one meta-analytic review ( k = 20) found no differences across CU traits and control groups in processing angry, disgusted, or happy cues (Marsh & Blair, 2008 ). In addition to dysfunctional parenting practices, it is this fearlessness that may explain why individuals with CU traits have been less successfully served by treatment efforts (Hawes et al., 2014a ; McMahon et al., 2021 ). Distinctions have also been made regarding the structure of CU traits, with studies having identified three (callousness, uncaring, unemotional; Ray & Frick, 2020 ) or two (callousness, uncaring; Hawes et al., 2014b ; Zheng et al., 2021 ) dimensions. Callousness tends to capture deficient empathy and guilt, uncaring reflects an aloof attitude regarding others, and unemotionality is defined as impoverished emotional experiences. Regardless of the structure, CU traits discriminates an etiologically and clinically atypical group (Frick et al., 2014 ). Traits and Symptoms in Adolescence Adolescence is a sensitive developmental period marked by profound cognitive, neurobiological, and interpersonal change (Dahl et al., 2018 ). In a large meta-analysis including 192 epidemiological studies, Solmi et al. ( 2022 ) found that the peak age of onset of any mental disorder was 14.5 years. Therefore, adolescence represents a prime time for understanding vulnerabilities and the development of psychopathological symptoms. Personality and psychopathology are typically considered distinct domains sometimes differentiated based on temporal stability. Specifically, some researchers have suggested that personality traits reflect long-term dispositions and that psychopathological symptoms are state-like instances of an experience or behavior (DeYoung et al., 2022 ). In childhood and adolescence, however, personality traits may also change over time (Soto et al., 2011 ). Indeed, some research has found little differences between mean-level change of several personality facets and internalizing symptoms across five 9-month follow-ups in a sample of female adolescents (Goldstein et al., 2022 ). Other evidence has shown that personality pathology tends to decline across adolescence (Álvarez-Tomás et al., 2019 ). In the CU traits literature, many panel studies have examined the long-term stability of CU traits with one review suggesting that CU traits show substantial rank-order stability and modest mean-level stability during childhood, adolescence, and into early adulthood (Frick et al., 2014 ). However, what cannot be ascertained from this work is whether adolescents with CU traits are consistent in their manifestation of CU traits. In other words, do CU traits fluctuate day-to-day (or moment-by-moment)? Intensive Longitudinal Methods Innovations in technology (e.g., smart phones and wearables) have generated a surge in studies investigating dynamic processes on micro timescales using a range of intensive longitudinal methods (for recent reviews, see Russell & Gajos, 2020 ; Urben et al., 2022 ). Broadly, intensive longitudinal methods, such as daily diary, experience sampling, ecological momentary assessment, and other ambulatory assessment designs, are characterized by many measurement occasions over a brief observation window. These methods focus on within-person variability, such that the measured variables may show deviations or fluctuations relative to the individual mean at specific observation occasions, but the individual mean is relatively stable over time compared to developmental growth or decrease in conventional longitudinal designs. Major advantages of intensive longitudinal methods also include the capability to assess real-time experiences and memory compared to traditional longitudinal designs, which can result in retrospective and recall biases (Russell & Gajos, 2020 ). These methods also enhance ecological validity, such that assessments are conducted when participants are in real-world settings relative to artificial laboratory experiments. Hence, intensive longitudinal data provide a high level of temporal granularity and inform our understanding of real-life short-term within-person dynamics as opposed to long-term developmental change. Testing Personality and Psychopathology with Intensive Longitudinal Methods Studies are beginning to apply intensive longitudinal designs to elucidate the dynamics of personality and psychopathology. For example, across three samples of adults, Edershile and Wright ( 2021 ) tested whether narcissism reflects a dynamic process and found variability in narcissistic states at a moment-to-moment level. Among 91 adults with a personality disorder, daily assessments of personality pathology including negative affect, detachment, impulsivity, and hostility predicted clinical characterizations of personality disorders and daily stress levels (Dotterer et al., 2019 ). In a sample of adolescent girls, daily negative affect, detachment, disinhibition, and psychoticism were associated with daily socio-affective processes (Kaurin et al., 2022 ). Such studies enhanced our understanding of state-level fluctuations in personality and psychopathology. Yet, no research to date has investigated whether CU traits also vary day-to-day or moment-to-moment. Only two intensive longitudinal studies have included a measure of CU traits using justice-involved adolescent samples; however, both studies only assessed CU traits as a total score at a single timepoint (De Ridder et al., 2016 ; Suter et al., 2017 ). Another study examined callousness in the context of interpersonal antagonism in samples of undergraduates, community adults, and psychiatric outpatients, but, similarly, callousness was only assessed at one timepoint as a predictor (Vize et al., 2022 ). Across these studies, higher levels of CU traits were associated with higher levels of perceived angry affect and self-reported antisocial behavior (De Ridder et al., 2016 ; Suter et al., 2017 ), and callousness was negatively associated with positive affect and empathy (Vize et al., 2022 ). In line with other personality constructs (Wright & Kaurin, 2020 ), much of our current understanding of CU traits is founded on structural or trait-based models (Frick et al., 2014 ). Research on dynamic modeling at micro timescales may help further inform functional narratives. As the affective domain of psychopathy, CU traits may demonstrate meaningful within-person fluctuations. Notably, the links between CU traits and conduct problems at the daily within-person level could possibly be instantiated by their intermediate and differential links involving daily positive and negative affect. Although most research places negatively valenced emotion processes at the center of personality and psychopathology, some research has implicated positive affect in the development of aggressive behaviors (Toro et al., 2020 ). In addition, affect and emotion are best conceptualized as dynamic processes underpinning a range of psychopathological outcomes (Houben et al., 2015 ). Other theoretical accounts suggest that certain forms of personality are derived and maintained by state-level fluctuations of the traits themselves (e.g., grandiosity and vulnerability in narcissism; Edershile & Wright, 2021 ). It may be the case that daily fluctuations in callousness and uncaring serves to perpetuate CU traits and associated conduct problems. Overall, state-level personality may be more malleable than traits, and thus, studying phenomena at the state-level can illuminate more effective treatment targets for this typically treatment-resistant group. The Present Study There is a lack of knowledge on CU traits on a micro timescale. In this exploratory work with data collected from a sample of adolescents via an intensive 30-day daily diary design, our aims were twofold. First, we aimed to explore whether CU traits (at the item level) demonstrated daily within-person fluctuations. In addition to the potential novel knowledge gained from investigating CU traits using intensive longitudinal designs, examining CU traits at the item level is congruent with recent research initiatives (e.g., Hierarchical Taxonomy of Psychopathology; Kotov et al., 2021 ) that emphasize fine-grained trait or symptom level analyses to advance mental health science. Second, we aimed to investigate whether daily CU traits (including callousness and uncaring) were associated with daily positive and negative affect, as well as emotional and conduct problems. To address these aims, we applied dynamic structural equation modeling (DSEM) to parse both within-person state-level fluctuations including autoregressive and cross-lagged associations (i.e., whether participants’ level at time t was significantly predicted by their level of the same or different construct, respectively, at time t -1) from stable between-person trait-like differences (Asparouhov et al., 2018 ; Hamaker et al., 2018 ). Given the exploratory nature of this study, we pose no specific a priori hypotheses.
Method Participants and Procedure Adolescents living in a western Canadian province were recruited between April 2019 and October 2020 to participate in an online study examining their psychological and behavioral adjustment (see Xu & Zheng, 2022 ). A total of 99 participants (12–17 years old, M = 14.60, SD = 1.76, 55.8% female, 51.5% white, 23.2% Asian, 8.1% multiracial, 4.0% Hispanic/Latinx, 2.0% Black, 6.1% Other) completed a baseline survey and at least one day of the 30-day daily diary surveys (2,108 total observations, M = 21.72 days, Range = 1–30, SD = 7.80). Among the adolescents, 81.1% reported living with both biological parents, 17.8% reported living with one biological parent, and 1.1% reported living with someone other than a biological parent. Most participants’ parents reported a personal annual income within the average (CAD $65,700) and median ($51,600) personal total incomes for groups of 25–54-year-olds in this Canadian province (17.2% had a personal annual income below $35,000, 9.1% were between $35,000–$45,000, 12.1% were between $45,000–$55,000, 17.2% were between $55,000–$65,000, and 38.4% had an annual income above $65,000). The research ethics committee at the University of Alberta approved the procedure and instruments for this study. Survey instruments were developed and administered via email through RedCap. Participants were recruited through newsletters, social media, and flyers posted or circulated in the western Canadian province. All adolescents were eligible for inclusion in the study. Interested participants were asked to contact the research team and were provided with information for the study. Participants received an online baseline survey (~45 min to complete) following the obtainment of assent online. Participants’ parents provided consent online for their children to participate in this study. Daily surveys (~10 min to complete) were sent out five days after completion of the baseline survey for 30 consecutive days. The daily survey was sent out at 5 pm each day and adolescents were asked to fill out the survey before going to sleep that night. Participants received a $45 e-gift card of their choosing as compensation for their participation in the baseline and daily surveys. Measures Callous-unemotional Traits Daily callous-unemotional traits were assessed with a shortened 12-item version of the 24-item Inventory of Callous Unemotional Traits (ICU; Hawes et al., 2014b ; Zheng et al., 2021 ; see Table 1 ). Each item is scored on a 4-point Likert scale (0 = not at all true , 1 = somewhat true , 2 = mostly true , 3 = definitely true ) in response to the stem “indicate how well the following statements described you today.” This ICU scale includes two subscales: callousness (6 items; ordinal ω w = 0.48, ordinal ω b = 0.91) and uncaring (5 items; ordinal ω w = 0.71, ordinal ω b = 0.91). Items that needed to be reverse-coded were done so prior to the analysis. Multilevel Confirmatory Factor Analyses (MLCFA) with the weighted least squares means and variances adjusted estimator demonstrated that the 2-factor model provided satisfactory and better model fit at within- and between-levels (CFI = 0.97, TLI = 0.97, RMSEA = 0.02, SRMR within(w) = 0.08, SRMR between(b) = 0.04), compared to the 1-factor model at both levels (CFI = 0.87, TLI = 0.83, RMSEA = 0.04, SRMR w = 0.15, SRMR b = 0.07). The estimated averaged standardized factor loading for the callousness factor was 0.60 and 0.84 at within- and between-levels, and for the uncaring factor was 0.70 and 0.80 at within- and between-levels. MLCFA using the maximum likelihood estimator with robust standard errors (MLR) demonstrated that the 2-factor model at within- and between-levels provided better model fit (Δ scaling correction = 3.68, Δχ 2 = 163.03, Δdfs = 2.00, p < 0.001) compared to the 1-factor model at both levels. For the first aim, all 12-items were used. However, in line with several studies that have found that item 3 ( does not show emotions ), the only item left from the original unemotional subscale, exhibits low factor loading (Colins et al., 2016 ; Zheng et al., 2021 ), this item was not included in calculations of subscales for the second aim. Positive and Negative Affect Positive (5 items; e.g., active , attentive ) and negative (5 items; e.g., ashamed , nervous ) affect were assessed daily (i.e., “indicate to what extent you have felt the way as described by each of the following words today”) with the short-form Positive and Negative Affect Schedule (PANAS-SF; Cooke et al., 2022 ; Thompson, 2007 ) on a 5-point Likert scale (1 = very slightly or not at all 2 = a little , 3 = moderately , 4 = quite a bit , 5 = extremely ). MLCFA model fit for the 2-factor model are as follows: CFI = 0.91, TLI = 0.88, RMSEA = 0.04, SRMR w = 0.13, SRMR b = 0.14. The estimated averaged standardized factor loading for the positive affect factor was 0.58 and 0.71 at within- and between-levels, and for the negative affect factor was 0.70 and 0.90 at within- and between-levels. Items were averaged within days with higher scores indicative of higher levels of positive (ω w = 0.63, ω b = 0.88, intra-class correlation [ICC] = 0.71) and negative (ω w = 0.74, ω b = 0.94, ICC = 0.61) affect. Emotional and Conduct Problems Participants reported their daily emotional and conduct problems using 5 items each from the emotional problems (e.g., I worry a lot ) and conduct problems (e.g., I take things that are not mine from home, school, or elsewhere ) subscales of the Strengths and Difficulties Questionnaire (Goodman et al., 1998 ). Items were rated on a 3-point Likert scale (0 = not true , 1 = somewhat true , 2 = certainly true ) and averaged within days. MLCFA model fit for the 2-factor model are as follows: CFI = 0.98, TLI = 0.98, RMSEA = 0.01, SRMR w = 0.10, SRMR b = 0.05. The estimated averaged standardized factor loading for the emotional problems factor was 0.57 and 0.88 at within- and between-levels, and for the conduct problems factor was 0.48 and 0.84 at within- and between-levels. Higher scores represented more emotional (ordinal ω w = 0.70, ordinal ω b = 0.94, ICC = 0.76) and conduct (ordinal ω w = 0.63, ordinal ω b = 0.92, ICC = 0.60) problems that day (i.e., “mark how the following items described you on the basis of how things have been for you today”). Analytic Approach In M plus version 8.6 (Muthén & Muthén, 2021 ), DSEM (Asparouhov et al., 2018 ) was conducted to examine both research aims (see Fig. 1 for DSEM schematic of aim 2). Combining time-series and structural equation modeling, this multilevel modeling approach tests temporal relations between measured variables by decomposing the data into within-person (Level 1) and between-person (Level 2) components (Asparouhov et al., 2018 ; Hamaker et al., 2018 ). At the within-person level, inertia , otherwise known as autoregression or carryover, is an estimate of the extent to which scores return to equilibrium (or persist) over measurement occasions, with higher scores indicating greater resistance to change (Kuppens et al., 2010 ). Regarding cross-lagged associations, augmentation describes positive cross-lagged associations, such that an increase in the level of one construct predicts an increase in another construct at a successive time point (Pe & Kuppens, 2012 ). By contrast, blunting characterizes negative cross-lagged associations, in which an increase in the level of one construct predicts a decrease in another construct at the next time point. In other words, these within-person components assess whether participants’ level at time t were significantly predicted by their level of the same or different construct, respectively, at time t -1. Contemporaneously, covariation specifies whether constructs at the same time point are associated with one another (Krone et al., 2018 ). Accordingly, at the within-person level (person-level centered), both autoregressive and cross-lagged parameters are estimated. Covariances among residuals between constructs within the same day after controlling for the previous day’s effect (i.e., autoregression and cross-lagged effects) were also estimated. At the between-person level, DSEM models means and (co)variances to examine between-person differences aggregated over the 30-day survey period (i.e., random intercepts). Bayesian Markov Chain Monte Carlo (MCMC) is used for estimation resulting in an entire distribution of possible values for each unknown parameter. Models were estimated using default priors and convergence was established by examining several diagnostic criteria including the Potential Scale Reduction (PSR) statistic, autocorrelation plots, and trace plots. Significance is determined by whether the 95% credible interval (CI) contains value 0 for each parameter. M plus code and output are supplied on the Open Science Framework: https://osf.io/n2rdf/?view_only=357ec650bfac4a63905e7ed34efd9255 .
Results Callous-unemotional Traits at the Item Level Descriptive statistics, ICCs, and within-level and between-level CU item correlations are shown in Supplementary Table S1 . ICC is the proportion of the total variance that is accounted for by stable or trait-like between-person differences plus measurement error (Hamaker et al., 2018 ). Accordingly, lower ICCs would indicate more within-person fluctuations. At the within-level, almost all items were significantly positively correlated with each other. At the between-level, items 2 ( feels bad or guilty when I have done something wrong ) and 3 ( does not show emotions ) showed few significant associations with other items and the highest ICC values (0.77 and 0.78, respectively). After controlling for the previous day’s autoregression and cross-lagged effects, item-level within-day residual correlations are shown in Table 2 . Similar to the within-level item correlations presented above, the majority of items were still positively correlated with each other (although magnitudes were smaller) after controlling for the previous day’s effect. Significant item-level autoregressive effects are shown in Fig. 2 (as indicated by curved arrows) and all item-level autoregressive estimates are described in Supplementary Table S2 . Except for items 6 ( does not care about doing things well ) and 10 ( shows no remorse when I have done something wrong ), all items showed significant autoregressive effects (βs = 0.08–0.23; 95% CI [0.02–0.18, 0.13–0.28]), suggesting that adolescents experiencing high levels of CU traits on the previous day were more likely to experience high levels the next day compared to their average level. More specifically, items 1 ( does not care who I hurt to get what I want ; β = 0.08; 95% CI [0.02, 0.14]), 4 ( concerned about the feelings of others ; β = 0.08; 95% CI [0.02, 0.15]), 8 ( apologizes to persons I have hurt ; β = 0.08; 95% CI [0.02, 0.14]), and 11 ( feelings of others are unimportant ; β = 0.08; 95% CI [0.02, 0.13]) showed the lowest levels of inertia and item 3 ( does not show emotions ; β = 0.23; 95% CI [0.18, 0.28]) revealed the largest inertia. Significant item-level cross-lagged effects are shown in Fig. 2 (as indicated by straight arrows) and all item-level cross-lagged estimates are described in Supplementary Table S3 . All items, except for item 10, demonstrated cross-lagged effects with at least one other item. However, it is notable that most cross-lagged associations involved only four items as outcomes: items 2 ( feels bad or guilty when I have done something wrong ; βs = -0.07–0.06 95% CIs [-0.12–0.01, -0.02–0.12]), 7 ( seems very cold and uncaring; βs = 0.06–0.10; 95% CIs [0.01–0.05, 0.11–0.14]), 9 ( tries not to hurt others’ feelings ; βs = -0.06–0.10; 95% CIs [-0.12–0.05, -0.00–0.14]), and 12 ( does things to make others feel good ; βs = 0.06–0.13; 95% CIs [0.00–0.07, 0.12–0.18]). These findings highlight several CU traits that adolescents reported as endorsing higher levels the next day compared to their average level. In addition, most items only showed two significant cross-lagged associations, with items 4, 8, and 9 displaying three cross-lagged effects. All significant cross-lagged associations were augmented, apart from four effects (i.e., 3 → 9, 7 → 2, 9 → 6, and 12 → 6), which revealed blunting effects. Between-level item correlations, means, and variances are shown in Table 2 . Overall, consistent with the between-level descriptive statistics, all items displayed significant positive associations with almost all other items, except items 2 ( feels bad or guilty when I have done something wrong ) and 3 ( does not show emotions ), which showed few significant associations. Callous-unemotional Traits and Associations with Subscales Descriptive statistics, ICCs, and within-level and between-level subscale correlations are shown in Supplementary Table S4 . Notably, at the within-level, callousness was significantly positively associated with all variables except positive affect, whereas uncaring was positively associated with negative affect and conduct problems, and negatively with positive affect. At the between-level, callousness was significantly positively associated with all variables and negatively associated with positive affect; uncaring was positively associated with conduct problems and negatively with positive affect. Also, emotional problems displayed the highest ICC (0.76) and uncaring had the lowest ICC (0.60). Table 3 shows within-day residual correlations after controlling for the previous day’s effects. Findings were similar to the within-level item correlations presented above. Significant autoregressive (curved arrows) and cross-lagged (straight arrows) effects are displayed in Fig. 3 (all estimates are shown in Supplementary Tables S5 and S6 , respectively). All variables showed significant autoregressive effects (βs = 0.13–0.35; 95% CIs [0.08–0.29, 0.18–0.41]). Thus, adolescents experiencing high levels of each variable on the previous day were more likely to experience high levels of each variable, respectively, the next day compared to their average level. Regarding cross-lagged effects, all significant associations were positive (i.e., augmentation). Specifically, adolescents who reported higher than their average levels of callousness also reported higher than their average levels of uncaring the next day (β = 0.08; 95% CI [0.03, 0.13]). Similar cross-day associations were observed between conduct problems and uncaring (β = 0.05; 95% CI [0.00, 0.10]), positive affect and callousness (β = 0.05; 95% CI [0.00, 0.10]), negative affect and emotional problems (β = 0.07; 95% CI [0.01, 0.13]), and, finally, emotional problems and negative affect (β = 0.09; 95% CI [0.03, 0.15]). Notably, uncaring did not show any significant cross-lagged effects. Between-level correlations, means, and variances are shown in Table 3 . Notably, adolescents with higher levels of callousness, on average, tended to also have higher average levels of uncaring, negative affect, conduct problems, and emotional problems and lower average levels of positive affect over the month. Those with higher average levels of uncaring also had higher average levels of conduct problems and lower average levels of positive affect over the month.
Discussion By collecting intensive longitudinal data and applying analytic approaches that elucidate temporal dynamics, we can expand our understanding of CU traits in a more ecologically valid way. Accordingly, this exploratory study examined whether adolescent CU traits manifest fluctuations across a 1-month observation window. Many CU traits items showed within-person autoregressive and cross-lagged links over the timeframe (i.e., participants’ levels at time t were significantly predicted by their levels at time t -1), and we described these effects using terminology from the affective dynamics literature. We also illustrated how daily CU traits were associated with daily affect and emotional and conduct problems. Overall, we observed several within- and between-person effects that serve as a starting point for further understanding of temporal dynamics and functional accounts of CU traits and related emotional and behavioral functioning. Callous-unemotional Traits in Daily Life In line with recent research calling for lower-order precision level analyses (Goulter et al., 2022 ; Kotov et al., 2021 ), we assessed CU traits at the item-level. Almost all items displayed within-person autoregressive effects (inertia). Of those items, two items that typically load onto callousness ( does not care who I hurt to get what I want , feelings of others are unimportant ) and two items that represent uncaring ( concerned about the feelings of others , apologizes to persons I have hurt ) had lower inertia and ICC scores (i.e., greater within-person fluctuations) relative to other items. Notably, these items all reflect CU traits related to behaviors directed toward others or others’ emotions or feelings. These items demonstrate greater variability likely because they tap interpersonal interactions with others, which could be less salient and stable relative to items regarding one’s own motivations and behaviors (e.g., does not care if I am in trouble , does not care about doing things well ). Conceptualizations of CU traits point to deficits in both affiliative capacity and emotional responding, and intensive longitudinal designs may prove especially useful in informing socio-emotional accounts of CU traits. Conversely, does not show emotions showed the largest inertia, ICC (i.e., the least within-person fluctuations), and highest between-level mean. This is a notable finding because this item is the only item retained from the 24-item ICU to the brief 12-item version presumably loading onto the unemotional subscale. Although this 12-item adaptation has been well-validated in child, adolescent, and young adult samples (e.g., Hawes et al., 2014b ; Zheng et al., 2021 ), others have emphasized the importance of unemotional items for understanding the broader CU construct (Ray & Frick, 2020 ). For example, one meta-analytic study found that although the unemotional subscale was only modestly associated with aggression ( r = 0.06), it was more highly associated with lower empathy ( r = -0.22; Cardinale & Marsh, 2020 ). These findings suggest that the unemotional scale may be more consequential for delineating “trait-like” impoverished emotion and empathy aspects of CU traits (Ray & Frick, 2020 ). Furthermore, both does not care about doing things well and shows no remorse when I have done something wrong did not show significant autoregressive effects. These results are somewhat expected as it is unlikely that our community sample committed transgressions every day, thus necessitating further research with higher-risk and clinical samples. A greater understanding of the short-term fluctuations in CU traits can also help inform ethical considerations in this field regarding labelling, stigmatization, and nomenclature. As a construct, CU traits was initially developed by downwardly extending the affective component of adult psychopathy to children and adolescents. Although definitions of psychopathy typically also comprise interpersonal and impulsive/antisocial dimensions, an extensive child and adult literature has revealed that the affective or CU dimension tends to distinguish those individuals with distinct neurocognitive correlates, as well as heightened risk for persistent antisocial behavior (Frick et al., 2014 ). However, there are reasonable concerns regarding the stigmatization of labelling children and adolescents with the term CU “traits” (Prasad & Kimonis, 2018 ). Several scholars have accordingly advocated for alternative language, such as CU behaviors , features , or symptoms , particularly in younger children when these characteristics may be less stable (e.g., Schuberth et al., 2019 ; Waller & Hyde, 2017 ). Others have made clear at the outset that although they use the term CU “traits” in young samples, they are not suggesting that these indicators are immutable (e.g., Fleming et al., 2022 ). As well-stated by Kaurin et al. ( 2022 , p. 23) “A stricter focus on dysregulated interpersonal and affective processes in daily life [therefore] may also reform the professional and public view on personality pathology, which is dominated by perceptions of destiny rather than manageable risk.” Overall, our item-level findings provide preliminary granular and temporal evidence of within-person “state” fluctuations of CU traits in real-life. Daily Callous-unemotional Traits and Emotional and Behavioral Functioning We also evaluated whether daily CU traits, including callousness and uncaring, were associated with several indicators of emotional and behavioral functioning. Both callousness and conduct problems augmented uncaring over time. One explanation for these findings may relate to how adolescents with elevated callousness appraise certain situations. It may be the case that adolescents high on callousness with a cognitive predisposition underpinned by fearlessness and deficits in emotional processing interpret certain situational features (e.g., negative interactions with parents or peers) in a way that results in higher levels of uncaring. Also noteworthy, uncaring did not predict any other construct. Structural differences in the ICU have been suggested to be a spurious by-product of method variance (Ray & Frick, 2020 ); however, our findings contribute some evidence pointing to meaningful differences between CU subscales. Taken together, these findings suggest that uncaring may be better conceptualized as an outcome rather than an antecedent, and callousness could represent a critical treatment target. An interesting finding was that positive affect was linked with callousness. Previous research found that callousness at baseline negatively predicted positive affect assessed six times per day across a 1-week period (Vize et al., 2022 ); however, callousness was assessed with a personality measure only at baseline. One explanation for our findings may be due to the specific measure of positive affect assessing high valence constructs, including being active, inspired, and determined (Cooke et al., 2022 ; Thompson, 2007 ). It may be the case that callousness also designates resolute individuals characterized by ambition. This notion goes against the DSM-5 “with Limited Prosocial Emotions” criterion specifying an absence of interest in school or work performance (American Psychiatric Association, 2013 ). However, the current uncaring subscale from the 12-item ICU is operationalized as a lack of concern for others (not activities). In addition, echoing Vize et al. ( 2022 ), such traits may be strongly associated with extroversion, which is linked to heightened positive affect (Watson et al., 2015 )–although these findings are derived from adult samples. Future studies should consider a broader range of positive affect items to replicate and extend current findings, as well as further research on personality in adolescent samples. Strengths and Limitations Notable strengths of the present study include the naturalistic 30-day diary design, which advances understanding of CU traits under higher ecological validity relative to traditional longitudinal studies. In addition, we applied an analytic approach that disentangles within-person fluctuations from between-person differences. We also evaluated whether daily CU traits were associated with multiple forms of daily emotional and behavioral functioning. However, current findings should be interpreted within the context of some limitations. First, the current sample was primarily comprised of community adolescents from families with higher annual income than the provincial median income. Participants also endorsed lower levels of CU traits ( M = 6.88; SD = 4.26) relative to other studies using the 12-item ICU scale. For example, Colins et al. ( 2016 ) found higher levels of CU traits in a sample of justice-involved adolescent girls ( M = 9.14; SD = 5.88). Thus, our results may not generalize to other populations, and as highlighted, it would be particularly important to examine daily CU traits in higher-risk or clinical samples. Second, all data are solely based on adolescent self-reports–we did not collect data from other informants (e.g., parents) on adolescent daily CU traits. Past research has revealed distinct associations between CU traits and conduct problems across adolescent- and parent-reports (e.g., Goulter & Moretti, 2021 ). Meta-analytic work has also shown less discrepancies in cross-informant correspondence when informants are reporting on observable behaviors (e.g., externalizing vs. internalizing) or when the behaviors are measured within the same context (De Los Reyes et al., 2015 ). Regarding CU traits, it may be more difficult for other informants to report on CU traits given these traits are not necessarily readily observable. Third, our sample was relatively small, although simulation studies have shown that sufficient power can be achieved for DSEM with a sample size of 100 and a minimum of 25 measured occasions (e.g., Schultzberg & Muthén, 2018 ). In particular, power at Level 2 is relatively small compared to conventional longitudinal studies, and thus, these between-person effects should be interpreted with caution. However, with over 2,000 observations at the daily level, power at Level 1 is sufficient for the current approach. Because of our sample size, we were not able to optimally investigate gender differences–an important direction for future research given most studies examining CU traits have relied predominantly on male samples. Future Directions and Implications By decomposing “trait” and “state”-like aspects of CU traits, current findings inform understanding of fluctuations in CU traits and lay a foundation for future intensive longitudinal designs to test critical questions in the field. For instance, adolescence represents a sensitive period marked by distinct change including within neural systems involved in social, emotional, and motivational processes (Dahl et al., 2018 ). Because of this, studies testing daily dynamics among adolescents may be particularly well-suited to reveal important information regarding socio-behavioral functioning. Atypical levels of CU traits have also been observed in samples as young as preschool age (Kimonis et al., 2016 ), and developmental models describe early childhood as the period in which the emergence of conscience and empathy typically occurs (Kochanska, 1993 ). Thus, future research should also investigate daily CU traits in younger samples. Ideally, CU traits would be examined at multiple developmental stages across multiple timescales, and measurement burst designs that combine micro and macro longitudinal methods could identify changes in short-term CU dynamics and their relations with long-term outcomes. Future research should also apply multiple measurement approaches (e.g., self-report, observational, psychophysiological) to better characterize the emotional components of daily experiences–a particularly important avenue in CU traits research given the distinct psychophysiological profiles among these samples (Fanti et al., 2019 ). It will also be crucial to investigate the role of specific situational features in CU traits fluctuations and employ different intensive longitudinal methods to test these relations. For example, the present study employed a time-based sampling scheme prompting participants at 5 pm each day. By contrast, other research may use a variable-interval scheme generating prompts at random throughout the day to further enhance ecological validity (Russell & Gajos, 2020 ). Future studies may also want to consider event contingent designs, which could be useful for examining negative interactions with parents, peers, or the broader environment. Theoretical and empirical works highlight the role of warm and supportive parent–child relationships in promoting empathy development and eliciting positive affect (Kochanska, 1993 ). Intensive longitudinal methods with parent–child dyads (e.g., Xu & Zheng, 2022 ) have the potential to further advance understanding of familial relations in the development of CU traits. Particularly, parenting behaviors, including parental warmth and harsh discipline, have been linked to child CU traits (Goulter et al., 2020 ; Waller et al., 2013 ; Zheng et al., 2017 ), and recent daily diary studies have demonstrated substantial daily fluctuations in parental warmth (Xu & Zheng, 2023 ). Thus, examining how daily parenting behaviors are associated with CU traits may provide modifiable targets in daily family processes for future treatment efforts. In adolescence, research may want to examine the role of peer affiliation and interactions in daily CU traits fluctuations. Although the present study assessed negative affect, a greater understanding of the dynamics of stress and associated stressful experiences would also be particularly informative.
Conclusion In this exploratory study, we observed meaningful fluctuations of CU traits at both item- and subscale-levels. These findings suggest that CU traits may be more susceptible to change and malleable than previously thought. In addition, callousness and uncaring subscales demonstrated distinct cross-day dynamics in relation to emotional and behavioral problems. Here, we have laid a foundation for future research focus applying intensive longitudinal methods to advance understanding of CU traits in daily life. Further functional information gained from these methods can potentially inform intervention efforts by offering modifiable targets situated in daily lives to promote personalized and just-in-time interventions. By conducting sampling in real-time in real-world situations, we can further inform personalized models of treatment targeting daily processes.
Intensive longitudinal methods (e.g., daily diary) inform understanding of dynamic processes by parsing within-person state-like fluctuations from stable between-person trait-like differences. In this exploratory study, we investigated whether self-reported callous-unemotional (CU) traits (callousness, uncaring) demonstrated daily fluctuations, as well as whether daily CU traits were associated with multiple forms of daily emotional and behavioral functioning. A sample of 99 adolescents (55.8% female; M age = 14.60 years) provided baseline information and completed a naturalistic 30-day diary reporting on CU traits, positive and negative affect, and emotional and conduct problems in their daily lives. Dynamic structural equation modeling revealed that many CU traits items showed within-person autoregressive and cross-lagged links; however, there was substantial between-person variation in within-person fluctuations across items. At the subscale level, cross-day associations were observed between callousness and uncaring, conduct problems and uncaring, positive affect and callousness, negative affect and emotional problems, and emotional problems and negative affect. By harnessing intensive longitudinal data, our findings provide preliminary state-level evidence of CU traits, as well as functional information with regards to CU traits and emotional and behavioral problems in daily life. We consider the implications of our findings in terms of informing future CU traits intensive longitudinal evaluations. Supplementary Information The online version contains supplementary material available at 10.1007/s10802-023-01077-6. Keywords
Supplementary Information Below is the link to the electronic supplementary material.
Acknowledgements We thank all the participating parents, adolescents, Elk Island and St. Albert public schools, and our research assistants. Study data were collected and managed using REDCap electronic data capture tools hosted and supported by the Women and Children’s Health Research Institute at the University of Alberta. Author Contribution Goulter, N: Conceptualization, Writing-Original Draft, Writing-Review/Editing; Cooke, EM: Formal Analysis, Writing-Original Draft, Writing-Review/Editing; Zheng, Y: Conceptualization, Funding Acquisition, Methodology, Writing-Review/Editing. Funding This research was supported partly with funding from the Social Sciences and Humanities Research Council (IDG 430-2018-00317 and 409-2020-00080) and Natural Sciences and Engineering Research Council (RGPIN-2020-04458 and DGECR-2020-00077) of Canada. EC was supported by a Mitacs Accelerate grant (IT18227) awarded to YZ. Data Availability Data are available by emailing the last author. Code and output are available on the Open Science Framework: https://osf.io/n2rdf/?view_only=357ec650bfac4a63905e7ed34efd9255 . Compliance with Ethical Standards Ethical Approval The study was approved by the institutional research ethics committee and was performed in accordance with the ethical standards as laid down in the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. Informed Consent Informed consent and assent for study participation and research publication was obtained from all individual participants included in the study. Conflicts of Interest The authors have no relevant or non-financial interests to disclose.
CC BY
no
2024-01-15 23:42:02
Res Child Adolesc Psychopathol. 2024 May 30; 52(1):51-63
oa_package/3c/1a/PMC10787886.tar.gz
PMC10787887
0
Introduction Social learning can be defined “as a change in behavior that follows the observation of another (typically a conspecific) perform a similar behavior, the products of the behavior, or even the products alone” (Zentall 2012 : 114). Which type of information is acquired through observation may differ strongly and is reflected in so-called social learning mechanisms. Motivational factors such as social facilitation (Zajonc 1965 ) promote the acquisition of information or change in behaviour by the observer via the mere presence of a demonstrator (Zentall 2001 ). Perceptual factors such as local (Roberts 1941 ) or stimulus enhancement (Galef 1988 ) can facilitate information acquisition by drawing the attention of the observer to either the location or stimulus of importance (Zentall 2001 ). In emulation (Tomasello 1998 ) the results of a demonstrated behaviour affect the observer who may strive to generate the same effect on the environment/objects as the demonstration did, without necessarily understanding the actions or reproducing the behaviour. Imitation is a specific learning mechanism, defined by a high degree of copying fidelity/response matching (Whiten and Ham 1992 ; Whiten et al. 2009 ), and cannot be explained by motivational, perceptual, or attentional factors alone (Zentall 2012 ), for a comprehensive overview of social learning mechanisms see Hoppitt and Laland ( 2013 ). Social learning is taxonomically widespread, ranging from insects to birds and mammals, possibly because it is a cost-effective way of acquiring information. Yet, social learning is not always advantageous (Giraldeau et al. 2002 ; Garcia-Nisa et al. 2023 ) and different social learning mechanisms may have different thresholds in this respect. Great apes, for instance, seem to be less prone to show imitation than other forms of social learning like emulation (Horner and Whiten 2005 ; Tennie et al. 2006 ; Clay and Tennie 2018 ). On the other hand, Marmosets ( Callithrix jacchus ) (Bugnyar and Huber 1997 ; Voelkl and Huber 2000 , 2007 ) and dogs ( Canis familiaris ) (Huber et al. 2009 , 2018 , 2020 ) show high-fidelity imitation despite not being closely related to humans ( Homo imitans as proposed by Meltzoff 1988 ). This suggests that high-fidelity imitation may be driven by natural ecology and social structure rather than phylogenetic relatedness. In fact, experimental evidence has illustrated that various avian species, namely budgerigars ( Melopsittacus undulates ) (Dawson and Foss 1965 ; Heyes and Saggerson 2002 ), European starlings ( Sturnus vulgaris ) (Fawcett et al. 2002 ), Japanese Quail ( Coturnix japonica ) (Akins and Zentall 1998 ; Akins et al. 2002 ), Common Ravens ( Corvus corax ) (Loretto et al. 2020 ) and Pigeons ( Columba livia ) (Nguyen et al. 2005 ) show (simple forms) of motor imitation. Parrots are renowned for their technical intelligence, vocal mimicry and social learning capacities (Pepperberg and Funk 1990 ; Huber et al. 2001 ; Funk 2002 ; Huber and Gajdon 2006 ; Werdenich and Huber 2006 ; Auersperg et al. 2009 , 2011 , 2012 , 2014 ; Miyata et al. 2011 ; Goodman et al. 2018 ; Klump et al. 2021 ; Smith et al. 2022 ). Yet surprisingly few parrot species have been tested on their motor imitation skills (budgerigars: Dawson and Foss 1965 ; Galef et al. 1986 ; Heyes and Saggerson 2002 ; grey parrots ( Psittacus erithacus ): Moore 1992 ; kea ( Nestor notabilis ): Huber et al. 2001 ; Suwandschieff et al. 2023 ; Goffin cockatoos ( Cacatua goffiniana ): Auersperg et al. 2012 ), revealing mixed results. Whereas most studies find evidence for motor imitation, the studies on kea remained inconclusive. Kea ( Nestor notabilis ) possess well-developed technical skills (Huber and Gajdon 2006 ), have long lifespans with multiple reproductive cycles, extended juvenile periods accompanied by considerable in-group tolerance, are highly neophilic and exploratory (Diamond and Bond 1999 ). They also have a very large number of documented food sources (Brejaart 1988 ; Clarke 1971 ; O’Donnell and Dilks 1994 ) many of which need to be extracted, which strongly suggest transfer of knowledge between individuals. All these characteristics facilitate the development of social learning (Gajdon et al. 2004 ), yet experimental evidence for imitation in this species is still missing. Therefore, we tested kea, for their social learning skills in a demonstrated sequence task. Specifically, we aimed at exploring kea’s imitative social learning capacities. We hypothesised that when confronted with a relatively complex two-step task, kea would pay attention to, and copy the behaviour of, a skilled conspecific. We thus predicted that observers would preferentially use the demonstrated opening side, sequence and colour whereas non-observing control individuals would apply trial-and-error learning to solve the task.
Method Subjects Eighteen kea from the Haidlhof Research Station (Bad Vöslau) participated in this study. All individuals were group-housed in an outdoor aviary equipped with perches, nesting areas, ponds, and various enrichment. All birds were fed three times a day, had access to water ad libitum and were not food deprived for testing. All individuals had prior experience with experimental testing and participated on a voluntary basis in the task. The testing compartment at the Haidlhof Research Station can be visually separated from the rest of the aviary and can be further divided into two different areas. The individuals were assigned to test groups of three and five individuals and two control groups of five individuals each. The distribution was sex and age balanced, for details see Table 1 of the supplementary material. Apparatus A rectangular box (44 × 18 × 18 cm) with two aluminium sliding lids, two pins, two strings and two rings served as the test box, see Fig. 1 . The test box was designed to provide a sequence function, requiring the subjects to pull a pin on top of the box, to then be able to open the opposite sliding lid by pulling a ring attached at the side of the box. The adjacent sliding lid remained locked. Hence, the only solving sequences for the test box were left pin–right ring or right pin–left ring. Electronics were added inside the box to provide the sequence function. If no pin or both pins were pulled the mechanism locked both lids and no rewards could be retrieved. Manual locking keys were added to provide a full and partial locking function for demonstrator training and demonstration sessions. The box was divided into equally sized reward sections underneath the sliding lids and a GoPro fixture was attached at the base. The pins were coloured in red and yellow, which is on the preferred spectrum for kea, and the strings and rings in green and blue (less preferred spectrum) respectively (Weser and Ross 2013 ). Each side and sequence had one preferred and one less preferred colour pairing, minimising the potential for a side bias based on colour preference alone. To increase salience, only those parts that needed direct manipulation were coloured, i.e., the pins, strings, and rings. Procedure To minimize the effects of individual learning over social learning, all test sessions consisted of one trial only. Each session/trial was terminated either after successful reward retrieval or a maximum of two minutes of exploration—i.e. pulling and touching different ring-pin combinations without opening success. A stopwatch was used to track the two minutes and all sessions were terminated by removing the individuals from the test compartment once either criterion (removal response or time-out) was met. Subjects from test groups received two experimental phases, whereas subjects from control groups received three phases (see below). In all phases, both sides of the box were baited and each sequence was rewarded. However, removing a pin and pulling on the wrong ring, pulling both pins or pulling on the rings without having removed the respective pin did not lead to any reward (as the doors were locked). Altogether four sessions (of one trial each) were tested on three consecutive days. On the first day, phase one (consisting of only one session) was tested in the morning and the first session of phase two (consisting of three sessions) in the afternoon. On the second and third day, each individual only received a morning test session of session two and three of phase two respectively. The experimenter wore mirrored sunglasses (as has been applied before by Bastos and Taylor 2020 ; Suwandschieff et al. 2023 ) and remained silent during testing (excluding direct commands such as “enter”, “exit” etc.) to avoid unintentional cueing. Phase 1: Forced failure task to all birds Pins were removed on this occasion and the test box could not be opened. All individuals (of all experimental groups) received access to the test box and were allowed to try and open it for two minutes. As the pins were missing all individuals were forced to fail this task. After two minutes passed the individual exited the testing compartment and the session was completed. This phase was introduced to prime individuals on the non-functioning of the task, focusing the attention towards the essential pins and increasing the motivation to follow a demonstration. Phase 2: Non-demonstrated task Control Group (CG) Control group individuals were allowed to try and solve the task by trial-and-error for three sessions (on three consecutive days). Phase 2: Demonstrated task Test Groups (TG) Test group individuals received three demonstration sessions of three trials each followed by three test sessions of one trial each (test following demonstration session in direct succession) on three consecutive days. Side/sequence and demonstrator assignment were counterbalanced across the two groups (see Table 1 of the supplementary material). Phase 3: Demonstrated task Control Test (CGTest) In the third phase, former control group birds (minus the two demonstrator birds) were randomly assigned to the two demonstrators and were tested again, this time as observer individuals. The test setup was identical to the phase two demonstrated task of the test group. Data scoring and analysis All experiments were videotaped from two sides, behind the observation compartment and directly above the test box (GoPro) within the test compartment, for the exact setup see Fig. 1 of the supplementary material. All GoPro videos of the test sessions were scored/coded with Solomon Coder (version beta 19.08.02) and one independent rater (blind to the study) scored 10% of all videos. Interobserver reliability was tested with Cohen’s Kappa for the categorical ( k = 0.93) and Intraclass Correlation Coefficient for the numerical data (ICC = 0.711, approach duration; ICC = 0.999, response duration; ICC = 1, solving latency), for more information see the supplementary material. A total of 74 sessions were analysed. Two individuals did not participate in two sessions and one session respectively, all other individuals completed all three of their test sessions. A binomial test was applied to compare the success rate of the test versus the control group, using the proportion of successful control birds as the baseline chance level of solving the setup spontaneously without demonstration. All other analyses were strictly descriptive.
Results After all subjects had experienced a non-functional apparatus (phase 1), they were allowed to engage in the task with or without prior demonstration of a skilled conspecific (phase 2). In sum, four individuals from the test group (out of a total of 8) successfully solved the task and two individuals from the control group (out of a total of 10). The exact binomial test resulted in a nearly significant trend between the test and control group ( x = 4, n = 8, p = 2/10; p-value = 0.056). The control group individuals participated in an additional round of testing (phase 3), in which they received a demonstration prior to getting access to the task again. Four individuals who had failed to solve the task without receiving a demonstration successfully solved the task after receiving a demonstration. In contrast, one of the individuals from the control test group, who had solved the task without receiving a demonstration (in the control group round), did not solve the task after receiving a demonstration (in the control test group round). Assessing the birds’ performance after receiving a demonstration (test group plus control test group) versus without a demonstration (control group), the exact binomial test revealed a significant difference between the groups with the test groups performing much better on average than the control group ( x = 8, n = 16, p = 2/10; p-value = 0.007). When comparing the within-subject design of the control group versus the control test group (repeated measure) twice as many individuals successfully solved the task after having seen a demonstration ( x = 4, n = 8, p = 2/10; p-value = 0.056). Half of the successful individuals (4/8) that saw a demonstration solved the task via the opposite (non-demonstrated) sequence and later reversed to the demonstrated sequence in subsequent sessions, see Table 1 .
Discussion We show that kea were able to solve a rather difficult two-step sequence task and that receiving a demonstration of a skilled conspecific had a positive effect on solving success. However, successful birds showed a high variation in their response topography and often abandoned faithfully copying the task in favour of exploration. This is particularly interesting in the case of John, who followed the demonstrated sequence twice and then generalised in the other direction. The effect of demonstration becomes particularly clear when combining the results from phase two and three. Eight out of 10 subjects that came to solve the task did so after a demonstration. That kea can profit from social learning is in line with previous studies (Huber et al. 2001 ; Huber 2002 ; Suwandschieff et al. 2023 ). However, the current study provides little evidence for motor imitation, despite various methodological differences from the other studies. For instance, previous experiments illustrated that the motivation to follow a demonstration was low. It was theorized that the complexity of the demonstrated actions could have contributed, as they likely were too simple to require a demonstration to solve the task (Suwandschieff et al. 2023 ), or too complex to follow the demonstration from afar (Huber et al. 2001 ) and reproduce with high fidelity. Therefore, the task was made more difficult than the basic two-choice task, while avoiding the complexity of the multi-lock box, and the forced failure in phase one was introduced to prime individuals on the task difficulty and to increase their motivation to follow a demonstration. Yet, these measures did not result in a higher copying fidelity than the other studies. On the one hand, half of the solvers used the opposite sequence to the demonstrators. On the other hand, solving success appeared not to reference solving consistency, as successful birds continued to explore the solving potential by applying different opening methods (see supplementary material for details). This variation in solving behaviour unfortunately made a statistical comparison not possible in terms of actual mechanisms, or quantifiable behavioural differences with previous studies. As all individuals, regardless of the experimental group, participated in the forced failure task, it is improbable that our results can be solely explained by social facilitation or local enhancement. Although we cannot rule out the possibility that the mere presence of a demonstrator motivated individuals to engage in the task, we have established that all individuals will do so even in the absence of a conspecific. Therefore, the presence of another individual does not appear to be the primary factor explaining our results. Additionally, since the location of the test box remained constant throughout the different phases, no additional information could have been gained from the demonstration. Consequently, this does not explain the discrepancy between the experimental groups. While the success rate of observer birds and the varied response patterns exhibited by successful individuals indicate emulation, we cannot dismiss the potential influence of stimulus enhancement. A test setup that clearly distinguishes these two mechanisms would have to be devised, i.e., one including a ghost control (Whiten and Ham 1992 ). Consistent with previous findings kea show high behavioural flexibility (Werdenich and Huber 2006 ; Auersperg et al. 2011 ; Laschober et al. 2021 ) and preferentially engage in exploratory behaviour, being more interested in potential affordances than feeding success (Diamond and Bond 1999 ; Huber et al. 2001 ; Smith et al. 2022 ; Suwandschieff et al. 2023 ). These results are in accordance with kea’s natural feeding strategies, as opportunistic group foragers, with kea paying close attention to what others feed on while engaging in individual manipulation strategies to obtain the resources (Diamond and Bond 1999 ). In addition, it corresponds with the characteristics of island-dwelling parrots, as described by Mettke-Hofmann and colleagues ( 2002 ) who found that island species spend significantly more time on exploratory behaviour especially in areas of seasonally fluctuating food availability. In conclusion, we find strong evidence that observing a conspecific opening an apparatus via two steps affected the solving success of observer kea. They thus profit from social learning, which aligns well with other studies on parrots and songbirds showing social information transmission and the spread of novel foraging techniques within captive groups and wild populations (e.g. Slagsvold and Wiebe 2011 ; Auersperg et al. 2014 ; Aplin et al. 2015 ; Klump et al. 2021 ). However, in our study, the response topography of solvers was variable and the copying fidelity was at a very low level, providing no indication of motor imitation over emulation or stimulus enhancement in kea. Therefore, our findings corroborate that kea display strong behavioural variability when attempting to solve a complex motor task. They also keep on exploring options after a successful solution and may rapidly shift solving strategies. Taken together, this makes kea a great model system to study behavioural flexibility but not so much for imitation.
Communicated by F. Bairlein. Social learning is an important aspect of dealing with the complexity of life. The transmission of information via the observation of other individuals is a cost-effective way of acquiring information. It is widespread within the animal kingdom but may differ strongly in the social learning mechanisms applied by the divergent species. Here we tested eighteen Kea ( Nestor notabilis ) parrots on their propensity to socially learn, and imitate, a demonstrated sequence of steps necessary to open an apparatus containing food. The demonstration by a conspecific led to more successful openings by observer birds, than control birds without a demonstration. However, all successful individuals showed great variation in their response topography and abandoned faithfully copying the task in favour of exploration. While the results provide little evidence for motor imitation they do provide further evidence for kea’s propensity towards exploration and rapidly shifting solving strategies, indicative of behavioural flexibility. Supplementary Information The online version contains supplementary material available at 10.1007/s10336-023-02127-y. Keas, Vögel der Vielseitigkeit. Kea-Papageien ( Nestor notabilis ) zeigen eine hohe Verhaltensflexibilität bei der Lösung einer demonstrierten Sequenzaufgabe. Soziales Lernen ist ein wichtiger Aspekt im Umgang mit der Komplexität des Lebens. Das Erlangen von Informationen durch die Beobachtung Anderer ist eine effiziente Möglichkeit der Informationsbeschaffung. Diese Art des Lernens ist im Tierreich weit verbreitet, kann sich jedoch hinsichtlich der unterschiedlichen sozialen Lernmechanismen, welcher sich die verschiedenen Arten bedienen, stark unterscheiden. Achtzehn Kea-Papageien ( Nestor notabilis ) wurden auf soziales Lernen, im Speziellen auf imitatives Lernen einer demonstrierten Abfolge von Schritten zum Öffnen eines mit Futter gefüllten Apparats, getestet. Die Resultate ergaben, dass eine Demonstration durch einen Artgenossen bei Beobachtervögeln zu erfolgreicheren Lösungsansätzen führten als bei Kontrollvögeln welche keine Demonstration erhielten. Allerdings zeigten alle erfolgreichen Individuen große Unterschiede in der Topographie ihrer Reaktion, der genauen Abfolge an Bewegungen. Ebenfalls gaben sie das getreue Kopieren der Aufgabe zugunsten von Erkundungs- und/oder individuellen Lösungsstrategien auf. Insgesamt zeigten unsere Ergebnisse, dass Keas starke Verhaltensvariabilität und Flexibilität zeigen, wenn sie eine komplexe motorische Aufgabe lösen. Damit stellen sie ein großartiges Modellsystem zur Untersuchung der Verhaltensflexibilität dar. Keywords Open access funding provided by Austrian Science Fund (FWF).
Supplementary Information Below is the link to the electronic supplementary material.
Acknowledgements We are thankful to Andràs Pèter for his help with devising the test box, Louise Mackie for her help with interobserver reliability check, Remco Folkertsma for his help with data analysis, the Austrian Science Fund (FWF) for funding the project (P 33507-B) and to the entire staff at the research station Haidlhof for their hard work and ongoing support. Author contributions ES: design, data collection, data curation and analysis, manuscript original draft, revisions, final manuscript; LH: conceptualisation, design, supervision, manuscript review; TB: conceptualisation, design, supervision, manuscript review; RS: conceptualisation, funding acquisition, design, supervision, manuscript review and editing. All authors read and approved the final manuscript. Funding Open access funding provided by Austrian Science Fund (FWF). This research was funded in whole, or in part, by the Austrian Science Fund (FWF) [P 33507-B]. For the purpose of open access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. Data availability All data generated or analysed during this study are included in this published article [supplementary material file: raw data]. Declarations Conflict of interest All authors have read the submitted version of this manuscript and declare no conflicts of interest. The authors have no relevant non-financial interests to disclose. Ethical approval The study is in accordance with the Good Scientific Practice guidelines and national legislation (ETK-10/11/2016) and has been approved by the institutional ethics and animal welfare committee at the University of Veterinary Medicine, Vienna.
CC BY
no
2024-01-15 23:42:02
J Ornithol. 2024 Dec 16; 165(1):49-55
oa_package/79/d5/PMC10787887.tar.gz
PMC10787888
38150097
Introduction Glucuronoyl esterases cleave ester-linked lignin-carbohydrate complexes (LCCs) in lignocellulosic biomass (Mosbech et al. 2018 ). In this way, their catalytic action assists in decomposing the plant cell wall matrix by removing some of the covalent interpolymeric linkages that sustain lignocellulose recalcitrance. As literature sees it today, ester-linked LCCs exist in vivo between the α-1,2-linked-D-glucuronoyl substitutions of glucuronoxylan and the aliphatic hydroxyl groups (Balakshin et al. 2011 ; Yuan et al. 2011 ). An important additional feature observed in many wood and cereal types of biomass, is the 4- O -methylation on said glucuronoyl, which creates a unique plant-based trademark for these aldouronic acids. We commonly agree on the carbohydrate side of the LCC, but the exact nature of the aliphatic alcohol on lignin remains a matter of debate (Giummarella et al. 2019 ; Sapouna & Lawoko 2021 ). In the most abundant lignin-substructures, two alcohol moieties present themselves as possible esterification points, namely the α- and the γ-positioned ones (Fig. 1 , structure 1 vs. structure 2), and whichever is most dominant in vivo is difficult to verify, because the occurrence is low, and because they prove difficult to isolate and annotate exactly. Ester bound acids tend to migrate (Puchart et al. 2020 ) and extensive sample handling in order to enrich these linkages may introduce artifacts. Hence, de novo ester synthesis in the plant cell may start in one position and migrate to a different one, as the cell wall matures (Li & Helm 1995 ). Glucuronoyl esterases of both fungal and bacterial origin has received much attention in research during the past ten to fifteen years since their discovery (Agger et al. 2023 ; Biely 2016 ; Larsbrink and Lo Leggio 2023 ). Activity has been demonstrated widely on both synthetic and natural substrates, and several crystal structures reveal the catalytic mechanism of these canonical α/β-hydrolases of the serine-esterase type (Baath et al. 2019 ; Charavgi et al. 2013 ; Ernst et al. 2020 ; Mazurkewich et al. 2019 ; Topakas et al. 2010 ). We know that fungal variants are generally more restricted in their substrate preferences than the bacterial enzymes, because most of the fungal ones are dependent on the 4- O -methyl-modification on the glucuronoyl moiety for recognition. Bacterial enzymes are broader in their substrate preferences in terms of the sugar-moiety, but it is still uncertain for all types of glucuronoyl esterases, if they have specificity towards the configuration of the lignin-alcohol and if so, which of the two possible variants they prefer. We have investigated the substrate specificity of a fungal glucuronoyl esterase from Cerrena unicolor ( Cu GE) by exploiting the fact that esterases often display transesterification capacities under the right reaction conditions. We hypothesize that the enzyme will preferably perform transesterification with the type of alcohol (α- or γ-positioned) that fits its substrate specificities best. We performed transesterifications with either methyl-glucuronate or methyl-4- O -methyl-glucuronate as donor substrates, and benzyl-alcohol or 3-phenyl-1-propanol as the alcohol-acceptors according to the four reactions outlined in Fig. 1 .
LC–MS method Reaction product profiles were analyzed by LC–MS. 5 μL of reaction mixture was injected onto a Hypercarb column (150 mm × 2.1 mm; 3μm, Thermo Fischer Scientific, Sunnyvale, CA, USA). The chromatography was performed on a Dionex UltiMate 3000 UPLC (Thermo Fischer Scientific, Sunnyvale, CA, USA) at 0.4 mL min −1 and 70 °C with a two-eluent system consisting of eluent A (acetonitrile) and eluent B (water). The elution was performed as follows (time indicated in min): 0–3, 0% A 100% B; 3–10, isocratic 40% A 60% B; 10–19, isocratic 0% A 100% B. Water was used as eluent B to avoid excessive spontaneous hydrolysis of either donor substrates or products on the LC-column. For the analysis of the transesterification of PhPrOH and Me-4- O -MeGlcA the chromatography was performed on the same system and condition of flow, temperature and eluent system but the elution was performed as follows (time indicated in min): 0–3, 0% A 100% B; 3–10, isocratic 70% A 30% B; 10–19, isocratic 0% A 100% B. The differences in elution system for this particular reaction was in order to elute the stronger retained transesterification product of this reaction compared to the other reactions within the same period. The HPLC was connected to an ESI-iontrap (model Amazon SL from Bruker Daltonics, Bremen, Germany) and the electrospray was operated in positive ultra scan mode using a target mass of 300 m/z. A scan range from 100 to 2000 m/z was selected and capillary voltage was set to 4.5 kV, end plate offset 0.5 kV, nebulizer pressure at 3.0 bar, dry gas flow at 12.0 L min −1 , and dry gas temperature at 280 °C.
Results and discussion Initial hydrolytic experiments with either Me-4- O -MeGlcA or Me-GlcA as substrate and in the absence of alcohol acceptors show substrate conversion as expected (Fig. 2 ), with a significantly higher level of conversion for the 4- O -methylated ester compared to the non-derivatized glucuronate ester within the first 100 min of reaction (Table 1 ). Enzyme concentrations were dosed to allow similar rates of catalysis in all experiments, independent of substrate, and consequently Cu GE was dosed 50 times less in the experiments where Me-4- O -MeGlcA was the substrate compared to the experiments with Me-GlcA as substrate. The fact that Cu GE prefers its substrates to have the 4- O -methyl-modification on the glucuronoyl is well-establised (d’Errico et al. 2015 ; Ernst et al. 2020 ; Monrad et al. 2018 ). As the 100 min reaction time progresses, it becomes evident that the hydrolytic reaction slows down for both substrates (Table 1 ), which again is not surprising as the affinity (K M ) for these synthetic model substrates is known to be relatively high (d’Errico et al. 2015 ). It is known, that glucuronoyl esterases prefer bulky alcohols and in that respect, the methyl-esters are poor substrates. Experiments with either benzyl-alcohol or 3-phenyl-1-propanol as the primary solvent and thereby dominating acceptor molecule (14% v/v water), show that the transesterification reaction occurring between Me-4- O -MeGlcA and PhPrOH (Fig. 1 , reaction 1) is significantly higher than any of the other combinations (Fig. 3 ). The reaction between Me-4- O -MeGlcA and BnzOH is second best (Fig. 1 , reaction 2) with about 3–4 fold less product compared to the PhPr-GlcA ester. Quantifications of the ester products are done relative to the standard of Bnz-GlcA dissolved either in BnzOH or PhPrOH depending on the relevant acceptor alcohol, and the calibrations indicate a certain level of ion suppression when PhPrOH is the solvent (Supplementary standards curves), hence leading to a potential underestimation in that respect. PhPrOH has similar retention time as the transesterification products. At the same time, compounds carrying the 4- O -methylation tends to ionize better, and therefore give stronger responses. In the reactions performed here, enzyme loadings vary according to formation of transesterification products, as the goal is to quantify transesterification products. Hence, at low enzyme loadings, hydrolysis is not observed for reactions containing Me-4- O -MeGlcA as donor (and relatively low water concentration), whereas increasing the enzyme concentration yields hydrolytic reactions in parallel to the transesterification (data not shown). Ultimately, the extent of hydrolysis is a competition for acceptor substrate, which appears favored in the case of the alternative alcohol when enzyme concentrations are low. Hydrolytic reactions are certainly present but becomes un-quantifiable at conditions where transesterification is dominating. Interestingly, the 4- O -methylation is a more important feature for catalysis than the nature of the acceptor-alcohol, since the Bnz-4- O -MeGlcA-ester forms faster than the PhPr-GlcA ester. A cautious estimation of the effect of the donor versus the effect of the acceptor demonstrates that the ratio between concentrations of products formed during the reaction time is higher when the donor is changed, compared to when the acceptor is changed (Fig. 4 ). Hence, the effect of changing the donor substrate between the 4- O -methylated and non-methylated counterparts is larger than the effect of changing the alcohol acceptor. This observation may not be valid for all glucuronoyl esterases, as it is common knowledge that not all GEs favor the 4- O -methylation. Recent QM/MM studies of the fungal Thermothelomyces thermophila glucuronoyl esterase ( Tt GE) show that the acylation step is the most energy demanding and hence rate-limiting step (Viegas et al. 2022 ), whereas similar studies with a bacterial Ot CE15A from Opitutus terrae show that the deacylation-step is rate-limiting (Zong et al. 2022 ). Without having investigated the energetic landscape of catalysis by Cu GE, the transesterification reactions we observe support the observation that the acylation-step is determining for reaction speed. However, it is well-known that the differences between fungal and bacterial GEs are quite large in terms of overall structure (Larsbrink and Lo Leggio 2023 ), and transesterification reactions of the kind presented here, may turn out different for bacterial GEs with respect to energetic fingerprints. The aim of this study is to investigate the preference for either an α- or a γ-positioned alcohol, and the results show that the γ-positioned alcohol is most favorable, yet the enzyme can perform the reaction with the α-alcohol. There are currently no univocal structural explanations for how the GEs interact with the alcohol moiety of the substrate. However, we have recently performed docking simulations with these exact two esters; Bnz-4- O -MeGlcA and PhPr-4- O -MeGlcA as ligands in Cu GE and guided by the sugar-moiety, which show steric obstacles when the α-benzyl constitutes the alcohol part of the ligand in contrary to the γ-linked ester (Agger et al. 2023 ). Exploiting the transesterification capacities of these enzymes potentially opens for other biotechnological applications than biomass deconstruction, such as functionalization of hemicellulose or smaller aldouronic acids with alcohols of different properties (hydrophobicity, other functional groups etc.). Future studies may explore the field of potential acceptors more broadly than here, and thereby determine the diversity of alcohols and properties that could be relevant to investigate further.
Results and discussion Initial hydrolytic experiments with either Me-4- O -MeGlcA or Me-GlcA as substrate and in the absence of alcohol acceptors show substrate conversion as expected (Fig. 2 ), with a significantly higher level of conversion for the 4- O -methylated ester compared to the non-derivatized glucuronate ester within the first 100 min of reaction (Table 1 ). Enzyme concentrations were dosed to allow similar rates of catalysis in all experiments, independent of substrate, and consequently Cu GE was dosed 50 times less in the experiments where Me-4- O -MeGlcA was the substrate compared to the experiments with Me-GlcA as substrate. The fact that Cu GE prefers its substrates to have the 4- O -methyl-modification on the glucuronoyl is well-establised (d’Errico et al. 2015 ; Ernst et al. 2020 ; Monrad et al. 2018 ). As the 100 min reaction time progresses, it becomes evident that the hydrolytic reaction slows down for both substrates (Table 1 ), which again is not surprising as the affinity (K M ) for these synthetic model substrates is known to be relatively high (d’Errico et al. 2015 ). It is known, that glucuronoyl esterases prefer bulky alcohols and in that respect, the methyl-esters are poor substrates. Experiments with either benzyl-alcohol or 3-phenyl-1-propanol as the primary solvent and thereby dominating acceptor molecule (14% v/v water), show that the transesterification reaction occurring between Me-4- O -MeGlcA and PhPrOH (Fig. 1 , reaction 1) is significantly higher than any of the other combinations (Fig. 3 ). The reaction between Me-4- O -MeGlcA and BnzOH is second best (Fig. 1 , reaction 2) with about 3–4 fold less product compared to the PhPr-GlcA ester. Quantifications of the ester products are done relative to the standard of Bnz-GlcA dissolved either in BnzOH or PhPrOH depending on the relevant acceptor alcohol, and the calibrations indicate a certain level of ion suppression when PhPrOH is the solvent (Supplementary standards curves), hence leading to a potential underestimation in that respect. PhPrOH has similar retention time as the transesterification products. At the same time, compounds carrying the 4- O -methylation tends to ionize better, and therefore give stronger responses. In the reactions performed here, enzyme loadings vary according to formation of transesterification products, as the goal is to quantify transesterification products. Hence, at low enzyme loadings, hydrolysis is not observed for reactions containing Me-4- O -MeGlcA as donor (and relatively low water concentration), whereas increasing the enzyme concentration yields hydrolytic reactions in parallel to the transesterification (data not shown). Ultimately, the extent of hydrolysis is a competition for acceptor substrate, which appears favored in the case of the alternative alcohol when enzyme concentrations are low. Hydrolytic reactions are certainly present but becomes un-quantifiable at conditions where transesterification is dominating. Interestingly, the 4- O -methylation is a more important feature for catalysis than the nature of the acceptor-alcohol, since the Bnz-4- O -MeGlcA-ester forms faster than the PhPr-GlcA ester. A cautious estimation of the effect of the donor versus the effect of the acceptor demonstrates that the ratio between concentrations of products formed during the reaction time is higher when the donor is changed, compared to when the acceptor is changed (Fig. 4 ). Hence, the effect of changing the donor substrate between the 4- O -methylated and non-methylated counterparts is larger than the effect of changing the alcohol acceptor. This observation may not be valid for all glucuronoyl esterases, as it is common knowledge that not all GEs favor the 4- O -methylation. Recent QM/MM studies of the fungal Thermothelomyces thermophila glucuronoyl esterase ( Tt GE) show that the acylation step is the most energy demanding and hence rate-limiting step (Viegas et al. 2022 ), whereas similar studies with a bacterial Ot CE15A from Opitutus terrae show that the deacylation-step is rate-limiting (Zong et al. 2022 ). Without having investigated the energetic landscape of catalysis by Cu GE, the transesterification reactions we observe support the observation that the acylation-step is determining for reaction speed. However, it is well-known that the differences between fungal and bacterial GEs are quite large in terms of overall structure (Larsbrink and Lo Leggio 2023 ), and transesterification reactions of the kind presented here, may turn out different for bacterial GEs with respect to energetic fingerprints. The aim of this study is to investigate the preference for either an α- or a γ-positioned alcohol, and the results show that the γ-positioned alcohol is most favorable, yet the enzyme can perform the reaction with the α-alcohol. There are currently no univocal structural explanations for how the GEs interact with the alcohol moiety of the substrate. However, we have recently performed docking simulations with these exact two esters; Bnz-4- O -MeGlcA and PhPr-4- O -MeGlcA as ligands in Cu GE and guided by the sugar-moiety, which show steric obstacles when the α-benzyl constitutes the alcohol part of the ligand in contrary to the γ-linked ester (Agger et al. 2023 ). Exploiting the transesterification capacities of these enzymes potentially opens for other biotechnological applications than biomass deconstruction, such as functionalization of hemicellulose or smaller aldouronic acids with alcohols of different properties (hydrophobicity, other functional groups etc.). Future studies may explore the field of potential acceptors more broadly than here, and thereby determine the diversity of alcohols and properties that could be relevant to investigate further.
Conclusion Cu GE prefers methyl-4- O -methyl-glucuronate as donor substrate and the γ-positioned acceptor according to the experiments conducted here, and evaluated based on yield and rate of product formation. It is important to emphasize that the results reported here illustrate the substrate preferences of Cu GE, which is not to be generalized for all GEs. These results exemplify a methodology for investigating substrate preferences, and future studies may include reverse hydrolytic reactions. We observe large differences in product formation rates during the first 100 min of reaction with the formation of the 4- O -methyl-glucuronoyl-3-propane-phenyl as fastest, and the results clearly indicate that Cu GE prefers the γ-positioned alcohol during transesterification. The most important factor for reaction though, continues to be the presence of the 4- O -methylation derivatization of the glucuronoyl moiety. These experiments demonstrate a methodology where the enzyme reveals its substrate preferences immediately by favoring product formation from the most suitable substrates. These results do not inform about the prevalence of either α- or γ-LCC esters in lignocellulosic biomass, but it certainly informs about the preferences of this particular enzyme. Given that enzymes evolve to tackle the linkages present in biomass, it is tempting to speculate that the γ-esters LCC is the more prevalent of the two types in biomass.
Purpose Glucuronoyl esterases (GE, family CE15) catalyse the cleavage of ester linkages in lignin-carbohydrate complexes (LCCs), and this study demonstrate how transesterification reactions with a fungal GE from Cerrena unicolor ( Cu GE) can reveal the enzyme’s preference for the alcohol-part of the ester-bond. Methods This alcohol-preference relates to where the ester-LCCs are located on the lignin molecule, and has consequences for how the enzymes potentially interact with lignin. It is unknown exactly what the enzymes prefer; either the α-benzyl or the γ-benzyl position. By providing the enzyme with a donor substrate (the methyl ester of either glucuronate or 4- O -methyl-glucuronate) and either one of two acceptor molecules (benzyl alcohol or 3-phenyl-1-propanol) we demonstrate that the enzyme can perform transesterification and it serves as a method for assessing the enzyme’s alcohol preferences. Conclusion Cu GE preferentially forms the γ-ester from the methyl ester of 4- O -methyl-glucuronate and 3-phenyl-1-propanol and the enzyme’s substrate preferences are primarily dictated by the presence of the 4- O -methylation on the glucuronoyl donor, and secondly on the type of alcohol. Supplementary Information The online version contains supplementary material available at 10.1007/s10529-023-03456-x. Keywords Open access funding provided by Technical University of Denmark
Materials and enzyme preparation Methyl 4- O -methyl-D-glucopyranosyluronate (Me-4-O-MeGlcA) was purchased from Institute of Chemistry, Slovak Academy of Sciences, Bratislava, Slovakia. Glucuronic acid methyl ester (Me-GlcA) and benzyl D-glucuronate (Bnz-GlcA) were purchased from Carbosynth. Benzyl alcohol (BnzOH), 3-phenyl-1-propanol (PhPrOH) and all other chemicals were purchased from Sigma. The construct containing the gene encoding for the CE15 glucuronoyl esterase from Cerrena unicolor ( Cu GE) was produced in P. pastoris as previously described (Mosbech et al. 2018 ). After fermentation the cells were separated by centrifugation and the fermentation broth (3 L) was sterile filtered and concentrated by ultrafiltration on a 10 kDa cut-off membrane to a final volume of approx. 80 mL. Cu GE was purified by affinity chromatography on an IMAC-column (HisTrap HP 5 mL column, GE Healthcare) using an Äkta Purifier 100 (GE Healthcare, Uppsala Sweden). Enzyme reactions The transesterification activity of Cu GE was tested in four different experimental setups using either Me-4- O -MeGlcA or Me-GlcA as donor substrates and benzyl alcohol (BnzOH) or 3-phenyl-1-propanol (PhPrOH) as acceptor substrates, allowing the assessment of the transesterification ability and preferences of Cu GE in either the α and γ position (Fig. 1 ). Transesterification with PhPrOH as acceptor molecule was performed in a 100 μL reaction mixture containing 0.45 mM of Me-4- O -MeGlcA or 0.48 mM of Me-GlcA (dissolved in 10 mM Na acetate buffer pH 6) and 6.32 M of PhPrOH in order to have an acceptor–donor ratio of approx. 14,000 and of 13,000 with Me-4- O -MeGlcA and Me-GlcA, respectively. The reaction was started in a HPLC vial by addition of 0.02 μM or 0.87 μM of Cu GE for the reaction with Me-4- O -MeGlcA and Me-GlcA, respectively. Transesterification with BnzOH as acceptor molecule was performed in a 100 μL reaction mixture containing 0.45 mM of Me-4- O -MeGlcA or 0.48 mM of Me-GlcA (dissolved in 10 mM Na acetate buffer pH 6) and 8.31 M of BnzOH in order to have an acceptor donor ratio of approx.. 18,500 and of 17,500 with Me-4- O -MeGlcA and Me-GlcA, respectively. The reaction was started in a HPLC vial by addition of 0.11 μM or 4.33 μM of Cu GE for the reaction with Me-4- O -MeGlcA and Me-GlcA, respectively. All transesterification reactions were run in triplicates. The reaction samples were placed in the UHPLC’s autosampler at 40 °C and the reaction evolution was followed directly on LC–MS by injecting 5 μL of the reaction mixture onto the column every 20 min followed by the LC–MS method described below. Control experiments containing 0.45 mM of Me-4- O -MeGlcA or 0.48 mM of Me-GlcA (dissolved in 10 mM Na acetate buffer pH 6) and 6.32 M of PhPrOH or 8.31 M of BnzOH and replacing the enzyme volume with an equal amount of 10 mM Na acetate buffer pH 6 were run to assess auto-hydrolysis during the reaction time. Chromatograms of all transesterification reactions and auto-hydrolytic control experiments are found in Supplementary Figs. S1-S8. Control experiments showing Cu GE’s hydrolytic activity were performed in 100 μL reaction mixture containing 0.45 mM of Me-4- O -MeGlcA or 0.48 mM of Me-GlcA dissolved in 10 mM Na acetate buffer pH 6. The reactions were started in the HPLC vial by the addition of 0.08 μM or 4.33 μM of Cu GE for the reaction with Me-4- O -MeGlcA and Me-GlcA, respectively. The hydrolytic reactions were performed in triplicates. Direct quantification results of substrate depletion are provided in Supplementary Table S1. Quantification method Quantification of all precursor ions was performed using Bruker TASQ software (Bruker Daltonics, Bremen, Germany). All ions were observed as [M + Na] + . Quantification of the Me-4- O -MeGlcA hydrolysis by Cu GE was performed by defining an Extracted Ion Chromatogram (EIC) of m/z 244.95 and m/z 467.03 (pseudo-double ion with one sodium), with a width of ± 0.5 and retention time 6.5 min ± 0.2 min, and reaction extend was quantified as substrate depletion relative to calibration with Me-4- O -MeGlcA. Quantification of the Me-GlcA hydrolysis by Cu GE was performed by defining an EIC of m/z 230.92 and m/z 439.07 (pseudo-double ion with one sodium), with a width of ± 0.5, retention time 3.0 min ± 0.2 min, and reaction extent was quantified as substrate depletion relative to calibration with Me-GlcA. Quantification of product formations of both Bnz-4- O -MeGlcA (EIC on m/z 321.08 with a width of ± 0.5, retention time 9.8 min) and Bnz-GlcA (EIC on m/z 307.04 and m/z 591.01 with a width of ± 0.5, retention time 8.0 min), were performed relative to a Bnz-GlcA standard dissolved in BnzOH (EIC on m/z 307.04 and m/z 591.01 with a width of ± 0.5 and retention time 8.0 min with a window of 0.2 min). Estimations of PhPr-4- O -MeGlcA (EIC on m/z 349.14, retention time 9.3 min) and PhPr-GlcA (EIC on m/z 335.12, retention time 7.7 min); products of Cu GE transesterification of PhPrOH with Me-4- O -MeGlcA or Me-GlcA, respectively, were performed using Bnz-GlcA diluted in PhPrOH as standard. Calibration curves was performed using 6–8 levels of concentrations and fitting the data with a quadratic curve for the Me-4- O -MeGlcA and Me-GlcA calibrations and with a linear curve for the Bnz-GlcA (Supplementary Figure S9). Supplementary Information Below is the link to the electronic supplementary material.
Author contributions Author VP performed the experiments and wrote parts of the paper. Author JWA designed the experiments and wrote the paper. Funding Open access funding provided by Technical University of Denmark. The work was funded by the Novo Nordisk Foundation “Emerging Investigator 2022 – Research within Industrial Biotechnology and Environmental Biotechnology” (grant number NNF22OC0074634) Lignin for the Future granted to JWA. Declarations Competing interests The author’s declare no competing interests.
CC BY
no
2024-01-15 23:42:02
Biotechnol Lett. 2024 Dec 27; 46(1):107-114
oa_package/ad/28/PMC10787888.tar.gz
PMC10787889
38217712
Introduction Adolescent and young adult cancer survivors (AYAs), those aged 15–39 years at initial cancer diagnosis, are considered a unique group with age-specific challenges, from cancer diagnosis until end of life [ 1 , 2 ]. This age range is, however, flexibly applied depending on the research question of interest, country, and health care system [ 3 ]. In the Netherlands, pediatric oncology is centralized and includes patients until 18 years of age at diagnosis. The Dutch AYA definition applies to all cancer patients initially diagnosed between 18 and 39 years. As the overall cancer incidence of AYAs has increased over the last decades and the 5-year relative survival is now exceeding 80%, a large part of this growing population will eventually become long-term survivors [ 4 ]. As a result of the cancer diagnosis and treatment, these AYAs are at increased risk of long-term (e.g., infertility) and late effects (e.g., secondary malignancies), and experience unmet (age-specific) needs related to finances and mental health for example [ 5 – 7 ]. AYAs are in a particularly exposed position for these risk factors due to their often invasive and long-lasting treatments [ 8 ], and being diagnosed during a complex phase of life, including many physical, emotional, and social transitions [ 9 ]. AYAs can have a long life ahead in which suffering from these long-term and late effects can have a significant impact on their health-related quality of life (HRQoL). HRQoL is defined by the survivor’s own perception of one’s health or well-being, including physical, mental, and social aspects [ 10 ]. Several studies have focused on the HRQoL issues among AYAs: literature shows that AYAs are at increased risk of fatigue [ 11 ], cognitive impairment [ 12 ], work and financial problems [ 12 , 13 ], psychological distress [ 14 ], and body image issues [ 15 ], which can result in a diminished HRQoL. In addition, studies described impacted physical health and functioning [ 16 – 19 ], and lower mental health [ 17 – 19 ] in AYAs compared to the general population/older cancer survivors [ 19 ]. In line, the systematic review of Quinn showed that AYAs are more likely to have impaired HRQoL compared to the general population, although QoL was difficult to measure due to their age-specific needs [ 20 ]. Many factors can independently impact HRQoL, yet they often co-occur in cancer patients [ 21 – 25 ]. To study the interconnectedness between these factors, a network analysis can be performed. It provides insight into the relationships among symptoms, risk factors, and protective factors. Network approaches involve the identification of symptoms and factors (network nodes) and the relations among them (positive or negative associations between nodes) [ 26 ]. Taking into account the dependence of factors, this type of analysis is more likely reflecting reality compared to focusing on these factors independently [ 27 ]. Although network analyses have been performed previously among cancer survivors in general [ 27 – 29 ], they are lacking among AYAs, especially when focusing on HRQoL outcomes. Gaining these insights is of importance to optimize supportive care and provide targeted interventions for this unique long-term surviving population. Within a large population-based sample of long-term AYAs, using an exploratory approach, we want to (1) assess HRQoL in long-term AYAs, (2) identify the most central nodes in a HRQoL network, and (3) determine how these nodes are linked to one another ((strengths of) interconnectedness).
Methods Study population and data collection Data of the population-based, cross-sectional SURVAYA study was used, which was approved by the Institutional Review Board (IRBd18122) and registered within clinical trial registration (NCT05379387). The study population is extensively described previously [ 30 ]. In short, the SURVAYA study was performed among AYAs (18–39 years old at time of initial cancer diagnosis) diagnosed with cancer between 1999 and 2015 (5–20 years post diagnosis at study invitation), treated at the Netherlands Cancer Institute or one of the University Medical Centers in the Netherlands, and registered within the Netherlands Cancer Registry (NCR). Survivors were invited to complete a one-time questionnaire within PROFILES (Patient Reported Outcomes Following Initial treatment and Long-term Evaluation of Survivorship) [ 31 ]. Measures Sociodemographic characteristics were obtained through a one-time questionnaire including age at time of questionnaire, sex at birth, marital status, and educational level. Clinical characteristics, obtained by the NCR, include tumor type, stage, primary treatment received, and time since diagnosis. Tumor type was classified according to the third International Classification of Diseases for Oncology (ICDO-3) [ 32 ]. Cancer stage was classified according to TNM or Ann Arbor Code (Hodgkin lymphoma and Non-Hodgkin lymphoma) [ 33 ]. The EORTC QLQ-SURV111 [ 34 ], a cancer core survivorship questionnaire that is currently being developed by the European Organization for Research and Treatment of Cancer (EORTC), was used for our network analysis. This questionnaire assesses long-term HRQoL outcomes, including physical, mental, and social HRQoL issues specifically relevant to cancer survivors. We selected 8 functioning scales (physical functioning, cognitive functioning, emotional functioning, role functioning, body image, symptom awareness, sexual functioning, and overall quality of life), 9 symptoms scales (fatigue, sleep problems, pain, social interference, health distress, negative health outlook, social isolation, symptom checklist, and sexual problems), and 3 single items (financial difficulties, worry cancer risk family, and treated differently). Scales and items measuring positive factors of HRQoL and items that were not applicable for all participants were excluded from the analysis. The rationale for this selection is that including these positive scales/items and optional items would result in difficulties with respectively interpreting network associations to intervene on and the development of a network structure. Participants scored the items on a 4-point Likert scale from 1 (not at all) to 4 (very much). Overall quality of life scores ranged from 1 (very poor) to 7 (excellent). All scales and single items were linearly transformed to a “0–100” scale [ 35 ]. A higher score on the functioning scales indicates better HRQoL/functioning, while a higher score for symptoms indicates more complaints. For our analysis, we transformed the scores of the symptom scales once again, so a higher score on the symptom scale indicates fewer complaints. Now, a higher score on all scales corresponds with better functioning, fewer complaints, and a better overall quality of life. Data analysis Data were analyzed using SPSS Statistics (IBM Corporation, version 26.0, Armonk, NY, USA) and R version 4.2.1 packages MVN, huge, qgraph, and bootnet. Descriptive statistics were calculated and presented as frequencies, percentages, means, and standard deviations. We used listwise deletion to exclude participants with missing items on the EORTC QLQ-SURV111 questionnaire, except for the items related to sexuality. This is because sexual functioning and problems can be an important aspect of HRQoL, and missing responses to sensitive questions like sexuality are common [ 36 ]. If half of the items from the two sexuality scales were answered, we assumed that the missing item had a value equal to the item that was present for that respondent according to the QLQ-SURV111 scoring manual. In case both items for the scale were missing, we used a copy mean imputation of the study population. We assessed the assumption of multivariate normality with Mardia’s test [ 37 ], which needs to be fulfilled prior to estimating the network [ 38 ]. Mardia’s multivariate skewness and kurtosis coefficients of the numeric scales were calculated [ 37 ]. In the case of multivariate normality, both p -values of skewness and kurtosis should be greater than 0.05 [ 37 ]. As the data were not multivariate normally distributed according to Mardia’s test ( p <0.05), a nonparanormal transformation was applied to relax the normality assumption [ 39 ]. In our network model, HRQoL is conceptualized as a network of mutually interrelated factors. Because data are continuous scales or items, we used Guassian Graphical model (GGM) [ 26 , 40 ]. Nodes represent the selected HRQoL scales and items, and edges (links connecting two nodes) represent the regularized partial correlation coefficients after controlling for all other nodes. The thickness of the edge visualizes the strength, and the color a positive (red) or negative (blue) partial correlation. The partial correlation is indicated as very small ( r <0.1), small (0.1≤ r <0.3), moderate (0.3≤ r <0.5), and large ( r >0.5) [ 41 ]. We applied graphical lasso tuned with the Extended Bayesian Information Criterion (EBIC) [ 38 , 42 ]. The EBIC hyperparameter, used to set the preferred simplicity of the model, was set to 0.5 to minimize spurious connections [ 38 ]. Graphical lasso is a form of lasso regularization to prevent that edges between two nodes are spurious because of other nodes (i.e., conditional independence association) and small edges were shrinked to zero by dropping them from the model [ 42 ]. In this way, the estimated network is not over fitted and interpretable. We estimated node strength (i.e., number and strength of edges between nodes), betweenness (i.e., how often a node lies in shortest path between any combination of two nodes), and closeness (i.e., average distance from one node to all other nodes, which indicates how fast a node can be reached), which are indices of node centrality [ 38 , 43 ]. Bootstrapping was performed to explore the accuracy and stability of the network [ 38 ]. To estimate the accuracy of edge weights, 95% bootstrapped confidence intervals (CIs) around each edge in the network were calculated. Non-parametric bootstrapping (1000 bootstrap samples) was used to construct CIs. To estimate the stability of node centrality, we applied case-dropping bootstrap (1000 bootstrap samples) to calculate the correlational stability coefficient (CS coefficient) [ 38 ]. This coefficient represents the maximum proportion of participants that can be dropped from the analysis with the correlation between the original centrality indices and the subset centrality indices of at least 0.7 with 95% probability [ 44 ]. The CS coefficients of at least above 0.25, but preferable above 0.5, are considered stable [ 44 ]. Additionally, the bootstrapped values were used to test the significance of edge weights and node strength [ 38 ]. These bootstrapped difference tests indicate the difference between two different edge weights or node strengths. A bootstrapped CI around these difference scores was calculated [ 38 ]. Detection of communities was performed using the Louvain clustering method, a hierarchical clustering method based on multi-level modularity optimization algorithm [ 45 ].
Results Characteristics AYA cancer survivors In total, 11296 AYAs were invited to participate in the study, of whom 4010 (36%) responded. After excluding 414 records with missing data, we included 3596 AYAs in our final analysis; their sociodemographic and clinical characteristics are described in Table 1 . AYAs were on average 31.5 years old at diagnosis and mostly female (61%). The average time since diagnosis was 12.4 years and the most common cancer types were breast cancer (24%), germ cell tumors (18%), lymphoid hematological malignancies (15%), and tumors of female genitalia (11%). The mean scores of the functioning and symptom scales of the QLQ-SURV111 are shown in Table 2 . The overall global quality of life score of AYAs was on average 77.3 (SD 18.7). The functioning scale with the highest score was physical functioning (mean 91.7; SD 13.9), whereas sexual functioning was the scale with the lowest score (mean 43.7; SD 25.5). AYAs scored the lowest on the symptom scales social isolation (mean 69.1; SD 30.7), fatigue (mean 70.3; SD 26.0), and negative health outlook (mean 74.2; SD 20.0). Network analysis Overall network The partial correlation network model is shown in Fig. 1 . In our sample, health distress had a strong partial correlation with negative health outlook ( r = 0.71) and moderate partial correlation with worries about family getting cancer ( r =0.48). Symptom checklist had a strong partial correlation with pain ( r =0.67). Role functioning was strongly partially correlated to physical functioning ( r =0.70) and to social interference ( r =0.72), and there was a moderate partial correlation between sexual functioning and sexual problems ( r =0.43). The bootstrapped confidence intervals of estimated edge weights (Fig. 2 ) show that the previously described strong/moderate correlations in our network are robust. The negative correlations in our network, on the other hand, are not reliable. To test if a correlation between two nodes was significantly different from other correlations, we used the edge difference test (Supplementary material Figure S1 ). This plot showed that the correlation between health distress and negative health outlook was significantly different from all other correlations. Cluster analysis Within our network model, we identified four clusters (Fig. 1 ); (1) the worriment cluster (orange), (2) the daily functioning cluster (yellow), (3) the psychological cluster (pink), and (4) sexual cluster (green). The worriment cluster consists of the nodes: health distress, negative health outlook, symptom awareness, social isolation, worried about family getting cancer, and people treating you differently. Role functioning, physical functioning, pain, symptom checklist, social interference, and financial difficulties were all part of the daily functioning cluster. Emotional functioning, fatigue, sleep problems, cognitive functioning, and overall quality of life formed the psychological cluster. The sexual cluster consists of sexual functioning, sexual problems, and body image. Network stability and centrality The CS coefficient for strength, closeness, and betweenness were 0.75, 0.75, and 0.75 respectively, indicating a stable and reliable network (Fig. 3 ). Regarding centrality, the nodes with the highest strength are the most central, and therefore the most important nodes of the network model. In our model, nodes with the highest strength were negative health outlook (standardized centrality estimates (SCE) = 1.60), role functioning (SCE = 1.40), health distress (SCE = 1.40), and emotional functioning (SCE = 1.30). In addition, health distress and negative health outlook had the highest betweenness, and emotional functioning and health distress had the highest closeness (Fig. 4 ). The centrality difference plot (Supplementary material Figure S2 ) demonstrates that the strength of the node negative health outlook significantly differed from the other nodes in the network model.
Discussion This study shows the results of a stable and reliable network analysis based on HRQoL data of 3596 long-term AYAs, as the first to our knowledge. Although the overall global quality of life score was 77.3 on average, the lowest functioning scale score was sexual functioning and the lowest symptom scale scores were social isolation, fatigue, and negative health outlook. This network showed several strong/moderate partial correlations, with the partial correlation between health distress and negative health outlook being the only one significantly different from all other. Also, the strength of the node negative health outlook was significantly different from all other nodes. In total, four clusters of negative symptom and functioning HRQoL scales were identified, including a worriment cluster, daily functioning cluster, psychological cluster, and sexual cluster. As this study is the first to apply a network analysis based on a wide range of HRQoL data of AYAs using the relatively new QLQ-SURV111 questionnaire, findings are difficult to compare with other studies. Although previous studies had similar aims, they mostly used the EORTC QLQ-C30 to assess HRQoL and studied adult cancer survivors [ 28 , 46 ]. This was also the case in the network analysis conducted by Rooij et al. in a heterogeneous sample of adult cancer survivors where the EORTC QLQ-C30 was used [ 28 ]. In their analysis, fatigue was consistently central and had moderate direct relationships with emotional symptoms, cognitive symptoms, appetite loss, dyspnea, and pain. Fatigue being the most central symptom was in contrast with our results, which might be explained by the much younger population in our study (31.5 years vs 61 years) and the difference in time since diagnosis, which was considerably longer in our study (12.4 years vs 4.2 years). In our network, negative health outlook was the node with the highest strength, which had a strong correlation with health distress within the worriment cluster. This suggests that psychological and emotional issues remain more of relevance to AYA cancer survivors also after long-term follow-up, and are better picked up by the QLQ-SURV111 questionnaire that covers a more complete range of relevant survivorship issues. Other network studies focused specifically on a construct (e.g., fear of recurrence) or predefined clusters of symptoms, for example, the network analysis on fear of cancer recurrence, anxiety, and depression in breast cancer patients of Yang et al. [ 46 ]. In their network, “having trouble relaxing” was the most central node, anxiety and depression were well-connected, and fear of cancer recurrence formed a distinct cluster. The use of the broad range of survivorship issues of the QLQ-SURV111, including psychological, social, physical symptoms and functioning, allowed us to explore clusters within our network as well. An interesting cluster that emerged was the worriment cluster, which consisted of negative health outlook, health distress, symptom awareness, social isolation, worried about family getting cancer, and people treating you differently. Although our questions regarding worries were different, one other network analysis on AYAs focused on the construct of fear of cancer recurrence [ 47 ]. Here, the researchers found fear of serious medical interventions as the most central symptom in their network, with the highest node strengths for fear of pain, fear of relying on strangers for activities of daily living, and fear of severe medical treatments. Based on their results, which emphasize the centrality of emotional issues among AYA patients, they stress the importance of prioritizing these symptoms for interventions. However, as stated previously, comparing the results of these studies with our results should be done cautiously as study populations, study designs, aims, and used questionnaires differ. The lack of studies among AYAs to compare our findings with stresses the need for more AYA-specialized research. The results of a network analysis can provide more insight in the HRQoL of AYAs, in which symptoms and functioning can influence each other, instead of perceiving them as individual factors [ 48 ]. Identifying the most important factors in a network can help to address these problems with targeted interventions and healthcare, and lead to novel research ideas. First, we recommend to replicate network analysis studies in other groups of AYAs to be able to make comparisons between studies with similar study populations and draw conclusions with more certainty—changing the exploratory approach into a confirmative approach. Ideally, longitudinal data needs to be collected to draw conclusions over time and adapt healthcare to these time-related changes where needed. This could even be specific to longitudinal ecological momentary assessment (EMA) in which participants are asked to complete (parts of) the expected momentary dynamic items of the HRQoL questionnaire multiple times a day during a study period. In this way, variations over time can be taken into account to a more detailed and individual level. Inter- and intra-individual differences over time might result in changes in prevalence (scores of the scales or items), partial correlations of nodes, centrality, and clusters formed. In addition, subgroup analysis should be performed to make healthcare interventions even more tailored. AYAs subgroup analysis could focus on age, gender, stage, treatment, and type of cancer. In order to establish tailored care, it is important to make the subgroups as specific as possible while remaining a stable and reliable network. However, it should also be noted that these results represent means and thus differences may exist on an individual level in clinical practice. Future research can lead to and optimize supportive care and targeted interventions, like psycho-oncological care and psychosocial, behavioral, and supportive interventions [ 49 ]. For example, the strong significant correlation between health distress and negative health outlook, which is part of the worriment cluster in this study, might be intervened on by distress screening and renewed/tailored, age-specific psycho-oncological aftercare [ 50 ]. For healthcare providers (HCPs) involved in AYA healthcare, and in general, visualization of the nodes, correlations, and clusters may help to understand the cohesion between different factors/symptoms and the influence they might have on each other, and more important, how a targeted intervention can influence several HRQoL outcomes simultaneously. This advocates for holistic and age-specific psycho-oncological aftercare in which multiple factors are targeted simultaneously by a multidisciplinary team of HCPs, to be as efficient and effective as possible.
Conclusions This innovative network analysis provides insight in the nodes, correlations, and clusters that could be targeted to improve the HRQoL outcomes of AYAs. Future studies with longitudinal data and subgroup analyses can tailor the interventions and provided healthcare even more, specifically for those at risk of poor HRQoL outcomes. With these insights, more targeted interventions and healthcare can be provided and developed.
Purpose Adolescent and young adult cancer survivors (AYAs) are at increased risk of long-term and late effects, and experience unmet needs, impacting their health-related quality of life (HRQoL). In order to provide and optimize supportive care and targeted interventions for this unique population, it is important to study HRQoL factors’ interconnectedness on a population level. Therefore, this network analysis was performed with the aim to explore the interconnectedness between HRQoL factors, in the analysis described as nodes, among long-term AYAs. Methods This population-based cohort study used cross-sectional survey data of long-term AYAs, who were identified by the Netherlands Cancer Registry (NCR). Participants completed a one-time survey (SURVAYA study), including the EORTC survivorship questionnaire (QLQ-SURV111) to assess their long-term HRQoL outcomes and sociodemographic characteristics. The NCR provided the clinical data. Descriptive statistics and a network analysis, including network clustering, were performed. Results In total, 3596 AYAs (on average 12.4 years post diagnosis) were included in our network analysis. The network was proven stable and reliable and, in total, four clusters were identified, including a worriment, daily functioning, psychological, and sexual cluster. Negative health outlook, part of the worriment cluster, was the node with the highest strength and its partial correlation with health distress was significantly different from all other partial correlations. Conclusion This study shows the results of a stable and reliable network analysis based on HRQoL data of long-term AYAs, and identified nodes, correlations, and clusters that could be intervened on to improve the HRQoL outcomes of AYAs. Supplementary Information The online version contains supplementary material available at 10.1007/s00520-023-08295-0. Keywords
Strengths and limitations This explorative study represents the very first network analysis using data on a range of HRQoL outcomes of long-term AYAs to our knowledge. Strengths include the large sample size, the establishment of a stable and reliable network, and the inclusion of a wide range of survivorship issues. However, the results should be interpreted with caution as the EORTC QLQ-SURV111 is not yet finalized and validated. In addition, our study included mostly females and over 40% was diagnosed with a stage I tumor. With a response rate of 36%, there are several subgroups (males, AYAs with a more aggressive disease, and AYAs diagnosed at the age of 18–24) underrepresented in this analysis who might have different HRQoL outcomes [ 30 ]. Results may therefore not be generalizable to the total AYAs population. Also, we have not taken a closer look at the outcomes of specific subgroups (between groups), as the study group as a whole was analyzed. As mentioned previously, the subgroup analyses should be part of future research to tailor interventions with a risk-based approach. In line with this, due to the methodology of this study, i.e., a cross-sectional questionnaire study, no causal pathways or changes in HRQoL factors over time can be assessed. In the future, this might be tackled by using longitudinal data instead of cross-sectional data. Supplementary information
Acknowledgements The authors wish to thank all the patients for their participation in the study and the registration team of the Netherlands Comprehensive Cancer Organisation (IKNL) for the collection of data for the Netherlands Cancer Registry. Author contribution Conceptualization: C.V., S.H.M.J., T.I.B., D.W., D.C.R., C.D., and O.H.; data curation: C.V. and T.I.B.; formal analysis: T.I.B., D.W., and C.D.; funding acquisition: W.T.A.G. and O.H.; investigation: T.I.B., D.W., C.V., D.C.R., C.D., and S.H.M.J.; methodology: T.I.B., D.W., C.V., D.C.R., C.D., and S.H.M.J.; project administration: C.V.; resources: T.I.B., D.W., C.V., D.C.R., C.D., and S.H.M.J.; software: T.I.B., D.W., C.V., D.C.R., C.D., and S.H.M.J.; supervision: W.T.A.G. and O.H.; validation: T.I.B., D.W., C.V., D.C.R., C.D., and S.H.M.J.; visualization: T.I.B., D.W., C.V., D.C.R., C.D., and S.H.M.J.; writing—original draft preparation: C.V., D.C.R., and S.H.M.J.; writing—review and editing: T.I.B., D.W., C.V., D.C.R., C.D., R.T., R.B., S.E.J.K., J.M.K., J.M.T., M.E.M.M.B., T.H., R.I.L., J.N., M.C.M.K., W.T.A.G., S.H.M.J., and O.H.; all authors have read and approved the final manuscript. Funding This study was funded by a VIDI grant (VIDI198.007) and an investment grant (#480-08-009) from the Netherlands Organization for Scientific Research. This research was also supported by an institutional grant of the Dutch Cancer Society and of the Dutch Ministry of Health, Welfare and Sport. Data availability The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy issues. Declarations Ethics approval The study was conducted in accordance with the Declaration of Helsinki, and approved by the Netherlands Cancer Institute Institutional Review Board (IRBd18122) on February 6, 2019. Consent to participate Informed consent was obtained from all subjects (responders) involved in the study. Competing interests The authors have no relevant financial or non-financial interests to disclose.
CC BY
no
2024-01-15 23:42:02
Support Care Cancer. 2024 Jan 13; 32(2):104
oa_package/4c/bd/PMC10787889.tar.gz
PMC10787890
37794263
Introduction Bioactive lipids are important for brain function. In particular, the homeostasis of the sphingolipids ceramide-1-phosphate, sphingosine, and sphingosine-1-phosphate (S1P) has been described to regulate proliferation, differentiation, cell growth and inflammation in the cells of the central nervous system (CNS) [ 1 ]. S1P is produced from ceramide by ceramidase followed by phosphorylation by sphingosine kinase (SphK1/2) [ 2 ]. S1P is abundantly produced by erythrocytes, platelets and endothelial cells [ 3 , 4 ]. Although, S1P synthesis in the CNS has also been reported, namely by astrocytes after fibroblast growth factor stimulation [ 5 ]. S1P in the extracellular space binds to its five specific G-coupled receptors S1PR1-5, which signal through diverse downstream pathways [ 1 ]. All S1P receptors are expressed in the CNS [ 6 ]. They participate in multiple important functions during CNS development, and further contribute to the development and/or resolution of neurodegeneration in pathological conditions such as ischemic stroke [ 7 ], multiple sclerosis [ 8 ], hearing loss [ 9 ] and seizures [ 10 , 11 ]. S1P receptors have received attention in the field of multiple sclerosis, for which immunomodulation is achieved by the oral drug Fingolimod (FTY720) that activates S1PR1,3–5 [ 12 ]. S1PR1 is the most studied S1P receptor with important roles in angiogenesis and neurogenesis [ 13 ]. S1PR2 is a modulator of neuronal excitability during neuronal development [ 14 ] and controls spontaneous activity of cultured neurons [ 6 ]. S1PR3 receptor controls activity of microglia and their participation in neuroinflammation [ 15 – 17 ], and is highly expressed in astrocytes, where it regulates astrogliosis via activation of the small GTPase RhoA [ 18 ]. Until recently, S1PR4 has received little attention in the CNS [ 19 ]. Our recent work proposed S1PR4 presence in synapses, and a role in neuronal activity control [ 6 ]. Furthermore, previous studies have demonstrated loss of sphingosine kinase activity and lowering of brain region-specific S1P levels early in the progression of Alzheimer’s disease [ 20 , 21 ]. Recent evidence also suggests an important role for S1P and its generating enzymes SphK1/2 in the development of diabetes [ 22 ] through adverse effects on endothelial function [ 17 , 23 ], hepatic insulin sensitivity and secretion [ 24 ], regulating insulin secretion in pancreatic β-cell [ 25 ], and contributing to diabetes-associated inflammation [ 23 ]. The sphingosine rheostat has been extensively studied in diabetes and insulin resistance, with special focus on ceramide and ceramide-1-phosphate [ 22 ]. Because S1P is involved in a plethora of cellular functions, a tight regulation of S1P homeostasis seems important for proper brain functioning, which would be disrupted in metabolic syndrome and diabetes. Diabetes impacts synapses with negative consequences for brain function, including memory performance [ 26 ]. Furthermore, rodent models of diabetes show altered neuromodulation systems operated by adenosine [ 27 – 29 ], ATP [ 30 ], or endocannabinoids [ 31 , 32 ], which control synaptic activity and brain energy metabolism, and might interact with brain insulin signaling [ 33 ]. These systems can also afford neuroprotection. For example, pharmacologically targeting the altered adenosinergic system components confers neuroprotection and improves memory performance in diabetes models [ 28 , 29 , 34 ]. It is hitherto unknown whether the neuromodulation system operated by S1P in the CNS is altered in diabetes and metabolic syndrome. In this study, we set out to test the hypothesis that T2D impacts the neuromodulation system operated by S1P in the CNS. Since S1PRs are present in nerve terminals of the cortex [ 6 ], we determined changes induced by T2D in the density of S1PR1-4 in cortical synaptosomes from insulin-resistant Goto-Kakizaki (GK) rats and from diet-induced obese mice, which are models with well-established brain dysfunction (e.g., [ 29 , 35 , 36 ]).
Methods Animals Experiments were performed according to EU Directive 2010/63/EU under approval of the Malmö/Lund Committee for Animal Experiment Ethics (permit numbers 994/2018 and 9987/2020) and are reported following the ARRIVE guidelines (Animal Research: Reporting In Vivo Experiments, NC3Rs initiative, UK). Male and female GK rats were obtained from a local colony, and male and female age-matched Wistar rats from Janvier (Saint Berthevin, France) were used as controls [ 29 ]. Wistar rats were housed in the facility for at least a month before experimentation. Eight weeks-old C57BL/6 J mice were purchased from Taconic (Ry, Denmark). Only male mice were used due to sex differences in responses to diet-induced obesity (see [ 35 ], and references therein). Animals were housed in ventilated cages enriched with a cylinder, wood toys and nesting material, at controlled temperature of 22 °C, 50–60% humidity and a 12:12-h light–dark cycle. Food and water were provided ad libitum. Rats were kept on a regular chow and were euthanized at 6 months of age (Fig. 1 A). Mice were fed a lard-based high-fat diet (HFD; 60% calories from fat) or a control diet (CD, 10% calories from fat) from Research diets (New Brunswick, NJ-USA), as previously described [ 35 ]. Mice were held on the diet for 1 week, 1 or 2 months, starting from 9 weeks of age (Fig. 1 B). Glucose Tolerance Test and Insulin Determination A glucose tolerance test (GTT) was performed 1–3 days before sacrifice. Food was removed at 08:00 for 6 h, and mice were put into clean cages to avoid coprophagy before and during the test. Before each test, a blood sample was collected from vena saphena into a heparinized tube for determination of plasma insulin concentration. Glycemia was measured from tail tip blood with the Accu-Chek Aviva glucometer (Roche, Solna, Sweden). Mice were then administered 2 g/kg glucose i.p. from a 20%(w/v) solution in saline, followed by determination of glucose levels after 15, 30, 60, 90 and 120 min. Plasma insulin was measured with ELISA kits from Mercodia (Uppsala, Sweden; #10-1250-01 for rats; #10-1247-10 for mice). Preparation of Nerve Terminal-Enriched Membranes After weighing and measuring tail-tip blood glucose, animals were anesthetized with isoflurane and quickly decapitated. Trunk blood was collected, and plasma was stored at − 80 °C for insulin and S1P determination. Brains were quickly dissected, frozen in N 2 (l) and stored at − 80 °C until further experiments. Synaptosomes were prepared as described previously [ 36 ]. Briefly, cortical tissue was homogenized (12 strokes at 800 rotations/min) with a glass-teflon Potter–Elvehjem homogenizer (rotor head InterMed/STIR20) in Sucrose-HEPES buffer (0.32 mol/L sucrose, 1 mmol/L EDTA, 10 mmol/L HEPES, 1 mg/mL bovine serum albumin, pH 7.4) at 4 °C. Homogenates were centrifuged at 3000 g for 10 min at 4 °C (Beckman Coulter/Avanti J-20 XP). After, supernatants were centrifuged at 14,000 g for 12 min at 4 °C. The pellet was re-suspended in 1 mL of 45% (v/v) Percoll (GE Healthcare, Uppsala, Sweden) solution prepared in Krebs-HEPES buffer (in mmol/L: 140 NaCl, 5 KCl, 10 HEPES, 1 EDTA, 5 glucose, pH 7.4), and then centrifuged at 21,000 g for 2 min at 4 °C. The top layer (rich in synaptosomes) was washed by re-suspending in Krebs-HEPES buffer and centrifuging again. The resulting pellets were re-suspended in Krebs-HEPES solution containing protease inhibitors (cOmplete cocktail from Roche, Mannheim, Germany), and stored at − 80 °C until immunoblotting. Immunoblotting Total protein content of the samples was measured with the bicinchoninic acid assay (kit from Pierce, ThermoFisher Scientific, Uppsala, Sweden). Then, Western blotting was carried out as detailed by Lizarbe et al. [ 36 ]. Briefly, samples were dissolved in sample buffer (#NP0007, Invitrogen, ThermoFisher), boiled at 95 °C for 5 min, and then separated by SDS-PAGE in 4–12% Bis–Tris gradient gels (#NP0336, Invitrogen), followed by transfer onto nitrocellulose membranes of 0.45-μm pore size (#GE10600002, GE Healthcare). The membranes were blocked for 1–2 h in Tris-buffered saline (in mmol/L: 20 Tris, 150 NaCl, pH 7.6) containing 5% (w/v) skim milk, 1% (v/v) Tween-20, and then sequentially incubated with primary and secondary antibodies (Table 1 ) diluted in this blocking solution. Immunoblots were developed with the chemiluminescence Super-Signal kit (#34,580, ThermoFisher). Whenever necessary, sensitivity was enhanced with the biotin-streptavidin kit VectaStain ABC-HRP according to manufacturer’s instructions (#PK-4000, Vectorlabs, CA-USA). Luminescence was detected using the Chemidoc XRS + interfaced to Image Lab 5.2.1 (Biorad, Stockholm, Sweden). Mass Spectrometry Plasma samples obtained from trunk blood, and cortical homogenates in phosphate-buffered saline (PBS; in mmol/L: 137 NaCl, 2.7 KCl, 1.5 KH 2 PO 4 , 8.1 Na 2 HPO 4 , pH 7.4) were spiked with deuterated S1P as internal standard (S1P-D7 > 99% deuterated; Avanti Polar Lipids/Merck, Darmstadt, Germany), and S1P was extracted as described in [ 37 ]. Extracts were dried under a nitrogen stream, dissolved in methanol, and subjected to liquid chromatography-coupled tandem mass spectrometry as previously described [ 38 , 39 ]. Statistics Results are shown as mean ± SD, and were analyzed with Prism 9.4.1 (GraphPad Software, San Diego, CA). Normal distribution was assessed with the Kolmogorov–Smirnov test. In the presence of normality deviations, results were analyzed using Mann–Whitney test or Kruskal–Wallis test followed by Dunn’s multiple comparisons. Normally distributed data was analyzed with unpaired, two-tailed Student t -test or ANOVA followed by independent comparisons with the Fisher’s least significant difference (LSD) test. Significance was accepted for P < 0.05.
Results We have previously detailed metabolic phenotypes of insulin-resistant GK rats [ 29 ] and HFD-induced obese mice [ 40 ]. In the present study, we have confirmed that GK rats displayed lower body weight, increased glycemia, and similar fed plasma insulin concentration, when compared to controls (Table 2 ). HFD-fed mice showed increased body weight, a tendency for increased fasting glycemia and plasma insulin concentrations, reduced glucose tolerance, and increased insulin resistance, when compared to controls (Table 3 ). Plasma concentrations of S1P were higher in GK than Wistar rats (+ 60%, P = 0.008, Fig. 1 A). However, S1P levels in homogenates from the cortex of GK rats were not significantly different from those in Wistar rats ( P > 0.05, Fig. 2 B). Nerve terminal-enriched membranes from GK rats showed lower immunoreactivity of S1PR1 (− 21%, P = 0.004), S1PR2 (− 26%, P = 0.006) and S1PR4 (− 40%, P = 0.046,) than those from control Wistar rats (Fig. 2 C–D). S1PR3 immunoreactivity was similar between the groups ( P > 0.05). Plasma concentrations of S1P were increased by both HFD exposure and age (interaction F(2,64) = 1.4, P > 0.05; diet F(1,64) = 7.2, P = 0.009; time F(2,64) = 17, P < 0.001). In particular, HFD-feeding increased plasma S1P levels by 11–16%, most prominently at 1 month after diet intervention (Fig. 3 A). In the cortex, S1P levels were also increased with HFD, and showed a variation with age that was unrelated to plasma S1P levels (interaction F(2,27) = 0.12, P > 0.05; diet F(1,27) = 5.5, P = 0.026; time F(2,27) = 4.4, P = 0.022, Fig. 3 B). The density of S1PRs was analyzed in nerve-terminal membranes from the cortex of mice fed a CD or HFD for 1 week, 1 month, and 2 months (Fig. 3 C–E). SNAP25 analyzed as constitutive protein showed no HFD-induced changes (Fig. 3 F). Relative to CD, HFD exposure reduced S1PR1 immunoreactivity (interaction F(2,30) = 0.64, P > 0.05; diet F(1,30) = 13, P = 0.001; time F(2,30) = 0.64, P > 0.05; Fig. 3 G), which is particularly prominent at 1 week (− 52%, P = 0.011) and 2 months of HFD feeding (− 50%, P = 0.014). The density of S1PR2 and S1PR3 was not modified by HFD feeding (Fig. 3 G). S1PR4 immunoreactivity in cortical nerve terminal membranes was lower in HFD-fed mice than controls (interaction F(2,31) = 1.6, P > 0.05; diet F(1,31) = 7.4, P = 0.011; time F(2,31) = 1.6, P > 0.05; Fig. 3 G), most prominently after 1 week of HFD feeding (− 54%, P = 0.005), and recovering to control levels at 2 months of HFD.
Discussion This study demonstrates for the first time that S1PR density in cortical nerve terminals is reduced in lean and obese models of insulin resistance, along with possible increases of S1P concentrations in the cortex, implying a role for S1PR signaling in the neuropathological process that occurs in T2D. Since S1PR activation generally attenuates neuronal activity [ 6 ], lower S1PR density might lead to overall cortical hyper-excitability. From those analyzed, S1PR1 is the receptor with the highest expression level in the CNS. Its density was decreased in cortical synapses of both insulin-resistant GK rats and HFD-fed mice, relative to their respective controls. Previous studies have reported similar S1PR1 level reductions in the hypothalamus of HFD-fed rats or mice, as well as in the leptin deficient ob/ob mice [ 41 ]. Silva et al. proposed that S1P is an appetite inhibitor through S1PR1 signaling, and reduced S1PR1 levels in the hypothalamus of obese models were associated with increased food intake. S1PR1 activation inhibits neuronal activity [ 6 ] and, therefore, the S1PR1 reduction observed in cortical synapses is likely to result in loss of S1P-dependent synaptic control. However, S1PR1 is ubiquitously expressed, and exert a multitude of regulatory actions, not being specific to synapses. Thus, a potential approach of pharmacologically targeting S1PR1 for synaptic modulation might also trigger a plethora of non-synaptic actions. Adding to the complexity of S1PR1 signaling, others have reported that specific S1PR1 activation increases excitability of some sensory neurons in the rat dorsal root ganglia [ 42 ]. Given the ubiquitous actions of S1P, any systemic interventions on S1PR1 should be taken with caution. The density of both S1PR2 and S1PR4 was decreased in cortical synapses from GK rats compared to controls. Considering the role of S1PR2 as neuromodulator in excitatory neurons [ 14 ], and its preferential localization at the pre-synaptic level in the cortex [ 6 ], dampening of S1PR2 signaling in GK rats is likely to contribute to the synaptic dysfunction and impaired synaptic plasticity [ 29 ]. Our previous study [ 6 ] showed that specifically S1PR2 and S1PR4 activation results in dampening of spontaneous spiking frequency of cultured primary neurons. A reduction in levels of synaptic S1PR2/4 could thus result in uncontrolled, excessive synaptic activity that might lead to excitotoxicity and synaptic damage [ 26 ]. We thus speculate that S1PR2/4 activation might allow for synaptic protection in diabetic conditions. HFD-fed mice did not reproduce the T2D-induced S1PR2 alterations observed in GK rats. Moreover, S1PR4 density was reduced in short- but not long-term HFD, when compared to the respective controls. Besides obesity, a key difference between the two models is the hyperglycemia in GK rats [ 43 , 44 ] that is negligible in HFD-fed mice [ 26 , 36 ]. On the other hand, both models become insulin resistant. In this regard, it needs to be noted that S1P signaling through S1PR2 has been shown to interact with insulin signaling, and might participate in the development of insulin resistance in peripheral cells [ 45 ]. Given the ability of S1PR2 to inhibit insulin-mediated signals, it is plausible that reduced S1PR2 in synaptic membranes results in enhanced insulin signaling. Thus, S1PR2 might contribute to the development of central insulin resistance. In turn, putatively increased S1PR2 signaling due to higher S1P concentrations in HFD-fed mice could also dampen synaptic insulin signaling. Loss of tonic insulin receptor signaling in nerve terminals is believed to contribute to memory impairment [ 33 , 46 ]. Although S1PR5 is present in the CNS, its relevance for neuromodulation at the synaptic level remains to be established ([ 6 ], and references therein). Therefore, we have not investigated diabetes-induced alterations of S1PR5 density in the present study. Concentrations of S1P measured by ELISA were found to increase in the liver of patients with T2D as well as in streptozotocin-induced T2D rats [ 47 ]. Plasma S1P concentration also increases in the ob/ob mouse model, mice exposed to HFD for 6 weeks, and in a population of young obese humans [ 48 ]. Interestingly, plasma S1P in humans was found to be associated with body fat, insulin levels, insulin resistance (by HOMA-IR), as well with cholesterol [ 48 ]. The same study reported an increase in concentrations of S1P in plasma after 12 h of fasting, suggesting a relation between S1P levels and increased inter-organ lipid flux. In our study, plasma samples for S1P determination were collected under fed state, which we therefore expect to depict diabetes-induced S1P changes rather than acute fasting-induced mobilization of lipid stores. While plasma S1P levels increase in non-diabetic young adults with obesity (< 30 years old; body mass index ~ 37 kg/m 2 ) relative to lean controls [ 48 ], others have reported that T2D is associated with a decrease of plasma S1P concentrations [ 49 , 50 ]. Namely, these studies found lower plasma S1P concentrations in individuals with T2D relative to healthy controls matched for age and body mass index (average: ~ 60 years old and ~ 29 kg/m 2 in Vaisar et al. [ 49 ]; 44–49 years old and 25–26 kg/m 2 in Sui et al. [ 50 ]). While reported effects of obesity and T2D on plasma S1P levels are opposite, these studies differ in the studied populations, age of the subjects, and methods for S1P extraction and detection. S1P levels in the cortex are reported to be lower than in other brain regions [ 51 ], and we now found S1P concentrations increased in the cortex of HFD-fed mice. To our knowledge, this is the first study investigating S1P concentration in the cerebral cortex of T2D/pre-diabetic animal models. We report an HFD-induced increase of S1P in cortical tissue that seems to be independent of HFD-induced increases in circulating S1P levels. In turn, in GK rats, the large variance of S1P concentrations measured in the cortex precluded determining the expected diabetes-induced increase in S1P levels. This contrasts the concurrent increases of brain and plasma S1P that have previously been reported in a mouse model of hypertension [ 52 ]. The assessment of S1PR density in mouse cortical synapses across the HFD treatment is limited to the effects of the exposure to the diet, and not the effect of age. This is due to the semi-quantitative nature of immunoblotting methods, and the fact that we have decided to analyze all age-matched samples in parallel. Nevertheless, mice were treated for a relatively short period of time (only 2 months), at an age range from about 2 to 4 months, during which changes of S1PR density might not be as important as effects of the diet. Another limitation of this study is that despite including both sexes in the GK rat part of the study, analysis of plasma S1P levels was only available for males. We are not able to determine sex effects on plasma S1P concentrations, although they were not apparent in cortical levels of S1P. Finally, another important limitation of this study is that the nerve-terminal enriched membranes obtained with our protocol also contain non-synaptic components, such as the perisynaptic astrocytic processes. In particular, S1PR1 is abundant in astrocytes, and the observed S1PR1 reduction could be due to alterations in astrocytic rather than synaptic compartments. However, in total membranes obtained as previously described [ 27 ], we have not observed any S1PR1 reduction in GK vs Wistar rats, or HFD-fed mice vs controls (data not shown). Thus, S1PR density alterations in the present study most likely take place within the synaptic membranes. In conclusion, we have observed a decrease in the protein levels of three S1P receptors in the cortex of the GK model that were likely unrelated to cortical concentrations of S1P. Two of the altered receptors, namely S1PR2 and S1PR4, have been proposed to control neuronal activity and synaptic transmission ([ 6 ], and references therein). Furthermore, S1PR4 levels were also decreased in the cortex of a diet-induced obese mouse model with metabolic syndrome, along with increased cortical levels of S1P. Altogether, these results point towards T2D-induced alterations of the neuromodulation system to which S1P signaling contributes. Our results open the door for testing brain-specific S1P-S1PR modulation as a potential neuroprotection strategy in diabetes.
Sphingosine-1-phosphate (S1P) is a phosphosphingolipid with pleiotropic biological functions. S1P acts as an intracellular second messenger, as well as extracellular ligand to five G-protein coupled receptors (S1PR1-5). In the brain, S1P regulates neuronal proliferation, apoptosis, synaptic activity and neuroglia activation. Moreover, S1P metabolism alterations have been reported in neurodegenerative disorders. We have previously reported that S1PRs are present in nerve terminals, exhibiting distinct sub-synaptic localization and neuromodulation actions. Since type 2 diabetes (T2D) causes synaptic dysfunction, we hypothesized that S1P signaling is modified in nerve terminals. In this study, we determined the density of S1PRs in cortical synaptosomes from insulin-resistant Goto-Kakizaki (GK) rats and Wistar controls, and from mice fed a high-fat diet (HFD) and low-fat-fed controls. Relative to their controls, GK rats showed similar cortical S1P concentration despite higher S1P levels in plasma, yet lower density of S1PR1, S1PR2 and S1PR4 in nerve-terminal-enriched membranes. HFD-fed mice exhibited increased plasma and cortical concentrations of S1P, and decreased density of S1PR1 and S1PR4. These findings point towards altered S1P signaling in synapses of insulin resistance and diet-induced obesity models, suggesting a role of S1P signaling in T2D-associated synaptic dysfunction. Keywords Open access funding provided by Lund University.
Abbreviations Central nervous system High-fat diet Goto-Kakizaki Glucose tolerance test Sphingosine 1-phosphate Sphingosine-1-phosphate receptors Sphingosine Sphingosine kinase Type 2 diabetes Acknowledgements The authors thank Nadiia Kravchenko for technical assistance with insulin determination, and Eugenia Cordero Concha for handling the GK rats. Author Contributions AM, JMND designed the study. CS, HE, LV, JPPV, FM conducted experiments, CS, JMND analyzed data. CS wrote the manuscript. AM, LE and JMND revised the manuscript. CS and JMND verified all data. All authors approved the final version of the manuscript. Funding Open access funding provided by Lund University. This work was supported by the Swedish foundation for International Cooperation in Research and Higher education (#BR2019- 8508), Swedish Research Council (#2019-01130 #2019-01406), Diabetesfonden (#Dia2019-440, #Dia2021-637, #Dia2022-723), Dementiafonden, and Direktör Albert Påhlssons Foundation. The authors acknowledge support from the Lund University Diabetes Center, which is funded by the Swedish Research Council (Strategic Research Area EXODIAB; #2009-1039) and the Swedish Foundation for Strategic Research (#IRC15-0067). Data Availability The datasets from the current study are available from the corresponding author on reasonable request. Declarations Conflict of interest The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:02
Neurochem Res. 2024 Oct 4; 49(2):338-347
oa_package/c1/70/PMC10787890.tar.gz
PMC10787891
0
Introduction Antibiotic resistance, a globalized public health peril in this new era of modern medicine, prevailed from the pathogenic microbial development of novel resistance mechanisms of genetic and epigenetic origins against the available antibiotics by circumventing the therapeutic actions of these drugs, leading to the failure of antimicrobial drugs curbing the menace of infection (Dar et al. 2017 ; Soares et al. 2021 ). The discovery, research, and clinical applications of immuno-peptides known as antimicrobial peptides (AMPs), such as defensins, LL-37, gramicidins D, histatins, Renalexin, and others with antibacterial properties, have been considered and proven as alternative antimicrobial agents to antibiotics (Perez-Perez et al. 2021 ; Peters et al. 2010 ; Zhang et al. 2021 ). Antimicrobial peptides are mostly cationic, amphipathic, and composed of 10–100 amino acid residues. These biological agents have shown novel therapeutic potencies against resistant bacterial pathogens elicited via the disruption of the cell wall and plasma membrane, inducing pore formation, upon the establishment of electrostatic interactions with the negatively charged membrane phospholipids, or inhibiting DNA replication and protein synthesis, thereby exhibiting a broad spectrum of therapeutic activities (Jindal et al. 2015 ; Ołdak and Zielińska 2017 ). Many AMPs possess a net charge of +2 to +9, confirming strong polarity to bacterial cell membrane surface structures, thereby conceding antibacterial activity against bacterial pathogens that pose global public health threats (Wei and Zhang 2022 ). AMPs are known for modulation and orchestration and as indispensable components of the innate immune system functioning as the first line of defense against bacterial attack in eukaryotes and often synthesized as a competitive strategy in prokaryotes to limit and outcompete the cellular growth of competitive microbes (Seyedjavadi et al. 2021 ). The mechanisms of therapeutic action of AMPs, including antibacterial peptides, have been extensively exploited over the years. Studies in both in vitro, in vivo, and model plasma membranes have confirmed novel and distinct modes of action compared to clinically prescribed antibiotics by extensively provoking plasma membrane incision and permeability, leading to membrane disruption and cell death (Erdem Büyükkiraz and Kesmen 2022 ; Wei and Zhang 2022 ). The recent applications of AMPs as disease control therapeutics, coupled with their growing interest as antimicrobial agents produced via recombinant expression systems, have provided a promising and safer platform for the expression and the clinical applications of AMPs ( Jindal et al. 2015 ; Montfort-Gardeazabal et al. 2021 ; Nuti et al. 2017 ). This study focused on two antimicrobial peptides, LL-37 and Renalexin. LL−37 is the only α-helix cathelicidin-based AMP with therapeutic action against bacterial, viral, and fungal infections (Kang et al. 2019 ). The biosynthesis of LL−37 occurs in most immune cells, including the mast cell, neutrophils, mucosal epithelial cell, keratinocytes, adipocytes, and the T and B lymphocytes. Cathelicidins (hCAP18) are distinct mammalian immune proteins that act as precursor molecules that undergo proteolytic cleavage at the C-terminal to release a short peptide with antimicrobial and immune-modulatory activity commonly known as LL-37 (Erdem Büyükkiraz and Kesmen 2022 ). The antibacterial activity of naturally and recombinantly purified LL−37 against multi-drug resistant (MDR) bacteria pathogens has been elucidated and published ( Dürr et al. 2006 ; Erdem Büyükkiraz and Kesmen 2022 ; Perez-Perez et al. 2021 ; Scott et al. 2002 ). Antimicrobial activity of single-peptide LL-37 showed promising results against Staphylococcus aureus (69% sensitivity) and Escherichia coli (64% sensitivity) at 50 μM peptide concentration (Perez-Perez et al. 2021 ). A study by Douglas Clark and colleagues in early 1994 led to the discovery of a novel antimicrobial peptide Renalexin (sometimes referred to as Ranalexin), isolated from the skin of the American Bullfrog Rana catesbeiana . It contains a single intramolecular disulfide bond between two cysteine amino acids at positions 14 and 20, which forms a heptapeptide ring within the molecule similar to that seen in the antibiotic Polymyxin B. Renalexin is initially synthesized as a precursor peptide with a putative signal sequence and an acidic amino acid–rich region at its N-terminal ( Aleinein et al. 2013 ; Clark et al. 1994 ). Renalexin has shown therapeutic actions against both gram-positive S. aureus and gram-negative E. coli bacteria at concentrations ranging from 50 μM and above by interacting with phospholipid membrane via electrostatic binding; in cases where the peptide traverses the membrane, it inhibits protein expression which leads to cell death (Dar et al. 2017 ; Nuti et al. 2017 ). Here, we report on an effective and reliable design and expression of soluble hybrid antimicrobial peptide LL-37_Renalexin in Escherichia coli . We started with synthetic DNA encoding for the target peptide and using CusF3H+ and SmbP as carrier proteins. The recombinant LL-37_Renalexin gene was cloned into the expression vector pET30a+. The fusion peptides CusF3H+_LL-37_Renalexin and SmbP_LL-37_Renalexin were expressed in E. coli with isopropyl β-D-1-thiogalactopyranoside induction under optimized conditions. The recombinant tag-free LL-37_Renalexin was purified via immobilized metal affinity chromatography after being released from the carrier proteins by enterokinase treatment. The in vitro antibacterial activities of the hybrid peptide were ascertained and evaluated.
Materials and methods Reagents, plasmid, enzymes, and bacteria strains Escherichia coli DH5α cells employed for routine plasmid propagation and subcloning experiments were supplied by New England Biolabs (NEB) (Ipswich MA, USA). Protease-deficient bacteria strains Escherichia coli BL21(DE3) and Escherichia coli SHuffle T7(DE3) also supplied by NEB were used as microbial expression hosts. The plasmid pET30a+ purchased from EMD Biosciences (Darmstadt, Germany) was chosen for the design and construction of plasmid expression vectors. The MEGAquick-spin DNA and plasmid purification kits were bought from iNtRON Biotechnology (Seoul, South Korea). Restriction enzymes Nde I and Xho I were purchased from NEB. T4 DNA ligase, Vent , and Taq DNA polymerases used for molecular cloning and amplification were provided from NEB as well. The synthetic DNA encoding for the hybrid peptide and the protease enterokinase were purchased from GenScript Inc. (Centennial, Piscataway, USA). Isopropyl-β-D-1-thiogalactopyranoside (IPTG) used for induction of expression was purchased from A.G. Scientific Inc. (San Diego, CA, USA). Kanamycin employed as a selective marker for transformed cells and as a positive control was supplied by Sigma-Aldrich (Darmstadt, Germany). Standard protein markers used for peptide size characterization were purchased from NEB and Bio Basic Inc. (Amherst, NY, USA). All culture media including Luria-Bertani, Tryptic Soy Broth, bacteriological agar, and Mueller-Hinton agar used for microbial cultivation, expression, enrichment, and antimicrobial activity were provided by Sigma-Aldrich (Darmstadt, Germany), and Legacy Biologicals (Mount Prospect, IL, USA). All clinical bacteria pathogens including S. aureus and E. coli used for the antimicrobial activities were supplied from the University Hospital, Department of Pathology, Clinical Microbiology Laboratory (UANL, Monterrey, Mexico). Design of the hybrid AMP LL-37_Renalexin The mature amino acid sequences of the single antimicrobial peptide LL-37 and Renalexin were retrieved from the AMP database ( https://APD3.unmc.edu/structure ) with the accession number AP0030/2K60 and AP00513/P39084, respectively, and were used for the design of a novel hybrid peptide LL-37_Renalexin via the application of our newly designed GS peptide linker and the protein tags CusF3H+ and SmbP (Fig. 1 ) that allow for expression and immobilized metal-affinity chromatography (IMAC) purification. The designed hybrid peptide amino acid sequence was flanked at respective positions with the amino acid sequences of protein tags SmbP or CusF3H+, enterokinase site, GS linker, Nde I, Kpn I, and Xho I restriction enzyme sites. The entire amino acid sequence of the designed complete hybrid peptide was optimized ( http://genomes.urv.es/CAIcal ) for efficient expression based on codon usage in E. coli as the microbial expression host. The biochemical properties and molecular structure of the designed hybrid peptide were elucidated and analyzed using Expasy and I-TASSER tools ( https://web.expasy.org/cgi-bin/protparam/protparam , https://zhanggroup.org/I-TASSER/ ) to access the efficiency and the reliability of production in microbial systems. Construction of pET30a+ expression vectors A 492- and 480-mer oligonucleotide DNA (gene) sequences (Fig. S1 ) encoding for SmbP_LL-37_Renalexin and CusF3H+_LL-37_Renalexin with the accession numbers OR356112 and OR356113, respectively, ( https://www.ncbi.nlm.nih.gov/Genbank.html ) were synthesized based on the optimized hybrid peptide sequence with reference to codon usage in E. coli . The gene was cloned into pUC57 and supplied to the protein expression and purification laboratory, UANL, Mexico. Both the pUC57 and pET30a+ plasmid vectors were digested with Nde I and Xho I restriction enzymes (Fig. S2 ). The restriction digestion products were visualized on a 1% agarose gel stained with 0.5 μg/ml ethidium bromide. The target DNA fragment was cut from the gel and purified using the MEGAquick-spin DNA purification kit. Purified DNA fragments were ligated with the T4 DNA ligase for the design of two plasmid expression vectors pET30a+_CusF3H+_LL-37_Renalexin and pET30a+_SmbP_LL-37_Renalexin. The resulting plasmid construct was transformed for propagation into E. coli DH5α by heat shock at 42 °C for 45 s. Transformant cells were selected on Luria-Bertani–kanamycin (30 μg/ml) agar plates. To confirm the presence of the gene in the designed plasmid constructs, the 5′-T7 promoter forward primer (5′-TAATACGACTCACTATAGGG-3′) and the 3′-T7 terminator reserve primer (3′-GCTAGTTATTGCTCACGG-5′) pairs were used for polymerase chain reaction (DNase-free water, 5 mM dNTPs, 10 ng DNA template, 0.5 μM primer, 5U Taq polymerase) under the following conditions: initial denaturation at 95 °C for 1 min, denaturation at 95 °C for 30 s, primer annealing at 55 °C for 45 s, elongation at 72 °C for 50 s, and final elongations at 72 °C for 5 min, 32 reaction cycles of amplification. The PCR amplicons were analyzed on a 1% agarose gel and visualized under a UV transilluminator (Fig. S3 ). The plasmid constructs were sequenced by STARSEQ GmbH (Instituto de Biotecnologia, UNAM, Mexico). FinchTV version 1.4 was employed to analyze and confirm the sequenced nucleotides using EMBOSS (data not shown) (Fig. S4 ). Expression and purification of recombinant fusion peptides CusF3H+_LL-37_Renalexin and SmbP_LL-37_Renalexin For expression, the designed recombinant plasmid DNA construct was transformed by heat shock at 42 °C for 45 s into E. coli BL21(DE3) and E. coli SHuffle T7(DE3) calcium competent cells. An inoculum of 5 ml LB broth (30 μg/ml kanamycin) was made, inoculated with a fresh single colony of transformed cells, and incubated at 37 °C overnight with shaking at 220 rpm. Overnight cultures confirming cell viability and plasmid stability were used to inoculate 1000 ml LB broth (30 μg/ml kanamycin) in a baffled flask and incubated at 37 °C, 220 rpm for about 3–4 h until optical cell density (OD 600nm ) of 0.4–0.6 was obtained. Recombinant fusion peptide expression was induced with 1 M isopropyl β-D-1-thiogalactopyranoside (IPTG) at a final concentration of 1 mM. Induced culture flasks were incubated under the optimized condition of 25 °C, 220 rpm for 16 h. Cell pellets harboring the expressed recombinant fusion peptide were harvested by centrifugation at 8500 × g, 4 °C for 15 min into sterilized 50-ml Eppendorf tubes. Collected pellets were resuspended in ice-cold lysis buffer (500 mM NaCl, 50 mM Tris-HCl pH 8.0) and lysed by vortexing on ice with 0.1 mm glass beads. Clear soluble cell lysate was obtained after centrifugation at 8500 × g, 4 °C for 15 min. Purification of recombinant fusion peptide was performed via IMAC using the ÄKTA Prime Plus System (GE Healthcare) for fast protein liquid chromatography (FPLC). Briefly, a 1-ml HisTrap FF agarose-resin Ni(II) charged column was employed for the IMAC purification. The column was equilibrated with 5 column volumes (CV) of equilibrating buffer (500 mM NaCl, 50 mM Tris-HCl pH 8.0). The clarified soluble lysate was loaded onto the column under the conditions of 0.5 MPa pressure and 0.5 ml/min flow rate. Subsequently, the column was washed with 3 CVs of washing buffer (500 mM NaCl, 2.5 mM Imidazole, 50 mM Tris-HCl pH 8.0). The recombinant fusion peptide was eluted by gradient elution with elution buffer (500 mM NaCl, 200 mM Imidazole, 50 mM Tris-HCl pH 8.0). A 10 μl cell lysate, column flow-through, and elution fractions were analyzed on a 15% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), and the purity of protein bands (Tab. S1 ) was measured by densitometry using ImageJ software (v.2.0) (Adamíková et al. 2019 ). Enterokinase cleavage and purification of LL-37_Renalexin To induce enterokinase treatment, the purified fusion peptide elution fractions were pooled together in a 6-cm dialysis membrane and dialyzed against dialysis buffer (1× PBS pH 7.2) to desalt the purified peptide and exchange the elution buffer. In brief, the dialysis setup was incubated at room temperature for 1 h with gentle shaking on a magnetic stirrer, followed by overnight incubation at 4 °C. The concentration of the dialyzed fusion peptide was estimated by Bradford analysis using the Bovine Serum Albumin (BSA) (Tab. S2 , Fig. S5 and Tab. S3 , Fig. S6 ) as standard protein (Colyer and Walker 1996 ). A 1 mg purified fusion peptide was exposed to 20 U enterokinase cleavage (5 U/μl) for the release of hybrid peptide LL-37_Renalexin (tag-free). The cleavage reaction was incubated at room temperature for 16 h, followed by enzyme inactivation at −20 °C for 3 h. 10 μl of cleavage mixture was analyzed on a 15% and 18% Tricine SDS-PAGE. In purifying the tag-free hybrid peptide by IMAC, a 1.0 × 10-cm chromatographic syringe column was loaded with 1 ml of agarose resins charged with 1 M Ni(II) ions. The column was equilibrated with 10 CV of 1× PBS pH 7.2, and the enterokinase cleavage mixture was loaded into the column, mixed, and incubated at 4 °C for 1.5 h for efficient binding of protein tags to the affinity column. Column flow-through was collected as tag-free hybrid AMP LL-37_Renalexin. The protein tags were released with elution buffer (500 mM NaCl, 200 mM Imidazole, 50 mM Tris-HCl pH 8.0). The purified tag-free hybrid peptide was characterized on a 15% Tricine SDS-PAGE, and the concentration was estimated by Bradford analysis and Nanodrop absorbance readings at 280 nm (Tab. S4 , Fig. S7 and Tab. S5 , Fig S8 ). Antimicrobial activity assay of recombinant hybrid peptide LL-37_Renalexin Culture media Tryptic soy broth (TSB): tryptone (pancreatic digest of casein) 17.0 g, soytone (peptic digest of soybean) 3.0 g, glucose (dextrose) 2.5 g, sodium chloride 5.0 g, and dipotassium phosphate 2.5 g, pH 7.3. Bacteriological agar: Mueller–Hinton agar (MHA), beef extract 2.0 g, acid hydrolysate of casein 17.5 g, starch 1.5 g, agar 17.0 g, pH 7.3. Phosphate-buffered saline (PBS); sodium chloride 8 g, potassium chloride 0.2 g, sodium phosphate dibasic 1.44 g, potassium phosphate monobasic 0.245 g, pH 7.4. All culture media and buffers were used for microbial cultivation of bacteria clinical isolates, inoculum suspension preparations, and antimicrobial activity tests. Inoculum preparation Single colonies of test bacteria clinical isolates were cultured in 5 ml TSB and incubated at 37 °C, 220 rpm for 16 h. A 20 μl overnight culture was inoculated into 5 ml TSB and incubated at 37 °C with shaking until an optical cell density (OD 600nm ) of 0.8–1.0 was obtained (mid-logarithmic growth). Cells at the log growth phase were serially diluted in 1× PBS pH 7.2 buffer and the dilution suspensions at 1 × 10 5 CFU/ml equivalent to 0.5 McFarland turbidity standard were employed for antimicrobial assay. Antimicrobial activity Dose-response assay: minimum inhibition concentration (MIC) determination The antimicrobial activity of the hybrid peptide was evaluated against gram-positive and gram-negative bacteria clinical isolates of Staphylococcus aureus , Escherichia coli , methicillin-resistant Staphylococcus aureus , and Klebsiella pneumoniae from the University Hospital. The minimum inhibition concentrations against the test pathogens were ascertained and estimated in accordance with the modification of the National Committee for Clinical Laboratory Standards (NCCLS) for the broth microdilution method (CLS 2022 ). Briefly, bacteria cell cultures in the mid-log growth phase in TSB were serially diluted in 1× PBS pH 7.2 buffer to 1 × 10 5 CFU/ml. A 2-fold broth microdilution assay was performed in a 96-well microtiter plate with a 200 μl assay volume per well composed of 100 μl diluted peptide at concentrations ranging from 0.5–33 μM, 80 μl bacteria suspensions, and 20 μl TSB medium. The plate was incubated at 37°C for 3 h with shaking. After 3 h incubation, 0.1 ml aliquot was taken per well and a 10-fold dilution was made from which a 100 μl aliquot was spread on tryptic soy agar (TSA). Inoculated plates were incubated at 37 °C for 20 h and the remaining colony-forming units were evaluated (Tab. S6 , S7 , S8 , S9 , S14 , and S15 ), and MICs were calculated as the lowest peptide concentration that obviate visible turbidity using a modified B. Gompertz function for the line of best fit in dose-response analysis (Lambert and Pearson 2000 ). Time-killing assay The antibacterial killing kinetics of the hybrid peptide was evaluated against two bacterial isolates by ascertaining the time course to kill test bacteria isolates suspension of S. aureus (gram+), and E. coli (gram−). In brief, bacterial cultures in the mid-logarithmic growth phase were incubated as described previously with the peptide LL-37_Renalexin at a concentration of approximately 2× MIC in a TSB medium. The 10 μl aliquot suspensions were taken at every 20 min interval until a period of 3 h incubation was observed. A 10-fold dilution in 1× PBS pH 7.2 was made, and 0.1 ml aliquots were inoculated on TSA medium. Inoculated plates were incubated at 37 °C for 20 h. Log remaining CFU/ml of test pathogens were taken and plotted against time (Tab. S10 , and S11 ). Two control samples were made, positive control (bacterial suspension and kanamycin at the same peptide concentration) and negative control (bacterial suspension and 1X PBS buffer). The average of the total remaining CFU/ml from each treatment was evaluated (Tab. S12 and S13 ) and analyzed via one-way analysis of variance (ANOVA). The hybrid peptide with a single disulfide linkage (S–S bond) expressed in E. coli SHuffle T7(DE3) showing relatively lower minimum inhibitory concentrations (MICs) was employed for time-killing kinetic assay.
Results Design of hybrid peptide and construction of recombinant plasmids In designing the hybrid peptide with the molecular gene map shown (Fig. 2 A), we employed the mature amino acid sequences that encode for peptides LL-37 and Renalexin retrieved from the antimicrobial peptide database http://aps.unmc.edu/AP/prediction/ with the accession number AP00310/2K60 and AP00513/P39084, respectively. The amino acid GS (glycine and serine) between LL-37 and Renalexin in the gene construct was employed as a novel, simple, flexible peptide linker allowing for the construction of the hybrid peptide LL-37_Renalexin. DDDDK amino acid sequence at the N-terminal of the hybrid peptide serves as an enterokinase site that allows for the molecular cleavage of the protein tags SmbP and CusF3H+ yielding a tag-free hybrid AMP LL-37_Renalexin. For efficient production and affinity purification, the carrier proteins CusF3H+ and SmbP were inserted between the LL-37_Renalexin gene sequence and the start codon. The molecular Riben structure (Fig. 2 B) of the designed hybrid peptide predicted in i-TASSER ( https://zhanggroup.org/I-TASSER/ ) suggests a secondary structural motif with efficient production in E. coli (Kesidis et al. 2020 ). The structure predicted and modeled in i-Tasser was confirmed in the Phyre2 server (Fig. S9 ) ( http://www.sbg.bio.ic.ac.uk/phyre2 ) with 99.9–99.3% confidence level under the model template identifier (i.d) c2K6oA (LL-37) and c2fcgF (Renalexin) (Kelley et al. 2016 ). The application of GS peptide linker facilitates the design of a novel hybrid peptide by maintaining the molecular α-helix structure in LL-37 and the heptapeptide ring of Renalexin ( https://APD3.unmc.edu/structure ). The target hybrid peptide has the following biochemical properties: 59 amino acid length, +9 net charge, pI of 10.3, 44% hydrophobicity and 56% hydrophilicity, GRAVY index of 0.00, instability index of 21.3, aliphatic index of 102.37, and a molecular weight of 6.740 kDa theoretically predicted in http://www.expasy.org/tools/proparameter . The physicochemical properties of the hybrid peptide suggest strong electrostatic cationic polarity, hydropathicity, stability, and good isoelectric potentials which provide theoretical evidence of strong antimicrobial activity of this hybrid peptide at a varying pH range compared to its counterpart single peptides. Cellular toxicity of the hybrid peptide against healthy human cells was ascertained using ToxIBTL advanced bioinformatic online server ( https://server.wei-group.net/ToxIBTL/server.html ). ToxIBTL is an online in silico peptide and protein toxicity prediction tool that operates using evolutionary information and the physicochemical properties of peptide sequence via the integration of bottleneck principle to predict peptide or protein toxicity level. Our in silico cytotoxicity analysis (Fig. S10 ) showed that the hybrid peptide has no toxic effect (zero toxicity) on normal human cells at the niche of infection at a toxicity score of 3.7139784e-05 (0.0000371) which is by far below the standard threshold of 0.5. The DNA nucleotide (gene) of the coding sequence (CDS) was chemically synthesized by GenScript based on the optimized amino acid sequence that encodes for the recombinant fusion proteins CusF3H+_LL-37_Renalexin and SmbP_LL-37_Renalexin. The gene map of the synthetic DNA constituted restriction sites for Nde I, Kpn I, Xho I, and enterokinase, a protein tag site at the N-terminal, with the hybrid peptide site between the protein tag site and a stop codon. The 492-bp (SmbP tag construct) and 480-bp (CusF3H+ tag construct) synthetic DNAs were molecularly cloned into pET30a+ at Nde I and Xho I restriction sites under the control of T7 promoter using the T4-DNA ligase for the design of two plasmid expression vectors pET30a+_CusF3H+_LL-37_Renalexin and pET30a+_SmbP_LL-37_Renalexin. The plasmid construct was confirmed by colony-based PCR screening using the T7 promoter and terminator-specific primers. After colony-based PCR screening, the correct recombinant plasmid sequence was further confirmed by DNA sequencing (data not shown). Expression and IMAC purification of recombinant fusion protein The protease-deficient bacterial strains E. coli BL21(DE3) and E. coli SHuffle T7(DE3) competent cells used as expression hosts encode for the T7 RNA polymerase allowing for the expression of the recombinant fusion peptides CusF3H+_LL-37_Renalexin and SmbP_LL-37_Renalexin under the influence of T7 promoter. IPTG induction and expression of fusion peptides successfully showed an efficient insertion of DNA under the T7 promoter in the designed plasmid constructs mentioned above. The carrier proteins CusF3H+ and SmbP show significant advantages in the production and purification of the recombinant fusion peptides expressed as soluble proteins with no formation of inclusion bodies (Perez-Perez et al. 2021 ; Vargas-Cortez et al. 2016 ). In evaluating small-scale expression of recombinant fusion proteins CusF3H+_LL-37_Renalexin and SmbP_LL-37_Renalexin (17 kDa) in E. coli BL21(DE3) and E. coli SHuffle T7(DE3), soluble cell lysates were prepared by lysing cell pellets collected from 2 ml of 1 mM IPTG-induced overnight cultures to access the presence of the target protein in soluble cell lysate. Analysis of 5 μl aliquot of clear lysate on a 15% SDS-PAGE showed successful evidence of expression of the target fusion proteins with 17–18 kDa-expected protein band compared to non-induced cells as negative control samples (Fig. 3 ). For the large-scale production of recombinant fusion proteins, E. coli strains E. coli BL21(DE3) and E. coli SHuffle T7(DE3) were used as expression hosts. From a 1-L expression volume, cell pellets were collected from 1 mM IPTG-induced overnight cultures and lysed by mechanical vortexing on ice with 0.1 mm glass beads. After cell lysis, clear soluble lysate was collected and employed as a protein source for the immobilized metal affinity chromatography. For the purification of fusion protein SmbP_LL-37_Renalexin and CusF3H+_LL-37_Renalexin, the ÄKTA Prime Plus system (GE Healthcare Systems) was employed for fast protein liquid chromatography (FPLC) by metal affinity chromatography. A 1-ml HisTrap FF column charged with Ni(II) was used to isolate the target recombinant peptide from the pool of cellular proteins present in the soluble lysate fractions collected. Analysis of IMAC purification fractions of the recombinant fusion protein on a 15% SDS PAGE (Fig. 4 ) showed evidence of target peptide in the elution fractions (200 mM imidazole) without any trace of peptide indications in the column flow-through. This result confirms the high affinity of CusF3H+ and SmbP to agarose-resin Ni(II)-charged column that facilitates the binding of the target proteins unto the column while the untargeted cellular proteins exit the column as flow-through (Montfort-Gardeazabal et al. 2021 ; Perez-Perez et al. 2021 ; Vargas-Cortez et al. 2017 ). Our Bradford quantification analysis using standard Bovine Serum Albumin (BSA) calibration equations (data not shown) indicated a peptide concentration of 3.136 mg/L for CusF3H+_LL-37_Renalexin produced in E. coli SHuffle T7(DE3) and 1.523 mg/L for SmbP_LL-37_Renalexin produced in E. coli BL21(DE3) with a purity of 90–95% matching the purity standard of commercially available synthetic therapeutic peptides. We observed a 2-fold higher recombinant fusion peptide yield in E. coli SHuffle T7(DE3) compared to production in E. coli BL21(DE3). Enterokinase cleavage and purification recombinant LL-37_Renalexin Our previous studies (Montfort-Gardeazabal et al. 2021 ; Perez-Perez et al. 2021 ) reported the abrogative effect on the bioactivity of recombinant antimicrobial peptides with attached protein tags. We employed the restriction enzyme enterokinase for the selective cleavage of protein tags CusF3H+ and SmbP from the above-purified fusion peptides. Electrophoretic analysis of inactivated enterokinase cleavage mixture revealed three protein bands of size 18 kDa (uncleaved fusion peptide), ≈13 kDa (CusF3H+ or SmbP), and ≈10 kDa (LL-37_Renalexin, tag-free) on Tricine SDS-PAGE (Fig. 5 ). The uncleaved fusion peptide detected was due to an incomplete enzymatic cleavage reaction. Finally, the tag-free LL-37_Renalexin was purified via a one-step IMAC purification using an agarose resin Ni(II)-charged syringe column unto which the uncleaved fusion peptide and the carrier proteins CusF3H+ and SmbP remain bounded, allowing for the elution of the target hybrid peptide LL-37_Renalexin as column flowthrough. A 2.16 mg/L for tag-free LL-37_Renalexin expressed in E. coli SHuffle T7(DE3) and 0.72 mg/L expressed in E. coli BL21(DE3) were obtained after Bradford and Nanodrop spectrometry (A280nm) quantification analysis. Antimicrobial activity of recombinant LL-37_Renalexin The antimicrobial potency of the purified hybrid peptide against clinical isolates of S. aureus , E. coli , MRSA, and K. pneumoniae was evaluated, and the minimum inhibitory concentrations (MICs) were determined via the broth microdilution assay as described by NCCLSI (Lacy et al. 2004 ). Data on remaining colony-forming units (CFU/ml) of test pathogens were taken and analyzed after overnight culture with the peptide. The MIC, defined as the minimum peptide concentration that prevented visible turbidity in the test pathogen, was calculated using a modified Benjamin Gompertz sigmoid function (Lambert and Pearson 2000 ) from the plot of peptide concentrations against the remaining CFU/ml (dose-response plot). The dose-response plots (Fig. 6 ) show the antibacterial activity of the hybrid AMP LL-37_Renalexin at MIC levels of 10–27 μM, much lower than the reported MICs of the single-peptide LL-37 and Renalexin (50–100 μM) (Aleinein et al. 2013 ; Kang et al. 2019 ; Perez-Perez et al. 2021 ). The MIC result we observed in this study affirms with data reported on related hybrid and dimeric peptides tested against the same bacterial pathogens (Cheng et al. 2021 ; Dürr et al. 2006 ; Kang et al. 2019 ; Wei and Zhang 2022 ; Seyedjavadi et al. 2021 ). The dose-response antimicrobial activity results indicated that the test bacterial pathogens were sensitive to the recombinant hybrid peptide at active peptide concentrations as low as 10 μM and 33 μM. Interestingly, we evaluated the bioactivity of the hybrid peptide without disulfide-linkage expressed in E. coli BL21(DE3); the results (Table 1 ) suggested no significant difference in the MICs compared to the hybrid peptide with disulfide-linkage expressed in E. coli SHuffle T7(DE3). Time-kill kinetics analysis The time-kill kinetic assay (Fig. 7 ) disclosed that the hybrid peptide shows a multifunctional antibacterial activity within 1.5 h via disruption of membrane integrity and membrane traversing, demonstrating a strong but relatively slow antibacterial potency against all investigated gram-positive and gram-negative bacterial pathogens as compared to the classical antibiotic kanamycin that exhibited its antibacterial activity within less than an hour. We observed that the antibacterial activity of the hybrid peptide in comparison to the known antibiotic kanamycin analyzed against total remaining CFU/ml showed a significant difference at p -values < 0.05 (Fig. 8 ), with the hybrid peptide showing relatively similar antibacterial actions as exhibited by kanamycin (Aleinein et al. 2013 ; Hanafiah et al. 2020 ; Montfort-Gardeazabal et al. 2021 ; Perez-Perez et al. 2021 ).
Discussion Antimicrobial peptides (AMPs) as drug candidates are being considered the new hope for the biomedical and pharmaceutical industries in conjunction with their multifunctional antibacterial pharmacological actions against infectious agents (Montfort-Gardeazabal et al. 2021 ; Moretta et al. 2021 ; Nuti et al. 2017 ). The production of AMPs as therapeutic peptides via the applications of recombinant DNA technology (rDNA) and the use of cost-effective microbial expression systems have facilitated the large-scale acquisition of bioactive AMPs, enhancing clinical and research applications (Akhtar et al. 2012 ; Dar et al. 2017 ). In addition to the successful application of rDNA, the efficient, fast, and easy growth of microbial expression hosts like E. coli BL21(DE3) and E. coli SHuffle T7(DE3) has allowed for commercial production of bioactive cationic AMPs independent of posttranslational modifications (Aleinein et al. 2013 ; Klubthawee et al. 2020 ; Montfort-Gardeazabal et al. 2021 ). Novel AMPs like Brevinin-2R, Renalexin, Cecropin A, and LL-37 as single peptides have been produced as recombinant protein-based drug candidates in microbial systems and have shown a promising antimicrobial effect (Aleinein et al. 2013 ; Chhetri et al. 2015 ; Zhang et al. 2018 ). The thriving commercial production of AMPs in E. coli largely depends on the design and use of protein tags, including but not limited to poly-histidine, maltose-binding protein (MBP), thioredoxin protein (THX), and glutathione S-transferase (GST), that allow for the expression of recombinant proteins as fusion proteins in benign forms (Akhtar et al. 2012 ; El-Gayar 2015 ; Gddoa Al-sahlany et al. 2020 ; Mo et al. 2018 ; Riguero et al. 2020 ). Recently, our newly designed small metal-binding proteins SmbP and CusF3H+ have been exploited as protein tags which aided in the production and purification of AMPs like LL-37 and Bin1b, and green fluorescence protein (GFP) in different E. coli strains (Montfort-Gardeazabal et al. 2021 ; Perez-Perez et al. 2021 ; Vargas-Cortez et al. 2016 ; Vargas-Cortez et al. 2017 ). Our previous study has unraveled the capabilities of these protein tags that facilitate the secretion, folding, and purification of expressed recombinant LL-37 and Bin1b as single peptides with intact bioactivity at purity above 80% (Montfort-Gardeazabal et al. 2021 ; Perez-Perez et al. 2021 ; Santos et al. 2019 ). Peptide hybridization has been considered an advanced technique for the design of novel AMPs having reliable peptide stability, long half-life with intact therapeutic activity with hybrid peptides like Cecropin A_Thanatin, and Indolicidin_Renalexin showing broad-spectrum antimicrobial activity against multidrug-resistant bacterial pathogens (Bayarbat et al. 2016 ; Seyedjavadi et al. 2021 ; Wade et al. 2019 ). The broad-spectrum antibacterial activity of LL-37 as a single peptide although at a higher peptide concentration has led us to design a recombinant hybrid peptide production strategy via the application of GS flexible peptide linker and a carrier proteins CusF3H+ and SmbP. The simple GS peptide linker enhances the expression of the hybrid peptide LL-37_Renalexin and maintains the spatial configuration within the hybrid peptide with intact and advanced bioactivity. This data strongly suggests that the GS peptide linker can be employed as a reliable, simple, flexible linker for the design and expression of recombinant therapeutic peptides as compared to the RGGPDGSGPDESGPDE flexible linker employed in the design of hybrid and dimeric peptides with primary structural modifications (Klubthawee et al. 2020 ; Seyedjavadi et al. 2021 ). In this study, we have efficiently employed the mature amino acids of LL-37 and Renalexin for the design of a novel hybrid peptide LL-37_Renalexin with zero cytotoxicity against healthy human cells. We reliably cloned the cDNA that encodes for the target peptide into pET30a+ under the T7 promoter and terminator regions and obtained a successful expression in E. coli BL21(DE3) and E. coli SHuffle T7(DE3) under the condition of 25 °C, for 16 h with 1 mM IPTG. In other studies, the microbial strains, induction, and temperature conditions demonstrated profound effect on the expression of both single, hybrid, and dimeric peptides with inclusion bodies indications (Chhetri et al. 2015 ; Montfort-Gardeazabal et al. 2021 ; Seyedjavadi et al. 2021 ; Shang et al. 2020 ; Wade et al. 2019 ; Xu et al. 2014 ). The expression condition and peptide isolation protocols observed in this study made it possible for efficient production coupled with higher yield. Recombinant fusion peptides CusF3H+_LL-37_Renalexin and SmbP_LL-37_Renalexin expression level and its presence in soluble cell lysate provide an evidential advantage of protein tag CusF3H+ and SmbP over others like glutathione S-transferase, maltose-binding protein, amyloid-β peptide, and thioredoxin tag (Aleinein et al. 2013 ; Chhetri et al. 2015 ; Zhang et al. 2018 ) yielding up to 95% peptide purity which matched the purity standard of commercially available synthetic therapeutic peptides (Zhongxuan et al. 2020 ). In this present study, Bradford quantification of purified and PBS-dialyzed protein elution fractions revealed a total recombinant peptide yield of 1.5–3.1 mg/L fusion proteins SmbP_LL-37_Renalexin and CusF3H+_LL-37_Renalexin expressed in BL21(DE3) and SHuffle T7(DE3), respectively. We observed a 2-fold higher peptide yield in E. coli SHuffle T7(DE3) for both SmbP and CusF3H+ tagged fusion proteins than in E. coli BL21(DE3), providing relevant supportive data on the usage of E. coli SHuffle T7(DE3) as microbial host for the production of either single or hybrid recombinant hybrid peptides with or without disulfide bonds (Montfort-Gardeazabal et al. 2021 ). Our result showed a higher expression level in recombinant hybrid peptide as soluble protein than reported from other studies, 0.9 mg/L by Seyedjavadi et al. ( 2021 ); 900 μg/L by Clement et al. ( 2015 ), and 0.3 mg/L by Cheng et al. ( 2021 ). We employed the enzyme enterokinase for the selective cleavage of protein tags CusF3H+ and SmbP from the purified fusion peptides due to the presence of an enterokinase site between the hybrid peptide and the protein tag. Analysis of inactivated enterokinase cleavage mixture revealed three protein bands of an approximate molecular size of 18 kDa (uncleaved fusion peptide), 13 kDa (CusF3H+ or SmbP), and 10 kDa (LL-37_Renalexin, tag-free) on Tricine SDS-PAGE with tag-free recombinant hybrid peptide showing slightly higher band size than the theoretically expected size (6.740 kDa). This result can be attributed to the inefficient enterokinase cleavage observed, which may be associated with the presence of phenylalanine, leucine, and isoleucine residues and heptapeptide motif (Rana box) in Renalexin that forms a cyclic disulfide bond aiding a molecular structural loop formation folded unto the N-terminal that is known to influence enzymatic cleavage and peptide reduction. Also, the high hydrophobicity of the peptide which prevents complete reduction with SDS and mercaptoethanol reagents, and the peptide molecular folding in aqueous systems all of which influence poor peptide mobility. These findings agree with previous studies reported where the single peptides LL-37 and Renalexin show higher protein band sizes than the theoretically determined sizes (Aleinein et al. 2013 ; Perez-Perez et al. 2021 ). A 0.7–2.1 mg/L LL-37_Renalexin (tag-free) peptide was obtained upon spectroscopic and Bradford quantification of the second IMAC purification elution fractions (Montfort-Gardeazabal et al. 2021 ; Perez-Perez et al. 2021 ; Vargas-Cortez et al. 2017 ). The relatively slow induction of the antibacterial activity of LL-37_Renalexin compared to the known antibiotic kanamycin we observed in the time-killing kinetic assay results can be related to the high hydrophobicity of the hybrid peptide which may cause partial exposure of hydrophilic regions to bacterial membrane (Wei and Zhang 2022 ) coupled with the presence of monovalent Na(I) and K(I) cations in the assay medium that are known to cause shielding effects between the cationic peptides and anionic bacterial membrane surface, hence, the delay in eliciting antibacterial activity in the case of the hybrid peptide as compared to kanamycin (Huan et al. 2020 ; Nuti et al. 2017 ). Our findings show that the hybrid peptide LL-37_Renalexin with 44% hydrophobicity and 56% hydrophilicity elicited near-microbicidal activity against all tested pathogens with above 85% reduction in bacteria colony-forming units at 33 μM peptide concentration. The reduction in CFU/ml observed suggests a bactericidal activity since the level of reduction ( R L ) in CFU/ml is more significant than three times the logarithm CFU/ml of bacteria (Klubthawee et al. 2020 ; Zhang et al. 2018 ). All S. aureus , E. coli , MRSA, and K. pneumoniae clinical isolates showed 85% sensitivity at 33 μM as minimum peptide bactericidal concentration with about 25% increment in sensitivity indicative of higher antibacterial potency of this novel hybrid peptide compared to its counterpart single-peptide LL-37 as reported from our previous study with 64% sensitivity against E. coli and 69% against S. aureus (Perez-Perez et al. 2021 ). We also observed approximately a 2-fold reduction with respect to the minimum inhibitory hybrid peptide concentration required to inhibit bacterial growth as compared to its single-peptide LL-37. We envisioned that the hybrid peptide’s antimicrobial effects are brought about through its ability to disrupt cell membranes, thanks to its pronounced cationic polarity which enables it to create a strong electrostatic bond with the negatively charged bacteria membrane. Additionally, it is thought that the hybrid peptide’s v-shaped structure, made possible by the presence of the GS flexible linker, allows it to effectively engage with bacterial chromosomal DNA. This interaction can lead to the formation of supercoils, ultimately hindering DNA replication and transcription (Zhang et al. 2021 ). In this study, we have designed, produced, and purified a novel multifunctional recombinant hybrid peptide LL-37_Renalexin for the first time via the application of newly designed flexible GS peptide linker and a characterized carrier proteins SmbP and CusF3H+. The small metal-binding protein tags SmbP and CusF3H+ provide an evidential advantage in cytoplasmic production and purification of the novel hybrid AMP LL-37_Renalexin with intact biochemical properties and can be applied as a new avenue to produce recombinant peptides and proteins. The purified tag-free hybrid peptide LL-37_Renalexin exhibited above 85% reduction in bacteria CFU/ml against S. aureus , E. coli , MRSA, and K. pneumoniae clinical isolates at lower minimum inhibition concentration levels of 10–33 μM as compared to its counterpart single-AMPs LL-37 and Renalexin of 50–100 μM reported (Aleinein et al. 2013 ; Kang et al. 2019 ; Perez-Perez et al. 2021 ), making it a competitive antimicrobial agent. From our antibacterial bioassay findings, we proposed that the newly designed hybrid peptide LL-37_Renalexin can be classified as an antibacterial peptide that may have no toxic effects against normal cells at the niche of infection as theoretically predicted. Previous investigations (Aleinein et al. 2013 ; Jindal et al. 2015 ; Kang et al. 2019 ; Perez-Perez et al. 2021 ) have confirmed the exclusive antibacterial effect of recombinant single-peptides LL-37 and Renalexin, respectively, showing no toxic effects against normal human cells. Notably, we successfully express the hybrid peptide in E. coli with intact bioactivity conferred by the conserved secondary α-helical structures as seen in the single peptides supporting the predicted secondary structure of LL-37_Renalexin clearly showing the conserved helical domains of LL-37 and Renalexin. It is imperative to note, however, that this study lacks empirical data from wet laboratory circular dichroism (CD) spectrometry and 3-(4,5-dimthylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay to support the theoretically predicted structure and toxicity. The recombinant DNA strategies used in the design, production, and purification of recombinant fusion proteins provide a reliable platform and protocol for the expression and purification of therapeutic recombinant proteins in E. coli BL21(DE3) and E. coli SHuffle T7(DE3) as microbial expression hosts.
Abstract An alarming global public health and economic peril has been the emergence of antibiotic resistance resulting from clinically relevant bacteria pathogens, including Enterococcus faecium , Staphylococcus aureus , Klebsiella pneumonia , Acinetobacter baumannii , Pseudomonas aeruginosa , and Enterobacter species constantly exhibiting intrinsic and extrinsic resistance mechanisms against last-resort antibiotics like gentamycin, ciprofloxacin, tetracycline, colistin, and standard ampicillin prescription in clinical practices. The discovery and applications of antimicrobial peptides (AMPs) with antibacterial properties have been considered and proven as alternative antimicrobial agents to antibiotics. In this study, we have designed, produced, and purified a recombinant novel multifunctional hybrid antimicrobial peptide LL-37_Renalexin for the first time via the application of newly designed flexible GS peptide linker coupled with the use of our previously characterized small metal-binding proteins SmbP and CusF3H+ as carrier proteins that allow for an enhanced bacterial expression, using BL21(DE3) and SHuffle T7(DE3) Escherichia coli strains, and purification of the hybrid peptide via immobilized metal affinity chromatography. The purified tag-free LL-37_Renalexin hybrid peptide exhibited above 85% reduction in bacteria colony-forming units and broad-spectrum antimicrobial effects against Staphylococcus aureus , Escherichia coli , Methicillin-resistant Staphylococcus aureus (MRSA), and Klebsiella pneumoniae bacteria clinical isolates at a lower minimum inhibition concentration level (10–33 μM) as compared to its counterpart single-AMPs LL-37 and Renalexin (50–100 μM). Key points • The hybrid antimicrobial peptide LL-37_Renalexin has been designed using a GS linker. • The peptide was expressed with the carrier proteins SmbP and CusF3H+. • The hybrid peptide shows antibacterial potency against clinical bacterial isolates. Supplementary Information The online version contains supplementary material available at 10.1007/s00253-023-12887-5. Keywords
Supplementary Information
Acknowledgements We thank Mexico’s Consejo Nacional de Humanidades Ciencias y Tecnologias (CONAHCYT) for the financial support to the graduate student JKN. Author contribution JKN made the DNA constructs, performed the experiments, and wrote the manuscript. NGC-V provided the fully characterized S. aureus , E. coli , MRSA, and K. pneumoniae strains from his laboratory at the University Hospital and assisted in the evaluation of the antimicrobial assays for MRSA and K. pneumoniae . XZ edited the manuscript, designed the experiments, and proposed the research as Principal Investigator. All authors read and approved the final manuscript. Funding This work was funded by grant UANL-PAICYT-330-CN-2022 awarded to XZ. Data availability All data supporting the findings of this study are available within the paper and its supplementary information. Declarations Ethics approval and consent to participate This in vitro research study does not involve human or animal models; hence, no ethical consent and approval were required. Conflict of interest The authors declare no competing interest.
CC BY
no
2024-01-15 23:42:02
Appl Microbiol Biotechnol. 2024 Jan 13; 108(1):1-15
oa_package/f9/42/PMC10787891.tar.gz
PMC10787892
37679534
Background According to the National Cancer Institute (NCI), an individual is considered a cancer survivor from diagnosis to the end of life [ 1 ]. In the United States (US), it is estimated that by 2026, the number of cancer survivors will surpass 20 million, which can be attributed to the ongoing innovation of treatment and early disease detection [ 2 , 3 ]. While this increase in survival is optimistic, this population requires additional healthcare services to prevent or manage chronic health conditions, sequelae of cancer treatment, and monitoring for cancer reoccurrence [ 3 ]. Moreover, given that at least 50% of cancer survivors will experience physical and mental health consequences due to their disease or treatment [ 4 ], is important to mitigate healthcare delays in this population. Overall, healthcare delays (due to the domains of accessibility, financial burden, and social support) can significantly impact cancer survivors, most notably individuals who are ethnic minorities, lower SES, or uninsured [ 5 ]. Cancer survivors who experience healthcare delays can significantly suffer the consequences as their care is typically time sensitive [ 6 ]. Because timely cancer care is associated with a favorable prognosis, barriers or delays to treatment can result in a more advanced stage of cancer at the time of eventual care, thus, resulting in poorer outcomes [ 7 ]. However, physical health is not the only aspect affected; mental health may also suffer at the hands of delayed care [ 8 ]. For instance, COVID-19 led to numerous appointment cancelations for cancer survivor patients, including self-cancelations that were caused by depression and anxiety symptoms surrounding the pandemic and safe access to care [ 9 ]. These missed appointments, in turn, exacerbated patients’ fears of cancer recurrence as their follow-up care were halted (e.g., laboratory testing, imaging, and appointments), impacting their overall well-being and physical and mental health [ 10 , 11 ]. One factor associated with increases in healthcare delays among cancer survivors is lower SES status, which is associated with all-cause mortality risk, poorer mental (e.g., depression), and physical health outcomes [ 12 , 13 ]. Similarly, SES has been linked to a range of cancer outcomes and higher SES (suggesting more financial resources and the ability to afford medical care) is associated with decreased in the length of healthcare delays [ 14 ]. Because of the long-term and specialized care needed for cancer survivors, they are at risk of experiencing higher financial burden, which in turn, impacts receipt of survivorship care and increases the risk of mortality, and worsens quality of life [ 15 ]. A second factor associated with healthcare delays among cancer survivors is a lack of health literacy, which presents a barrier in properly understanding, communicating, and obtaining information required to navigate the complexity of the healthcare systems efficiently to obtain the care needed and to make educated health decisions [ 16 , 17 ]. Among the general population, low health literacy has been positively associated with various care delays, including seeking treatment, forgoing care, and struggling with accessibility to needed care and providers [ 18 ]. As cancer survivors have complex healthcare needs, mastering the skills required for their continual care is important. Being foreign-born (immigrant) brings an additional barrier to accessing healthcare and consequently promotes healthcare delays [ 19 ]. These barriers among foreign-born may be partially attributed to a higher likelihood of lack of insurance, healthcare cultural perception, and English skills proficiency [ 20 ]. Evidence suggests that cancer survivors who are immigrants have a lower quality of life and higher depression symptomology compared to native-born [ 21 ]. Moreover, foreign-born individuals may experience unique barriers (e.g., language, discrimination, laws and regulations to qualify for services) that impact the quality of care they can receive. For instance, language barriers and insurance difficulties caused by laws and policies may make it much more difficult for an immigrant patient than a US-born patient to receive adequate needed care [ 22 ]. While it is well established that SES barriers and a lack of health literacy are associated with healthcare delays, there is limited and inconsistent knowledge of how these associations differ by nativity. To guide the selection of our variables for our model, we used the theoretical framework from Wafula and Snipes for barriers to healthcare among Black immigrants in the US [ 23 ]. Additionally, we adapted their framework to focus on assessing the association between the barrier factors (i.e., SES barriers, health literacy) and general healthcare delays among cancer survivors and tested the moderating effects of nativity between SES barriers and health literacy with general healthcare delays (Fig. 1 ). Thus, this study aims to contribute to the current literature by examining whether nativity status modifies the relationship between a combination of SES and health literacy barriers, and healthcare delays in a large national cohort.
Methods Data collection and sample Cross-sectional data for this study were obtained from the “All of Us” research program collected by online survey between May 2018 and April 2021. Briefly, this program is open to individuals who are 18 and over and are living in the US. Participants signed a consent form following the Declaration of Helsinki for data collection. The participants’ data used are de-identified and available to approved researchers. The All of Us program was approved by the National Institutes of Health (NIH) Institutional Review Board (IRB). In this study, cancer survivors were defined as those participants who indicated that they had ever been diagnosed with cancer. Inclusion criteria for our cohort included participants who were ever told by their healthcare provider that had/have cancer. Skin cancer is one of the most prevalent cancers in the US with most cases being reasonably benign basal cells, not often tracked on most cancer registries, and having over 90% 5-year survival rate [ 24 ]. In addition, previous studies have excluded these cancer survivors as their follow-up care is often reasonably minor [ 25 ]. Thus, we excluded those with skin cancer and participants with missing data on the healthcare delay survey questions.
Results The median age of the study population ( n = 10,020) was approximately 64 (interquartile range [IQR Q1, Q3] 55.5, 71.8) years. The majority of participants were female (66.1%), US-born (92%), and self-identified as White (82.3%). There was a higher distribution of females vs males and other sex, foreign-born vs US-born, and Black cancer survivors vs all other race/ethnicity categories that had three or more SES barriers (Table 1 ). While a higher proportion of females vs males and other sex, foreign-born vs US-born, and Hispanic cancer survivors vs all other race/ethnicity cancer survivors had one or more healthcare delays (Table 2 ). Results from the multivariable-adjusted model showed that neither nativity (OR 1.04, 95% CI [0.87, 1.25]) nor health literacy (OR 1.20, 95% CI [0.89, 1.59]) were statistically significantly associated with healthcare delays (see Table 3 ). However, when assessing for a p-trend for health literacy, for every one-unit increase in health literacy there was an 8% (OR 0.92, 95% CI 0.89, 0.95) decrease in the likelihood to experience healthcare delays. Furthermore, compared to those who did not have any SES barriers, those who reported two or three or more were 65% (OR 1.65, 95% CI [1.43, 1.90]) and 118% (OR 2.18, 95% CI [1.84, 2.58]) more likely to experience delays in healthcare, respectively. In addition, using SES barriers as a continuous measure to test for a p-trend, we found that for every one additional barrier increase, there was a 29% increase (OR 1.29, 95% CI [1.23, 1.36]) in the likelihood of experiencing healthcare delays. The association between SES barriers and healthcare delays differs by nativity status ( p interaction = 0.02). The stratified model by nativity showed that among those who were foreign-born, those who experienced two or three or more SES barriers were almost three times (OR 4.35, 95% CI [2.61, 7.34] vs OR 1.53, 95% CI [1.32, 1.78]) and two times (OR 3.83, 95% CI [2.14, 6.98] vs OR 2.10, 95% CI [1.76, 2.50]) as likely to experience healthcare delays as their US-born counterparts who experienced the same levels of SES barriers, respectively (see Table 3 ). Assessing for p-trend in the stratified model by nativity, we found that for every additional SES barrier experienced among foreign-born individuals, they were 72% (OR 1.72, 95% CI [1.43, 2.08] vs OR 1.27, 95% CI [1.21, 1.34]) more likely to experience healthcare delays compared to their US counterparts. Finally, in the stratified model, low health literacy was associated with a 41% (OR 1.41, 95% CI [1.02, 1.97]) increase in the likelihood of healthcare delays, and each one-point increase in health literacy score was associated with a 9% (OR 0.91, 95% CI [0.88, 0.94]) decrease in the odds of healthcare delays among US-born cancer survivors. While for foreign-born cancer survivors, low health literacy was not statistically significantly associated with experiencing healthcare delays (OR 0.66, 95% CI [0.34, 1.25]), and for every one-unit increase in health literacy score there was a 2% (OR 0.98, 95% CI [0.90, 1.07]) decrease in the odds of healthcare delays. Although this association did not reach statistical significance (see Table 3 ).
Discussion Using data from the All of Us research cohort, we aimed to investigate the associations between SES barriers and health literacy with healthcare delays. We also explored whether there were any differences in these associations by nativity. We found that among all cancer survivors, health literacy (binary) and nativity were not statistically significantly associated with healthcare delays. We also found that experiencing 2 or 3+ SES barriers was significantly associated with an increased likelihood of healthcare delays. Further, at equal levels of SES barriers, foreign-born individuals had significantly higher odds of healthcare delays when compared to US-born individuals. Lastly, in our separate models by nativity status assessing for a trend, we found that health literacy was inversely associated with healthcare delays among US-born cancer survivors only. Socioeconomic barriers Our multivariable model suggested that cancer survivors who experienced more than one SES barrier experienced an increase in the likelihood of healthcare delays compared to cancer survivors who experienced no SES barriers. These findings are consistent with previous research that found that low SES cancer survivors are more likely to not receive appropriate follow-up care [ 30 ] and that those who experience financial, housing, and employment barriers have a greater likelihood of delaying needed care [ 31 – 33 ]. Similarly, in another study, cancer survivors from low SES who reported lower income and education levels were less likely to have follow-up care discussions with their medical providers [ 30 ]. This lack of follow-up discussions can potentially contribute to prolonged delays in preventative care. Regarding education, in a previous study by Gonzalez and colleagues, cancer survivors who had a college degree or higher were more likely to have higher access to care but experienced more delays than those with less than or equal to a high school diploma [ 34 ]. Perhaps belonging to a higher educational level improves health literacy, enabling cancer survivors to adequately make an informed decision when seeking the appropriate care needed. Among cancer survivors, those with low SES barriers are more likely to delay medical care, preventive care (dental and vision care), and not fill prescription medications due to cost-related concerns [ 32 , 35 , 36 ]. Furthermore, as cancer survivors are met with the unexpected financial costs of cancer treatment, they may worry about struggling to meet their housing and household bills payments, food insecurities, and retirement [ 37 ], thus, potentially forgoing or delaying the crucial care they are required to enhance their survival. Similarly, housing insecurities are linked with negative health outcomes and poor access and quality of healthcare [ 31 , 38 ]. Lastly, the impact of modifiable factors such as education and SES are inversely associated with experiencing more unmet healthcare needs [ 39 ] and a lack of health insurance [ 39 , 40 ]. Hence, uninsured cancer survivors have a higher risk for comorbidities, bearing a greater mortality risk than uninsured non-cancer survivors [ 41 ]. Nativity differences Previous research among cancer survivors has shown nativity to be a factor in cost-related barriers among US-born Hispanic cancer survivors [ 42 ]. In our study, we found that foreign-born individuals who experienced the same level of SES barriers had higher likelihood of experiencing healthcare delays than their native-born counterparts when compared to those who experienced no SES barriers. Previous research that explored nativity differences among cancer survivors is limited and inconsistent. For example, although the results did not reach statistical significance, Diamant and colleagues reported the opposite in a sample of non-cancer survivors, showing that non-native-born were less likely to report healthcare delays compared to native-born individuals [ 43 ]. Whereas, in a study of female cancer survivors that assessed disparities in healthcare access and utilization, they found that non-US-born females were less likely to report having a routine place to go to meet their healthcare compared to US-born cancer survivors [ 44 ]. This is important, as not having a primary healthcare office to seek care or have routine services can promote delaying accessing the extended care cancer survivors need. In addition, a higher prevalence of sociodemographic and SES barriers (e.g., income, education) was found among foreign-born individuals than among native-born in two North American countries (US and Canada), and disparities in healthcare access were higher among foreign- compared to native-born [ 20 ]. Thus, supporting the role of SES barriers among foreign-born individuals. Health literacy Our study found that, after adjusting for confounding sociodemographic, nativity, and SES barriers, health literacy was not statistically significantly associated with healthcare delays in our entire study cohort. However, we found that solely among US-born participants, limited health literacy was associated with an increased likelihood of healthcare delays compared to adequate health literacy. We also saw a statistically significant monotonic relationship between increased health literacy scores and decreased odds of healthcare delays among US-born. While our findings only reached statistical significance among US-born cancer survivors, the magnitude of our findings was in the same direction among foreign-born cancer survivors. This indicates that regardless of nativity, increasing health literacy could mitigate the impact of healthcare delays. While we controlled for nativity, sociodemographic, and SES barriers, our findings suggest that health literacy could also be associated with cultural differences and language that we were not able to control in our study. For example, cancer survivors with low health literacy may suffer difficulties when trying to decode their symptoms or understand their diagnoses through communication with providers, which could result in healthcare delays and a later stage of disease at diagnosis [ 45 ]. In the case of cancer survivors, there is a great need for complex health services after they have completed their primary treatment for their illness and subsequent management of their health [ 46 ]. These deficiencies in communicating efficiently with their providers could increase the risk of healthcare delays. Strengths and limitations An important strength of this study is using data from the All of Us research program, as it has a high proportion of enrolled underrepresented minority populations, which increases our sample size and access to geographically and ethnically diverse populations. Similarly, we were able to analyze a relatively large sample with complete socioeconomic and sociodemographic data of foreign-born cancer survivors, which helps to expand the current cancer survivor, cancer health disparities, and health disparities research. Lastly, our results can be generalized to cancer survivors and individuals with similar characteristics and settings that experienced similar SES barriers in the US. This study is not without limitations. The cross-sectional nature of the design does not allow us to establish a temporal or causal relationship. There is potential for misclassification for some of the factors included in our main independent variables as this data is from self-reported questionnaires. Our sample was also comprised of a higher distribution of individuals with a college degree or higher and a higher income. We were unable able to assess for acculturation; thus, future studies should account for it as it can be a potential confounder in these associations. Using a complete case (CC) analysis method may have introduced bias to our results. We addressed this concern by conducting a sensitivity analysis using multiple imputation (MI) method for our outcome variable. This analysis revealed that there was a slight overestimation in the relationship between SES barriers and healthcare delays among foreign-born cancer survivors, compared to their US-born counterparts with the same level of SES barriers. Despite the small differences observed in the effect estimates, the overall trend and interpretation of the results remained consistent across both the MI and CC analyses (Supplemental Table 3). Although the All of Us collects data nationwide, these results cannot be generalizable to all cancer survivors in the United States. A clear example of this limitation can be seen in that our sample reported roughly 90% some college or higher degree, whereas in 2021, those who had completed some college, or more were approximately 63% in the general US population [ 47 ]. Lastly, an additional limitation is that during our study period, the COVID-19 pandemic may have worsened SES barriers and healthcare delays that may have contributed to our findings and may need further exploring. Implications and future direction Our results can help guide policymakers to promote the development of policies that aim at eliminating SES barriers. For example, many of the variables that are part of our SES index are system-modifiable factors. Implementation of laws that make education equitable, job creation and training, housing affordability, and universal healthcare are ways in which policies can aid in mitigating these SES barriers. At the healthcare system level, practitioners and systems should recognize that these SES barriers exist and promote solutions. For example, systems can offer transportation services to those who are experiencing SES barriers to lessen healthcare delays [ 48 ]. Similarly, providing adult- and childcare services can help avoid delays in seeking care [ 49 ]. As the All of Us continues to enroll participants and participants complete all surveys, future studies should reassess this association to determine if our findings remain true. Moreover, in future analyses, it is important to consider adjusting for acculturation, as well as other types of stressors such as discrimination, as these experiences can contribute to healthcare delays. Additionally, future studies should aim to explore racial differences between US-born and foreign-born cancer survivors. It is crucial to recognize that races and ethnicities such as Black, Hispanic, and Asian are heterogeneous, varying across cultural and socioeconomic aspects. Thus, understanding the SES barriers associated with healthcare delays among US-born ethnic minorities from different racial and ethnic backgrounds (e.g., Mexican-US-born, Guatemalan-US-born, Chinese-US-born, Nigerian-US-born) compared to foreign-born counterparts (e.g., Mexican-foreign-born, Guatemalan-foreign-born, Chinese-foreign-born, Nigerian-foreign-born) is of extreme importance.
Conclusion Using data from the All of Us research program, we found that SES-related barriers are significantly associated with healthcare delays in cancer survivors in our study. However, a greater impact was observed among those who were foreign-born. Similarly, we observed a possible protective effect of health literacy on healthcare delays among US-born only. Our study highlights that to mitigate the impact of delayed healthcare, both policymakers and healthcare providers must prioritize addressing the social determinants of health and promoting health literacy in these populations.
Purpose We aimed to assess whether nativity differences in socioeconomic (SES) barriers and health literacy were associated with healthcare delays among US cancer survivors. Methods “All of Us” survey data were analyzed among adult participants ever diagnosed with cancer. A binary measure of healthcare delay (1+ delays versus no delays) was created. Health literacy was assessed using the Brief Health Literacy Screen. A composite measure of SES barriers (education, employment, housing, income, and insurance statuses) was created as 0, 1, 2, or 3+. Multivariable logistic regression model tested the associations of (1) SES barriers and health literacy with healthcare delays, and (2) whether nativity modified this relationship. Results Median participant age was 64 years ( n = 10,020), with 8% foreign-born and 18% ethnic minorities. Compared to survivors with no SES barriers, those with 3+ had higher likelihood of experiencing healthcare delays (OR 2.18, 95% CI 1.84, 2.58). For every additional barrier, the odds of healthcare delays were greater among foreign-born (1.72, 1.43, 2.08) than US-born (1.27, 1.21, 1.34). For every 1-unit increase in health literacy among US-born, the odds of healthcare delay decreased by 9% (0.91, 0.89, 0.94). Conclusion We found that SES barriers to healthcare delays have a greater impact among foreign-born than US-born cancer survivors. Higher health literacy may mitigate healthcare delays among US cancer survivors. Healthcare providers, systems and policymakers should assess and address social determinants of health and promote health literacy as a way to minimize healthcare delays among both foreign- and US-born cancer survivors. Supplementary Information The online version contains supplementary material available at 10.1007/s10552-023-01782-z. Keywords Open access funding provided by SCELC, Statewide California Electronic Library Consortium
Measures Demographics Personal level characteristics accounted for in this study were age (at survey completion), sex (male vs female), race (Asian, Black, Hispanic, White, multiracial/biracial, other [includes those who selected: none of this, another population, and prefer not to answer]), marital status (married [includes those who selected: married and living with a partner] vs single [includes those who selected: Single, divorced, widowed, and separated]), nativity (US- vs foreign-born), annual income (using quintiles, the lowest quintile were those who reported income of < 35 K vs quintile 2–5 ≥ 35 K), education (college or more vs ≤ high school or equivalent), insured (yes vs no), housing status (own vs rent/other arrangements), employed (yes vs no), current treatment (yes vs no), and cancer type (range of multiple cancer sites). Health literacy Health Literacy was assessed using the three-item Brief Health Literacy Screen (BHLS) [ 26 , 27 ], which measures individual needs for help with filling out forms (“How confident are you filling out medical forms by yourself?”), reading health-related documents (“How often do you have someone help you read health-related materials?”), and difficulty learning due to a lack of understanding of written medical documents (“How often do you have problems learning about your medical condition because of difficulty understanding written information?”). Response options were on a five-point Likert scale, with options for reading and understanding health documents including “Always,” “Often,” “Sometimes,” Occasionally,” and “Never” and for the need for help to fill forms were, “Extremely,” “Quite a bit,” “Somewhat,” “A little bit,” “Not at all.” The survey question measuring if the participant required help with forms was reversed coded. All items were then summed to create a composite score with higher scores (max = 15), indicating fewer health literacy problems. Following a previous study that assessed health literacy using the BHLS scale by Willens and colleagues, we dichotomized this score, with those who scored ≤ 9 as having limited health literacy and scores > 9 as having adequate health literacy [ 28 ] (Supplemental Table 1). Socioeconomic barriers Five SES factors (education, income, insurance, housing, and employment status) were dichotomized to create a composite measure following a previous study [ 29 ]. If individuals selected an income “ ≥ 35 K,” an education level of “college or more,” being insured, owning a home, and being employed, they were coded as having no SES barriers (0). Those who selected either one of the following: an income between “ < 35 k,” an educational level of “ ≤ high school or equivalent,” not being insured, not being employed, and/or having a housing status as rent/another arrangement, they were given an additive score ranging from 1 to 5 [ 29 ]. Due to sparse counts in categories of four and five SES barriers, scores were truncated to range from 0 to 3 or more SES barriers (Supplemental Table 2). Healthcare delays Nine questions were used to assess healthcare delays. These questions were obtained from the National Health Interview Survey asking participants if they experienced delays in any healthcare received due to various reasons in the past 12 months [ 25 ]. These reasons include transportation, living in a rural area where healthcare providers are too far, nervousness about seeing a healthcare provider, could not get time off work, could not get childcare, cannot leave adult unattended due to being a caretaker, could not afford copays, deductible was too high, and could not afford it or had to pay out of pocket for some or all procedures. Response options were “no,” “yes,” or “don’t know.” A dichotomized measure was created with those who responded “yes” to one or more reasons as having experienced healthcare delays, and those who responded “no” to all reasons as having experienced no delays, those who reported “don’t know” were not counted in this measure. Statistical methods To characterize the study population, descriptive statistics were calculated for all demographic variables and variables of interest (nativity, SES barriers, and health literacy). Listwise deletion method was used to address missing data (15.9%). We conducted a post-hoc sensitivity analysis to determine the direction of the potential bias introduced from our listwise deletion method for a complete case analysis. We used a multiple imputation analysis using chained equations (MICE). This approach allowed us to impute missing values based on observed data and estimate relationships between variables. To satisfy the safe data sharing policy of “All of Us,” groups with less than 20 participants are reported in tables as ≤ 20 or < x% with another category in the same column/row also showing ≥ (%) to ensure that another count value cannot be used to derive the exact count that is less than n = 20 in the suppressed group. First, a multivariable logistic regression model tested the hypothesized independent relationships between SES barriers, and health literacy with healthcare delays adjusted for covariates (age, sex, ethnicity/race, marital status, treatment status, and cancer type). Secondly, to assess nativity differences in the association between SES barriers and healthcare delays, a product interaction term for nativity and SES barriers (SES barriers*nativity) was included in the model. Furthermore, we assessed for a p-trend in the adjusted and stratified model by introducing SES barriers and health literacy as continuous measures in the model. All statistical analyses were performed using R Jupyter Notebooks embedded in the “All of Us” workbench, with a significance level at alpha 0.05. Odds ratios (ORs) with 95% confidence intervals (CI) and p value are reported. Supplementary Information Below is the link to the electronic supplementary material.
The authors would like to thank all of the participants of the All of Us Research Program, Drs. Cecilia Patino-Sutton, and Elizabeth Burner. The All of Us Research Program is supported by the National Institutes of Health, Office of the Director: Regional Medical Centers: 1 OT2 OD026549; 1 OT2 OD026554; 1 OT2 OD026557; 1 OT2 OD026556; 1 OT2 OD026550; 1 OT2 OD 026552; 1 OT2 OD026553; 1 OT2 OD026548; 1 OT2 OD026551; 1 OT2 OD026555; IAA #: AOD 16037; Federally Qualified Health Centers: HHSN 263201600085U; Data and Research Center: 5 U2C OD023196; Biobank: 1 U24 OD023121; The Participant Center: U24 OD023176; Participant Technology Systems Center: 1 U24 OD023163; Communications and Engagement: 3 OT2 OD023205; 3 OT2 OD023206; and Community Partners: 1 OT2 OD025277; 3 OT2 OD025315; 1 OT2 OD025337; 1 OT2 OD025276. In addition, the All of Us Research Program would not be possible without the partnership of its participants. Author contributions AA: Conception/design, assembly of data, data analysis, writing, and approval of the final version. SN: Conception/design, writing and editing of the article, and approval of the final version. CYO: Conception/design, writing and editing the article, and approval of the final version. CR: Writing the article and approval of the final version. SEK: Writing and editing the article and approval of the final version. AJF: Conception/design, assembly of data, writing and editing, and approval of the final version. Funding Open access funding provided by SCELC, Statewide California Electronic Library Consortium. Carol Ochoa was supported by the National Cancer Institute to conduct this study (K00CA264294-02). Data availability Data from the All of Us Research Program can only be accessed through the Researcher Workbench ( https://workbench.researchallofus.org/login ) as per the informed consent of program participants. The investigators are prohibited to share raw-level data of participants in accordance with the user agreement established by this program. Therefore, it is not possible to provide a de-identified dataset for this manuscript. Declarations Competing interests The authors declare no competing interests. Ethical approval The All of Us Research Program Institutional Review Board (IRB) established that registered tier data available on the All of Us’ Workbench meets criteria for non-Human Subject Research. This project used registered tier data and required no IRB review. https://www.researchallofus.org/faq/do-i-need-institutional-review-board-irb-approval-from-my-own-institution-in-order-to-access-this-data-through-the-researcher-workbench/ .
CC BY
no
2024-01-15 23:42:02
Cancer Causes Control. 2024 Sep 7; 35(2):203-214
oa_package/b4/1d/PMC10787892.tar.gz
PMC10787893
0
Introduction As a major discovery in the twentieth century, antibiotics play an irreplaceable role in human health, animal epidemic prevention, and disease treatment (Coleman et al. 2015 ). However, due to the overuse of antibiotics, a variety of bacteria have developed drug resistance, including the intensively studied Staphylococcus aureus ( S. aureus ) through a variety of escape routes to avoid drug access, which can cause various animal diseases such as mastitis, endometritis, sepsis, and other systemic symptoms, seriously endangering the production of livestock and poultry and the development of animal husbandry (Coleman et al. 2015 ; Kim et al. 2019 ). Drug resistance was transmitted through the environment, humans, and livestock (Davies and Davies 2010 ), causing huge economic losses (El-Sayed Ahmed et al. 2020 ). Vancomycin is considered the last line of defense against infection by drug-resistant Gram-positive bacteria (G + ), but resistance has quickly developed with the increase in drug use in clinical practice (Rao 1995 ). Therefore, there is an urgent need for an antibiotic substitution to address this issue. Antimicrobial peptides (AMPs), as an important part of the natural immune barrier, exhibit antibacterial characteristics, immune regulation, and multiple other functions (Shinohara et al. 1995 ; Huttner and Bevins 1999 ; Peters et al. 2010 ; Wang et al. 2021a , b ). The DBAASP database contains 20,523 antibacterial peptides and derivatives (Pirtskhalava et al. 2021 ), being considered a rich resource for AMPs screening (Pinto et al. 2019 ). AMPs generally have the characteristics of cationic, hydrophobic, short sequence length (< 50aa), and unique three-dimensional (3D) structure including α-helix, β-sheet, and β-turn or coil; these play a vital role in the functional activity of AMPs for better understanding of the structure-activity relationship (Cao et al. 2015 ; Wang et al. 2018 ; Wang et al. 2021a , b ; Wu et al. 2022 ). The AMPs, consisting of only amino acid in term of narrow definition, have multiple antibacterial mechanisms including destroying the cell membrane or wall or causing the contents to leak (Miao et al. 2016 ; Glukhov et al. 2008 ; Jhong et al. 2019 ; Lee et al. 2011 ; Ma et al. 2017 ); this is different from traditional heterocyclic peptide antibiotics, for which it is not easy to produce drug resistance. Therefore, they are becoming a research hotspot for the substitution of antibiotics (Gao et al. 2021 ; Pinto et al. 2019 ; Wu et al. 2021 ). Plectasin is a fungal defensin, isolated from Pseudoplectania nigrella , which can inhibit the synthesis of the bacterial cell wall by binding to the cell wall precursor Lipid II. It has a strong killing effect on G + (Mygind et al. 2005 ; Schneider et al. 2010 ) and the ability to treat peritonitis and pneumonia caused by Streptococcus pneumoniae and S.aureus (Mygind et al. 2005 ). However, its high cost and low druggability limit its clinical applications. In order to solve the above problems, 12 kinds of plectasin-derived peptides were screened, analyzed, and verified in the DBAASP database (Zhang et al. 2014 , 2021 ; Othman et al. 2018 ; Wang et al. 2021a , b ; Huang et al. 2022 ), AP138 (DBAASPS_12115) created by Lociuro was predicted to be the best one among plectasin-derived peptides with high antimicrobial activity and a long postantibiotic effect (PAE) against G + bacteria including S. aureus and methicillin-resistant Staphylococcus aureus (MRSA) (Lociuro et al. 2015 ; Groo et al. 2018 ). Compared with plectasin, there are three main differences in AP138: (i) Five amino acids, 9D, 13 M, 14Q, 17N, and 26 K, are mutated to 9S, 13L, 14R, 17R, and 26R, respectively; (ii) the positive charge increased from + 1 to + 4.5; (iii) the second structure α-helix (30.75 to 20.8%) and β-sheet (20.8 to 12.5%), β-turn (95.8 to 120.8%) changed obviously (Table 1 and S1 ). These studies and bioinformatics analysis all indicated that AP138 has research and development value in clinic therapeutics. However, AP138 is chemically synthesized due to D-type Arg at 26th site in sequence, so suffer from the heterologous expression difficulty and high cost (Koo and Seo 2019 ). Heterologous expressions may solve the problem of high cost, but the amino acid composition will be natural L-type amino acids (Lee et al. 2011 ; Lin et al. 2018 ). Therefore, the potential new different properties and mechanism of L-type AP138 (defined as AP138L-arg26 (DBAASPS_12115)) should be characterized as one of the major goals of this work (Groo et al. 2018 ; Pinto et al. 2019 ; Patel et al. 2009 ; Umerska et al. 2016 ; McEwen and Collignon 2018 ). In this study, plectasin-derived peptide AP138L-arg26 was expressed in Pichia pastoris ( P. pastoris ) cells X-33 with a 5-L fermenter, and in vitro analysis of structure, antibacterial activity, stability, drug resistance, toxicity, and safety was performed. Finally, its bactericidal mechanism against S. aureus was revealed.
Materials and methods Strains and cell lines Gram-positive bacteria: S. aureus (ATCC 25923, 43300), Staphylococcus epidermidis ( S. epidermidis ) (ATCC 12228, 35984), and Streptococcus agalactiae ( S. agalactiae ) ATCC 13813 were purchased from American Type Culture Collection (ATCC). S. aureus CVCC 546 and Streptococcus dysgalactiae ( S. dysgalactiae ) CVCC 3938 were purchased from the China Veterinary Culture Collection Center (CVCC). S. aureus E48 was donated by Northwest Agriculture and Forestry University . S. agalactiae CAU-FRI-2022-01 and S. agalactiae CAU-FRI-2022-02 were donated by China Agricultural University. Gram-negative bacteria: Escherichia coli ( E. coli ) ATCC 25922 were purchased from ATCC. Salmonella enteritidis ( S. enteritidis ) CVCC 3377 and Salmonella pullorum ( S. pullorum ) CVCC1789 were purchased from CVCC. Shigella flexneri ( S. flexneri ) (CMCC 3926, 51571) were purchased from National Center for Medical Culture Collections (CMCC). E. coli O157 (CICC 21530), Pseudomonas aeruginosa ( P. aeruginosa ) (CICC 21625, CICC21630) were purchased from China Center of Industrial Culture Collection (CICC). S. aureus CAAS-FRI-2023-01 and CAAS-FRI-2023-02 were separated from Tianjin Aoxin Animal Husbandry Sheep Farm and Huanxian Sheep Farm, respectively. E. coli DH5α, P. pastoris X-33, and pPICZαA were purchased from Invitrogen (Beijing, China). RAW 264.7 mice macrophages were obtained from Peking Union Medical College. Bovine endometrial epithelial cell line BNCC35923 was purchased from by BeNa Culture Collection (Beijing, China). Reagents The recombinant plasmid pPICZαA-AP138L-arg26 was synthesized by Sangon Biotech Co., Ltd. (Shanghai, China). Plasmid extraction kits, antibiotics (vancomycin and ceftiofur sodium), Dulbecco’s modified Eagle medium (DMEM), and fetal bovine serum (FBS) were purchased from Tiangen Co., Ltd, China Meilungel and Gibco (China), respectively. Other reagents were analytical grade. Model animal The female ICR mice (SPF, 6–8 weeks, 20–25 g/mouse) were purchased from the Vital River Laboratories (VRL, Beijing, China). Mice acclimated to the environment for 1 week before the experiment. Animal experiments strictly complied with the requirements for animal handling and welfare of the Laboratory Animal Ethical Committee and its Inspection of the Feed Research Institute of Chinese Academy of Agricultural Sciences (CAAS) (AEC-CAAS-20090609).
Results Analysis of AP138 and designation of AP138L-arg26 Plectasin was searched for as a keyword in the DBAASP database, and 12 sequences were retrieved; these derived peptides, AP138, NZ2114, NZX, MP1102, and MP1106, and other reported peptides DLP4, ID3, and P2 were also collected. Their physicochemical properties are shown in Table 1 . AP138L-arg26 (GFGCNGPWSEDDLRCHRHCKSIKGYR L26 G GYCAKGGFVCKCY) was predicted to have the highest AMP possibility (0.992) and charge (+ 4.5) (Table 1 ) and so was considered the plectasin-derived peptide. Through secondary and 3D structure analysis, we found that AP138L-arg26 conformation changed greatly with reduced α-helix/β-sheet and increased β-turn compared with the parent peptide plectasin (Table 1 ). Meanwhile, the electron cloud of amino acids at the 9th, 13th, 14th, 17th, and 26th positions changed significantly with the mutant amino acids; especially at the 13th, 14th, and 17th positions, more than 2 electron clouds changed at α-Helix, which may affect the function of peptides (Table 1 and Fig. 1 ). Design, expression, and purification of AP138L-arg26 The recombinant plasmid pPICZαA-AP138L-arg26 was linearized and transferred into P. pastoris X-33 competent cells. The recombinant plasmid sequence length was 3000–5000 bp (Fig. 2 a). The positive transformants were screened using the inhibition zone test, and transformants AP138L-arg26-8, AP138L-arg26-28, AP138L-arg26-68, and AP138L-arg26-94 had a good antibacterial effect with a large and clear inhibition zone (Fig. 2 b). Tricine-SDS-PAGE analysis of AP138L-arg26 expression in shaking flask level showed that the protein band was around 4.6 kDa, which was consistent with the predicted value of 4.46 kDa (Fig. 2 c). The positive transformants AP138L-arg26 with the highest effect (AP138L-arg26-8) was chosen and expressed with a 5-L fermenter, which displayed high production compared with the shaking flask (Fig. 2 d, e). The total protein of fermentation supernatant was 3.1 mg/mL and the biomass was 0.36 g/mL after induction in a 5-L fermenter at 120 h (Fig. 2 f). The peptide was purified using the cation-exchange column (AKTAxpress system), and only a target peptide was detected at around 4.6 kDa through Tricine-SDS-PAGE band and a single peak was detected in matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF/TOF MS) of 4464.21 Da (Fig. 2 g, h). MICs and MBCs The results of MICs and MBCs showed that AP138L-arg26 had potent antimicrobial activity against G + bacteria such as S. aureus (2–16 μg/mL), Streptococcus (4 μg/mL), and S. epidermidi s (4 or 8 μg/mL) with MIC values of 4–16 μg/mL, and the MBCs were around fourfold higher than MICs (Table 2 ). Antimicrobial activity was slightly lower than AP138 (Groo et al 2018 ) and vancomycin, but AP138L-arg26 retained good activity. These results show that AP138L-arg26 has potential for drug development against G + . Time killing curves and PAE As shown in Fig. 3 a, all selected concentrations of AP138L-arg26 (1 × , 2 × , 4 × MIC) could kill 99.99% S. aureus ATCC 43300 within 1.5 h, while the antibiotic vancomycin needed at least 6 h. Meanwhile, AP138L-arg26 had longer PAE with 0.9 h, 1.25 h, and 1.91 h at 1 × , 2 × , 4 × MIC than vancomycin (0.27, 0.45, 1.18) (Fig. 3 b). These results suggest that AP138L-arg26 had a faster and longer bactericidal effect than vancomycin, even within a low concentration (1 × MIC), which may be related to the different bactericidal mechanisms of AP138L-arg26 and vancomycin. Intracellular activity of AP138L-arg26 against S. aureus As shown in Fig. 3 c, d, AP138L-arg26 could kill intracellular S. aureus CVCC 546 in mouse macrophages RAW264.7 and bovine endometrial epithelial cells BNCC359233. The peptide of AP138L-arg26 could kill more than 70% of bacteria at a concentration of 5 × MIC, and the bactericidal rate was up to 85% at 50 × MIC. Therefore, AP138L-arg26 exhibited potent intracellular bactericidal ability, laying the foundation for in vivo applications. The synergism assay of AP138L-arg26 with antibiotic As shown in Table 3 , the FICI values of peptides AP138L-arg26 and vancomycin, ceftiofur sodium, ampicillin, or streptomycin sulfate ranged from 0.375 to 0.75, showing a synergistic effect. AP138L-arg26 had the lowest FICI value (0.375) with ceftiofur sodium. The bacterial growth curve results showed that AP138L-arg26 (8μg/mL, 2 × MIC) and sublethal levels of ceftiofur sodium alone (0.25 μg/mL, 1/8 MIC) had little effect on the growth of S. aureus CVCC 546 (Fig. 3 e), but their combination sharply inhibited the growth of S. aureus CVCC 546. Therefore, the combination of AP138L-arg26 and ceftiofur sodium had the potential to be used to cure infection by S. aureus . Stability in vitro assay These results show that AP138L-arg26 maintained its original MIC value (8 μg/mL) after incubation at different temperatures (20 °C, 40 °C, 60 °C, 80 °C), but the antibacterial activity was lost (> 64 μg/mL) under 100 °C. The antimicrobial activity of AP138L-arg26 was unaffected by gastric juice and different pH values (2–10). AP138L-arg26 was sensitive to trypsin, losting activity within 30 min after incubation (> 64 μg/mL) (Table 4 ), this would be the key limitation factor at least special for its oral administration. Structure analysis The secondary structures of AP138L-arg26 were measured in a different environment of 20 mM SDS and 50% TFE solution, which was used to simulate the hydrophobic environment of eukaryotic cells and the bacterial cell membrane environment, respectively. The results showed a positive peak at 196 nm and two negative peaks at 208 nm and 228 in 50% TFE solution, indicating an α/β spatial structure in these environments (Figure S2 ). However, the second structure of AP138L-arg26 had obviously changed (α-Helix, 4.72 to 6.06%; β-sheet, 47.78 to 42.82%; β-turn, 15.6 to 17.4%) in 20 mM SDS, especially for the increased α-Helix (1.34% increase), which could enhance membrane interactions between bacteria and AP138L-arg26 (Table S1 ). High safety of AP138L-arg26 in vitro and vivo As shown in Fig. 4 a, b, there was only 3% hemolysis and over 75% cell viability at a high concentration of AP138L-arg26 (256 μg/mL) in vitro, which indicated that AP138L-arg26 had low hemolysis and cytotoxicity. In vivo, mice (ICR, 6-8 weeks) were intraperitoneally injected with AP138L-arg26 at a concentration of 10 mg/kg (Fig. 4 c–g). Compared with the control group, there were no significant differences in the whole blood detection indexes and biochemical indexes of the AP138L-arg26 treatment group. Moreover, the tissues of the AP138L-arg26 treatment group such as the heart, kidney, spleen, lung, and liver had no hyperemia, bleeding, hyperplasia, or other lesions in the full field, being almost same as those of the control group (Fig. 4 d). These data indicate that the AP138L-arg26 peptide has excellent safety and good potential as clinical therapeutic drugs. The mechanism of AP138L-arg26 against S. aureus CVCC 546 SEM The morphology changes of bacteria after treatment with AP138L-arg26 were observed using SEM. As shown in Fig. 5 a, the surface of S. aureus CVCC 546 (nearly 100%) was smooth and spherical without peptide treatment. However, 60–80% of cell structures changed significantly with surface roughness, granular secretion, perforation, and deformation after treatment with 2 × MIC peptide AP138 at 0.5, 1, and 2 h. More than 80% of S. aureus CVCC 546 had significantly changed structure at 1 h. The results showed that AP138L-arg26 could destroy the structure of S. aureus . The effect of AP138L-arg26 on the cell membrane According to the SEM results demonstrated that AP138L-arg26 caused disruption to the cell membrane or wall. Thus, a PI/SYTO9 assay was conducted to prove the destructive effect of AP138L-arg26 on S. aureus CVCC 546. The untreated group was almost stained green by SYTO9 (over 99%), demonstrating that they were predominantly living cells. The AP138L-arg26 treatment group could disrupt the cell membrane integrity with more than 50% cells stained red by PI (Fig. 5 b). K + leakage further illustrates the disruptive effect of AP138L-arg26 on bacterial membrane, the results showed that the AP138L-arg26 treated group significantly increased the extracellular K + leakage level (0.05 mg/L) compared to the untreated bacteria (0.02 mg/L), similar to the positive control (0.1% TritonX-100) (Fig. 5 d). It was also found that AP138L-arg26 could affect the membrane potential, as measured using DiSC 3 (5) fluorescence staining, the relative fluorescence units (RFU) values of the untreated and 1 × MIC AP138L-arg26-treated groups were around 2000, and the 4 × MIC AP138L-arg26 and nisin (positive control) groups were above 3000 (a 1000 increase) (Fig. 5 c), which was mainly due to the depolarization caused by the destructive effect of the AP138L-arg26 on the bacterial membrane when the special compounds insert into lipid bilayers, they can change the membrane fluidity and disrupt the normal plasma membrane fluidity homeostasis, which further leads to leakage of cellular components and bacterial death. Therefore, a dye, Laurdan was used to measure the S. aureus membrane fluidity and quantified it using the Laurdan GP index. The results showed that the Laurdan GP of S. aureus decreased from 0.22 to 0.18 after treatment with 2 × MIC AP138L-arg26, which indicated an increase in the fluidity of the bacterial membrane due to the interaction of the AP138L-arg26 with the membrane (Fig. 5 e). These results suggested that AP138L-arg26 could destroy the cell membrane of S. aureus . The effect of AP138L-arg26 on bacterial metabolism In order to further study whether AP138L-arg26 could affect the metabolism to kill bacteria, bacterial respiration was detected using an ATP kit. The results showed that AP138L-arg26 could increase intracellular ATP: the RLU values of 1 × , 2 × , 4 × MIC AP138L-arg26-treated groups (10,700, 11,500, and 11,900, respectively) were 7.6, 8.2, and 8.5 times higher than that of the untreated group (1400), which indicated a concentration-dependent pattern (Fig. 6 a). The increase in ROS causes a series of cascading reactions such as oxidation of lipids, proteins, and damage to DNA, inducing bacteria death. After treatment with 1 × , 2 × , and 4 × MIC AP138L-arg26, the fluorescence values increased from 600 (blank control, CK) to 930, 1000, and 1400 with concentration-dependent (Fig. 6 b). LDH levels decreased after antimicrobial peptide incubation. The LDH activity % of CK, 1 × , 2 × , and 4 × MIC AP138L-arg26 were 100%, 24%, 16%, and 16% respectively, which indicated a concentration-dependent pattern (Fig. 6 c). The results showed that AP138 could induce the ROS generation of S. aureus CVCC 546, ultimately promoting cell death and reducing the probability of resistance generation.
Discussion S . aureus is a pathogenic bacterium that greatly threatens human health and the production efficiency of livestock and poultry. Many bacteria have developed drug resistance, particularly MRSA (Pinto et al. 2019 ). The rise of drug resistance has triggered a public health crisis; therefore, there is an urgent need to develop new antibacterial agents to alleviate bacterial resistance (Peters et al. 2010 ; Li et al. 2020 ). The unique bactericidal mechanism and multiple functions of AMPs have been extensively studied as antibiotic substitutes (Rao 1995 ; Peters et al. 2010 ). Among them, plectasin is the first fungal defensin extracted from Pseudoplectania nigrella (Saprophytic ascomycetes) ; it kills G + bacteria but has low activity and some cytotoxicity. Therefore, researchers have conducted in-depth studies on the peptides derived from plectasin (Mygind et al. 2005 ; Schneider et al. 2010 ). AP138 is a derived peptide with potential for clinical development, obtained by chemical synthesis due to the D type Arg at the 26th position of the sequence (Stecher et al. 2014 ; Groo et al. 2018 ). In this work, natural amino acid sequence (L-type amino acid) was obtained using heterologous expression to reduce the high cost of chemical synthesis, and the activity and bactericidal mechanism of the L-type AP138 were analyzed. As can be seen in Table 1 , it was predicted that AP138L-arg26 would have a high charge (+4.5) and antimicrobial peptide potential. Compared with plectasin, the substitution of five amino acids (D9S, M13L, Q14R, N17R, and K26R) significantly changed the physicochemical properties of AP138L-arg26 including an increased positive charge (+1 to +4.5) and hydrophobicity (32 to 33%) (Table 1 ), making it a good AMP possibility. Although D-type modification of the Arg in the sequence (D-type AP138) could effectively improve the tolerance of the antimicrobial peptide to trypsin which this property cannot be maintained in our AP138L-arg26, its chemical synthesis is difficult on a large scale and at low cost, increasing the challenges and difficulties of clinical development (Stecher et al. 2014 ; Umerska et al. 2016 ; Groo et al. 2018 ; Koo and Seo 2019 . Heterologous expression can achieve high yields, thus reducing the cost of developing the target protein. The exogenous expression technology of proteins in a y east expression system or E. coli expression system is relatively mature, but the expression of small molecule peptides (< 50 aa) is relatively difficult execpt few successful cases from fungal defensins and others (Zhang et al. 2011 ; Zhang et al. 2014 ; Cao et al. 2015 ; Yang et al. 2019 ; Shen et al. 2021 ; Gries 2022 ; Zeng et al. 2022 ). In addition, the AMPs Retrocyclin-101, and Protegrin-1, and Abaecin was also expressed in a prokaryotic expression system with a low yield (Lee et al. 2011 ). More details in our previous works, many antimicrobial peptides were successfully expressed in P. pastoris with high yield at level of 1.0-3.0 g/L (supernatant) and even higher level via high-density fermentation such as NZ2114, NZX, ID3 and NZL, with continous optimisation including enhancing the microbial biomass, controlling the fermentation and induction conditions, the increasing production capacity of equipment per unit volume and modificating expression system (Zhang et al. 2014 . Liu et al. 2020 ; Feng et al. 2012 ; Li et al 2020 ; Shen et al. 2021 ; Hao et al. 2023 ; Jin et al. 2023 ). In this study, the recombinant vector AP138L-arg26 was successfully constructed and expressed for the first time at high levels in P. pastoris . The total protein concentrations of the fermentation supernatant and microbial biomass were up to 3.1 mg/mL (95% purity) and 0.36 g/mL, respectively, after 120 h high density fermentation in a 5-L fermenter (Fig. 2 ), higher than those of NZ2114 (2.390 mg/mL, 94.8% purity) (Zhang et al. 2014 ), and the production cost of AP138L-arg26 was significantly lower than that of chemical synthesis (solid phase). The MIC is one of the key indicators for the early screening of active AMPs, for which values of MICs under 16 μg/mL may be clinically relevant (Rao 1995 ; Patel et al. 2009 ; Oh et al. 2019 ). The MICs of AP138L-arg26 were 2–16 μg/mL (0.45–3.6 μM) against selected standard and clinical S. aureus , S. epidermidis , S. dysgalactiae , and S. agalactiae (Table 2 ). Although the MIC values were lower than those of D-type AP138 (0.125–4 μg/mL), it is still superior to lincomycin, which is mainly used clinically against G + bacteria. The structure and function of AMPs are closely related, and in-depth investigation of the structure-activity relationship is an essential step in studying AMPs (Rost and Sander 1994 ; Rao 1995 ). It was analyzed that AP138L-arg26 has a special CSαβ structure with +4.5 net charge and 33% hydrophobic ratio in a normal environment with ddH 2 O. In the simulated bacterial membrane structure (20 mM SDS), AP138L-arg26 showed significant secondary structure changes, increased α-helix structure (4.72 to 6.06%), which may be related to the bactericidal function of the peptide, while it does not change significantly in the simulated eukaryotic cell membrane structure (50% TFE buffer), suggesting no damage to cells (Figure S2 ). The killing curves showed that 1 × , 2 × , 4 × MIC of AP138L-arg26 could kill all S. aureus ATCC 43300 within 1.5 h, which was shorter than that of vancomycin (6 h) (Fig. 3 a). These results showed that the AMP AP138L-arg26 could rapidly killed bacteria. The PAE is an important guide for the rational use of clinical drugs, and the re-evaluation of adverse effects of antibiotics and combination drugs. In this study, the PAE of AP138L-arg26 was longer than that of antibiotics (Fig. 3 b), indicating that low doses and long dosing intervals are feasible, which could reduce the amount of drugs used and alleviate the problem of drug resistance caused by the misuse of antibiotics (Li et al. 2020 ). Staphylococcus pathgens can evade antibiotics by entering cells (Wang et al. 2018 ). AP138L-arg26 had the ability to enter mammalian cells and can kill infected bacteria within cells such as S. aureus CVCC 546 in mouse macrophages RAW264.7 and bovine uterine epithelial cells in this study (Fig. 4 c, d). Stability and safety are the factors that must be controlled when drugs enter the clinic, as high toxicity and low stability will hinder AMPs’ application (Koo and Seo 2019 ). AP138L-arg26 retained its antimicrobial activity in different concentrations of salt ions, pH, temperature, and pepsin, but it was sensitive to trypsin and lost antimicrobial activity for 30 min. These results showed that it remained the antimicrobial activity (MIC against S. aureus ATCC 43300, 8 μg/mL) and has the potential for topical drug development (Table 2 ), whereas it may need to be encapsulated for oral administration (Umerska et al. 2017 ; Groo et al. 2018 ). AP138L-arg26 had a better safety rating in vitro and in vivo, as reflected in the high cell survival (over 75% for mouse macrophages RAW264.7) and the low hemolysis (less than 3%) at a high concentration of 256 μg/mL of AP138L-arg26 in vitro (Fig. 4 a, b). After intraperitoneal injection of 10 mg/kg AP138L-arg26 for 1 week, it was found that there were no obvious differences in whole blood detection indexes (< 4.5%), biochemical index (< 4%). The histological sections of the heart, liver, spleen, lung, and kidney were not damaged and retained a normal structure (no bleeding, hyperemia, or hyperplasia) indicating the safety of AP138L-arg26 in vivo (Fig. 4 c–g). These results indicate that AP138L-arg26 has a high safety profile as a drug in vivo and in vitro . In general, the researchers suggest that the unique positive charge and hydrophobic properties of most AMPs could interact with bacterial cell membranes and exert bactericidal functions (Gao et al. 2021 ; Pinto et al. 2019 ; Zheng et al. 2022 ). Changes in bacterial morphology were first observed using SEM after treatment with 2 × MIC AP138L-arg26 for 1 h, and most bacteria had rough surfaces, depressions, and granular secretions (Fig. 5 a). It was further shown that AP138L-arg26 had a directly destructive effect on the cell membrane through PI staining, K + leakage, and membrane fluidity assays (Fig. 5 a–e). AP138L-arg26 might mainly be considered to interact with the negatively charged components on the surface of the bacterial membrane through itself charges, and then insert into the cell membrane, interfering with its orderly arrangement, causing further destruction (She et al. 2022 ; Wang et al. 2021a , b ). This was similar to the case with the antimicrobial peptide 5j and L007-0069 (She et al. 2022 ). AP138L-arg26 also affect the bacterial metabolism: (i) After the incubation of AP138L-arg26 with bacteria, intracellular ATP levels were elevated (maximum: 8.5-fold), which may cause bacteria to switch from a dormant state to an active state that is more conducive to them being killed (Shi et al. 2022 ). (ii) Intracellular ROS were increased twofold with 2 × MIC of AP138L-arg26, which suggests that bacteria could be damaged through bacterial auto-oxidation (Shi et al. 2022 ), this is an important way in which drugs could damage bacteria. (iii) At the same time, the amount of LDH decreased (lowest, 16%), which may limit the level of respiratory metabolism due to LDH being an essential enzyme in the respiratory chain (Wang et al. 2021a , b ) (Fig. 6 ). All in all, AP138L-arg26 may kill bacteria by damaging cell membranes and influencing bacterial metabolism (ATP, ROS, LDH). Finally, we changed the D-type AP138D to L-type AP138L-arg26, and it was successfully expressed in P. pastoris with high production. It was demonstrated that AP138L-arg26 had a faster and longer bactericidal effect compared with conventional antibiotics, high stability, and excellent safety in vivo and in vitro. It was revealed that AP138L-arg26 has multiple bactericidal mechanisms including membrane rupture and metabolism imbalance, which is also an important reason that antimicrobial peptides do not tend to develop resistance. The better derived AMPs of plectasin will be worth of developing in the future so that the excellent property of high resistance to trypsin hydrolysis in AP138 construct could be merged and maintained in new AMP derivative with a high expression level.
Abstract The low activity and yield of antimicrobial peptides (AMPs) are pressing problems. The improvement of activity and yield through modification and heterologous expression, a potential way to solve the problem, is a research hot-pot. In this work, a new plectasin-derived variant L-type AP138 (AP138L-arg26) was constructed for the study of recombination expression and druggablity. As a result, the total protein concentration of AP138L-arg26 was 3.1 mg/mL in Pichia pastoris X-33 supernatant after 5 days of induction expression in a 5-L fermenter. The recombinant peptide AP138L-arg26 has potential antibacterial activity against selected standard and clinical Gram-positive bacteria (G + , minimum inhibitory concentration (MIC) 2–16 μg/mL) and high stability under different conditions (temperature, pH, ion concentration) and 2 × MIC of AP138L-arg26 could rapidly kill Staphylococcus aureus ( S. aureus ) (> 99.99%) within 1.5 h. It showed a high safety in vivo and in vivo and a long post-antibiotic effect (PAE, 1.91 h) compared with vancomycin (1.2 h). Furthermore, the bactericidal mechanism was revealed from two dimensions related to its disruption of the cell membrane resulting in intracellular potassium leakage (2.5-fold higher than control), and an increase in intracellular adenosine triphosphate (ATP), and reactive oxygen species (ROS), the decrease of lactate dehydrogenase (LDH) and further intervening metabolism in S. aureus . These results indicate that AP138L-arg26 as a new peptide candidate could be used for more in-depth development in the future. Key points • The AP138L-arg26 was expressed in the P. pastoris expression system with high yield • The AP138 L-arg26 showed high stability and safety in vitro and in vivo • The AP138L-arg26 killed S. aureus by affecting cell membranes and metabolism Supplementary Information The online version contains supplementary material available at 10.1007/s00253-023-12947-w. Keywords
Biological information analysis of AP138L-arg26 AP138L-arg26 and plectasin derived AMP sequences were obtained from DBAASP database ( https://www.dbaasp.org/home ) (Pirtskhalava et al. 2021 ), their physical and chemical properties, 3D structure, and antimicrobial peptide possibility were predicted by APD database ( https://aps.unmc.edu/AP/ ) (Wang et al. 2016 ), I-TASSER ( https://zhanggroup.org/I-TASSER/ ) (Yang and Zhang 2015 ), and CAMP database ( http://www.camp.bicnirrh.res.in/ ), respectively (Waghu et al. 2014 ). Expression, purification, and identification of AP138L-arg26 The nucleic acid sequence of AP138L-arg26 (GenBank ID: 2736833) was subjected to codon preference using the Reverse Translate Tool ( http://www.bioinformati cs.org/sms2/rev_trans.html), then a Kex2 signaling peptide cleavage site was added at its N-terminus (Figure S1 ) and inserted into the eukaryotic expression vector pPICZαA between the double enzyme ( Xho I and Xba I) cleavage site, to construct the plasmid pPICZαA-AP138L-arg26 . Additionally, the pPICZαA-AP138L-arg26 was digested using Pme I and transformed into competent P. pastoris X-33 cells via electroporation. Peptide purification was carried out with the AKTAxpress system. Expression of AMP AP138L-arg26 was identified using Tricine-SDS-PAGE firstly, and the purified AP138L-arg26 was identified using MALDI-TOF/TOF MS (Ultraflextreme, Bruker, Germany), finally, the concentration was detected using Bradford assay kits. Antimicrobial activity and pharmacodynamic analysis in vitro Minimum inhibitory concentration (MIC) and minimum bactericidal concentration (MBC) To determine the MIC values of AP138L-arg26, we used the broth microdilution technique as previously described (Andrews. 2001 ; Wiegand et al. 2008 ). In brief, AP138L-arg26 peptide were diluted from 1280 to 2.5 μg/mL and added to a 96-cells plate with 10 μL per well. The bacteria in the logarithmic stage were diluted to 10 5 CFU/mL with 90 μL per well. All plates were cultured in a 37 °C constant-temperature incubator; no visible growth of bacteria was the MIC after 18 h incubation. The MBC method was also based on the CLSI 2021 guidelines. In brief, after the bacteria were incubated at 1 × , 2 × , 4 × , and 8 × MIC of AP138L-arg26, the cultures were coated with Mueller-Hinton Agar (MHA) plates, the concentrations of AP138L-arg26 with 99.99% the bacteria killed was defined as MBC. Time-killing curve assay The time-killing curve of AP138L-arg26 against MRSA ATCC 43300 was used to analyze the pharmacodynamics. The method was based on previous laboratory experiments (Flamm et al. 2019 ; Yang et al. 2019 ). Briefly, exponential-phase bacteria were diluted to 1 × 10 5 CFU/mL, and peptides underwent a twofold gradient dilution setting of 4 × , 2 × , and 1 × MIC. There was incubation in a 37 °C thermostatic shaker at 200 rpm for 0–24 h to collect samples, which were plated on MHA plates. Colony numbers were recorded at all collected time points. PAE assay The method of PAE of AP138L-arg26 against MRSA ATCC 43300 or S. aureus CVCC 546 was described previously (Wang et al. 2018 ). The formula is PAE = T − C ( T is the time required for the number of colonies in the sample treatment group to increase by tenfold, and C is the time required for the untreated group). Intracellular antibacterial activity The method of intracellular bactericidal effect was described previously (Wang et al. 2018 ). Only the number of cells and bacteria changed from 2.5 × 10 5 to 2 × 10 5 cells/mL. Synergism with antibiotics Fractional inhibitory concentration index (FICI) The synergistic effects of AP138L-arg26 with different antibiotics were evaluated using a checkerboard assay. The synergistic effect was calculated using the FICI as follows: FICI = FIC of AP138L-arg26 + FIC of antibiotic; FIC = MIC c /MIC a , where MIC c is the MIC of the peptide and antibiotic in combination, and MIC a is the MIC of the peptide/antibiotic alone (Blier et al. 2010 ). Three parallel experiments were performed for each group. The efficacy of combination therapy was defined as FICI ≤ 0.5, 0.5 < FICI ≤ 1, 1 < FICI ≤ 4, and FICI > 4 indicating synergy, addition, no difference, and antagonism, respectively. Growth curve of S. aureus Samples were taken during the logarithmic growth of S. aureus and diluted to 1 × 10 5 CFU/mL (Delpech et al. 2019 ). Different concentrations of AP138L-arg26, ceftiofur sodium, or a combination of them with an equal volume of bacteria were added to a 96-well microplate. The growth curves were recorded using a fully automatic growth curve recording instrument. Stability analysis Artificial gastric and intestinal juice stability AP138L-arg26 was incubated in artificial gastric and intestinal juice (Beijing Coolaber Technology Co., Ltd., Beijing, China) for 0.083, 0.167, 0.25, 0.75, and 1 h. The antimicrobial activity of AP138L-arg26 against MRSA ATCC 43300 was tested through the MIC assay. Thermal, pH, and salt stability The thermal, pH, and salt stability of AMPs were studied previously (Zhang et al. 2011 ). Briefly, the AP138L-arg26 was incubated at different temperatures (4/20/40/60/80/100 °C), pH values (2/4/6/8/10), and kinds of salt ions (50/100/150/200/300 mM NaCl, 1.25/2.5/5 mM KCl, MgCl 2 , CaCl 2 ) for 1 h at 37 °C, respectively. The antimicrobial activity of AP138L-arg26 against MRSA ATCC 43300 was tested using MIC assay. Circular dichroism (CD) spectrum assay The secondary structures of AP138L-arg26 were detected using CD (Bio-Logic MOS450 spectropolarimeter, France) in a simulated eukaryotic cell environment (50% trifluoroethanol, TFE) and bacterial membrane environment (20 mM sodium dodecyl sulfate, SDS) (Yao et al. 2018 ). Safety evaluation in vitro and in vivo Hemolysis activity The hemolysis of AP138L-arg26 to fresh ICR mouse erythrocytes was described previously (Zheng et al. 2021 ). Briefly, 8% of mouse erythrocytes were incubated with different concentrations of antimicrobial peptide AP138L-arg26 (256, 128, 64, 32, 16, 8, 4, 2, 1, 0.5 μg/mL) in equal volumes. The blank and positive control was phosphate buffered saline (PBS) and 0.1% Triton X-100, respectively. Cytotoxicity The cytotoxicity of AP138L-arg26 to mouse macrophages RAW264.7 was detected by the thiazolyl blue tetrazolium bromide (MTT) method as described previously (Shen et al. 2021 ). Acute toxicity in mice AP138L-arg26 was administrated by intraperitoneal injection ( n = 4 per group, 10 mg/kg, body weight 25 g) every day for 1 week (Shi et al. 2022 ). After 7 days, the mice were euthanized to collect anticoagulant whole-blood and tissues (liver, spleen, kidney, and lung), which were used for whole blood and biochemical detection, or tissue sections (hematoxylin-eosin (HE) staining), respectively. Antimicrobial mechanism of peptides Scanning electron microscopy (SEM) assay The effect of peptides on changes in bacterial morphology was observed by scanning electron microscopy. Briefly the exponential phase S. aureus CVCC 546 cells (1 × 10 9 CFU/mL) were incubated with 2 × MIC peptides at 37 °C for 0.5 h, 1 h, and 2 h. The untreated bacteria were the negative control. The processing methods and steps of the samples were described in detail in our previous study (Li et al. 2020 ). Briefly, there were two important steps: (1) Sample preparation: incubation, cleaning, fixation, drying, gold spraying. (2) Observation using microscope. Only one concentration was used to deal with bacteria at different times (0.5, 1, and 2 h). SYTO9/propidium iodide (PI) assay To explore the disruption of bacterial cell membranes by the AP138L-arg26, a SYTO9/PI kit was used to detect the integrity of cell membranes (L7007 LIVE/DEAD R BacLightTM Bacerial Viability Kits). The unique feature of this kit is that SYTO9 alone can pass through integral cell membranes, while PI can only penetrate damaged cell membranes. Briefly, the S. aureus CVCC 546 cells (1 × 10 9 CFU/mL) were incubated with AP138L-arg26 (2 × MIC, 37 °C for 1 h), and a volume of 3 μL SYTO9/PI (1.5 μL: 1.5 μL) was added to 1 mL of bacterial suspension. Finally, the results were observed using fluorescence microscopy. Membrane fluidity assay The final concentration of 10 μM Laurdan was used to detect the effect of AP138L-arg26 on bacterial cell membrane fluidity (Shi et al. 2022 ). Briefly, (1) co-incubation of bacteria with Laurdan; (2) the stained bacteria are co-incubated with the peptide; (3) spectrophotometer (Tecan, Männedorf, Switzerland) detection. The calculation formula: generalized polarization (GP) = (I435 − I490)/(I435 + I490). Membrane depolarization assay To further explore the effect of AP138L-arg26 on the bacterial membrane, the membrane probe DiSC 3 (5) was selected to detect the change in membrane potential (Wang et al. 2020 ). Briefly, S. aureus was washed and resuspended in PBS to 1 × 10 8 CFU/mL and incubated with 0.5 mM membrane dyes DiSC 3 (5) at 37 °C for 1 h in the dark. Then, 90 μL of stained bacterial suspension and 10 μL of AP138 L-arg26 (1 × , 2 × , 4 × MIC) were mixed and added to black and clean 96-cell plates and the fluorescence intensity was determined using a spectrophotometer (excitation wavelength 622/emission wavelength 670 nm, Infinite M200). Potassium ion (K + ) leakage The integrity effect of antimicrobial peptide AP138L-arg26 on bacterial cell membranes was further verified, and the K + leakage assay was previously described (Li et al. 2020 ). Briefly, the main procedures were as follows: Firstly, prepare the S. aureus CVCC 546 bacteria suspension at a concentration of 1 × 10 8 CFU/mL. Secondly, incubate the bacterial suspension with 2 × MIC AP138L-arg26 at different times (15/30/60/90/120 min) at 37 °C, with untreated cells and nisin used as negative and positive controls. Finally, the supernatants were detected using inductively coupled plasma mass spectrometry (ICP-MS) (SantaClara, CA, USA). Intracellular adenosine triphosphate (ATP) determination As ATP is the most direct source of energy for living organisms, the effect of antimicrobial peptides on ATP was explored with an ATP assay kit (Beyotime, Shanghai, China). Briefly, exponential-stage S. aureus CVCC 546 (1 × 10 8 CFU/mL) were incubated with different concentrations of AP138L-arg26 (1 × , 2 × , 4 × MIC) for 1 h, and the pellets were collected, lysed, and centrifuged to harvest the intracellular supernatant. The luminescence was detected using an Infinite M200 Microplate reader (Tecan, Luminescence signals). Lactic dehydrogenase (LDH) activity As disruption of the cell membrane structure leads to the release of LDH from the cytoplasm into the culture medium, the detection of LDH in bacteria can further reveal the antibacterial mechanism (Shi et al. 2022 ). Briefly, S. aureus cells were co-incubated with AP138L-arg26 for 6 h. Later, the pellets were collected and sonicated (3 s/10 s, 30 times). The intracellular LDH activity was detected using an Infinite M200 Microplate reader (Tecan, Luminescence signals). The calculated results of LDH% = treated cells/the control. Reactive oxygen species (ROS) measurements The probe 2′,7′-Dichlorodihydrofluorescein diacetate (DCFH-DA) was used to measure the level of ROS in bacteria after incubated with AP138L-arg26 (Wang et al. 2021a , b ; Shi et al. 2022 ). Briefly, exponential stage bacteria were incubated with DCFH-DA (10 μM, 37 °C, 0.5 h), and the stained bacteria were treated with AP138L-arg26 (1 × , 2 × , 4 × MIC) for 1 h. The ROS were detected at excitation wavelength (488 nm) and emission wavelength (525 nm) using a microplate reader (Tecan, Männedorf, Switzerland). Statistical analysis The software GraphPad Prism (version 8, USA) was used to analyze all data, and ANOVA was the method to determine the statistical significance. The results are presented as means ± standard deviation (SD). A P value of < 0.05 was considered statistically significant. Supplementary Information Below is the link to the electronic supplementary material.
Acknowledgements We acknowledge Chunli Li from the Core Facility at the Institute of Microbiology at the Chinese Academy of Sciences (CAS) for the technical support with SEM and Tong Zhao for her technical support with FACS analysis. Author contribution KZ and JW conceived and designed the research. DT, RM, NY, and YH conducted experiments. KZ, NY, and JW evaluated data. KZ, NY, and JW wrote and revised the manuscript. All authors read and approved the manuscript. Funding This work was supported by the National Natural Science Foundation of China (Grant No. 31872393), National Key Research and Development Plan—High Expression of Thiopeptides and their Analogs (Grant No. 2022YFC2105000-03, 2022–2026), and the Innovation Program of Agricultural Science and Technology (ASTIP) in CAAS (Grant No. CAAS-ASTIP-2017-FRI-02) and its key projects (Grant No. CAAS-ZDRW202111 and Grant No. CAAS-ZDXT 201808). Data availability All data generated or analyzed during this study are included in this published article (and its supplementary information files). Declarations Ethical approval The mouse experiment was performed according to the Animal Care and Use Committee of the Feed Research Institute of the Chinese Academy of Agricultural Sciences (CAAS) and approved by the Laboratory Animal Ethical Committee and its Inspection of the Feed Research Institute of CAAS (AEC-CAAS-20090609). Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:02
Appl Microbiol Biotechnol. 2024 Jan 13; 108(1):1-18
oa_package/45/d3/PMC10787893.tar.gz
PMC10787894
38217709
Introduction Neprilysin (NEP) is a ubiquitous membrane-bound zinc-dependent metalloprotease that hydrolyzes numerous regulatory peptides including vasoactive peptides such as natriuretic peptides, adrenomedullin, angiotensins I and II, bradykinin, and endothelin-1 as well as non-vasoactive peptides such as beta amyloid, glucagon, enkephalins, and substance P [reviewed in 1 – 3 ]. NEP’s broad tissue distribution and wide range of substrates portend its regulatory functions and involvement in the pathology of cardiovascular, renal, respiratory, brain, and nervous systems. NEP has a short cytoplasmic N-terminal domain of 27 amino acids, a single transmembrane region of 23 residues and a large C-terminal ectodomain harboring the catalytic site that extends into the extracellular space. As such, a catalytically active and soluble form of NEP (sNEP) can be released into the circulation by proteolytic ectodomain shedding [ 4 ]. NEP-bound exosomes may also be released from endothelial cells and human adipose-derived mesenchymal stem cells [ 4 , 5 ]. Recent mass spectrometry analysis of NEP antibody pull-down material from human plasma detected peptides that map to the N-terminal cytoplasmic domain, further corroborating possible NEP entry into circulation via exosomes [ 6 ]. NEP is an emerging biomarker for various diseases including heart failure (HF), cardiovascular diseases, diabetic kidney disease, and metabolic syndrome [ 7 ]. Most studies use commercially available research-use-only (RUO) ELISA kits for quantifying sNEP in circulation and data comparisons show strikingly poor correlations between assays [ 1 ]. It is not surprising, therefore, that published findings on the concentration of plasma sNEP and its clinical associations in health and disease lack agreement [ 1 , 6 , 8 ]. Although discrepant findings may in part be attributed to different cohort populations or disease subtypes studied in these reports, two important issues need to be addressed to help resolve discordant observations. First, significant areas of concern exist with NEP quantification, and in fact with many other protein biomarkers, where RUO ELISAs are used for measurement. These assays are often employed with little or no manufacturer’s information on the immunogen sequence, antibody type (polyclonal versus monoclonal and animal of origin), antibody specificity and binding site on the target, and nature of the calibrator (protein sequence and source). Second, little is known about the circulating forms of NEP in humans. A previous study suggested the presence of both catalytically active and inactive forms of NEP in circulation [ 6 ]. The quaternary structure of NEP in humans is also unclear although it appears to exist as a monomer in rabbits [ 9 ], while non-covalently associated homodimers have been reported in pigs [ 10 ]. NEP is also heavily glycosylated with experimentally published N-linked glycosylation sites at N145, N285, N294, N325, and N628 [ 11 – 14 ]. Differences in glycosylation contribute to the range of NEP molecular weights (85–110 kDa) observed in different tissue sources [ 15 ]. Clearly, correct interpretation of immunoassay data requires the availability of highly specific antibodies with well-defined binding footprints and at least some information on the circulating NEP moieties detected by the antibody pairs. Previously, we described an epitope-directed monoclonal antibody (mAb) production method that facilitated antibody characterization and validation [ 16 ]. Using a similar approach, we generated and characterized mAbs directed against multiple non-overlapping preselected peptide regions on NEP. To facilitate the generation of glycosylation-sensitive mAbs, an epitope harboring experimentally verified glycosites was deliberately selected for immunogen preparation. Such mAbs may constitute useful tools for probing deficiencies in glycosylation site occupancy and elucidating how this property relates to health and in disease. The best mAb pair was validated for applications in western blotting and ELISA. The profile of circulating NEP moieties was elucidated using these two mAbs and compared against that obtained with a previously validated commercial polyclonal antibody (PE pAb) [ 6 ]. These newly developed binding agents provide fresh insights into circulating forms of NEP and are useful tools for future research into mechanistic pathways of NEP regulation and its relationship in disease states.
Materials and methods Peptide design and purification The protein sequence of human NEP (GenBank Accession no. NP000893) was analyzed using the B Cell Linear Epitope Prediction tool on the Immune Epitope Database Analysis Resource server ( http://www.iedb.org ). Four putative antigenic sequences, VATENWEQKYGASW (residues 169–182; designated AgNEP-1), YIKKNGEEKLL (residues 665–675; designated AgNEP-2), EIANATAKPEDRNDP (residues 282–296; designated AgNEP-3) and EKVDKDEWIS (residues 528–537; designated AgNEP-4) were chosen. Each sequence was fused as independent three-copy inserts separated by short glycine-serine linkers into the surface-exposed active loop of thioredoxin (Trx) harboring a C-terminal polyhistidine tag as described elsewhere [ 16 ]. These constructs were used for preparation of the immunogens. A further composite construct for antibody screening, designated Trx-AgNEP(1–4), containing single copies of all four antigenic sequences arranged in tandem within the active loop of thioredoxin was also generated. The coding sequences of all constructs were verified by DNA sequencing. All thioredoxin fusion proteins were expressed in Escherichia coli BL21 (DE3) trxB and purified by immobilized metal affinity chromatography (IMAC) under native or denaturing conditions as previously described [ 16 , 17 ]. Antibody generation and characterization A cocktail comprising equimolar concentrations of Trx-AgNEP-1, -2, -3 and -4 (final total concentration 1 mg/ml) was added to an equal volume of Sigma Adjuvant System (Sigma Aldrich) and mixed to homogeneity for immunization into Balb/c mice. The immunization scheme and method for generating, screening, and selecting the hybridoma clonal cell lines are as previously described [ 16 ]. All approved animal experiments were performed in compliance with A*STAR Institutional Animal Care and Use Committee (IACUC) regulations. Antibody purification, isotyping, characterization by surface plasmon resonance (SPR), and epitope mapping by alanine scan analysis were performed as previously described [ 16 ]. The association (k a ), dissociation (k d ), and equilibrium (KD) constants of mAbs were determined using the ProteOn XPR36 (Bio-Rad, Hercules, USA) and tested against recombinant human sNEP (HEK293 human cell line-derived NEP ectodomain spanning Y52-W750; Aviscera Bioscience, USA; product code 00724-06-10) and the cognate thioredoxin-fused AgNEP antigen. Critical residues for mAb binding were determined using the Multipin system (Mimotopes, Australia) comprising a library of peptides corresponding to the wild-type sequences of the NEP antigens and their analogs containing one alanine substitution in the peptide sequence. Immunoprecipitation from 40 ml of pooled HF human plasma (NEP concentration at ~ 600 pg/ml as determined by sandwich ELISA using the mAb pair of 17E11 and 31E1 described below) was performed using biotinylated mAb 17E11 (raised against Trx-AgNEP-4)/streptavidin magnetic beads as described previously [ 16 ]. Proteomic processing (reduction, alkylation and trypsin digestion) was performed on the pull-down eluate using the S-TRAP Micro column (Protifi) following manufacturer’s instructions. Peptides were then acidified to a final concentration of 0.1% formic acid. For online liquid chromatography–mass spectrometry (LCMS) analysis, 1 μg of peptides were injected into an eksigent 425 nano-LC system fitted with a ProteoCol C18P trap column (3 μm 120 Å, 300 μm × 10 mm; Trajan) and an Acclaim PepMap100 C18 analytical column (3 μm 100 Å, 75 μm × 250 mm; Thermo Scientific). The peptides were separated at 300 nL/min, using 0.1% formic acid in water and 0.1% formic acid in acetonitrile as mobile phase A and B, respectively, using a gradient of 5–15% B over 60 min and 15–30% B over the next 60 min. MS data were acquired on a TripleTOF 6600 mass spectrometer (SCIEX) in high-resolution multiple reaction monitoring (MRM HR ) mode. The MRM HR analysis consisted of a TOF–MS scan across 400–1600 m /z with 50 ms accumulation time, followed by 15 product ion scans across 100–1800 m /z with 85 ms accumulation time each. The target NEP peptides were chosen based on predicted tryptic peptides that are close to the epitope of mAb 17E11. The product ion scan parameters are detailed in Table 1 . MRM HR data were processed using the PeakView 2.2 software with the Bio Tool Kit 2.2 plugin (SCIEX). Extracted ion chromatograms (XIC) of all predicted b and y fragment ions from each targeted NEP peptide were generated with 0.05 Da width. Spectrum matching was performed using a mass tolerance of 0.05 Da, and against all charge states of b and y fragment ions. Sandwich ELISA Antibody pairs that gave the strongest positive signal readout against Trx-AgNEP(1–4) at 8000 pg/ml were identified by ELISA using a checkerboard screening method as described previously [ 16 ]. With capture and detection antibody concentrations kept constant, Trx-AgNEP(1–4) was titrated to generate a seven-point standard curve ranging between 125 and 8000 pg/ml. Colorimetric signal was generated with streptavidin–HRP in conjunction with the chromogenic substrate, tetramethylbenzidine (TMB). To find the antibody pair yielding the maximum signal-to-noise ratio, absorbance values measured at 450 nm on the Enspire Multi-mode microplate reader (Perkin Elmer) with background subtraction at 570 nm were plotted against Trx-AgNEP(1–4) concentrations and fitted against a five-parameter logistic (5PL) model using the Enspire® software. The optimal antibody pair comprising mAb 17E11 as capture antibody and biotinylated mAb 31E1 for detection was used to develop a two-site sandwich ELISA for human NEP. Calibration curves were generated with eight calibrators at 8000, 3200, 1280, 512, 205, 81.9, 32.8, and 13.1 pg/ml of Trx-AgNEP(1–4) prepared in 2.7% bovine serum albumin (BSA)/phosphate buffered saline (PBS). Assessment of assay cross-reactivity to structurally similar M13 family endopeptidases is as described previously [ 6 ]: human endothelin converting enzyme 1 (ECE-1), human endothelin converting enzyme 2 (ECE-2), human phosphate-regulating neutral endopeptidase (PHEX), and human endothelin converting enzyme like 1 (ECEL-1). Spike and recovery testing was performed using recombinant Trx-AgNEP(1–4) stock solution and a plasma sample with very high endogenous NEP (23,000 pg/ml) spiked at two different concentrations into three plasma samples. Parallelism experiments to ascertain that the binding characteristics of the antibody pair to endogenous NEP and Trx-AgNEP(1–4) are comparable were performed using non-HF and HF plasma EDTA samples. Six plasma samples with endogenous NEP levels within the assay range when diluted with 2.7% BSA/PBS in the range of 2–24-folds were tested. Human test samples used were as previously described and in accord with relevant ethics approval and written informed consent [ 6 ]. Western blotting The reactivity of the mAbs to various protein sample types was determined by simple western analysis using the Jess fully automated system (ProteinSimple; Bio-Techne, USA) following the manufacturer’s instructions. Recombinant human sNEP (2 ng/well; Aviscera Bioscience, USA), purified recombinant Trx-AgNEP(1–4) (2 ng/well), whole cell lysates (1.2 μg/well) from human prostate adenocarcinoma LNCaP (Santa Cruz Biotechnology, USA; product number sc-2231) and human prostate cancer PC-3 (Santa Cruz Biotechnology, USA; product number sc-2220) cell lines, and human plasma samples (both individual and pooled ( n > 100) samples; 1.2 μg/well) were used for western analysis. The previously validated polyclonal antibody (PE pAb) from the Neprilysin AlphaLISA® kit (Perkin Elmer, MA, USA; Cat# AL337HV) was used as the positive control for comparing immunoreactivity to the same samples [ 6 ]. The chemiluminescence assay was used following the manufacturer’s instructions. Samples were processed as follows: 5 × master mix was prepared using reagents provided by Bio-Techne (EZ Standard Pack 1, cat. no.: PS-ST01EZ-8). Protein samples were mixed 1:4 with 5 × master mix, heated at 95 °C for 5 min and stored in ice. Samples were loaded onto the 12–230 kDa Jess/Wes Separation Module where protein separation and immobilization, immunoprobing, washing, and detection take place in capillary tubes. The incubation time of the primary and the secondary antibodies was 30 min. Concentration of all mAbs used was 5 μg/ml (when probing against recombinant sNEP and Trx-AgNEP(1–4)) and 10 μg/ml (when probing against whole cell lysates and human plasma samples). Concentration of PE pAb used was 5 μg/ml for all protein sample types except human plasma samples where antibody concentration at 10 μg/ml was used. The amount of total plasma proteins loaded in each capillary was 4 μg. All other samples were loaded at total protein of 1.2 μg per capillary. For the secondary antibody, ready-to-use HRP-conjugated anti-mouse or anti-goat antibody was used, depending on the primary antibody. Peptide competition assays were performed to validate the immunodetected bands in human plasma samples and Trx-AgNEP(1–4) was included as a positive control. Western blotting was performed exactly as described above except that the primary antibodies (mAbs 31E1 and 17E11) were pre-blocked overnight with matched peptides (AgNEP3 and AgNEP4, respectively) or an unmatched peptide (DLNAKDREGDTPLH) at tenfold excess concentration relative to that of the primary antibody. Peptide competition assays by two-site ELISA (mAb 17E11/31E1 as capture and detection antibodies and the reverse pairing) were also performed to corroborate the effect of matched/unmatched peptide blocking of mAb 31E1 and 17E11 on the detection of Trx-AgNEP(1–4) using 10X excess concentration of peptide relative to the detection antibody. Glycosylation sensitivity of mAb 31E1 To provide direct evidence that mAb 31E1 binds only to glycan-deficient recombinant sNEP, PNGase F (New England Biolabs, Cat# P0704S) was used to fully remove N-glycan moieties from HEK293 cell line-derived recombinant sNEP (Aviscera Bioscience, USA) under denaturing conditions according to manufacturer’s instructions. PNGase F-treated recombinant sNEP and a paired sample with no glycosidase as control was separated on 12% SDS-PAGE and transblotted onto a PVDF membrane. Detection of recombinant sNEP was performed using purified mAb 31E1 (1.5 μg/ml) and a goat anti-mouse polyclonal antibody conjugated to horse-radish peroxidase (diluted 1:20,000; Abcam cat# AB205719). Protein bands were visualized using SupersignalTM West Pico Plus chemiluminescent substrate (Thermo Fisher Scientific, USA).
Results Recombinant protein design and preparation Sixteen antigenic sites (8–24 amino acids long) on human NEP were identified using the BepiPred B-cell linear epitope prediction tool [ 18 ]. Four of these (designated AgNEP-1, -2, -3, and -4) were selected on the basis of their non-overlapping surface locations with respect to the 3D structure of NEP modeled against template 6suk.1.A using SWISS-MODEL (Fig. 1 a). AgNEP-1, -2, and -4 do not contain any experimentally verified or predicted post-translational modifications. AgNEP-3 was deliberately selected as it contained two experimentally-verified N-glycosylation sites at positions N285 and N294 and the mAbs generated could be useful in revealing subtle effects of NEP glycosylation on its biological actions, clinical associations, and circulating concentrations. Figure 1 b shows the secondary structure elements (helix, strand, and others) and solvent accessibility (buried versus exposed residues) of these selected epitopes using the PredictProtein analysis tool [ 19 ]. All Trx-tripeptide proteins were highly expressed in E. coli and accounted for 21–29% of total bacterial protein (Fig. S1 ). Trx-AgNEP-1 was produced in the insoluble form while Trx-AgNEP-2 accumulated equally in both soluble and insoluble forms. Trx-AgNEP-3 and -4 were mainly expressed in the soluble form. Trx-AgNEP-1 was purified by denaturing IMAC and refolded into phosphate buffer while the other three antigens were purified by native IMAC to recover only the soluble form. Protein yields ranged from 15 to 60 mg/l bacterial culture (Table S1 ). These four antigens were then mixed in equimolar concentrations and used as an immunogen cocktail for animal immunization. The composite construct, Trx-AgNEP(1–4), was produced primarily as a soluble protein with yield at 42 mg/l of bacterial culture. Hybridoma screening and selection A single fusion experiment of splenocytes from two mouse spleens with SP2/0-Ag14 myeloma cells produced a total of 3840 hybridoma clones. Initial screening was performed using validated 96-well DEXT microplates as described previously [ 16 ]. Hybridoma clones producing mAbs that reacted strongly to only one of the four coated antigens were selected for further expansion. A total of 7, 41, 77, and 67 clones were found to react strongly to Trx-AgNEP-1, -2, -3 or -4, respectively. Of these, ~ 90% were unstable and exhibited significant loss of reactivity after a second round of sub-culturing. Finally, a total of 16 stable primary parent clones of high specificity to their cognate antigen (Trx-AgNEP-3 or -4) were obtained. No stable clones with specificity against Trx-AgNEP-1 and -2 were obtained. Ten stable high-yielding hybridoma cell lines were chosen and sub-cloned for further analysis. Antibodies were purified by Protein A affinity chromatography with yields ranging between 32 and 92 μg/ml of hybridoma culture supernatant. Monoclonal antibody characterization Functional characteristics of the selected mAbs are summarized in Table 2 . All ten mAbs were of the IgG-kappa class and could be classified into three isotypes, IgG 1 , IgG 2a , and IgG 2b . Binding characteristics of the mAbs to both recombinant sNEP as well as their cognate Trx-fused peptide (AgNEP-3 or AgNEP-4) were evaluated by surface plasmon resonance (SPR) analysis (Fig. S2). Two out of five mAbs (33G5 and 31E1) against AgNEP-3 were able to bind Trx-AgNEP-3 while all five mAbs against AgNEP-4 were able to bind Trx-AgNEP-4. Of these, three mAbs (33G5, 17E11, and 3E12) were able to bind recombinant human sNEP with monovalent binding affinities (KD) measuring 4.8 nM or lower. The importance of individual amino acid side chains on the epitopes was assessed by performing an alanine scan, allowing for direct identification of residues that are critical for antigen–antibody binding (Fig. S3). It was observed that the functional epitopes of the mAbs involved 2–6 key interacting residues. Immunoprecipitation from human plasma was performed using biotinylated mAb 17E11 and streptavidin magnetic beads. Targeted MRM HR mass spectrometry analysis of the antibody pull-down material detected 2 putative hits out of 15 NEP peptides targeted as determined by co-eluting peaks in the extracted ion chromatogram: DEWISGAAVVNAFYSSGR, which partially coincides with the epitope for mAb 17E11, and an upstream sequence, IGYPDDIVSNDNK. However, upon manual curation, these peptide matches were considered of low confidence due to the poor peak matching assignments (Fig. S4). ELISA development Antibody pairing Twelve positive antibody pairs (signal-to-noise ratio > 10) were found from 10 × 10 capture/detector mAb checkboard combinations by ELISA using Trx-AgNEP(1–4) as test analyte (Table S2). The antibody pair that gave the highest signal comprised AgNEP-4/mAb 17E11 as capture and AgNEP-3/mAb 31E1 as detector while the pair that gave the highest signal-to-noise ratio comprised AgNEP-3/mAb 31E1 as capture and AgNEP-4/mAb 12A10 as detector. AgNEP-3/mAb 31E1, AgNEP-4/mAb 12A10, and AgNEP-4/mAb 17E11 displayed dual utilities as both capture and detector while AgNEP-3/mAb 25A5 was more useful as a capture antibody only. Among the 12 positive antibody pairs, 5 pairs displayed raw absorbance (450 nm) values greater than 1.0 (for the highest concentration of 8000 pg/ml) and were further analyzed. All five antibody pairs demonstrated a linear dose–response relationship over a concentration range of 125–8000 pg/ml with R-squared values exceeding 0.99 (Fig. 2 ). No antibody pairs were found when using cell-line-derived recombinant human sNEP as test analyte. Hence, Trx-AgNEP(1–4) was used as calibrator in the ELISA developed and represents a surrogate for relative quantification of circulating NEP. Assay performance and validation The antibody pair with the highest signal comprising AgNEP-4/mAb 17E11 as capture and AgNEP-3/mAb 31E1 as detection antibody was chosen for the development of an immunoassay to measure plasma NEP. The assay does not cross-react with all tested closely related endopeptidases spiked at concentrations between 5 and 100 ng/ml (Fig. 3 a). The assay working range of 13.1–8000 pg/ml was established based on the precision profile whereby the intra-assay coefficient of variation (CV) for each Trx-AgNEP(1–4) calibrator point did not exceed 20% in 18 independent assays performed. The limit of detection (LOD), defined as the concentration derived from the mean absorbance values of 18 zero standard replicates, was 2.15 pg/ml. The lower limit of quantification (LLOQ), defined as the lowest analyte concentration at which intra-assay CV was below 20% over 18 independent assays, was 13.1 pg/ml. Intra-assay CVs ranged from 1.7 to 3.2% for three different human plasma samples between the range of 151 and 1678 pg/ml of NEP. Inter-assay CVs ( n = 18) for NEP concentrations between 142 and 2076 pg/ml were 10.8–15.4%. Spike and recovery analyses showed poor recovery of between 51 and 64% when plasma samples were spiked with Trx-AgNEP(1–4). However, acceptable recovery of 108 to 118% was obtained when spiking with endogenous NEP from a high concentration plasma sample (Table 3 ). One possible explanation for poor recovery of Trx-AgNEP(1–4) is that the analyte from an exogenous source was mostly bound by unknown matrix component(s) rendering it unavailable for antibody binding. The dilution curves for six human plasma EDTA samples were parallel to the Trx-AgNEP(1–4) calibrator response curve, indicating that the recombinant calibrator was suitable for measuring endogenous human NEP (Fig. 3 b). NEP ELISA characteristics and assay performance are summarized in Table 4 . Western blot Reactivity toward recombinant sNEP, Trx-AgNEP(1–4), and endogenous cellular NEP All ten mAbs were tested for their reactivity toward recombinant human sNEP, recombinant Trx-AgNEP(1–4) (Fig. 4 a), and endogenous human NEP in LNCaP (human prostate adenocarcinoma) cell lysate (Fig. 4 b and c) using capillary western blot. Two of the mAbs (12A10 and 17E11) were reactive toward recombinant sNEP (~ 109 kDa) as well as endogenous human NEP (~ 130 kDa) while three of the mAbs (12A10, 17E11, and 31E1) were reactive toward Trx-AgNEP(1–4) detected at the ~ 28 kDa molecular weight position. The oxidizing environment of the mutant trxB bacteria host used for Trx-AgNEP(1–4) expression allowed dimerization via intermolecular disulphide bond formation and a second minor band at ~ 51 kDa was also detected. The apparent deviation in estimated molecular weight of immunodetected Trx-AgNEP(1–4) by capillary electrophoresis (CE) on the Jess/Wes Separation Module compared with SDS-PAGE (Fig. S1 i; theoretical molecular weight ~ 20.4 kDa) may be attributed to the different CE separation matrix and running conditions used in the former. Although molecular weight determination by CE and SDS-PAGE is generally congruent, deviation by as much as 30–40% of calculated molecular weight based on amino acid sequence can occur for some proteins [ 20 ]. PE pAb, previously validated by immunoprecipitation-mass spectrometry analysis to pull down NEP from human plasma [ 6 ], was used as the positive control antibody. All mAbs and PE pAb showed no immunoreactivity to the NEP-negative PC3-cell lysate. Results are summarized in Table 5 . Reactivity toward circulating NEP The circulating profile of NEP in human plasma was probed with mAb 17E11 and 31E1. PE pAb was also included for comparison. Plasma samples tested were obtained from pooled as well as individual HF patients with reduced ejection fraction (HFrEF) or preserved ejection fraction (HFpEF) and non-HF controls (CTRL). All three antibodies immunodetected four common bands at 57–60, 101, 143, and 177 kDa, albeit at different band intensities with weakest detection for the high-molecular-weight moieties by mAb 31E1 (Fig. 5 a). mAb 17E11 and mAb 31E1 share a common band at 33 kDa that was not apparent with PE pAb. A unique band at the 84 kDa position was recognized by mAb 17E11 alone. Fluctuating band thickness was observed for the immunodetected 57–60 kDa moiety which could be suggestive of variable post-translational modification and/or antibody reactivity against this fragment. Blocking of mAb 17E11 with the matched peptide completely abolished detection of Trx-AgNEP(1–4) as well as the 101, 143, and 177 kDa bands of the plasma samples while weak traces of the 33 and 57 kDa bands remained (Fig. 5 b). In the case of mAb 31E1, the matched peptide significantly blocked detection of only the 57–60 kDa band, albeit not completely to leave a weak band at 57 kDa. All other immunoreactive plasma sample bands were still weakly detected. Surprisingly, blocking of mAb 31E1 with the matched peptide marginally reduced detection of Trx-AgNEP(1–4). Increasing the concentration of the blocking peptide did not change the results for the plasma samples (Fig. S5). However, a reduction in the intensity of Trx-AgNEP(1–4) band was observed when mAb 31E1 was blocked at higher matched peptide concentration. ELISA peptide competition experiments corroborate the ability of the matched peptide to completely block Trx-AgNEP(1–4) detection by mAb 17E11 and only partially by mAb 31E1 (Fig. 6 ). Reactivity of mAb 31E1 to glycan-deficient recombinant sNEP HEK293 cell-line-derived recombinant sNEP is expected to be fully glycosylated and this is corroborated by its migration at a higher molecular weight position than its theoretical molecular mass of ~ 79.8 kDa on SDS-PAGE (Fig. 7 a, untreated). The band appears smeared between 90 and 110 kDa due to heterogeneity of the glycosites and glycan structures at each site. Following digestion with PNGase F, recombinant sNEP migrates as a sharp band at a lower molecular weight position (~ 78.2 kDa) compared with the untreated sample, confirming complete N-glycan removal. Western blot analysis confirms that mAb 31E1 does not bind to recombinant sNEP without PNGase F treatment but is able to detect deglycosylated recombinant sNEP in a concentration-dependent manner (Fig. 7 b).
Discussion Careful selection of peptide immunogens is an important first step in the entire antibody production workflow from antigen preparation to antibody characterization and validation. The B-cell linear epitope prediction tool used in this study is trained to locate antigenic peptides that tend to be surface accessible, hydrophilic, and flexible. These physicochemical properties coincide with parameters for soluble expression in E. coli , an important advantage for easy protein production and purification, and is corroborated in our previous work where epitope-predicted peptides on human ankyrin repeat domain 1 (hANKRD1) were all produced primarily as soluble Trx fusion proteins in the bacteria host [ 16 ]. Surprisingly, Trx-AgNEP-1 was produced primarily in the insoluble form while Trx-AgNEP-2 and -3 were partitioned into both soluble and insoluble fractions, with the former showing greater insolubility. We used the NetSolP analysis tool [ 21 ], a protein language model for in silico prediction of the solubility of expressed proteins in E. coli , to cross-check the predicted solubility of the Trx-AgNEP antigens. Indeed, Trx-AgNEP-1, -2, -3, and -4 achieved comparable high solubility scores of 0.8634, 0.8812, 0.8740, and 0.8817, respectively. We posit that the presence of a helical secondary element, composed of at least four amino acids sufficient to form one helical turn and located in a central position on the antigenic peptide as in the case of AgNEP-1 (see Fig. 1 b), contributes to the propensity for insoluble protein expression in E. coli . Although AgNEP-2 and -3 contained 3–4 amino acids that overlap from a predicted alpha-helix structure in the native protein, these are located to one end of the antigenic peptides and presumably exerts less influence in promoting protein insolubility. Concordant with the supposition that the presence of a helical secondary element contributes to insoluble protein expression, Trx-AgNEP-4 contained no predicted alpha-helix and was produced almost entirely in the soluble form. In further support of this postulation, re-examination of our three previously reported hANKRD1 antigenic peptides using the PredictProtein analysis tool showed only one peptide containing three amino acids that overlap from an alpha-helix region in the native hANKRD1 protein and hence is not expected to have a strong effect on Trx-peptide solubility. Indeed, all three Trx-peptide fusions were highly soluble in E. coli expression [ 16 ]. In view of these observations, we suggest that secondary structure predictions could also be incorporated prior to final selection of in silico predicted antigenic peptides to increase the likelihood of soluble recombinant immunogen production, making the protein expression and purification workflow more straightforward and efficient. We show that our previously reported Trx-tripeptide fusion strategy offers an effective antigen presentation scheme that consistently leads to successful generation of protein-reactive anti-peptide mAbs that exhibit high specificity and affinity under native and denaturing conditions. In this study, we applied a similar approach to generating protein-reactive mAbs targeting multiple non-overlapping sites on human NEP. The selected in silico predicted epitopes corresponding to short (10–15 residues) surface-exposed peptides initially yielded a total of 192 hybridoma clones (7, 41, 77, and 67 for AgNEP-1, -2, -3, and -4, respectively) that reacted strongly only with the peptide that they were generated against. However, after a second round of sub-culturing, only 16 hybridoma clones remained stable. All 16 clones were reactive toward either Trx-AgNEP-3 or -4. Of these, 10 mAbs were selected for detailed evaluation and application-specific validation. A few mAbs demonstrated protein reactivity toward recombinant sNEP by SPR (mAb 33G5, mAb 3E12 and mAb 17E11) and/or western blot (mAb 12A10 and mAb 17E11). Interestingly, although mAb 33G5 demonstrates strong affinity toward recombinant sNEP and Trx-AgNEP-3 by SPR, it was unable to detect these targets under denaturing conditions on western blot conditions. In contrast, mAb 12A10 detected only Trx-AgNEP-4 and showed no binding to sNEP by SPR but was able to detect both targets by Western blotting. These observations highlight differences in antibody reactivity to the same target in its native and denatured forms, and emphasizes the importance of careful selection and application-specific validation of antibodies for downstream deployment. In addition, 12 antibody pairs were found to be positive for two-site ELISA with Trx-AgNEP(1–4) but not recombinant sNEP as target analyte. Since mAb 33G5 and mAb 17E11 bind to recombinant sNEP in SPR, it is logical to expect some signal response in ELISA with this antibody pair. However, mAb 33G5 does not work well whether as a capture or detection antibody in ELISA as indicated by the checkerboard antibody screening results. When used as a capture antibody, poor performance may relate to partial denaturation of mAb 33G5 as it spreads over the microplate surface during passive adsorption via hydrophobic and hydrophilic interactions. Poor analyte binding by mAb 33G5 as a detection antibody may be attributed to changes in analyte conformation after its interaction with the capture antibody as exemplified in a precedent example [ 17 ]. We examined our NEP mAbs for suitability in Western immunoassay applications. Although all ten selected mAbs were able to bind to their cognate peptide in ELISA testing with the Multipin system, only three mAbs (31E1, 12A10, and 17E11) were able to detect denatured Trx-AgNEP(1–4) by western analysis. The high number of mAbs failing to immunodetect Trx-AgNEP(1–4) may be attributed to the preferred conformational fold adopted by the immobilized cognate peptide, whether influenced by other flanking peptides alone or in combination with the Trx carrier, lacking steric complementarity with the antibody combining site [ 22 , 23 ]. This result reiterates earlier observations that mAbs differ in their ability to bind to native and denatured forms of their antigen peptide sequence and this property governs their suitability for downstream applications. More importantly, the practical usefulness of anti-peptide antibodies resides in their ability to bind to the native protein from which its peptide sequence was derived. Two mAbs raised against AgNEP-4 (12A10 and 17E11) were able to detect Trx-AgNEP(1–4), recombinant sNEP and endogenous LNCaP NEP. However, mAb 31E1 derived from the AgNEP-3 peptide antigen detected Trx-AgNEP(1–4) but not recombinant sNEP or endogenous NEP. It is noteworthy that two reported glycosylation sites at N285 (in LNCaP [ 12 ] and B cell lymphoma cell lines [ 14 ]) and N294 (in B cell lymphoma cell lines [ 14 ]) are present on the epitope of mAb 31E1. Glycosylation at N294 would likely have a greater influence on antibody binding since it is flanked on both sides by the critical binding residues of the antibody. However, published mass spectrometry evidence for the N294 glycosite is weak as the fragment ion matches are quite sparse and the fragment ion for deglycosylated N294 was not detected [Supplemental data in reference 14]. Furthermore, the sequence context of N294 does not fall within the canonical glycosylation consensus motif (N-X-T/S where X is not a proline) nor to known atypical sequons [ 24 , 25 ]. Despite the uncertainty of glycosylation at N294, modification at the N285 glycosite might still be sufficient to abolish mAb 31E1 binding to NEP derived from eukaryotic sources since glycosylation at nearby sites outside of an epitope have been shown to be able to completely hinder antibody binding [ 26 ]. In addition, western blot analysis provided direct evidence that mAb 31E1 can bind to HEK293 cell-line-derived sNEP only after N-glycan removal by PNGase F digestion. Proteins that are glycosylated at multiple sites can assume a variety of glycoforms depending on site occupancy and glycan structure at each glycosite [ 27 ]. Proteins from different tissues of origin and disease states have been shown to exhibit variable glycosylation site occupancy [ 28 , 29 ]. Hence, the sensitivity of mAb 31E1 to glycosylation is of special interest in probing the significance of N285 and/or N294 glycan-deficient NEP in health and disease. At this juncture, it is worth highlighting a seeming conundrum in the differences in peptide and protein reactivity of mAb 33G5 and 31E1. Both mAbs were raised against the same AgNEP-3 peptide antigen and share exactly the same critical binding residues on the epitope. Yet, mAb 33G5 was able to detect cell-line-derived recombinant sNEP and its cognate Trx-peptide by SPR but not under western blot conditions. This antibody also showed poor utility, whether as a capture or detection antibody, in two-site ELISA with low signal response to Trx-AgNEP(1–4) as analyte. In contrast, mAb 31E1 exhibited good binding only to Trx-AgNEP(1–4), but not recombinant sNEP, by SPR, western blotting, and ELISA. How could the inhibitory effect of glycosylation on mAb 31E1 protein reactivity be reconciled with the immunity of mAb 33G5 to do so by SPR? First, glycosylation does not necessarily always impose steric hindrance to antibody binding but can even enhance antibody affinity to its antigen. It has been reported that glycosylation at one specific location on the MUC-16 epitope increased antibody recognition while modifications at other positions within the epitope were inhibitory to antibody binding [ 26 ]. Second, N-glycosylation has been observed to occur on the antibody variable light-chain domain. Such modifications can change the conformation of the antigen-binding region and increase or decrease subsequent antibody–antigen interactions [ 30 , 31 ]. Thus, it is possible that mAb 33G5 and 31E1 may differ in terms of the glycosylation profile of their variable light chains which in turn alters their peptide/protein-reactive properties and glycosylation sensitivity. We next investigated whether the mAbs 17E11 and 31E1 (the antibody pair used to develop the NEP ELISA) were able to detect circulating NEP in plasma by western blot analysis. Interestingly, both mAbs directed against different regions on NEP share four common immunodetected bands with the validated PE pAb, adding to confidence in antibody specificity for the target. Since NEP is highly glycosylated and can conceptually exist as monomers or dimers of the full-length protein or its ectodomain, it is difficult to interpret what the high-molecular-weight bands (> 100 kDa) represent. Since recombinant sNEP and LNCaP-derived endogenous NEP was found to migrate at the 109 and 130 kDa position, respectively, it is reasonable to postulate that the 101, 143, and 177 kDa bands detected in plasma might be glycosylated forms of monomeric NEP ectodomain, monomeric full-length NEP, and dimeric NEP ectodomain, respectively. PE pAb, mAb 31E1, and 17E11 consistently detected one major broad band at the 57–60 kDa position in all samples tested. This band exhibited slight differences in mobility and band thickness between samples, possibly reflecting a plethora of subtle nuances in post-translational modifications and/or antibody reactivity. Clear differences in antibody reactivity are exemplified in the case of mAb 17E11 and 31E1 binding to Trx-AgNEP(1–4) whereby more intense and broader immunodetected bands were obtained with mAb 31E1 under the same image exposure setting, especially for the higher molecular weight dimeric form of the recombinant protein. Peptide competition with the matched peptide for mAb 31E1 and 17E11 resulted in significant blocking of the 57–60 kDa band, though a weak residual band at the 57 kDa position could still be observed even at 25-fold excess of blocking peptide used. The poor ability of the free matched peptide to block the reaction between mAb 31E1 and the bound Trx-AgNEP(1–4), whether fixed within capillary tubes in western experiments or bound by the capture antibody in ELISA, suggests that the immobilized epitope sequence takes on a conformation that displayed much stronger affinity for mAb 31E1 [ 32 ]. Identifying the biological processes that generate moieties in the 57–60 kDa region may reveal important information on the regulatory cascades modulating NEP activity and implications in health and disease. Thus, efforts to elucidate the identity of this fragment constitute a worthwhile pursuit in further work. Finally, mAb 17E11 and 31E1 shared a band immunodetected at the ~ 33 kDa position that is apparently not recognized by PE pAb. The validated sandwich ELISA developed from this mAb pair will likely be predominantly measuring the 33 and 57–60 kDa NEP fragments in circulation. Using the SitePrediction tool [ 33 ], we searched for candidate protease cut sites predicted from cleavage site entries in the MEROPS database [ 34 , 35 ] with species specificity restricted to Homo sapiens. The analysis suggests that NEP can potentially be cleaved by numerous endopeptidases such as matrix metalloproteinases, cathepsins, and caspases at multiple sites. The SitePrediction output predicted matrix metalloproteinase 2 (MMP 2) to cleave at S 65 AAR↓LI 70 , A 187 IAQ↓LN 192 , S 251 VAR↓LI 256 (all three sites with > 99% specificity) and F 555 PAG↓IL 560 (> 99.9% specificity) (Fig. 8 ). Cleavage at S 65 AAR↓LI 70 and F 555 PAG↓IL 560 would result in a predicted 56.2 kDa (calculated from the amino acid sequence alone with no N-linked glycosylation at N145, N285, N294 or N325) fragment that could account for the 57 kDa immunodetected band. On the other hand, cleavage at A 187 IAQ↓LN 192 or S 251 VAR↓LI 256 and F 555 PAG↓IL 560 is expected to produce fragments of calculated molecular weights of 42.5 and 35.2 kDa, respectively, with the balance mass to make up the final mass of 57–60 kDa accounted for by the presence glycan chains at any of the abovementioned N-linked glycosylation sites. Cathepsin K (CatK), a cysteine protease with important functional roles in bone resorption and far-reaching action on other organs including the cardiovascular system [ 36 ], was also predicted to cleave NEP at ten different sites with a specificity ≥ 99%. Two of these cleavage sites at L 262 PID↓EN 267 and V 554 FPA↓GI 559 would generate a ~ 33.7 kDa (assuming no N-linked glycosylation at N285 or N294) fragment, representing the minimal sequence that contains both the epitopes of mAbs 17E11 and 31E1 (Fig. 8 ). This also harmonizes with the direct evidence from PNGase F experiments that mAb 31E1 is glycosylation sensitive and would bind only to glycan-deficient NEP fragments. Hence, the ~ 33 and 57–60 kDa NEP bands recognized by both mAbs could, respectively, be generated by CatK or MMP2 cleavage alone or in combination with each other. It can be further deduced that NEP concentration measured by our two-site ELISA will not correlate with NEP activity since the predominantly detected fragments do not harbor residues that make up the active site. A schematic representation of NEP and its fragments in circulation is depicted in Fig. 9 . The complexity of circulating forms of NEP may help explain why previous NEP studies gave conflicting results from immunoassay data. For example, it was reported that circulating NEP concentrations were elevated in HF patients compared with non-HF controls [ 6 ] and was positively associated with adverse outcomes [ 37 , 38 ]. Yet, in other studies, soluble NEP levels were found to be significantly lower in HF patients compared with non-HF controls [ 39 ] and not a prognosticator of adverse outcomes [ 40 , 41 ]. One of the contributing factors for these contradictory results could be the use of different ELISA kits in the various studies. Hence, depending on the antibody pair used in the immunoassay, different results may be obtained. Clearly, our data showed that different NEP antibodies recognized different subsets of circulating NEP forms. Disease severity may be correlated with a reduction or increase in N-linked glycosylation site occupancy of serum proteins [ 29 ]. In addition, N-linked glycosylation is known to modulate protein function by altering the interaction of a cell-surface protein with its ligand [ 42 ]. The new mAb pair generated in this work will add to the toolbox of well-defined binders to probe dynamic changes in NEP concentration and glycosylation site occupancy in health and disease. It remains to be determined which specific fragment(s) of NEP is useful as a biomarker and this would certainly be an important area requiring further research.
Conclusion We show that the thioredoxin scaffold constitutes an effective strategy for surface presentation of in silic o predicted peptides as immunogens to generate high-affinity mAbs with well-defined epitope binding footprint. The validated antibody pair comprising mAb 17E11 and mAb 31E1 can be applied in Western blot and ELISA applications, adding to the arsenal of much needed reliable binders to unravel NEP biology, regulation, and function in the context of health and disease. Successful generation of glycosylation-sensitive mAb 31E1 demonstrates the power of epitope-directed antibody production by allowing peptides harboring post-translational modifications of interest to be preselected to meet experimental objectives. New information on the complexity of circulating forms of NEP and the consequent implications for correctly interpreting immunoassay data have been highlighted. We are now poised for applying our NEP ELISA on clinical HF samples to elucidate dynamic changes in NEP concentration in disease and how it might associate with adverse outcomes.
Neprilysin (NEP) is an emerging biomarker for various diseases including heart failure (HF). However, major inter-assay inconsistency in the reported concentrations of circulating NEP and uncertainty with respect to its correlations with type and severity of disease are in part attributed to poorly characterized antibodies supplied in commercial ELISA kits. Validated antibodies with well-defined binding footprints are critical for understanding the biological and clinical context of NEP immunoassay data. To achieve this, we applied in silico epitope prediction and rational peptide selection to generate monoclonal antibodies (mAbs) against spatially distant sites on NEP. One of the selected epitopes contained published N-linked glycosylation sites at N285 and N294. The best antibody pair, mAb 17E11 and 31E1 (glycosylation-sensitive), were characterized by surface plasmon resonance, isotyping, epitope mapping, and western blotting. A validated two-site sandwich NEP ELISA with a limit of detection of 2.15 pg/ml and working range of 13.1–8000 pg/ml was developed with these mAbs. Western analysis using a validated commercial polyclonal antibody (PE pAb) and our mAbs revealed that non-HF and HF plasma NEP circulates as a heterogenous mix of moieties that possibly reflect proteolytic processing, post-translational modifications and homo-dimerization. Both our mAbs detected a ~ 33 kDa NEP fragment which was not apparent with PE pAb, as well as a common ~ 57–60 kDa moiety. These antibodies exhibit different affinities for the various NEP targets. Immunoassay results are dependent on NEP epitopes variably detected by the antibody pairs used, explaining the current discordant NEP measurements derived from different ELISA kits. Supplementary Information The online version contains supplementary material available at 10.1007/s00018-023-05083-1. Keywords
Supplementary Information Below is the link to the electronic supplementary material.
Abbreviations Neprilysin Heart failure Research-use-only Enzyme-linked immunosorbent assay Monoclonal antibody Polyclonal antibody Thioredoxin Immobilized metal affinity chromatography Surface plasmon resonance High-resolution multiple reaction monitoring Bovine serum albumin Phosphate buffered saline Human endothelin converting enzyme 1 Human endothelin converting enzyme 2 Human phosphate-regulating neutral endopeptidase Human endothelin converting enzyme like 1 Limit of detection Lower limit of quantitation Coefficient of variation Heart failure Reduced ejection fraction Preserved ejection fraction Non-HF control Capillary electrophoresis Acknowledgements Funding support for this work was provided under the Centre Grant (NMRC/CG21APR1008) awarded to A. Mark Richards by the National Medical Research Council, Singapore. Author contribution OWL and SSML designed and developed the antibody production methodologies, performed most of the experiments, analyzed the data, and wrote the paper; LS, JYX, JPCC performed hybridoma screening and analysis of results; QFL, XEY, TKL, and QSL performed the tryptic digest and mass spectrometry analysis of immunoprecipitated eluates for antibody verification; AMR contributed key intellectual inputs to the manuscript and is the principal investigator of the grant that supported the projected. All authors have contributed, discussed, and approved the final version of the manuscript. Funding Funding support for this work was provided under the Centre Grant (NMRC/CG21APR1008) awarded to A. Mark Richards by the National Medical Research Council, Singapore. Data availability All data are available from the corresponding author upon reasonable request. Declarations Conflict of interests All authors declare no competing interests.
CC BY
no
2024-01-15 23:42:02
Cell Mol Life Sci. 2024 Jan 13; 81(1):42
oa_package/2d/2b/PMC10787894.tar.gz
PMC10787895
38051494
Introduction Interferons (IFN) are a widely expressed family of cytokines. They are categorised, based on their receptor signalling, into types I, II, and III [ 1 ]. IFN-I signal via a heterodimeric receptor composed of two distinct multi-chain structures, IFN-α receptor 1 and 2 (IFNAR-1 and IFNAR-2). The former is constitutively associated with tyrosine kinase 2 (TYK2) and the latter associated with Janus Kinase 1 (JAK1) [ 2 ]. IFN-I are produced as part of the innate immune response to infection and possess potent antiviral effects [ 2 ]. Triggers of IFN-I production and subsequent downstream signalling have been recently reviewed in [ 3 ] and is summarised in Fig. 1 . Similarly, IFN-II and IFN-III signal via their own unique heterodimeric receptors composed of IFN-γ receptors 1 and 2 (IFNGR-1 and IFNGR-2), IFNLR1 (IFN lambda receptor-1), and IL-10R2 (interleukin-10 receptor 2) subunits, respectively [ 4 ••]. Both of which subsequently lead to downstream signalling and potential induction of interferon inducible genes. In this review, we explore what role IFN-I, particularly IFN-α, may play in rheumatoid arthritis (RA) pathophysiology.
Conclusions There is growing evidence that IFN-α plays an important role in early RA pathophysiology and Fig. 5 summarises a working paradigm on IFN-α influencing RA progression. However, the triggers of IFN-α production and its timing in relation to early immune dysregulation or symptom onset remain unclear. Further work focusing on early disease or at-risk populations with a focus on genetic and epigenetic factors is likely to be informative. Despite mechanistic uncertainties, there is clear rationale to further test IFN-α targeting therapies in early RA, potentially using the IGS as a theragnostic biomarker, or to use the IGS as a biomarker for more intensive initial therapy. The heterogeneity and variety of IGSs remain challenging with regard to clinical utility, but recent progress in the international community on IGS stratification and uniform application of standardised measures of IFN-I signalling is encouraging [ 4 ••, 5 ••, 124 , 125 ], and its use in this capacity may be on the horizon.
Purpose of Review Type 1 interferons (IFN-I) are of increasing interest across a wide range of autoimmune rheumatic diseases. Historically, research into their role in rheumatoid arthritis (RA) has been relatively neglected, but recent work continues to highlight a potential contribution to RA pathophysiology. Recent Findings We emphasise the importance of disease stage when examining IFN-I in RA and provide an overview on how IFN-I may have a direct role on a variety of relevant cellular functions. We explore how clinical trajectory may be influenced by increased IFN-I signalling, and also, the limitations of scores composed of interferon response genes. Relevant environmental triggers and inheritable RA genetic risk relating to IFN-I signalling are explored with emphasis on intriguing data potentially linking IFN-I exposure, epigenetic changes, and disease relevant processes. Summary Whilst these data cumulatively illustrate a likely role for IFN-I in RA, they also highlight the knowledge gaps, particularly in populations at risk for RA, and suggest directions for future research to both better understand IFN-I biology and inform targeted therapeutic strategies. Supplementary Information The online version contains supplementary material available at 10.1007/s11926-023-01125-6. Keywords
The Interferon Gene Signature (IGS) Measuring IFN-α protein in vivo has been historically challenging due to low circulating levels being frequently below the detection thresholds of standard assays. A solution was to infer IFN-I exposure, and hence levels, by measuring transcripts that reflect interferon stimulated or response genes (IRGs), and their cumulative expression was termed the interferon gene signature (IGS) (see Fig. 2 ). However, there are over 2000 IRGs and which IRGs are chosen to generate an IGS is lacking consensus across studies [ 5 ••]. Despite this, an IGS is widely reported in autoimmune rheumatic diseases, and there are mutual IRGs increased in RA and other rheumatic diseases [ 6 ]. Nevertheless, some propose an exclusive and highly diverse IRG transcriptional profile in RA peripheral whole blood [ 7 ] as well as in synovial biopsy samples [ 8 ], distinct from that found in SLE. However, IRG expression, and subsequently the calculated IGS, may vary between different cell types, suggesting that differences seen amongst related autoimmune diseases could be secondary to different immune cell proportions and signalling pathway activation [ 9 ]. Indeed, variation is seen in flow cytometry detected STAT class phosphorylation in CD4+ T cells, CD8+ T cells, B cells, and monocytes following IFN-I stimulation [ 10 ]. As IFN-I, IFN-II, or even IFN-III can induce IRGs (see Fig. 1 ), there has been a historical lack of clarity as to which IFN class was responsible for the IGS in RA. Indeed, it remains a controversial topic as IRG expression may be modulated by additional stimuli, such as TNF-α, with variable effects reported in monocytes vs T cells [ 11 ]. Nevertheless, in established RA, there was reportedly equal contribution of IFN-α and IFN-β to the whole blood IGS vs IFN-α exposure being dominant in SLE [ 9 ]. However, in a cohort of nearly 200 early drug naïve RA patients, circulating IFN-α protein and not IFN-β, IFN-II, or IFN-III nor any other circulating inflammatory cytokine uniquely correlated with the whole blood IGS [ 12 ••]. This work remains to be validated, and reported differences may reflect disease stages, but does implicate predominantly IFN-α with the IGS in early RA. Despite these caveats regarding its calculation, the IGS remains a useful tool in dissecting the role of IFN-I in RA, as explored below. The IGS by Disease Phase It is increasingly appreciated that disease processes in early RA are likely to be distinct from established RA. In early RA, a raised IGS ( MxA , OAS1 , ISG15 , IFI44L , IFI6 ) was more prevalent compared with established RA, approximately 50% vs 20% of patients, respectively [ 13 ], and fell with the initiation of therapy [ 12 ••, 13 ]. Therapeutics may contribute to a reduced incidence in established RA as glucocorticoids, as well as disease modifying anti-rheumatoid drugs (DMARDs), can modify the IGS [ 14 ]. Notably, this increase in early RA persists even after accounting for potential confounders such as disease stage dependant variation in cell subset proportions [ 15 ]. Corroborating the raised IGS noted at disease onset, there is emerging data that IFNs may contribute to the transition from preclinical to sustained clinical disease. In ontology studies and network pathway analyses, the IGS distinguished DMARD-naïve early arthritis patients that developed a persistent inflammatory arthritis from those that had a self-limiting course [ 16 ]. In ACPA+ arthralgia populations, i.e. those who are at risk for developing RA, an IGS increases the chance of progression to synovitis, and its inclusion in outcome models improved its predictive capacity [ 17 , 18 ]. Even in healthy asymptomatic CCP+ individuals, there was evidence of increased IFN-α signalling which mirrored what was seen in early RA cohorts, and this, with other parameters, was able to differentiate progressors with a median of 4.1 years before symptom onset, from controls [ 19 , 20 •]. In seropositive and seronegative RA, as well as in high-risk seropositive arthralgia patients, there was an overlap in circulating cytokine profiles with IFN-α, as well as IL-5, and TNF-α upregulated up to 50% in seropositive arthralgia and seropositive RA patients but not in seronegative RA [ 21 ] with an odds ratio (OR) of 21 for RA development in seropositive arthralgia patients [ 18 , 21 ]. Clinical Characteristics and the IGS There has been conflicting evidence around the impact of an IGS/IFN-I signalling on autoantibody production in RA. In established RA, there is a significant correlation between the IGS and ACPA titres and anti-carbamylated protein (anti-CarP) antibodies as well as with genes linked to B cell differentiation and antibody production [ 22 ]. Conversely, others found no relation between the IGS and the presence and/or titres of ACPA and RF in established disease [ 23 ]. Similarly, a 2016 systematic analysis, involving patients with established RA, found that there was no difference seen in the IGS between ACPA negative and ACPA positive patients [ 24 ]. Conversely, rheumatoid factor (RF) demonstrated a positive association between either the IGS or circulating IFN-α levels in both established and early RA as well as across several autoimmune rheumatic diseases [ 12 ••, 13 , 25 ]. These differences may reflect disease stage but may also reflect variability in the IRGs chosen to represent the IGS, with some using a combination of 19 IRGs [ 24 ] and others using only 6 (IFI27, IFI44L, IFIT1, ISG15, RSAD2 and SIGLEC1) [ 25 ] for example. Multiple observational studies in established RA have found no association between the IGS and disease activity [ 13 , 24 ]. This contrasts with early drug naïve RA where in a number of prospective observational studies, a higher IGS in early drug naïve RA, were associated with increased baseline disease activity as well as a poorer response to initial therapies [ 12 , 13 , 15 ] which was validated in additional cohorts for specific IGSs [ 26 ]. RA, including early disease, is a risk factor for cardiovascular disease (CVD). [ 27 , 28 ]. In mouse and human in vitro models, IFN-α stimulation of macrophages resulted in significant metabolic rewiring with over 500 metabolic genes, including those related to key processes in the pathophysiology of CVD such as glycolysis, oxidative phosphorylation, fatty acid synthesis, and lipid metabolism [ 29 ]. In addition, endothelial progenitor cells (EPCs), involved in vasculogenesis and repair, have impaired function in RA, with IFN-α implicated in both in vitro and in vivo studies [ 30 – 32 ]. In murine lupus models, prolonged and enhanced IFN-I exposure significantly reduced EPC numbers, with acute exposure affecting only EPC differentiation but not the cellular number [ 30 ]. Finally, IFN-α may also influence CVD by promoting insulin resistance given, as early as the 1980s, IFN-α was shown to impair glucose tolerance and insulin sensitivity [ 33 ] with reversal of this effect in IFNAR-/- mouse models [ 34 ]. IFN-I and Its Effects on Cellular Function B and T Cells IFN-I can widely influence B cell activity which may contribute to RA pathophysiology, for example, by supporting B cell survival via increased monocyte B-lymphocyte stimulator (BLyS) production, by direct stimulation of B cells, and indirectly through T cell and Dendritic Cells (DCs) stimulation [ 35 ]. Prolonged B cell survival can lead to increased differentiation into memory and plasma cells, immunoglobulin isotype switching, and autoantibody formation [ 36 , 37 ]. Furthermore, IFN-α modifies the plasma cell transcriptome towards a proinflammatory phenotype [ 38 ]. IFN-I regulates BCR signalling, specifically via IFN-αR, which in turn may promote pathways involved in antibody formation and germinal centre development in murine models [ 39 ]. IFN-I can also influence the differentiation of CD4+ T cells towards a Th1 response [ 40 ], fostering B cell activation and subsequent activity [ 41 ]. IFN-I also promotes CD8+ T cell survival and CD8+ cytotoxic T cell activity as well as prolonging the proliferation and expansion of CD8+ antigen specific T cells via inhibition of apoptosis [ 42 ]. Dendritic Cells Dendritic cells (DCs) upregulate HLA-DR, CD40, CD80, and CD86 expression upon IFN-I exposure [ 43 ]. DC maturation and enhanced antigen presentation, in the context of increased co-stimulatory molecules, can result in the induction of autoimmunity in predisposed individuals via self-antigen presentation to low affinity autoreactive T cells [ 44 ]. In SLE susceptible mice, IFN-I-treated DCs showed relative apoptosis resistance, this activated DC longevity potentially contributing to the development of autoimmunity [ 45 ]. Conversely, in early drug naive RA, there was no association between the IGS and circulating CD1c or pDC frequency, but there was an inverse association with CD141+ DC frequency [ 46 ]. This highlights the DC subset dependant complexity of IFN-I signalling in vivo . Monocytes Classical and non-classical monocytes have been implicated in RA pathogenesis [ 47 ]. How the IGS affects monocyte function in vivo in RA remains to be fully examined but, when exposed to IFN-Is in vitro , monocytes upregulate TLR7 and IRF expression, resulting in increased responsiveness to subsequent immunostimulatory ligands [ 48 ]. IFN-I exposure also increases expression of CD40, CD80, and CD86 and HLA-DR, ultimately promoting differentiation into a monocyte-derived dendritic cell, or mo-DC, with high capacity for antigen presentation [ 43 , 49 ]. Mo-DCs are also known to be increased in the RA synovial compartment and promote Th17 differentiation [ 50 ]. However, as with DCs, what happens in vivo may be subset dependant as highlighted by enhanced responsiveness to IFN-α in murine proinflammatory monocytes secondary to increased IFNAR expression when compared with anti-inflammatory monocytes [ 51 ]. Neutrophils Neutrophils are one of the first cell types to enter the RA joint and may play an important role in the development and progression of RA [ 52 ]. They are a major contributor to the whole blood IGS in RA, attributed to their uniquely upregulated IFNAR expression, a phenomenon not seen in either healthy controls or RA PBMCs [ 53 •]. Indeed, next generation sequencing of isolated blood neutrophils has found significantly upregulated IRGs in RA neutrophils compared to healthy controls [ 54 ]. How this increased sensitivity to IFN-I influences neutrophil function is being explored, but, intriguingly, the pathogenic phenotype proposed for RA consists of delayed neutrophil apoptosis, increased ROS production and chemokine expression which, in part, can be recapitulated by IFN-I exposure in vitro [ 55 •]. Fibroblasts Synovial fibroblasts are resident cells in the stroma of joints [ 56 ], and we recently demonstrated comparable IFN-α levels in serum and early RA synovial fluid [ 12 ••]. In RA, these fibroblasts have an activated phenotype, characterised by resistance to apoptosis, and increased proliferation and production of inflammatory mediators that promote immune cell differentiation and survival [ 57 ]. Histology and RNA sequencing of early RA synovial tissue demonstrated three distinct pathotypes: fibroblastic pauci-immune, macrophage-rich diffuse myeloid, and a lympho-myeloid pathotype [ 58 ••]. In the lympho-myeloid pathotype, seven out of the eight differentially expressed blood transcripts in synovial versus whole blood were IFN-I responses genes (IFI27, ISG15, IFI44L, OASL, USP18, RSAD2, LY6E) [ 58 •]. In addition, a pathogenic subset of sub-lining fibroblasts ( THY1 + HLA − DR high ) have increased IRG expression [ 59 ]. However, this may not directly be secondary to IFN-I as TNF-α induced signalling co-opts the mTOR pathway to shift fibroblast like synoviocytes towards an IFN response [ 60 ] which has been shown to be via secondary autocrine production of IFNβ and subsequent activation of the IRF1-IFNβ-IFNAR-JAK-STAT1 axis [ 61 ]. Nevertheless, the role of IFN-α on fibroblast function in RA remains an important research question. Cumulatively, these effects are likely to contribute to a highly activated and potentially autoimmune prone phenotype as summarised in Fig. 3 . Source of IFN-I in RA pDCs, particularly in their immature state, are the main IFN-I producing cell, however, whether they are the primary source of IFN-α in RA remains unclear. In SLE, there is an element of so-called pDC fatigue, where the ability of the pDC to produce IFN-I reduces and other cells take over production [ 62 ]. In early RA, circulating pDCs were not the primary source of IFNA transcript, with comparable expression in circulating lymphocytes. However, circulating pDC numbers were reduced with increased CCR7 expression inferring increased migration to the synovial compartment and target tissue [ 46 ]. Indeed, in established RA patients, the synovial compartment has increased numbers of pDCs with reduced numbers seen in peripheral blood. However, those that remained in the circulation were immature with inferred increased IFN-I producing capacity [ 63 ]. Nevertheless, RA synovial pDCs are potent producers of IFN-α [ 64 ] and, in mice, intraarticular transfer of IFN-I producing dendritic cells was sufficient to propagate a persistent inflammatory arthritis [ 65 ]. Conversely, after arthritogenic serum transfer in K/BxN serum-induced arthritis, collagen-induced arthritis, and human TNF transgene insertion, only pDC deficient mice showed exacerbations of symptoms and signs of inflammatory arthritis [ 66 ] and topical imiquimod, a TLR7 agonist, increased pDC recruitment and activity which subsequently improved arthritis [ 66 ]. Furthermore, transcriptomic analysis of circulating pDCs in early RA suggested enhanced tolerogenic function [ 46 ]. These discrepancies may arise due to the complexities of DC development [ 67 ] and cellular differences across species. Given their relative paucity in vivo , pDCs have been relatively neglected in RA research; however, better understanding of their complexity, particularly in relation to any location specific function, will help inform their role in RA and role in IFN-I production. Potential Triggers of IFN Production What drives the observed increased IGS/IFN-α in RA remains unclear; however, triggers may include viral infections or microbial DNA or antigen fragments, with these elements repeatedly reported in the joints of RA patients [ 68 – 70 ]. Retroelements are non-protein encoding portions of DNA derived from ancient transposable elements, such as retroviruses, that have been historically incorporated into the genome. Their activity can trigger intracellular viral sensors and thus promote local IFN-I production [ 71 ]. In SLE and primary Sjogren’s syndrome, increased retrotransposon activity in disease relevant tissue associated with increased local IFN-α production [ 72 •], and, in established RA synovium, there is also increased retroelement expression [ 73 , 74 ]. Furthermore, in a subgroup of RA patients, a transcriptional profile was documented, reminiscent of a viral infection, which associated with both IFN-I signalling as well as increased ACPA titres [ 75 , 76 ]. How these retroelements may influence IFN-I production in RA remains to be seen. Cell-free nucleic acids have been extensively implicated in IFN-I generation in SLE [ 77 ] and monogenic interferonopathies [ 78 ]. Mouse models with DNA clearance defects develop autoantibody-mediated chronic polyarthritis, resembling human RA [ 79 ]. This corroborates RA human observational data, where evidence of raised levels of circulating cell -free DNA have been found in both peripheral blood [ 80 , 81 ] and synovial fluid [ 82 ]. Although direct links with IFN-I were not made in these human RA studies, a similar mechanism to that described in SLE may be present . Neutrophils, found in high numbers in RA synovium, can undergo NETosis, a unique form of cell death which has been proposed as a potential trigger for IFN-I production [ 83 ]. DNA from these NETs form complexes with antimicrobial peptides including LL37, secretory leukocyte protease inhibitor (SLPI), or with immunoglobulins to form immune complexes which facilitate pDC TLR7/9 signalling ultimately culminating in IFN-α production [ 84 – 86 ]. In RA, links between NETs and ACPA have been reported [ 87 , 88 ] and known pathogenic cytokines in RA, such as TNF-α and IL-17A, as well as IFN-a itself [ 89 ], can also induce NETosis, potentially creating a vicious cycle of inflammation and disease activity [ 88 ]. Other potential triggers include lifestyle and environmental factors. An inverse correlation between physical activity and IFN-I signalling has been reported [ 90 ]. In addition, physical activity was associated with downregulation of TLR and IL-17R signalling and reduced inflammatory cytokines production, including IFN-I [ 90 ]. Potential triggers are summarised in Fig. 4 ; however, much of the above involves extrapolation from other diseases, such as SLE, and caveats exist including differences in IRG expression and genetic risk between these diseases [ 91 ]. Further work is needed to explore these pathways in RA specifically. Heritable Genetic Risk and IFN-I Signalling As genome-wide association studies (GWAS), and relevant data sets, become more available, numerous single nucleotide polymorphisms (SNPs) have been identified as contributing to the genetic risk of RA. Interestingly, a number of these SNPs are in genes related to the IFN-I response pathway including DNA-sensing proteins, toll-like receptors, and JAK-STAT protein mediators. These are summarised in Table 1 . However, the functional consequences of these polymorphisms in RA with regards to IFN-I production or signalling are yet to be elucidated. Nevertheless, an overlap of certain at-risk genes associated with increased IFN-I signalling in SLE has also been linked to RA, for example SNPs in IRF5, STAT4, and PTPN22 [ 92 ]. Some of these RA risk SNPs, such as IRF5 polymorphisms, associate with more severe or erosive disease [ 93 , 94 ], which may corroborate the clinical refractory disease phenotype observed in IGS high early RA patients [ 12 ••]. Further work is needed to elucidate both the role of IFN-I on susceptible genetic backgrounds as well as the contribution of these SNPs to IFN-I production. IFN-I and Epigenetics Epigenetic changes are modifications that regulate genome activity, independent of DNA sequence. This occurs via molecular factors and processes, such as DNA methylation of CPG sites or chromatin conformational changes, which subsequently modulate transcription. They are frequently triggered by environmental factors or exposure to inflammatory stimuli, such as cytokines. Methylation changes are noted early in RA progression and vary by cell subset [ 95 ••]. Furthermore, differential methylation has been implicated in initial response to methotrexate in early drug naïve RA patients as well as to certain biologics in established disease [ 95 ••, 96 – 98 ], and these processes are emerging as important modifiers of RA clinical progression and phenotype [ 99 ]. Analysis of B and CD4 T cells from early drug naïve RA patients demonstrated differentially methylated CPG sites at disease relevant genes, such as PARP9, STAT1, and EPSTI between IGS high and low patients. It also implicated altered transcription factor binding, cumulatively promoting increased lymphocyte activation, and a proliferative phenotype in the IGS high cohort [ 12 ••]. These data suggest that these changes may be IFN-α induced, and negatively influence clinical trajectory. In undifferentiated arthritis (UA) monocytes, methylation changes, which associated with disease progression and a poor prognosis, were partially recapitulated by monocyte exposure to IFN-α [ 100 •]. Furthermore, IFN-α treatment causes methylation changes in monocytes similar to those seen in established RA, which in vivo were themselves associated with increased disease activity [ 101 ]. Intriguingly, in models of type 1 diabetes, where IFN-I plays a key part in disease initiation, exposure to IFN-α triggered increased TET2 expression. This prompted hypomethylation changes in genes controlling inflammatory and immune pathways, ultimately resulting in their increased expression and disease acceleration [ 102 ]. TET proteins are key players in demethylation and are also increased in early drug naïve RA circulating lymphocytes [ 103 ], however whether this is secondary to IFN-α is unknown. It is important to acknowledge that CPG sites in IRGs themselves are frequently hypomethylated in autoimmune conditions, including RA [ 104 , 105 ]. In twin studies of CD4 T cells, hypomethylation of IRGs IFIT1 , IRF7 , MX1 , OAS1 , USP18 , RSAD2, and IFI44L has even been proposed as biomarkers of progression to RA [ 104 ]. This questions whether the IGS could be an artefact of altered gene expression secondary to hypomethylation caused by other circulating inflammatory cytokines, such as IL6, or due to increased IFN-α signalling itself. IFN-α protein is increased in early RA and uniquely correlates with the IGS [ 12 ••], so the reality is likely to involve both mechanisms. Although less extensively investigated, IFN-I-associated chromatin conformational changes may also be relevant to RA pathophysiology. There is variation in chromatin accessibility in RA synovial fibroblasts which is likely influenced by the synovial environment [ 106 ], where IFN-α is known to be present [ 12 ••]. In early RA, chromatin conformation changes in IFNAR2 were associated with poorer outcomes [ 107 ]. Furthermore, monocytes stimulated with IFN-α have increased trimethylated histone H3 Lys 4 (H3K4me3) which enhances transcription at promotors of genes that encode inflammatory mediators. In a more representative in vivo environment, incubation of monocytes with both IFN-α and TNF-α was associated with increased H3K4me3 that reduced monocyte tolerization to LPS and promoted an enhanced response to subsequent environmental challenges [ 108 ]. This intriguingly implicates IFN-I, and chromatin-mediated modifications, with the induction of inflammatory genes beyond canonical IRGs. Indeed, instances where prior exposure to IFN-α can influence cellular response to additional stimuli are increasingly being reported [ 102 , 109 – 111 ] and remain a potential mechanism whereby IFN-I can influence disease development in RA. The IGS/IFN as a Therapeutic Target Anifrolumab targets IFNAR1 and therefore blocks IFN-α and IFN-β signalling [ 112 ]. In a pilot trial, seven established RA patients, all with a high IGS ( IFI27, IFI44, IFI44L and RSAD2 ), and active diseases were randomised to anifrolumab or placebo [ 113 •]. The primary endpoint of an American College of Rheumatology (ACR) response of ≥ 20% after 24 weeks was achieved in patients receiving anifrolumab although only one patient in each arm completed the study despite a safety profile similar to that reported in SLE [ 114 ]. Reasons for early discontinuation in the treatment group included lack of efficacy, hypersensitivity reaction, and infection whilst the control group participants stopped due to insufficient therapeutic response [ 113 •]. Larger trials are needed to assess the efficacy of this drug in RA. Alternatively, JAK inhibitors (JAKi) suppress phosphorylation of STAT and thus affect downstream IFN signalling and reduce IRG expression [ 92 ]. Indeed, in vitro JAKi reduce IFN-I driven plasmablast differentiation [ 115 ], synovial BAFF expression, monocyte-derived DCs costimulatory molecule CD80/CD86 expression, and T cell differentiation into Th1 and Th17 cells [ 115 , 116 ] [ 61 ]. The efficacy of JAKi in the treatment of established RA has been widely reported [ 117 ], although how the IGS impacts on its effect has not been comprehensively examined. However, analysis of baricitinib SLE trial data demonstrated that clinical effect was independent of IGS reduction [ 118 ]. Other inhibitors of downstream IFN-I signalling include a novel small molecule selective for JAK3/JAK1/TBK1 (tank-1 binding kinase), which, in mouse models, suppressed IFN-I production and osteoclast formation via TBK1 inhibition [ 119 ]. Autoantibody dependent collagen-induced arthritis mice models confirmed the clinical benefit of TBK1 inhibition [ 120 , 121 ] and TBK1 deficient mice have reduced IRG and protein expression [ 122 ]. These findings are yet to be reproduced in human studies, but given the interest in cancer regarding TBK1 inhibition [ 123 ], this may provide a novel therapeutic approach. Supplementary Information
Funding This work was supported by the Research into Inflammatory Arthritis Centre Versus Arthritis (RACE) (grant number 22072) and the National Institute for Health and Care Research (NIHR) Newcastle Biomedical Research Centre for Ageing and Long-Term Conditions; views expressed are the authors’ and not necessarily those of the National Health Service, the National Institute of Health and Care Research, or the Department of Health. Compliance with Ethical Standards Conflict of Interest JDI discloses research grants from Pfizer, Janssen, and GSK; conference support from Eli Lilly and Gilead; speaker/consulting fees from AbbVie, BMS, Gilead, Roche, and UCB. FAHC discloses speaker fees from AstraZeneca. The remaining authors have no competing interests. Human and Animal Rights and Informed Consent This article does not contain any studies with human or animal subjects performed by any of the authors.
CC BY
no
2024-01-15 23:42:02
Curr Rheumatol Rep. 2024 Dec 5; 26(2):37-52
oa_package/5b/23/PMC10787895.tar.gz
PMC10787896
38217570
Introduction Various technical and procedural modifications have been implemented intraoperatively to reduce fluoroscopy exposure to both urologists and patients, including proper shielding, low power fluoroscopy settings, and conservative use of fluoroscopy [ 1 ]. Although these measures significantly reduce radiation exposure to the operating room staff and patient, the surgeon’s hand remains susceptible to radiation exposure using these methods. This is concerning as ionizing radiation is a risk factor for malignancy and many other adverse effects [ 2 , 3 ]. Surgeons who frequently work within the direct radiation beam may experience skin and nail pigment abnormalities, joint pain, and osteoarthritic changes [ 4 – 6 ]. Furthermore, the full understanding of the health effects of radiation to the hand are currently limited in existing literature [ 7 , 8 ]. Compared to shock wave lithotripsy (SWL) and ureteroscopy, percutaneous nephrolithotomy (PCNL) confers the highest radiation to both the surgeon and patient [ 9 – 11 ]. Radiation exposure to the surgeon includes both scatter and direct radiation with direct exposure being exponentially greater. The most common scenario when surgeons experience direct radiation exposure is while holding the needle during percutaneous access [ 9 , 12 ]. This becomes even more important as the indications for PCNL expand and as a greater number of urologists obtain their own access [ 13 ]. To mitigate this exposure, radiation-attenuating gloves and needle holders have been proposed as potential protective measures by reducing penetrating radiation and enabling removal of the hand from the direct radiation beam. However, their effectiveness during percutaneous renal access for PCNL has not been investigated. The aim of this study was to evaluate the effectiveness of a radiation-attenuating glove compared to a novel needle holder in reducing radiation exposure to the surgeon's hand.
Methods Study design and set up After approval from the Loma Linda University’s Department of Pathology and Human Anatomy, and in compliance with institutional policies for use of anatomical specimens in research, a simulated percutaneous renal access for PCNL was performed. A male cadaver (body mass index 36.1) was positioned prone and draped in a manner typical for PCNL. A separate cadaveric right upper extremity, representing the surgeon's hand, was positioned to simulate percutaneous right renal access with an access needle (Fig. 1 ). Radiation dose measurements Landauer nanoDot optically stimulated luminescence dosimeter (OSLD) chips (Glenwood, Illinois) were affixed to four locations on the surgeon's hand: the thumb, middle finger, hypothenar eminence, and forearm (Fig. 2 ). Additionally, two OSLD chips were placed on the patient: one on the ventral surface and one on the dorsal surface of the skin directly in line with the right kidney. The index finger was positioned 7.6 cm above the skin overlying the right kidney. Fluoroscopy was performed using a GE OEC 9900 portable C-arm system (GE Medical system, Inc., Salt Lake City, UT) using the default automatic exposure control (AEC) for all trials. The AEC adjusts the milliampere-seconds (mAs) and peak kilovoltage (kVp) based on the target density to provide optimal image quality [ 14 ]. The C-arm was positioned over the right kidney with the X-ray source below the table and the image intensifier above the patient at a skin-to-source distance of 20 cm. The study consisted of three groups. The first group served as the control with radiation exposure tested on the surgeon’s hand wearing conventional polyisoprene surgical gloves (Mölnlycke, Gothenburg, Sweden) directly holding the access needle. In the first experimental group, the renal access needle was held directly by the cadaveric arm using radiation-attenuating gloves (AliMed, Dedham, Massachusetts). The second experimental group employed a novel low radiodensity needle holder (Fig. 1 ), designed to facilitate PCNL access while ensuring that the surgeon’s hand is not directly in the line of the radiation beam. Each treatment arm underwent five trials, with each trial having a fluoroscopy time of 300 s. This duration was chosen based on previous studies that reported the average fluoroscopy time during renal access for PCNL [ 15 – 18 ]. The OSLD chips used in the study were read with a microSTARii Dosimetry System (LANDAUER, Glenwood, Illinois). The absorbed dose measurements by OSLD chips were converted to equivalent doses in millisieverts (mSv) using the radiation weighting factor for fluoroscopy specified by the International Commission on Radiological Protection (ICRP) ( w R = 1) [ 19 ]. The radiation attenuating surgical gloves are constructed from a radiopaque proprietary material, which does not contain lead or latex. The interior is coated to allow easy donning. They are specifically designed to minimize the ionizing radiation penetrating the surgeon’s hand [ 20 ]. The needle holder employed in this study is 3D-printed with a 9-inch-long handle and a very low profile on fluoroscopy. The holder was specifically designed for use with a novel needle which includes a hub that securely interfaces with the needle holder for easy maneuverability and the potential for insertion using the handle. For consistency, this same needle was used in each of the three arms to eliminate it as a potential confounding variable. The primary objective was to quantitatively measure and compare the radiation dose to the surgeon’s hand using the three different techniques for obtaining percutaneous renal access. The secondary outcomes were dose received by the patient, as well as the C-arm recorded cumulative radiation dose in mGy, current in mA, and voltage in kVp for each of the three arms. Statistical analysis The data was statistically analyzed using one-way analysis of variance (ANOVA) and Tukey’s B post hoc test using SPSS version 24 (IBM, Armonk, NY). The significance threshold was set at p < 0.05.
Results During the 300 s of fluoroscopy with AEC settings, the surgeon’s hand in the control group received an average equivalent dose of 3.92 mSv (Table 1 ). Compared to the control, both the radiation-attenuating glove (2.48 mSv) and needle holder (1.37 mSv) reduced the average equivalent dose to the surgeon’s hand ( p < 0.001; Table 2 and Fig. 3 ). The needle holder resulted in a significantly lower dose to the surgeon's hand than the radiation-attenuating glove ( p < 0.001). At all locations on the surgeon's hand, both the radiation-attenuating glove and needle holder had lower doses compared to the control ( p < 0.05 for all). The needle holder demonstrated a reduced dose that was significant at all locations except the middle finger ( p = 0.082). Within the control group, the hypothenar eminence received the highest dose (5.64 mSv), while the forearm received the lowest (2.10 mSv; p < 0.001; Fig. 3 ). The doses for the middle finger (4.19 mSv) and the thumb (3.74 mSv) were similar ( p = 0.207). In the radiation-attenuating glove condition, only the hypothenar eminence received a significantly higher dose compared to the other locations (Table 1 ). In the needle holder condition, the middle finger received the highest dose compared to the first digit and forearm, but it was not significantly greater than the hypothenar eminence (Table 1 ). Compared to the control (8.17 mSv), the mean equivalent dose to the dorsal surface of the patient was greater when using a radiation-attenuating glove (8.43 mSv) and less when using the needle holder (7.03 mSv), though these differences were not significant ( p > 0.05 for all; Table 1 and Fig. 4 ). However, the radiation-attenuating glove resulted in a significant increase in equivalent dose to the dorsal surface of the patient compared to the needle holder (8.43 vs 7.03 mSv; p = 0.027). No significant differences were found in the equivalent dose to the ventral surface of the patient between surgeon hand conditions ( p > 0.05 for all). Overall, the ventral surface had a significantly higher mean equivalent dose than the dorsal surface of the patient for the control (874.48 vs 8.17 mSv), radiation-attenuating glove (676.24 vs 8.43 mSv), and needle holder (632.75 vs 7.03 mSv) conditions ( p < 0.001 for all). Using a radiation-attenuating glove resulted in a significant increase in the radiation generated by the fluoroscopy machine compared to using a needle holder (83.49 vs 69.22 mGy; p = 0.019; Fig. 5 ). Compared to the control (79.00 mGy), the radiation produced was higher with a radiation-attenuating glove and lower with a needle holder ( p < 0.05 for all; Table 3 ). Although the kVp and mA were higher when using the radiation-attenuating glove and lower when using the needle holder compared to the control, there was no statistically significant difference (Table 3 ).
Discussion The existing literature extensively covers the various impacts of ionizing radiation such as DNA damage, malignancy, cataracts, dermatologic changes, and delayed wound healing [ 2 – 6 , 21 ]. Despite this research evidence, there remains a significant gap in our understanding of the full health effects of chronic low-level radiation exposure [ 7 , 8 ]. One area of specific concern is the potential damage to the hands of surgeons due to prolonged exposure to significant doses of ionizing radiation during procedures involving fluoroscopic-guided imaging. Urologists, who frequently use fluoroscopy to perform essential procedures, are particularly at risk for repeated low-level radiation exposure. Consequently, it becomes imperative to acknowledge the potential harm associated with all levels of ionizing radiation and take every possible protective measure to reduce exposure to as low as reasonably achievable (ALARA) [ 22 ]. In addition to posing a significant risk of joint, vascular, and skin pathologies, any adverse effects on a surgeon’s hand can significantly impact their ability to provide effective care and perform surgical procedures safely. To address these concerns and minimize potential harm, the ICRP guidelines suggest not exceeding 20 mSv of average radiation exposure per year over a 5-year period and an annual limit of 50 mSv. Additionally, the ICRP advises maintaining an annual radiation dose limit of 500 mSv to the skin and extremities [ 19 ]. Our study aimed to investigate the efficacy of radiation-attenuating gloves and a novel needle-holder in reducing ionizing radiation exposure to the surgeon’s hands during urological procedures. Using a cadaver model, we evaluated different techniques to reduce radiation exposure during simulated percutaneous renal access and observed significant exposure reductions to the surgeon’s hand. The use of a radiation-attenuating glove resulted in a 37% reduction, while employing a needle holder led to a 65% reduction in the mean equivalent dose to the surgeon’s hand (Table 1 ). The 37% radiation reduction for the radiation-attenuating glove at 95.2 kVp is comparable to the manufacturer specifications of 33% reduction at 100 kVp [ 20 ]. The decreased radiation dose associated with the radiation-attenuating glove may be attributed to the material’s high attenuation coefficient which reduces penetrating radiation. Similar reductions in exposure have been reported in orthopedics, where radiation-attenuating gloves were found to decrease exposure by 61% in an anthropomorphic model [ 21 ]. However, it is worth noting that their study used a mini-C arm with lower kVp and mA settings, unlike the GE OEC 9900 system at AEC settings used in our research. On the other hand, the needle holder further reduced exposure, likely by enabling the surgeon to remove their hand from the direct radiation beam. Notably, there is currently a lack of studies investigating the radiation-reducing effects of a needle holder during percutaneous renal access. Simulated PCNL access demonstrated a mean radiation dose of 3.92 mSv to the hand without protection. Recent studies have reported similar hand radiation exposure during PCNL of 0.36–4.36 mSv [ 9 , 23 ]. This wide range of hand exposure could be attributed to variations in fluoroscopy use, settings, and surgeon experience. Using the ICRP guidelines, approximately 127 PCNLs may be performed annually using surgical gloves without exceeding the dose limit for extremities. This number increases to 201 with radiation-attenuating gloves and 364 with the needle holder. Custom-made needle holders have been specifically designed for the conventional “bullseye” technique in percutaneous renal access [ 24 , 25 ]. As demonstrated in our study, these needle holders can potentially reduce radiation exposure to both the surgeon's hand and patient when constructed from a low radiation density material. However, literature on specialized needle holders for PCNL access is sparse. In contrast, the practice of utilizing needle holders has been extensively studied and more commonly employed in fields that frequently rely on fluoroscopic-guided imaging, such as interventional radiology. Studies investigating the use of improvised metal and custom-made plastic needle holders during fluoroscopic-guided interventions demonstrated significantly reduced radiation exposure to the user’s hand [ 26 , 27 ]. In the field of endourology, the utilization of specialized needle holders is not yet widespread, requiring further research to determine their efficacy and safety. Considerable research has been dedicated to reducing radiation exposure during fluoroscopic procedures in the operating room [ 1 ]. Implementation of simple yet effective measures, such as appropriate shielding, can lead to a significant reduction of up to 70-fold in radiation exposure [ 1 ]. Similarly, operating the fluoroscopy machine at lower power settings is another effective strategy [ 1 ]. However, adoption of these practices is not universal. A survey among endourologists revealed that lead aprons were worn in 99.3% of cases, thyroid shields in 98.7%, and radiation-attenuating gloves in only 9.7% [ 28 ]. The underuse of radiation-attenuating gloves is most likely multifactorial in nature, potentially due to cost, unacquaintance, inconvenience, or believing further protection is unnecessary due to current occupational dose limit guidelines. For fluoroscopy settings, the AEC setting remains the most commonly used mode due to its ability to obtain optimal quality images [ 29 ]. However, low dose modes and pulsed fluoroscopy are sufficient for many procedures [ 1 ]. The radiation reduction techniques explored in this study had implications not only for the surgeon but also for the patient. While the use of a radiation-attenuating glove reduced radiation exposure to the surgeon's hand by 37%, it resulted in a 3% increase in dose to the patient's dorsal surface (Table 1 ). In contrast, utilizing the needle holder reduced exposure for both the surgeon's hand by 65% and the patient's dorsal surface by 14%. The use of radiation-attenuating gloves may offer protection for the surgeon, but at the cost of increasing the dose to the patient. This is likely due to the effect of introducing hyperdense objects, such as radiation-attenuating gloves, into the path of the fluoroscopy beam. This has been shown to increase the radiation produced by the machine when it is operating in the AEC setting [ 14 ]. On the other hand, the needle holder is made from a low-density material and has a slim contour, which may have allowed the fluoroscopy machine to generate a lower radiation dose. This is supported by the findings regarding radiation dose, kVp, and mAs generated by the fluoroscopy machine in our study, which were found to be higher for the radiation-attenuating glove compared to the needle holder (Table 3 ). Limitations of our study include the use of a cadaver model for simulated percutaneous renal access which cannot entirely replicate all aspects of the working environment encountered in a live PCNL. Nonetheless, this approach provided a controlled testing environment that allowed for accurate comparisons of radiation reduction techniques without resultant undue radiation exposure to human subjects. An additional limitation of our study is that it was designed to compare three different methods of holding the needle during fluoroscopic-guided access and did not include a comparison of ultrasound-guided access. We also used the AEC setting and a preset fluoroscopy time for simulated renal access. While the AEC setting is the most commonly used fluoroscopy setting, low-dose settings and pulsed fluoroscopy are also used in practice [ 1 , 29 ]. The preset fluoroscopy time of 5 min in our study will not be representative of all practices and institutions. Finally, it is important to note that this needle holder and radiation-attenuating glove were only tested in a prone PCNL model, where the surgeon’s hand receives direct radiation exposure. In triangulation and supine PCNL the surgeon’s hand is less likely to encounter direct radiation exposure, and subsequently, these were not tested in our model. Despite these limitations, to our knowledge, this is the first study to assess and compare hand radiation reduction techniques during percutaneous renal access for PCNL in a controlled cadaver model.
Conclusion Protective measures during percutaneous renal access for PCNL can effectively reduce radiation exposure to a surgeon’s hand. Radiation-attenuating gloves showed a significant reduction in hand exposure but increased patient dose, while needle holders resulted in decreased exposure for both the surgeon and the patient. However, further research is needed to determine the efficacy and safety of radiation-attenuating gloves and needle holders in clinical urologic practice. These findings emphasize the importance of implementing radiation reduction techniques to enhance occupational safety during PCNL access.
Percutaneous nephrolithotomy confers the highest radiation to the urologist’s hands compared to other urologic procedures. This study compares radiation exposure to the surgeon’s hand and patient’s body when utilizing three different techniques for needle insertion during renal access. Simulated percutaneous renal access was performed using a cadaveric patient and separate cadaveric forearm representing the surgeon’s hand. Three different needle-holding techniques were compared: conventional glove (control), a radiation-attenuating glove, and a novel needle holder. Five 300-s fluoroscopy trials were performed per treatment arm. The primary outcome was radiation dose (mSv) to the surgeon’s hand. The secondary outcome was radiation dose to the patient. One-way ANOVA and Tukey’s B post-hoc tests were performed with p < 0.05 considered significant. Compared to the control (3.92 mSv), both the radiation-attenuating glove (2.48 mSv) and the needle holder (1.37 mSv) reduced hand radiation exposure ( p < 0.001). The needle holder reduced hand radiation compared to the radiation-attenuating glove ( p < 0.001). The radiation-attenuating glove resulted in greater radiation produced by the C-arm compared to the needle holder (83.49 vs 69.22 mGy; p = 0.019). Patient radiation exposure was significantly higher with the radiation-attenuating glove compared to the needle holder (8.43 vs 7.03 mSv; p = 0.027). Though radiation-attenuating gloves decreased hand radiation dose by 37%, this came at the price of a 3% increase in patient exposure. In contrast, the needle holder reduced exposure to both the surgeon’s hand by 65% and the patient by 14%. Thus, a well-designed low-density needle holder could optimize radiation safety for both surgeon and patient. Keywords Open access funding provided by SCELC, Statewide California Electronic Library Consortium
Author contributions RC: Methodology, Formal Analysis, Investigation, Writing—Original Draft, Writing—Review and Editing. EJ: Methodology, Formal Analysis, Investigation, Writing—Original Draft, Project Administration. CB: Methodology, Investigation. JH: Investigation. ASA: Conceptualization, Methodology, Formal Analysis, Resources, Investigation, Writing—Review and Editing. KS: Writing—Review and Editing. JDB: Conceptualization, Methodology, Resources, Investigation. CR: Resources, Investigation. EAB: Investigation. ZO: Writing—Review and Editing. AF: Writing—Review and Editing. DDB: Conceptualization, Methodology, Validation, Resources, Investigation, Writing—Review and Editing, Supervision. Funding Open access funding provided by SCELC, Statewide California Electronic Library Consortium. No funding was received for this study. Declarations Conflict of interest D. D. B. is currently developing the needle and needler holder utilized in this study. No other authors have anything to disclose.
CC BY
no
2024-01-15 23:42:02
Urolithiasis. 2024 Jan 13; 52(1):27
oa_package/3b/6b/PMC10787896.tar.gz
PMC10787897
38217735
Introduction The genome integrity of all living organisms is constantly threatened by endogenous and exogenous sources of DNA damage (Lindahl 1993 ). Endogenous sources include oxidation by reactive oxygen species (ROS), alkylation, mismatch of DNA bases, topoisomerase-DNA complexes, spontaneous base deamination and abasic (apurinic/apyrimidinic, AP) sites. Important exogenous sources are ionizing radiation (IR), ultraviolet (UV) radiation and exposure to chemical agents with genotoxic capacity (Chatterjee and Walker 2017 ). To cope with such variety of DNA insults, cells have developed a robust DNA damage response (DDR) that activates several DNA repair pathways and DNA damage checkpoints. In addition, certain types of DNA lesions are substrates for DNA damage tolerance pathways (Chatterjee and Walker 2017 ). The major DNA repair pathways comprise direct reversal of DNA damage, base excision repair (BER), single-strand break repair (SSBR), nucleotide excision repair (NER), mismatch repair (MMR) and double-strand break repair (DSBR) by homologous recombination (HR) or non-homologous end joining (NHEJ). Amongst such mechanisms, BER is required for the repair of a broad range of modified DNA bases (alkylated, oxidized or deaminated) and AP sites (Krokan and Bjoras 2013 ). BER is initiated by DNA glycosylases that remove the damaged or modified bases generating AP sites, that are repaired either by AP endonucleases or by the AP lyase activity associated with bifunctional DNA glycosylases (Demple and Harrison 1994 ; Krokan and Bjoras 2013 ; Seeberg et al. 2000 ). AP endonucleases hydrolyze DNA at the 5′-side of the AP site, leaving 3′-hydroxyl (3′-OH) and 5′-deoxyribose phosphate (5′-dRP) termini (Levin and Demple 1990 ). AP lyases cleave 3′ to the AP site by β-elimination, generating 3′-phospho-α, β-unsaturated aldehyde (3′-PUA) and 5′-phosphate (5′-P) termini. A subset of AP lyases catalyze β-δ-elimination and generate 3′-phosphate (3′-P) termini (Levin and Demple 1990 ). Therefore, AP endonucleases and AP lyases generate single-nucleotide gaps with 5′- and 3′-blocked ends, respectively. Once the blocked termini have been processed to canonical 5′-P and 3′-OH ends, gap filling may proceed by insertion of either one nucleotide (short-patch BER, SP-BER) or several nucleotides (long-patch BER, LP-BER) (Cordoba-Cañero et al. 2009 ; Fortini and Dogliotti 2007 ). In mammals, DNA polymerase β is involved in gap filling during SP-BER (Srivastava et al. 1998 ), whereas LP-BER requires replicative DNA polymerases Pol δ and Pol ε (Levin et al. 1997 ). The last BER step is DNA ligation that, in mammals, is carried out by the complex of XRCC1 and LigIIIα during SP-BER (Nash et al. 1997 ). In LP-BER Pol δ/ε displace the strand containing the 5′-dRP terminus generating a flap structure that is processed by the flap endonuclease (FEN1), creating a nick that is sealed by LIG1 (Levin et al. 1997 ). Since AP endonucleases generate a 5′-dRP blocked end, SP-BER requires a 5′-dRP lyase activity to produce 5′-P ends amenable to DNA ligation. In mammals, the mayor dRP lyase activity is intrinsic to Pol β (Srivastava et al. 1998 ), through an N-terminal 8-kDa domain characteristic of X family of DNA polymerases (Beard and Wilson 2000 ). The members of the mammalian X family of DNA polymerases are Pol β, Pol λ, Pol μ and terminal deoxyribonucleotidyl transferase (TdT), and they are mainly involved in gap filling during DNA repair (Beard and Wilson 2000 ). The X family proteins are distributive polymerases involved in synthesis of short segments of DNA. All these enzymes are composed of a DNA polymerization domain at the C-terminus, a DNA binding domain in the central region of the protein and, except for Pol β, a BRCT (BRCA1 C-terminal) domain at the N-terminus (Uchiyama et al. 2009 ). Both mammalian Pol β and Pol λ display DNA polymerase and 5′-dRP lyase activity (Garcia-Diaz et al. 2001 ; Matsumoto and Kim 1995 ), which is needed for SP-BER (Braithwaite et al. 2010 ). Pol λ, Pol μ and TdT, through their BRCT domain, are involved in NHEJ to repair double-strand DNA breaks generated by DNA damage, and/or V(D)J recombination to create diversity in the immunoglobulins (Fan and Wu 2004 ; Lee et al. 2004 ; Mahajan et al. 2002 ; Nick McElhinny et al. 2005 ). Plants possess homologs of most BER genes found in other organisms, but there are exceptions. For instance, Pol β and Lig III, involved in SP-BER in mammals, are absent in plants. On the other hand, some BER proteins are only found in plants, indicating that there are some plant-specific BER features that have appeared during evolution (Roldan-Arjona et al. 2019 ). Since there are no plant homologs of Pol β and LigIII, it was initially believed that plants, unlike mammals, do not use SP-BER to repair damaged bases (Uchiyama et al. 2008 ). However, it has been reported that Arabidopsis cell extracts catalyze DNA repair of uracil and AP sites by both LP- and SP-BER (Cordoba-Cañero et al. 2009 ). However, the identity of the DNA polymerase(s) involved in plant BER is still unknown. The only member of the X family of DNA polymerases present in higher plants is Pol λ (Uchiyama et al. 2004 ). By contrast, in the unicellular alga Chlamydomonas reinhardtii , in addition to Pol λ, sequences with similarity to X family members Pol μ and TdT have been identified (Morales-Ruiz et al. 2018 ) . The possible role of plant Pol λ in DNA repair has been studied both in rice and Arabidopsis (Amoroso et al. 2011 ; Furukawa et al. 2015 ; Garcia-Diaz et al. 2000 ; Roy et al. 2011 ; Uchiyama et al. 2004 ). In Arabidopsis it has been reported that, as its mammalian homolog, Pol λ performs error-free translesion synthesis past 8-oxoguanine (8-oxoG) (Amoroso et al. 2011 ; Maga et al. 2008 ). Both Arabidopsis and rice Pol λ have been described to have a highly conserved PCNA-interacting protein-box (PIP box) motif. In Arabidopsis the PIP box is involved in mediating the interaction of Pol λ with PCNA2 (proliferating cell nuclear antigen 2) to enhance the efficiency and fidelity of translesion synthesis (Amoroso et al. 2011 ). Moreover, it has been shown that Arabidopsis Pol λ plays a role in NER of UV-B induced DNA damage (Roy et al. 2011 ). It has been suggested that Arabidopsis Pol λ also participates in double-strand break repair. Specifically, an Arabidopsis Pol λ T-DNA insertion mutant showed sensitivity to both gamma-irradiation and treatment with radiomimetic agents, such as bleomycin, but not to others DNA damaging treatments including methyl methanesulfonate (MMS) and mitomycin-c (MMC) (Furukawa et al. 2015 ). All these reports suggest that Pol λ is an important component of the plant DNA damage response. However, our knowledge about the role of plant Pol λ in specific DNA repair processes, such as BER, is still very limited. The biochemical characterization of rice Pol λ indicates that it displays dRP lyase activity (Uchiyama et al. 2004 ). Nonetheless, although some biochemical properties of Arabidopsis Pol λ have been described, there is no evidence reported of its dRP lyase activity (Amoroso et al. 2011 ; Roy et al. 2011 ). In this work, we have biochemically characterized the enzymatic activity of Arabidopsis Pol λ and, importantly, we have found two residues, K248 and K255, within the 8-kDa domain characteristic of the X family of DNA polymerases, that are needed to repair 5′-dRP blocked ends. Our results show that Arabidopsis Pol λ displays both DNA polymerization and dRP lyase activities on DNA substrates mimicking DNA repair intermediates and suggest it could play an important role in plant BER.
Material and methods DNA substrates Oligonucleotides used (Supplementary Table S1 ) were synthesized by Integrated DNA Technologies (IDT) and purified by PAGE or dual HPLC before use. Double-stranded DNA substrates were prepared by mixing a 5 μM solution of a fluorescein (Fl) or alexa (Al) labelled oligonucleotide with a 10 μM solution of an unlabelled complementary oligonucleotide. DNA substrates with one nucleotide gap with a 3′-OH end were prepared by mixing a 5 μM solution of a 5′-alexa (Al) labelled oligonucleotide (Al_28-OH) with a 10 μM solution of both unlabelled oligonucleotide, corresponding to the complementary strand (CGR_G, CGR_A, CGR_T or CGR_C) and to the oligonucleotide containing a 5′-P, 5′-OH or 5′-THF terminus (P_30-51, OH_30-51 or THF_30-51, respectively). Annealing was carried out by heating at 95 °C for 5 min followed by slowly cooling to room temperature. DNA substrates with a 5′-dRP end were generated by incubating an oligonucleotide duplex containing a U:G mispair with 0.5 U of Escherichia coli uracil DNA glycosylase (UDG) (New England Biolabs) and 10 U of human AP endonuclease 1 (APE-1) (New England Biolabs, NEB). Protein expression and purification The full-length Arabidopsis thaliana Pol λ (AtPol λ) cDNA, obtained from the Arabidopsis Biological Resource Center (pENTR_221-At1G10520), was subcloned into pET28a expression vector (Novagen) to add a polyhistidine (His 6 ) tag at the N-terminus of AtPol λ protein. Expression was carried out in E. coli BL21 (DE3) dcm − Codon Plus cells (Stratagene) by adding 1 mM isopropyl-1-thio-β-D-galactopyranoside. His-AtPol λ was purified by affinity chromatography on a Ni 2+ -Sepharose column (HisTrap HP; GE Healthcare). Protein was eluted with a 5 mM to 1 M gradient of imidazole and analyzed by SDS/PAGE (10%) using broad-range molecular weight standards (Bio-Rad). Other enzymes and reagents Escherichia coli UDG and human APE1 were obtained from NEB. Human Pol β was purchased from Trevigen, T4 DNA ligase was obtained from Promega and Taq DNA Polymerase from Bioline. Anti-AtPol λ antibodies were generated by injecting rabbits with His-AtPol λ. Site-directed mutagenesis Site-directed mutagenesis was performed using the Quick-Change II XL kit (Stratagene). The mutations were introduced into the expression vector pET28a (Novagen) containing the full-length wild-type (WT) AtPol λ cDNA using specific oligonucleotides (Supplementary Table S2). Mutational changes were confirmed by DNA sequencing and the constructs were used to transform Escherichia coli BL21 (DE3) dcm − Codon Plus cells (Stratagene). Mutant proteins were expressed and purified as describe above. Gap-filling assay Reactions (10 μl) contained 50 mM Tris HCl pH 7.5, 1 mM DTT, 0.2 mg/ml BSA, 2% glycerol, 10 mM MgCl 2 , 100 nM DNA substrate, the indicated amount of AtPol λ and dCTP, dNTPs (deoxynucleotides) or ddNTPs (dideoxynucleotides). When indicated, AtPol λ was pre-incubated with pre-immune serum or anti-AtPol λ antibody for one hour at 4 °C. After incubation at 37 °C for the indicated times, reactions were stopped by adding 20 mM EDTA, 0.6% SDS and 0.5 mg/ml proteinase K, and mixtures were incubated at 37 °C for 30 min. DNA was extracted with phenol:chloroform:isoamyl alcohol (25:24:1) and ethanol precipitated at -20 °C in the presence of 0.3 mM NaCl and 16 mg/ml glycogen. Samples were resuspended in 10 μl of 90% formamide and heated at 95 °C for 5 min. Reaction products were separated in a 12% denaturing polyacrylamide gel containing 7 M urea. Labelled DNA was visualized using FLA-5100 imager (Fujifilm) and analyzed using Multi Gauge software version 3.0 (Fujifilm). dRP lyase assay Reactions (50 μl) contained 45 mM HEPES–KOH pH 7.8, 70 mM KCl, 5 mM MgCl 2, 1 mM DTT, 0.4 mM EDTA, 2 mM ATP, 36 μg BSA, 1 mM NAD, 0.2% glycerol, 22 mM phosphocreatine, 2.5 ng creatine kinase, 0.02 mM dCTP, and a DNA substrate with a 5′-dRP (100 nM) prepared as described above. When indicated, human Pol β (2.4 U) or AtPol λ (10 nM), were added. After incubation at 30 °C for 90 min, reaction products were stabilized by the addition of freshly prepared sodium borohydride (NaBH 4 ; Sigma-Aldrich) to a final concentration of 300 mM and incubated on ice for 30 min. Subsequently, the samples were desalted using microspin G-25 columns (GE Healthcare). The reactions were stopped, DNA was extracted, and samples were processed as described above. In vitro reconstitution of base excision repair BER reactions (50 μl) were performed in 45 mM HEPES–KOH pH 7.8, 70 mM KCl, 5 mM MgCl 2 , 1 mM DTT, 0.4 mM EDTA, 2 mM ATP, 36 μg BSA, 1 mM NAD, 0.2% glycerol, 22 mM phosphocreatine, 2.5 ng creatine kinase, 0.02 mM dCTP, and a DNA duplex containing a U:G mispair (100 nM). All reactions contained UDG (0.5 U) and APE1 (10 U). When indicated, human Pol β (2.4 U), AtPol λ (10 nM), and/or T4 DNA ligase (1.5 U), were added. Mixtures were incubated at 30 °C for 90 min, and the reactions were stopped, DNA was extracted, and samples were processed as described above.
Results Characterization of AtPol λ gap filling activity In order to characterize the catalytic activity of Arabidopsis thaliana Pol λ, a recombinant protein (AtPol λ) was expressed in E. coli and purified. Subsequently, optimal conditions for detecting polymerase activity were determined by performing polymerization assays with varying enzyme concentrations. A DNA duplex with the upper strand labelled at the 5′-end and containing a single nucleotide gap flanked by 3′-OH and 5′-P ends was used as substrate (Fig. 1 a). AtPol λ already showed DNA polymerase activity in a 10 min reaction at a 2 nM concentration of recombinant protein (Fig. 1 b, lane 6). To confirm that the polymerase activity detected was intrinsic to AtPol λ and not due to contamination with bacterial polymerases, further polymerization assays were carried out in which AtPol λ was pre-incubated with serum generated against AtPol λ (anti-AtPol λ) (Fig. 1 c). We found a complete inhibition of the polymerase gap-filling capacity in the presence of the anti-AtPol λ serum (Fig. 1 c, lane 4), suggesting that the DNA polymerase activity is intrinsic to AtPol λ. To determine whether AtPol λ is able to insert the correct nucleotide in a gapped DNA mimicking a BER intermediate, we performed polymerization assays using a DNA substrate containing a single-nucleotide gap opposite G (Fig. 1 a) in the presence of dATP, dCTP, dGTP, dTTP or a mixture of all four dNTPs (Fig. 1 d). AtPol λ only incorporated a nucleotide opposite G when dCTP or dNTPs were present in the reaction (Fig. 1 d, lanes 3 and 6). We also examined how the structure of the deoxyribose affects nucleotide incorporation by AtPol λ by analyzing the effect of dideoxynucleotides (ddNTPs) on its DNA polymerase activity. DNA polymerization is blocked during replication when a ddNTP is incorporated due to the lack of a canonical 3′-OH. While replicative DNA polymerases are resistant to ddNTP inhibition because they fail to bind the nucleotide analogue, human Pol β is ddNTP-sensitive because its active site can accommodate ddNTPs and incorporate them as well as dNTPs (Cavanaugh et al. 2010 ). Previous studies have shown that human and O. sativa Pol λ are also sensitive to ddNTPs and therefore unable to discriminate between dNTPs and ddNTPs (Garcia-Diaz et al. 2002 ; Uchiyama et al. 2004 ). Additionally, we have previously shown that BER catalyzed by Arabidopsis whole cell extracts is partially sensitive to ddNTPs (Cordoba-Cañero et al. 2009 ). To investigate whether AtPol λ is affected by ddNTPs, a polymerization assay was performed using a primer-template DNA substrate in which dGMP insertion requires the previous incorporation of dCMP (Fig. 2 ). In the absence of ddCTP, AtPol λ generated products of 29 and 30 nucleotides, corresponding to the insertion of dCMP and dGMP, respectively (Fig. 2 , lane 3). However, in the presence of increasing ddCTP:dCTP ratios, a near complete inhibition of AtPol λ activity was observed (Fig. 2 , lanes 4–8). A strong inhibition was already detected with a 1:1 ddCTP:dCTP ratio (Fig. 2 , lane 4). Therefore, AtPol λ exhibits high sensitivity to ddNTPs, in agreement with the weak discrimination of the 3′-OH group of the nucleotide to be inserted that is displayed by the X family of DNA polymerases (Banos et al. 2008 ; Garcia-Diaz et al. 2002 ; Prasad et al. 1993 ). Effect of the template base on gap filling activity of AtPol λ To further analyze the DNA polymerase activity of AtPol λ, we performed a time-course DNA polymerization assay using DNA duplexes with a gap flanked by 3′-OH and 5′-P ends containing different bases (A, T, C, or G) on the complementary strand (Fig. 3 a). As expected, AtPol λ performed gap-filling in all four substrates, although the efficiency of DNA synthesis was somewhat different depending on the template base (Fig. 3 a-b). We found that AtPol λ displays a significant preference for C as template in comparison with A (Fig. 3 b; t-test p < 0,05 and < 0,01 at 20 and 40 min, respectively). In the DNA substrate containing C as template, the presence of another C downstream in the complementary strand allowed limited strand displacement accompanied of a second dGMP insertion (Fig. 3 a, lanes 12–13). Some DNA polymerases involved in BER are capable of bypassing lesions on the template strand, resulting in both error-prone and error-free DNA synthesis (Krokan and Bjoras 2013 ). The ubiquitous oxidative DNA lesion 8-oxoG is highly mutagenic, as it can pair with both cytosine and adenine, leading to G:C → T:A transversions after replication (Krokan and Bjoras 2013 ). To analyze the translesion DNA synthesis capacity of AtPol λ, we performed a gap-filling assay with a DNA substrate containing 8-oxoG opposite the gap in the presence of either dCTP or dATP (Fig. 3 c). The results indicate that AtPol λ can effectively bypass 8-oxoG by incorporating dCMP or dAMP with similar efficiencies (Fig. 3 c). Effect of the 5′-end group on the gap-filling activity of AtPol λ Human Pol λ and Pol β possess high affinity for DNA substrates containing gaps flanked by 3′-OH and 5′-P ends. It has been suggested that this preference is due to the interaction between the 8-kDa domain of the protein and the phosphate group at the 5′-end of the gap (Garcia-Diaz et al. 2002 ; Singhal and Wilson 1993 ). In mammalian SP-BER initiated by AP endonucleases, the processing of the 5′-dRP group to generate a 5′-P end is a limiting step (Srivastava et al. 1998 ) and it has been reported that the presence of a 5′-dRP group moderately decreases the DNA polymerization activity of human Pol λ (Duym et al. 2006 ). To analyze the effect of the 5′-end group on the gap-filling activity of AtPol λ, we carried out DNA polymerization assays on gaps with a 5′-phosphate (5′-P), a 5′-hydroxyl (5′-OH) or a tetrahydrofuran residue (5′-THF) mimicking a 5′-dRP, which is resistant to dRP lyase activity. Time-course reactions with each substrate were performed in the presence of different dCTP concentrations (Fig. 4 a–c). When a canonical 5′-P end was used (Fig. 4 a), AtPol λ incorporated a nucleotide at the shortest tested time (5 min) and at the higher dCTP concentrations of 100 and 10 μM (Fig. 4 a, lanes 2 and 6), while at a lower dCTP concentrations, longer times were needed for the gap filling to occur with the same efficiency (Fig. 4 a, lanes 11 and 17). However, in the presence of DNA substrates containing a 5′-OH end (Fig. 4 b), AtPol λ required a longer time (20 min) for gap filling at 100 μM of dCTP (Fig. 4 b, lane 4). Moreover, at lower dCTP concentrations (10, 5 and 1 μM) the percentage of product was around 80%, 60% and 20%, respectively, at the longest tested time (Fig. 4 b, lanes 9, 13 and 17). When AtPol λ was incubated with the substrate containing a 5′-THF end and the lowest dCTP concentration (1 μM) the polymerase was able to fill the gap as well as with the 5′-P end (Fig. 4 c, lane 17), and the minor differences were not statistically significant. At higher dCTP concentrations (100, 10 and 5 μM), the polymerase performed gap filling generating a product of 29 nt in 5, 10 and 20 min, respectively (Fig. 4 c, lanes 2, 7 and 12). Altogether, these results suggest that the gap-filling capacity of AtPol λ is highest on a gap with a 5′-P end, and strongly inhibited by a 5′-OH end (Fig. 4 d). Importantly, the presence of a THF group at the 5′-side of the gap did not have any inhibitory effect on the enzyme efficiency, suggesting that AtPol λ can effectively fill DNA repair gaps with a 5′-dRP. AtPol λ possesses an intrinsic dRP lyase activity A limiting step of SP-BER is the removal of the 5′-dRP generated after the processing of abasic sites by AP endonucleases. In humans, the removal of this end is mainly carried out by the dRP lyase activity of polymerase β (Beard and Wilson 2000 ). No homologs of Pol β have been described in plants and AtPol λ is the only member of the X family described so far (Uchiyama et al. 2009 ), suggesting that this polymerase might play a role in SP-BER. The 8-kDa domain present in all members of this family is responsible for the dRP lyase activity, which is catalyzed through β-elimination via a Schiff base intermediate between a nucleophile lysine residue and DNA (Matsumoto and Kim 1995 ). Specific lysine residues responsible for this activity have been identified in human Pol β (K72) (Deterding et al. 2000 ) and Pol λ (K312) (Garcia-Diaz et al. 2001 ). In order to test whether AtPol λ has intrinsic dRP lyase activity, we first identified candidate nucleophile residues in its 8-kDa domain. By multiple sequence alignment we found that AtPol λ has a histidine (H260) at the orthologous position of human Pol β K72 and Pol λ K312 (Supplementary Fig. S1 ). We therefore focused on two AtPol λ lysine residues located nearby. One of such residues (K255) is conserved in all Pol λ orthologs, but not in Pol β enzymes, whereas the other one (K248) is only conserved in plant Pol λ proteins (Supplementary Fig. S1 ). To examine the role of K248 and K255, two mutant derivatives (AtPol λ K248A and AtPol λ K255A) were expressed and purified as recombinant His tagged proteins. Subsequently, dRP lyase activity assays were performed using AtPol λ and its mutant versions (Fig. 5 a). The removal of the 5′-dRP group generates a fragment with a 5′-P end that exhibits greater electrophoretic mobility. Although the non-processed 5′dRP end was stabilized by treatment with borohydride, spontaneous appearance of 5′-P was observed in the absence of any enzyme (Fig. 5 b, lane 1) and in the presence of a DNA polymerase without dRP lyase activity (Fig. 5 b, lane 2), suggesting that some spontaneous 5′-dRP hydrolysis takes place during the incubation time. We found that AtPol λ exhibited a clearly detectable dRP lyase activity (Fig. 5 b, lane 6) that was significantly reduced in the mutant versions K248A and, particularly, K255A (Fig. 5 b-c). These results indicate that AtPol λ possesses an intrinsic dRP lyase activity and suggest that both K248 and K255 are involved in this enzymatic function. Next, we asked whether mutations in K248 or K255 have any effect on the DNA polymerase activity of AtPol λ. For this purpose, a polymerization assay was carried out using a DNA substrate containing a gap flanked by 3′-OH and 5′-P ends (Fig. 6 ). As compared with Klenow DNA polymerase, which achieved full-length DNA synthesis by displacement of the top strand (Fig. 6 , lane 9), AtPol λ exhibited a limited capacity for strand displacement inserting only between 8 and 10 nucleotides (Fig. 6 , lane 5). AtPol λ did not exhibited 3′-5′ exonuclease activity when the incubations were carried out in the absence of dNTPs (Fig. 6 , lane 4). In comparison, mutant AtPol λ K248 showed a DNA polymerization activity similar to that of the WT protein, but a significantly lower capacity for strand displacement (Fig. 6 , lane 3). Also, it exhibited a detectable 3′-5′ exonuclease activity in the absence of dNTPs (Fig. 6 , lane 2). The K255 mutant version of AtPol λ displayed a stronger reduction in strand displacement capacity (Fig. 6 , lane 7), but did not show detectable 3′-5′ exonuclease activity (Fig. 6 , lane 6). These results suggest that residues K248 and K255 are not required for DNA polymerase activity but modulate the capacity of AtPol λ to perform strand displacement during gap-filling. AtPol λ 5′-dRP lyase activity is required for efficient completion of SP-BER in vitro The dRP lyase activity of human Pol β, which removes the blocking dRP group at the 5′-end of the gap, is required to allow repair completion during SP-BER (Srivastava et al. 1998 ). To determine whether AtPol λ also carries out this function, an in vitro SP-BER reconstitution assay was performed using recombinant proteins and a DNA substrate containing a U:G mispair (Fig. 7 ). In this assay, uracil is removed by UDG leaving an intact AP site that is cleaved by APE1 at the 5′-side generating a gap flanked by 3′-OH and 5′-dRP ends. After nucleotide insertion by a DNA polymerase, a dRP lyase activity is required before DNA ligation completes repair (Fig. 7 a). The results show that reactions containing both DNA ligase and AtPol λ, but not those with DNA ligase alone, generated a fully repaired product detected as a 51-nucleotide fragment (Fig. 7 b, lanes 1 and 4). Although lower, repair levels achieved with AtPol λ were close to those observed with Pol β (Fig. 7 c). Importantly, we found that the levels of fully repaired products were significantly lower in reactions containing the AtPol λ K255A mutant protein (Fig. 7 c). Taken together, our results suggest that AtPol λ can support SP-BER, and that its dRP lyase activity is required for efficient completion of repair.
Discussion In this work we have performed a biochemical characterization of A. thaliana DNA polymerase λ and explored its possible role in BER. Firstly, we have shown that, as the rice and human orthologs (Garcia-Diaz et al. 2002 ; Uchiyama et al. 2004 ), AtPol λ inserts the correct nucleotide in a single-nucleotide gap. We have also shown that AtPol λ is able to catalyze DNA synthesis at the lowest dNTP concentration tested (1 μM), which is consistent with the high affinity for dNTPs observed in human Pol λ (Garcia-Diaz et al. 2002 ). We also showed that AtPol λ is inhibited by ddCTP, even at low ddCTP: dCTP ratios. This result suggests a low discrimination for the presence of the 3′-OH group in the sugar of the incoming deoxynucleotide, as it has been found in other X family DNA polymerases (Banos et al. 2008 ; Garcia-Diaz et al. 2002 ; Prasad et al. 1993 ). We have found that, like human Pol β and Pol λ, AtPol λ does not possess an intrinsic 3′-5′ exonuclease activity that might remove a nucleotide with the wrong sugar. The ddNTP resistance of high-fidelity DNA polymerases is not due to their 3′-5′-exonuclease activity, but to an exquisite structural selectivity that prevents productive binding to deoxynucleotides lacking hydroxyl at C3′ (Wang et al. 2012 ). Such selectivity is absent in ddNTP-sensitive DNA polymerases. For Pol β and other members of the X family, it has been proposed that a conserved Arg residue (R183 in human Pol β and R420 in human Pol λ) is sufficient to stabilize the incoming nucleotide in the absence of O3′ (Cavanaugh et al. 2010 ). Interestingly, this residue is also conserved in AtPol λ (R370), which suggests that the structural basis for low deoxynucleotide discrimination is conserved across Pol β and Pol λ proteins. Interestingly, we found an effect of the base opposite the gap on the polymerization efficiency of AtPol λ. The rate of nucleotide incorporation catalyzed by the enzyme was higher in substrates containing a C in the template strand, as compared to G, T or A. Such preference might be related to the type of initial DNA damage that leads the generation of the repair gap. A cytosine opposite the gap is found in DNA intermediates arising during repair of 8-oxoG. This suggests that AtPol λ could participate, preferably but not exclusively, in the repair of certain DNA lesions. In human cells, Pol β has been implicated in 8-oxoG BER initiated by OGG1 DNA glycosylase (Dianov et al. 1998 ; Fortini et al. 1999 ). Plants possess orthologs of OGG1 and FPG, both of which are involved in 8-oxoG repair (Cordoba-Cañero et al. 2014 ). Therefore, it will be interesting to analyse whether AtPol λ functions in DNA repair of 8-oxoG initiated by OGG1 and/or FPG. Besides DNA repair functions, Pol λ enzymes are thought to play a role in DNA damage tolerance through DNA translesion synthesis. Thus, mammalian and Arabidopsis Pol λ have been reported to perform both error-free and error-prone bypass of 8-oxoG lesions (Amoroso et al. 2011 ; Picher and Blanco 2007 ). Here, we have confirmed that AtPol λ can catalyze both the correct incorporation of C and misincorporation of A opposite 8-oxoG. In our analysis, we have used a one-nucleotide gap mimicking a DNA intermediate generated during BER of 8-oxoG:A mispairs. In human cells such mispairs are targets of MUTYH DNA glycosylase, which specifically excises the misincorporated A and leaves 8-oxoG in the repair-template strand (Slupska et al. 1996 ). Since Arabidopsis possesses a MUTYH ortholog (Roldan-Arjona et al. 2019 ), the capacity of AtPol λ to misincorporate A opposite 8-oxoG must be somehow reduced in vivo in order to avoid futile cycles of 8-oxoG:A repair. In fact, it has been reported that PCNA2 increases AtPol λ fidelity in translesion DNA synthesis (Amoroso et al. 2011 ). It has been previously shown that human Pol λ is a distributive DNA polymerase on a template-primer substrate but processive in short gaps containing a phosphate group at its 5′-end (Garcia-Diaz et al. 2002 ). This is believed to be a consequence of the additional contacts that are made between the N-terminal 8-kDa domain of the protein and the 5′-end of the downstream strand in the gapped substrate (Garcia-Diaz et al. 2004 ). Here we have shown that AtPol λ is able to perform gap-filling in the presence of different groups at the 5′-end of the gap but shows the highest polymerization efficiency when the 5′-end contains a phosphate group. This finding suggests that plant Pol λ is not specialized in DNA substrates with a canonical 5′-P end and might fulfils diverse functions by being able to process substrates with different 5′-termini. We found a much-reduced, but still significant, gap-filling activity in gapped substrates containing an OH group at the 5′-terminus of the downstream strand. Such 5′-OH blocked ends are found at DNA breaks induced by abortive activity of topoisomerase I (TOP1) which are generated by TOP1 inhibitors, reactive oxygen species, and other genotoxins (Caldecott 2008 ; Pommier et al. 2003 ). TOP1 cleavage complexes are repaired by tyrosyl-DNA phosphodiesterase 1 (TDP1), that removes topoisomerase I from the 3′-termini at the gap, generating a gap with 3′-P and 5′-OH ends that are converted to 3′-OH and 5′-P by the DNA phosphatase and kinase activities of PNKP, respectively (Caldecott 2008 ; Jilani et al. 1999 ; Pouliot et al. 2001 ). In mammals, the repair is completed in most cases by short patch gap filling mediated by Pol β (Krokan and Bjoras 2013 ; Kubota et al. 1996 ; Srivastava et al. 1998 ). The Arabidopsis ortholog of PNKP lacks kinase activity (Martinez-Macias et al. 2012 ; Petrucco et al. 2002 ), and the mechanism responsible for processing 5′-OH ends in gapped DNA repair intermediates in plants remains unidentified. It is also unknown if AtPol λ plays a role in the processing of TOP1 cleavage complexes. Interestingly, we found that the nucleotide insertion efficiency of AtPol λ when the gap contains a 5′-dRP-mimicking group (THF), is comparable to that observed with a 5′-P end. BER DNA intermediates frequently harbour 5′-dRP blocking termini that arise upon cleavage of abasic sites by APE1 (Krokan and Bjoras 2013 ; Levin and Demple 1990 ) and are processed by the dRP lyase activity of Pol β (Matsumoto and Kim 1995 ; Srivastava et al. 1998 ). However, it has been reported that the dRP lyase activity of Pol β is rate-limiting during BER, and that its DNA polymerase activity performs gap-filling prior to removal of the dRP group (Srivastava et al. 1998 ). Our results suggest that AtPol λ may perform efficient DNA synthesis in gapped DNA repair intermediates before elimination of the 5′-dRP moiety. Like Pol β, human Pol λ possesses an 8-kDa domain responsible for dRP lyase activity (Garcia-Diaz et al. 2000 ). The human Pol λ K312 residue, which is structurally homologous to Pol β K72, is crucial for this enzymatic function, suggesting that this is the main nucleophile responsible for the reaction (Garcia-Diaz et al. 2001 ). Interestingly, our multiple sequence analysis revealed that human Pol λ K312 and Pol β K72 are replaced by histidine in plant Pol λ proteins. In our search for alternative candidate residues, we have found that AtPol λ K248 and K255 are important for the dRP lyase reaction. Replacement of these lysine residues by alanine caused a significant reduction of the dRP lyase activity of AtPol λ, which was nearly abolished in the K255A mutant protein. Alanine substitution for K248 resulted in a lesser, but still important reduction of activity. These results suggest that K255 is the preferred but no required nucleophile responsible for the dRP lyase reaction catalyzed by AtPol λ. Interestingly, AtPol λ K255 is conserved in plant and animal Pol λ enzymes, but not in Pol β proteins, whereas AtPol λ K248 is conserved in plant Pol λ and in Pol β proteins, but not in metazoan Pol λ. The residue aligned with AtPol λ K255 in human Pol β is K60, and its replacement by alanine also results in a significant reduction in Pol β dRP lyase activity (Prasad et al. 1998 ). We found that the DNA polymerase activity of AtPol λ K248A and K255A mutant proteins was not affected, but their ability to extend DNA synthesis in a single-nucleotide gap, with the consequent displacement of the downstream strand, was impaired. This result agrees with the fact that residues in the 8-kDa domain of human Pol λ make specific contacts with the 5′-end of the downstream strand in gapped substrates (Garcia-Diaz et al. 2004 ). The strand-displacement ability of AtPol λ might allow its participation in LP-BER. Although this has already been suggested for human Pol λ by some authors (Garcia-Diaz et al. 2001 ), to date there is no evidence that AtPol λ is involved in plant LP-BER. In this work we have also shown that the dRP lyase activity of AtPol λ is required to complete SP-BER of uracil in a reconstituted repair reaction. In the BER reconstitution assay, AtPol λ was able to perform both the gap-filling and dRP removal steps that allow the final DNA ligation of the processed strand. In contrast, the ability of AtPol λ K248A and K255A mutant proteins to promote the repair of the BER intermediate was significantly impaired. Importantly, the capacity of the mutant AtPol λ versions to support SP-BER was inversely correlated with their dRP lyase activity and was strongly reduced in the AtPol λ K255A protein. In mammalian cells, removal of the dRP group is a critical BER step in vivo, since only the dRP lyase activity of Pol β, and not its DNA polymerization capacity, is required to reverse the DNA damage sensitivity of Pol β-null cells (Sobol et al. 2000 ). Since plants lack Pol β homologues, our results suggest that the dRP lyase activity of AtPol λ might play a role in Arabidopsis SP-BER in vivo.
Base excision repair (BER) generates gapped DNA intermediates containing a 5′-terminal 2-deoxyribose-5-phosphate (5′-dRP) group. In mammalian cells, gap filling and dRP removal are catalyzed by Pol β, which belongs to the X family of DNA polymerases. In higher plants, the only member of the X family of DNA polymerases is Pol λ. Although it is generally believed that plant Pol λ participates in BER, there is limited experimental evidence for this hypothesis. Here we have characterized the biochemical properties of Arabidopsis thaliana Pol λ (AtPol λ) in a BER context, using a variety of DNA repair intermediates. We have found that AtPol λ performs gap filling inserting the correct nucleotide, and that the rate of nucleotide incorporation is higher in substrates containing a C in the template strand. Gap filling catalyzed by AtPol λ is most efficient with a phosphate at the 5′-end of the gap and is not inhibited by the presence of a 5′-dRP mimic. We also show that AtPol λ possesses an intrinsic dRP lyase activity that is reduced by mutations at two lysine residues in its 8-kDa domain, one of which is present in Pol λ exclusively and not in any Pol β homolog. Importantly, we also found that the dRP lyase activity of AtPol λ allows efficient completion of uracil repair in a reconstituted short-patch BER reaction. These results suggest that AtPol λ plays an important role in plant BER. Supplementary Information The online version contains supplementary material available at 10.1007/s11103-023-01407-8. Key message Pol λ, the only X family DNA polymerase present in plants, possesses a dRP lyase activity that is required for completion of base excision repair in Arabidopsis . Supplementary Information The online version contains supplementary material available at 10.1007/s11103-023-01407-8. Keywords Funding for open access publishing: Universidad de Córdoba/CBUA
Supplementary Information Below is the link to the electronic supplementary material.
Acknowledgements We are grateful to members of our lab for helpful criticism and advice. We thank Dr. Luis Blanco (Centro de Biología Molecular Severo Ochoa, CSIC-UAM, Madrid, Spain) for his advice on amino acid conservation analysis in Pol λ proteins. Author contributions DCC, TMR, TRA and RRA designed research; CBM, DOP and JALM. performed experiments; DCC, TMR, MIMM, RRA. and TRA analyzed data; DCC, TMR, MIMM, TRA and RRA wrote the manuscript. All authors read and approved the final manuscript. Funding Funding for open access publishing: Universidad de Córdoba/CBUA. This work was supported by the Spanish Ministry of Science and Innovation (MICINN) under grant number PID2019-109967 GB-I00. Funding for open access charge: Universidad de Córdoba / CBUA. Data availability The authors confirm that the data supporting the findings of this study are available within the article and its supplementary materials. Declarations Conflict of interest The authors have no financial interests to disclose.
CC BY
no
2024-01-15 23:42:02
Plant Mol Biol. 2024 Jan 13; 114(1):3
oa_package/d6/08/PMC10787897.tar.gz
PMC10787898
38217571
Introduction Sensing and regulating the tension of the plasma membrane is crucial for cells to function effectively, especially when challenged by changes in the local micromechanical environment. Whether during malignant transformation, cytokinesis or morphogenesis, altered cell–cell and cell–substrate adhesivity are coupled to the dynamic cell signalling state defining these processes [ 28 , 48 ]. A remodelled micro-environment can influence the cell membrane's mechanical state, changing its flexibility or tension. Processes involved in membrane homeostasis would ensure the maintenance of membrane tension or conservation of the microscopic fluctuations, providing the plasma membrane’s effective tautness or excess area (difference between microscopic area and projected area per unit microscopic area) [ 46 , 64 ]. While enhancing tension can affect endocytosis [ 18 , 29 ] and pit-formation [ 9 ], as displayed for clathrin-mediated endocytosis (CME) [ 2 ], conversely, cells can also alter their endocytosis/exocytosis rates to regulate tension. Studies have addressed the role of membrane trafficking in tension regulation, either in general [ 65 ] or focussing on particular pathways such as the CLIC-GEEC (CG) pathway [ 74 ], CME [ 17 , 39 ] or those using caveolae [ 45 ]. Flattening of caveolae can regulate tension surge [ 69 ] suggesting that pits that participate in endocytosis may directly regulate tension. Since pit-formation and internalization are common to all pathways and sometimes also regulated similarly, it becomes essential to understand whether the formation of endocytic pits in general can initiate tension regulation or their subsequent internalization is also essential. Addressing such questions would require mechanical perturbation of cells and following membrane mechanics and endocytosis state over time. Additionally, perturbing molecules necessary for pit formation or internalization can further elucidate their individual roles. To quantify endocytosis, specific cargoes (e.g. Transferrin for CME) are usually labelled to quantify its corresponding pathways. However, Rab5 labels early endosomes of most pathways and, thus, reports the general state of endocytic trafficking [ 8 ]. A fraction of endosomes are recycled back to the plasma membrane using a fast-recycling arm, Rab4 [ 71 ] which can accumulate if either excess endosomes are formed, or the fusion with the plasma membrane is hampered [ 30 ]. Therefore, following labelled cargo (like Transferrin), as well as Rab5 and Rab4, together with well-established Transferrin uptake assays, can collectively describe the trafficking state of the basal plasma membrane. For perturbing endocytosis, molecules utilized by multiple pathways can be targeted. For example, many constitutive pathways in HeLa cells (e.g. CME, caveolae) require dynamin for the scission of their endocytic pits, making it an important target. Dynamin's GTPase action can be inhibited by Dynasore, an agent that can prevent scission and, thus, internalization of pits in these pathways without affecting pit formation [ 40 ]. However, certain constitutive pathways, like CG, could still be operational even on Dynasore treatment—although its presence in HeLa cells is debated [ 75 ], while other dynamin-independent constitutive pathways are not well classified. Knocking down adaptor protein AP2 would directly impact the pit-formation step of CME [ 38 , 50 , 77 ]. Cholesterol is required very early for the primary clustering of lipids/proteins in several pathways [ 32 ]. The formation of endocytic carriers critically requires cholesterol for pathways such as CG pathway and caveolae, while assembly of clathrin and membrane curving in CME mostly requires adapters like AP2 [ 11 , 27 , 77 ]. Depletion of Cholesterol (by MβCD) thus, can arrest the formation of new pits of the CG pathway and caveolae. However, it can also enhance tension [ 7 ], and has been reported to cause a substantial reduction in the formation and budding of deep pits of CME without completely blocking CME [ 2 , 4 , 58 , 73 ]. The use of ATP-depletion can reduce both the pinching-off of a majority of pathways as well as the formation of pits in pathways like CME [ 62 ] and caveolae [ 69 ]. Thus, ATP depletion could suppress the completion of a multitude of constitutive endocytic pathways and, therefore, could be used to study the role of scission in general. The alterations of these endocytic events need to be understood in response to perturbations to the mechanical microenvironment. Mounting evidence [ 19 , 36 , 79 ] shows significant crosstalk between endocytic machinery and cell adhesion molecules and its role in regulating cellular homeostasis [ 80 ]. While cell–substrate adhesion profiles of single cells have been reported to determine the distribution of endocytosis sites [ 22 ], studies have also addressed if de-adhesion can trigger mechano-regulation via endocytosis [ 74 ] or blebbing [ 49 ]. Although measurements of apparent tension have been reported during cell spreading [ 20 , 69 ] similar studies involving tension measurements during de-adhesion have not been performed, to the best of our knowledge. Measurement of membrane mechanics can be either done at specific points by optical tweezer-based tether pulling or by studying the spontaneous fluctuations of the membrane measured using interference reflection microscopy (IRM). Several studies [ 7 , 66 ] have proposed spatial heterogeneities in tension. Multi-point measurements would enable accessing the local distribution of excess membrane undulations or tension in cells during different phases of de-adhesion. In this study, we address the contribution of endocytosis to mechano-regulation during de-adhesion by employing the IRM-based measurement of effective fluctuation–tension of de-adhering HeLa cells. IRM [ 1 , 10 , 12 , 21 , 37 , 54 ] primarily provides information about the distance of the basal membrane from the substrate—also called the membrane ‘height’. Spatio-temporal measurements of height provide fluctuation amplitude (SD time ) excess area and enable deriving mechanical parameters like fluctuation-tension [ 7 , 67 ]. Spatial maps of fluctuation-tension (also termed tension in the rest of the manuscript) and their temporal evolution help us quantify the changes. We use total internal reflection fluorescence (TIRF) microscopy to image Rab5 and Rab4 labelled structures, labelling the early endosomes and rapidly recycling endosomes, respectively. This measures how de-adhesion alters internalization during endocytosis. Fluorescent cargo—transferrin—is used to further validate the involvement of endocytosis while Dynasore, dynamin-mutants, knockdown of AP2, Cholesterol and ATP depletion are used to perturb steps/pathways of endocytosis to assess their contribution to the mechano-regulation.
Methods Cell line HeLa cell line (CCL-2, ATCC) was used to perform all the experimental studies. Cell culture HeLa cells were grown in Dulbecco’s Modified Essential Medium (DMEM, Gibco, Life Technologies, USA) with 10% foetal bovine serum (FBS, Gibco) and 1% Anti-Anti (Gibco) at 95% humidity, 5% CO 2 and 37 °C. Experiments were always performed after 16–18 h of cell seeding. Pharmacological treatments To de-adhere cells from the substrate, HeLa cells were incubated with 0.05% or 0.25% Trypsin–EDTA solution (Gibco) at 37 °C on the onstage microscope incubator. To inhibit all dynamin-dependent endocytic pathways, we incubated cells with Dynasore hydrate (80 μM; Sigma) in serum-free media for 20 min [ 5 , 31 , 40 ]. HeLa cells are incubated with ML-141 (10 μM; Sigma) for 30 min in serum-free media to inhibit CLIC-GEEC endocytic pathways [ 74 ]. For ATP depletion, cells are incubated for 1 h with 10 mM Sodium Azide (Sigma-Aldrich) and 10 mM 2- deoxy-D- glucose (Sigma-Aldrich) dissolved in M1 media composed of 150 mM NaCl (Sigma-Aldrich), 1 mM MgCl 2 (Merck) , 20 mM HEPES (Sigma) [ 6 , 78 ]. 10 mM Methyl-ß-cyclodextrin (Sigma-Aldrich) was used in serum-free media for 50 min to deplete Cholesterol [ 7 ]. For inhibiting Actin filament polymerization, cells were kept in 5 μM Cytochalasin D (Sigma-Aldrich) for 1 h in serum-free media [ 6 ]. Cells were de-adhered by TrypLE Express (Gibco). This was used as an alternative to Trypsin–EDTA for de-adhesion experiments [ 74 ]. 1 mM EDTA was used for de-adhering cells [ 34 ]. All the treatments were incubated at 37 °C inside the incubator. During imaging and de-adhesion, all treatments were maintained at the same concentration. For Calcein AM staining, 2 μM Calcein AM was added in serum free media and incubated for 30. After incubation media was discarded and fresh media was added for the imaging. For endocytosis uptake assay, 10 μg/ml Transferrin Alexa fluor 568 was incubated for 5 min at 37 °C. For stopping endocytosis, HEPES based buffer was used at ice cold temperature. To remove external Tf fluorescence, ascorbate buffer was used at 4 °C [ 74 ]. Cells were fixed using ice cold 4% Paraformaldehyde. To follow Transferrin uptake during de-adhesion or in non-de-adhered cells (in normal or dynamin-inhibited conditions), cells were first incubated with 25 nM of Transferrin for 5 min. Subsequently, Transferrin was washed off and de-adhesion was followed. For dynamin inhibition cells were pre-treated with 80 μM Dynasore hydrate and then Transferrin was added. Immunostaining For immunostaining, cells were first fixed with 4% paraformaldehyde (Sigma-Aldrich) for 15 min and subsequently washed twice with phosphate-buffered saline (PBS, Sigma-Aldrich). Next, cells were incubated in 0.1 M glycine (Sigma-Aldrich) for 5 min and then washed again with PBS. Triton-X was used for 2 min and then washed with PBS. For blocking, cells were incubated with 3 ml of 0.2% Gelatin (Sigma-Aldrich) solution for 3 h at room temperature. Primary antibody treatment was done with Recombinant Anti-Rab4 antibody (Abcam) at 1:200 dilution in Gelatin and kept overnight at 4 °C to mark recycling endosomes. Goat Anti -Rabbit IgG H&L, Alexa Fluor 488 secondary antibody (Abcam) was used at 1:500 dilution at Gelatin for 2 h after washing with PBS. Subsequently, cells were imaged in 2 ml of PBS. Transfection EGFP-Rab5 (Addgene) was a gift from Arnab Gupta. Cells were transfected with 1 μg of Rab5 plasmid DNA to label early endosomes, respectively, by using Lipofectamine 3000 (Invitrogen). EGFP-CAAX [ 41 ] was a gift from Lei Lu. Cells were transfected with 1 μg EGFP-CAAX (Addgene) plasmid to mark the membrane. Cells were transfected with 1 μg of mCherry- Clathrin LC-15 (Addgene) to label clathrin-coated vesicles. To mark fast recycling endosomes, cells were transfected with 1 μg of mCherry- Rab 4a plasmid. Other treatments, if required, were performed 16 h after transfection. Fixation of cells was performed by using 4% paraformaldehyde (Sigma-Aldrich) for 15 min at 37°C temperature. siRNA-based knockdown HeLa cells are transfected with 10 nM of AP2 siRNA total for 72 h. At 0 h first siRNA transfection was done and at 48 h fresh media was added with 10 nM of AP2 siRNA booster dose. To validate the knockdown of AP2, cells are immunostained with Anti AP2 antibody to check intensity at the basal membrane by TIRF microscopy. Western blotting was also performed using established protocols to validate the knockdown. IRM imaging Cells were imaged in Nikon Eclipse Ti-E motorized inverted microscope (Nikon, Japan) equipped with adjustable field and aperture diaphragms, 60X Plan Apo (NA 1.22, water immersion) and a 1.5X external magnification on an onstage 37 °C incubator (Tokai Hit, Japan). Either an EMCCD (Evolve 512 Delta, Photometrics, USA) or an s-CMOS camera (ORCA Flash 4.0, Hamamatsu, Japan) was used for imaging. A 100 W mercury arc lamp, an interference filter (546 ± 12 nm) and a 50–50 beam splitter were used [ 6 ]. For IRM, movies consisted of 2048 frames (19.91 frames/s, EMCCD and 20 frames/s for s-CMOS) recorded at EM 30 and an exposure time of 50 ms. Optical trap experiment For optical trap-based tether-pulling experiments, a 2 μm polystyrene bead was trapped by focusing a 1064 nm Laser (Coherent, Sweden) of 1000 W power at source using a 100 × objective. The back-aperture of the objective was over filled after beam expansion and using mirrors and a 50/50 beam-splitter for beam manipulation. The trap stiffness ( k ) was calibrated from trajectories of the trapped bead using the equipartition approach where x is the displacement of the bead from the trap centre, k B is the Boltzmann constant and T the temperature in the Kelvin scale. For analysis, the bead was detected as an object (MATLAB, Fiji), and its centre was tracked with time. The bead’s displacement from the centre of the trap and the spring constant of the trap was used to get the force ( ). For every cell, a tether was pulled at a constant velocity of 0.5 μm/s up to a distance of 40 μm (LabVIEW, National Instruments, USA). Tether force was calculated from the average bead position during the period when it was parked with the tether pulled (Fig. S3c) for ~ 50 s. Imaging was done at 200 frames per sec. The apparent membrane tension of the apical section of the cell was derived from the force using the Canham-Helfrich equation , where κ is the bending rigidity and taken to be 15 k B T [ 13 ], and σ A denotes the apparent membrane tension. TIRF imaging For TIRF Microscopy, an inverted microscope (Olympus IX-83, Olympus, Japan) was used with a 100X 1.49 NA oil immersion TIRF objective (PlanApo, Olympus). An s-CMOS camera (ORCA Flash 4.0, Hamamatsu, Japan) and 488 nm, as well as 561 nm laser sources, were used. Images were acquired using an exposure time of 300 ms with ~ 70 nm penetration depth. Confocal imaging For confocal imaging, a Leica confocal microscope (Leica SP8) was used with a 63X oil objective lens (NA 1.4). A step size (in z) of 250 nm is used for imaging with a pixel size of 45 nm and deconvoluted (Leica Lightning Software). STED imaging An Abberior Facility Line system with an Olympus IX83 microscope (Abberior Instruments) was used for STED imaging. Abberior autoalignment sample was utilized for alignment of the STED and confocal channels. 15 nm pixel size was maintained during imaging. A pulsed STED line at 775 nm was used for depletion and STAR Red and Alexa 568 conjugated secondary antibodies were used for imaging Clathrin and AP2, respectively. Analysis of IRM images The intensity of IRM images was converted to height (wherever applicable) as reported [ 6 ]. The amplitude of spatial undulations spatial (SD space ) was obtained from the standard deviation (SD) of relative heights across all pixels in an FBR after averaging it over 20 frames. For obtaining the amplitude of temporal fluctuations, SD time , the SD of the relative heights over 2048 frames in each pixel was calculated and averaged across all pixels in an FBR. The power spectral density (PSD) of individual pixels was obtained from the temporal relative height time series using either the FFT method or the covariance method (MATLAB). PSDs of all pixels in an FBR were averaged to obtain the PSD for that FBR. For obtaining mechanical parameters, the PSDs were fitted to [ 3 , 6 , 23 ] where active temperature (A), effective cytoplasmic viscosity ( η eff ), confinement ( γ ) and membrane tension ( σ ) were used as fitting parameters. The bending rigidity ( κ ) was fixed at 15 k B T [ 68 ]. For obtaining excess area fraction ( ) over an FBR, the flat area of the FBR is taken as A P (= L 2 when patch/FBR is a square of side L ) and A is the sum of all dA calculated for each pixel by comparing the height at that pixel with its neighbours ( ) [ 16 ]. The excess area is represented in figures as the percentage excess area ( ). The activity per FBR is calculated as the lower bound of the entropy generation rate at each FBR. For obtaining the entropy generation rate, data pooled from all pixels of the FBR were taken through dimensional reduction using principal component analysis. The time series (2048 frames) of one of the principal components were built and analyzed as described recently [ 43 ] using the short-time inference scheme reported earlier [ 44 ]. Tension mapping For tension mapping (Fig. 2 c), PSD was calculated for either for each pixel, and either directly used (pixel-wise tension mapping) or averaged over all pixels in each FBR (FBR-wise tension mapping). PSDs, thus, obtained are fitted and every parameter extracted from fits—including tension and R 2 were mapped on to the same location as the pixel/FBR. Fluorescence image analysis For endosomal or puncta counting (MATLAB), first, a Gaussian blur operation was performed to spatially average out the image. The Gaussian-blur is next subtracted from the original image, thus enhancing local contrast. The subtracted image is normalized between 0 and 1, and thresholding was performed using the appropriate threshold, resulting in a binary image. Next, a mask was applied over the image to select the cell. Single pixels were removed using serial erosion and dilation, and subsequently, the binary image was used to detect objects. The area fraction is calculated by dividing the total area of detected objects ( ) by the total area of the ROI or cell ( ). Similar approach was used for calculating colocalization. Objects were detected for each channel separately. Total area of the pixels overlapping in the two binary images was considered to be colocalizing area ( ). Colocalizing area divided by the total area of the ROI was used as colocalizing area fraction ( ). Colocalizing area divided by object-covered area of any particular was used as the percentage colocalization of that channel ( ). Mander’s coefficient was obtained using ImageJ. Counting tubules from confocal z-slices For counting tubules, Z-stack images of cells were captured. Line scans were drawn at the cortex regions of cells (ImageJ/Fiji), and the intensities of these line scans were taken from the intensity plot profile of each line scan to plot the internal intensity. Peak analysis (MATLAB) involved evaluating the intensity line scans for peaks of minimum prominence of ~ 6 and width of ~ 5. Peaks were counted as tubules. Length of tubules were obtained by drawing line ROIs along tubules and measuring their length. Statistical analysis Every IRM experiment was preceded by imaging beads. Every experiment was repeated at least thrice involving multiple cells and FBRs (Table S1 ). A Mann–Whitney U test was performed for statistical significance testing (ns denotes p > 0.05, *denotes p < 0.05, **denoted p < 0.001). When indicated, a linear mixed effect model (LMM, MATLAB) was also used (using the fixed effect of time and random effects grouped under replicate set number of the experiments and cell number) to quantify the significance of the observed changes in logarithm of tension values (Table S2). This helped avoid the effect of the high sample size of FBRs that could influence hypothesis testing. LMM has been used for comparisons in other high-sampling mechanical measurements [ 24 , 56 ].
Results De-adhesion-mediated increase in membrane fluctuations is actin-dependent IRM images reflect the distance of the cellular basal membrane from the glass coverslip and, thus, can be used to study the spatiotemporal basal PM height fluctuations in adherent cells [ 6 ]. Qualitatively, darker regions were closer to the coverslip while intensity increases with height till ~ 100 nm and periodically oscillates. Calibration with standards (beads) was performed to quantify the relative heights, followed by selecting regions where the intensity-height conversion was possible—usually, regions that are ~ 100 nm from the coverslip and fall under the first branch of the intensity profile (that displays bands) [ 6 ]. The selected regions (square membrane patches) were termed first-branch regions (FBRs). Multiple such regions were marked for every cell. The time series of membrane height fluctuations (obtained from single pixels) were used to get mean height and standard deviation (SD) of height and termed SD time . Pixel-wise measurements were averaged over neighbouring pixels to get “FBR-wise” data. SD space was obtained from a snapshot—comparing height across NxN pixels of any FBR where N was usually 12 in this study (see " Methods "). These parameters depicted the amplitude of fluctuations. Averaging over all FBRs in a cell and pooling such data from all cells provided the “cell-wise” data. We compared these parameters between control and treated cells to understand how a mechanical perturbation like de-adhesion altered membrane mechanics. IRM provides the ability to map fluctuation-tension, get its distribution in the cell, and thus differentiate between large global and local changes. In this study, we have compared changes at the level of whole cells (termed cell-wise) and those that show up in the local distribution of tension in single cells. However, while comparing local data, to ensure that the repeated measurements per cell do not falsely strengthen the statistics, we have employed linear mixed models (LMM) as usually used [ 24 , 56 ] to account for the nested grouping of replicate measurements. Finally, it is important to note certain limitations. Fluctuations reported by IRM are mainly thermal but can have contributions from active (ATP-dependent, non-thermal) processes in cells [ 76 ]. Using a model-free method, a recently developed algorithm [ 43 , 44 ], quantified the effect of active forces on membrane fluctuations (termed activity) and revealed local membrane fluctuations to be weakly active. However, fluctuation-tension solely should not be used for drawing inferences. We deal with this primarily by using direct measurements of fluctuations amplitude. We also compare activity at different time points of de-adhesion to record the level of deviation of the measured fluctuations from equilibrium. Finally, we corroborate the main effect of de-adhesion on fluctuation-tension with apparent membrane tension measurement using the generally accepted optical-trap-based tether extraction method. HeLa cells were treated with either a low (0.05% Trypsin–EDTA) or high concentration (0.25% Trypsin–EDTA) of de-adhering solution and imaged by IRM (Fig. 1 a) to study the changes in membrane fluctuations and mechanics in the basal membrane during de-adhesion. We observed high variability in the time cells took to de-adhere (Fig. 1 b). To quantify the rate of de-adhesion, the time taken by each cell to reduce their spread area to 67% of the initial was calculated (Fig. 1 c). We found that, among the various treatments used, Cytochalasin D (Cyto D)—an agent reducing actin polymerization [ 61 ] of the actin cortex resulted in a very fast de-adhesion. Therefore, a lower concentration of de-adhering solution of Trypsin (0.05%) was used for Cyto D experiments. This was in line with the understanding that during de-adhesion, the contractile cortex caused lateral retraction [ 34 , 63 ]. At lower trypsin concentrations, the process could be slowed down such that the decay times for Control and Cyto D were similar (Fig. 1 c). In general, de-adhesion caused an increase in temporal fluctuations (Fig. 1 a—7 min), followed by lateral retraction (Fig. 1 a—9–18 min). The increase in fluctuation amplitude (SD time ) could be visualized from the maps (Fig. 1 a bottom, right). Following the lateral retraction of the edge using a kymograph) of the IRM intensity (along the white line, Fig. 1 d), we found that as the edges retract inwards, intensity patterns lying inward also moved. Membrane height (IRM intensity) increased in a cluster close to the edge (white oval, Fig. 1 d) as the retraction progressed. This accumulation faded away with time (section below the oval, Fig. 1 d). While the membrane undulations were locally enhanced in control cells, such local increase was less prominent in Cyto D-treated cells (Fig. 1 d). Thus, as de-adhesion progressed, cytoskeleton-dependent transient accumulation of membrane folds point to the ongoing membrane remodelling. We proceeded to measure the fluctuations and effective membrane-mechanical parameters from the fluctuations during the different phases and over time to underpin the effect of de-adhesion on membrane mechanics quantitatively. The initial rise in fluctuations is regulated back Since the rates of de-adhesion in cells were different, we classified the data in this and the following sections based on the level of reduction in spread area. For every cell, we divided the process of de-adhesion into four phases. The “C” phase was defined as the period for which the cell had not been treated with the de-adhering agent (Trypsin). The “P1” phase was demarcated as the slow de-adhesion phase before the exponential fall of the spread area started. Typically, the spread area in this phase remained within 90% of the initial cell spread area. The “P2” phase was defined as the period when the spread area exponentially reduced, at least to ~ 67% of the original value. The “P3” phase marked the period when the low spread area had stabilized to ~ 40–20% of the initial (Fig. 1 b). Height fluctuations were measured before (phase: C) and then every few minutes after adding de-adhering medium from image stacks acquired at every time-point, where each movie lasted for ~ 102 s. Only membrane regions that remained adhered through the movies were analyzed. On following a representative cell in time (Fig. 2 a; same cell as shown in Fig. 1 a) or averaging over a population (Fig. 2 b), we found that the amplitude of temporal height fluctuations (SD time ) first increased, followed by a decrease/saturation on de-adhesion. The reverse was observed for fluctuation-tension (Fig. 2 a). Maps of FBR-wise fluctuation-tension (Fig. 2 c) revealed the lowered tension state of the cell followed by an increase in the regions that remained adhered. Strictly for visualization purposes, we also created a pixel-wise map of tension (Fig. 2 c). Mapping tension showed a reduction in the initial heterogeneity when cells are at phase P2. The global trend (Fig. 2 b ) was corroborated by the distribution of local (FBR-wise) fluctuation amplitude and tension for single cells. Clearly, the changes were greater than the error calculated while averaging the distributions over multiple cells and repeats (Fig. 2 d). We confirmed that in the absence of de-adhesion media, the fluctuations or tension of these cells did not change over 20 min (Fig. S1 ). Following the same regions in single cells through de-adhesion also showed the dip and recovery of tension (Fig. S2), confirming that regions that remain adhered also underwent these changes. Although the spatial height variation (SD space ) and excess area showed a decreasing trend in contrast to SD time (Fig. S3a), using a gentler substrate detachment reagent—TrypLE (Fig. S3b) or at lower trypsin concentration, their trends matched [Fig. S3 c, d (0.05% Tryp)]. The corresponding entropy-generation rate obtained from fluctuations of de-adhering cells changed mildly (Fig S3a), implying that the measured fluctuations did not capture any major enhancement in non-equilibrium activity in the frequencies assayed. Since actin remodelling was expected during de-adhesion, its impact on tension reduction was next studied. The initial rise in fluctuations is weaker on cortex disruption We used Cyto D to weaken the cortical actin. At 0.05% Trypsin, the spread area reduction rate was similar for Control and Cyto D, but cells de-adhered properly. Tension reduction in P2 was not substantial in the presence of Cyto D or reduced filamentous cortical actin (Fig. 2 e, f). Although fluctuations were enhanced weakly (than Control (Fig. 2 f, S4) at P2, by P3, unlike in Control, Cyto D showed similar fluctuations as C. In line with Fig. 1 d, these data also suggest that disruption of cortical actin reduced membrane accumulation or enhancement of fluctuations. De-adhesion only occurs at the basal membrane. To study its effect on the apical membrane, we next extracted membrane tethers from cells and measured tether forces. These were measured on single cells—before and after initiating de-adhesion (Fig. 2 f top, Fig. S5) within the first—3–7 min. There was a reduction in force for 8/10 cells, which showed a 2–52% reduction in force (or effectively a 4–77% reduction in apparent tension). Next, we examined whether the tension reduction and subsequent regulation correlated with endocytic activity. De-adhesion increases early endocytic and fast recycling endosomes near the basal plasma membrane To quantify endocytosis, we utilized three strategies—labelling the cargo of an endocytic pathway, labelling early endosomes using Rab5 and labelling recycling endosomes with Rab4 (Fig. 3 a). For analyzing fluorescent puncta imaged in TIRF, the puncta were detected as objects and analyzed (Fig S6a). Since nearby objects could be distinguished from connected ones, we scored for the area fraction or the area covered by the puncta per μm 2 of the analyzed area. At first, we followed a cargo of CME—Transferrin (Tf)—added in Control and de-adhering cells (Fig. 3 b, c) and found a clear increase in Tf's internalization (Fig. 3 b) as well incorporation (per μm 2 ) in the plasma membrane as clusters in de-adhering cells (Fig. 3 c, d, Fig S6 b–d). Analyzing their disappearance indicated enhanced short-time dynamics on de-adhesion (Fig. S6. E–i) validating increase in pits not plaques [ 60 ]. The involvement of the endocytic machinery was next confirmed by labelling early endosomes by Rab5 [ 8 ] and the short-loop (fast) recycling endosomes by Rab4 [ 71 ]. Near the plasma membrane, endosomes mainly contain these two labels [ 71 ]. The Rab4-containing endosomes emerge from the same endosomes as Rab5 [ 71 ] and thus are studied to evaluate the state of endocytic and recycling activity. We imaged EGFP-Rab5 and Rab4 -mCherry using TIRF microscopy (Fig. 3 e, f, Fig S6). The area fraction of early endosomes (Fig. S6), calculated for whole cells, showed an increase and a subsequent tapering off/decrease (Fig. 3 f). The increase was anti-correlated initially with the reduction in spread area (Fig. 3 f). However, it was difficult to discount the effect of heterogenous de-adhesion in the observed trend since some regions had denser features than others. Thus, we looked at regions that stayed on through the observed time scale (15 min). We followed the area fraction of Rab5 for the same sub-cellular region of interest (ROI) (Fig. 3 g). The rise and saturation were found to be consistent (Fig. 3 g). To further validate if endocytosis was ramped up, we analyzed clathrin-coated pits (Fig. S6j). Imaging cells fixed before or after faster de-adhesion. Although an increase and saturation in the area fraction of these pits were observed, the changes were not appreciable. However, on evaluating with super-resolution using Stimulated Emission Depletion (STED) microscopy, we found that clathrin was less colocalized with its adaptor AP2 (Fig. 3 h) on de-adhesion. As reported earlier, AP2 was observed either well-colocalized with smaller clathrin clusters or at the edges of larger structures (Fig. 3 h, zoomed-in sections) and expected to be more curved [ 70 ]. On increasing tension, previous studies report a higher fraction of AP2 with clathrin indicative of flat structures [ 9 ]. Our observation of AP2 localized at edges (Fig. 3 i, j) and showing lesser colocalization (Fig. 3 k, l) might indicate that more fraction of clathrin structures are pits on de-adhesion. We also find an increase in the distance between clathrin and AP2 clusters after de-adhesion, comparing pairs already with 525 nm of each other (Fig. 3 m) or higher. The significance holds true even while comparing pairs within 165 nm but does not show any difference when only those lying within 150 nm (close to the size of well-formed pits, Fig. 3 i , yellow arrow) are compared, clearly indicating the population with AP2 at edges start making this difference significant. Together, the data showed that triggering de-adhesion reduced tension, and cells ramped up the frequency of their endocytosis events. We also know from following the fluctuations that the lowering of tension was also stalled or recovered back at later stages of de-adhesion (Fig. 2 b). To understand if endocytosis affected tension regulation and how we next used pharmacological agents to block endocytosis and followed treated cells on de-adhesion. Blocking dynamin-dependent pathways reduces de-adhesion-triggered endocytosis We studied the effect of blocking dynamin-dependent pathways (Fig. 4 , Fig. S6, S7) by treating HeLa cells with Dynasore [ 31 ]. Specifically, it does not stop the formation of pits (clathrin-coated or caveolae, among others) but prevents dynamin’s function in the scission of pits (Fig. 4 a). Through Tf-uptake assay Dynasore’s effect on stalling of pit scission is clear (Fig. S6 e, h, i). Instead of the rise in Rab5’s area fraction on de-adhesion, an initial drop was observed with de-adhesion as expected since early endosomes were expected to contain Rab5 [ 53 ]. (Fig. 4 b, c, Fig. S7 a, b). Failure of fission caused tubes to form, which also contained Rab5 (arrows, Fig. 4 b, Fig. S7a). The increase in Tf puncta in Dynasore-treated cells was enhanced in de-adhering cells (Fig. 4 c, d). Rab4 labelling (immunofluorescence) revealed an increase during the de-adhesion (Fig. 4 b, Fig. S7b). This data suggested that the intermittent accumulation could be due to its inability to fuse normally with the plasma membrane (with reduced tension). This aligns with studies that have reported reduced recycling on inhibiting Dynamin [ 14 , 15 ]. Thus, Dynasore drastically reduced the formation of new early endosomes while also inhibiting the fusion of recycling endosomes with the plasma membrane. Together, these indicate that Dynasore effectively brought down de-adhesion-triggered endocytosis in the cell. We next studied how such blocking of endocytosis would affect the tension regulation during de-adhesion. Inactivating dynamin does not stop tension recovery, but knocking down AP2 does The single cell distribution of fluctuation amplitude and tension clearly changes as de-adhesion progresses to P2, implying that the tension reduced on de-adhesion of Dynasore treated cells (Fig. 4 e–f, Fig S7d–i). Interestingly, instead of preventing tension recovery, it was enhanced in Dynasore-treated cells (Fig. 4 e) in comparison to untreated cells (Fig. 2 d). Normalized cell averages (Fig. 4 g, Fig S7 f) point to the augmented mechanical regulation operating from P2 to P3 in Dynasore-treated cells. Maps help us visualize this (Fig. 4 f, Fig S7e). On checking the state of activity (entropy generation rate), we found a lowering of activity on Dynasore treatment (Fig. S7g). To further validate, we performed experiments with a dominant-mutant of dynamin that was transiently transfected in cells. We found that even in these cells, the increase in tension or reduction in SD time was more substantial than in the control condition (Fig. S8). Clearly, the tension at P3 was higher than C, unlike in control sets. The data imply that although endosome formation is reduced, membrane mechanics regulation during de-adhesion is not stopped. This clearly implies that the formation of pits has a direct role in tension recovery. To verify, we next compared the excess areas from IRM images to find the impact of Dynasore on membrane smoothness. There was a significant reduction in the excess area in Dynasore-treated cells, implying a smoother membrane (Fig. 4 g), supporting the hypothesis that pit formation can reduce excess area and enhance the effective tension. As a control, we checked how excess area changed on inhibiting Cdc42 by ML141. Cdc42 is required for the formation and scission of endocytic pits of the CG pathway. Inhibiting it for 30 min enhanced the excess area (Fig. S7h), further supporting the hypothesis that pit-formation reduces excess area. To validate further the role of pit-formation in tension recovery, we used siRNA-mediated knocked-down AP2. The knock-down was confirmed by immunofluorescence (Fig. S7j) and western blot (Fig. S7k) using control cells and cells treated with either scrambled siRNA (Scramble) or AP2-siRNA (AP2-siRNA). Depletion of AP2 by siRNA has been shown [ 25 ] to reduce membrane clathrin as well drastically reduce coated pits on the plasma membrane. We observe that AP2-siRNA significantly reduced tension and enhanced fluctuations and excess area (Fig. 4 i top) of cells. During de-adhesion (Fig. 4 i bottom), the tension reduced in the P2 phase but failed to be regulated back in the P3 phase (Fig. 4 h, i). SD time increased with de-adhesion (Table S2), although when compared using cell-wise statistics, no significant change was noted (Fig. 4 i). However, it was clearly not regulated back. The response of AP2-siRNA-treated cells were marked different than control cells or cells treated with scramble siRNA. Together, the Dynasore, dynamin mutant and AP2 siRNA data clearly establish the pit formation step of clathrin-mediated endocytosis to contribute to tension regulation during de-adhesion. Having provided evidence suggesting pit formation as the critical step in endocytosis to cause tension increase, we next aimed to understand the nature of processes used for the initial pit formation. Dynamin-dependent pathways are not limited to major pathways like CME/caveolae. To understand the dependency on pathways that use key energy-consuming molecules like dynamin or actin, we next check the regulation’s dependency on ATP. ATP-dependent-pit formation is common (CME and caveolae [ 69 ], for example). However, multiple pathways, like once induced by Shiga toxins, are ATP-independent [ 57 ]. ATP-depletion does not block tension regulation We used ATP depletion prior to de-adhesion to investigate the role of active processes and actin polymerization in the mechanical changes observed during de-adhesion (Fig. 5 a, b). ATP-depleted cells showed low initial fluctuations but a similar trend of increase in fluctuations followed by a decrease as observed for control cells (Fig. 5 a, b, Fig. S9a, Table S1 ). The effective tension reduced and then increased. The distribution of local values in single cells also captured the changes which were found to be significant. No appreciable change was observed in the area fraction of Rab5 in ATP-depleted cells as de-adhesion progressed (Fig. 5 c, d, Fig S9b). The area fraction of RAb5 in ATP-depleted cells was lower than in control cells (Fig. S9b), expected from reduction in active endocytosis. Therefore, the data here suggest the use of mechanisms that start recovering the tension drop without critically requiring ATP. As the first steps to identify the involved pathway, we next checked the dependence of tension regulation on cholesterol in the ATP-compromised condition. We do so because known pathways (like FEME used to internalize Shiga toxin) using ATP-independent pit formation are cholesterol-dependent and depend on the ability of cholesterol to cluster lipids and initiate the creation of invaginations [ 33 ]. Tension regulation in ATP-depleted cells is cholesterol-dependent We used ATP depletion and cholesterol depletion as controls (Fig. 6 a, S9e) to assess the role of cholesterol in tension regulation in ATP-depleted cells during de-adhesion. Comparing single-cell distributions or cell-averaged values (Fig. 6 b–d) showed that while the initial (C-P2) fluctuation (SD time ) increase displayed the same trend as for the controls, fluctuations from P2 to P3 were not reduced efficiently on combined depletion of ATP and cholesterol. Depleting ATP or cholesterol enhanced the recovery of tension (from Control), but on dual depletion of ATP and cholesterol, there was a further reduction (Fig. 6 a, c, Fig S9 c, d) [clear from distributions of local measurements, tension maps (Fig. S9e) and the LMM analysis of the local values (Fig. 6 d)]. Together, we showed (Fig. 6 e) that the tension drop was significant even on perturbing scission, ATP or cholesterol content and faster in the case of the latter two. The recovery was also significant and faster for these perturbations but could not be effective under reduced ATP and cholesterol-depleted conditions. Treating tension reached by the control cells at the P3 phase as a reference, Dynasore treatment effectively tilted the balance towards a larger tension recovery rate while dual depletion of ATP and cholesterol resulted in lower and, therefore, appreciably slower recovery. Therefore, the observations strengthen the hypothesis that in the absence of ATP, cells also use cholesterol-dependent processes for mechano-regulation. Since in pathways involved in Shiga-toxin's endocytosis, toxin-rich tubules have been reported to be formed even in ATP-depleted cells/giant unilamellar vesicles [ 59 ], we next checked if ATP-depleted cells formed tubules when tension was lowered by de-adhesion. For this, we imaged cells transfected with EGFP-CAAX and quantified the membrane cross-section at higher planes in ATP-depleted and ATP and cholesterol-depleted cells using confocal microscopy. On de-adhesion, ATP-depleted cells make more cholesterol-dependent tubular invaginations Cells were transfected with EGFP-CAAX [ 41 ] to label the plasma membrane and taken through different treatments (Fig. 7 , S10a). They were subsequently fixed (at 3, 6 min after treatment). Since fixation is not instantaneous, we measured the spread area of cells after fixation to determine the decrease in spread area (Fig. S10b). We observed a reduction to ~ 40% of initial when fixed at 3 min and ~ 17% of initial when fixed at 6 min. Therefore, they were classified as P2 and P3. Cells were subsequently imaged in confocal microscopy at a final resolution of ~ 120 nm in x and y directions (Fig. 7 a, b, S10). From the intensities obtained from scans under the membrane (Fig. 7 b, Fig. S10 c, d) along multiple line regions of interest (ROI), each of length ~ 4 μm, peaks were detected with larger widths and heights (from basal intensity) than set thresholds from the line scans (Fig. 7 c). After de-adhesion, ATP-depleted cells displayed significantly more internal surface-connected structures than ATP + cholesterol-depleted cells (Fig. 7 d) at the P2 and P3 phases. It should be noted that on cholesterol deletion (without ATP depletion, Fig. S10 e) or in control cells (no drug treatment, Fig. S10 f), there were much fewer tubules (Fig. S10 g), whose number also did not appreciably increase on de-adhesion (Fig. 7 d, Fig. S10 e–g). The number of peaks transiently increased with de-adhesion in ATP-depleted cells but, in contrast, decreased for cells also depleted of cholesterol (Fig. 7 e, Fig. S10). Tubule length also increased from before trypsinization to 3 min post-trypsinization. Hence, we concluded that the ATP-depleted cells might gain more cholesterol-dependent tubules on de-adhesion, thereby causing the tension surge. In conclusion, we show that de-adhesion induces a tension drop in the whole cell due to an altered adhesion state. The regulation sets in soon and engages endocytic pathways that internalize the membrane but also populate a recycling pool. Tension reduction is stalled but must be balanced by the recycling back of the membrane, because perturbations blocking internalization enhance the tension recovery rate. ATP and cholesterol depletion inhibits tension recovery, and this inhibition cannot be achieved by either preventing internalization or depleting only ATP or cholesterol individually.
Discussion Loss or remodelling of the adhesion machinery are known determinants of malignant transformation [ 42 , 47 ], while regulated de-adhesion has been shown to aid tumour invasion [ 26 ]. In this study, we aimed to understand the basic steps of endocytosis-mediated tension regulation during de-adhesion using IRM as the primary tool. It should be noted that tension was derived using a model that does not account for the frequency-dependent activity factor. We have evidenced that membrane fluctuations assayed at such small patches (0.75 μm 2 ) has overall weak activity [ 43 ] as inferred from measurements of the lower bound of the entropy generation rate. Using the same algorithm [ 43 ], change in activity is clear when endocytic activity is blocked by Dynasore (Fig. S7g), indicating that the technique can resolve small changes. The excess area was observed to be reduced, and fluctuation-tension was enhanced (Fig. S7). While incorporating the frequency dependence of activity factor would be ideal while extracting fluctuation-tension, such formulations are currently not yet feasible for easy application to our data. Hence, direct measurements of fluctuations (SD time ) are first used to comment on the state of the membrane and the effective fluctuation-tension, excess area and functional state of the membrane (endocytic activity) used for further inferences. We also observed that during de-adhesion, the active nature of fluctuations did not significantly increase (Fig. S3c). SD time supported inferences about fluctuation-tension, while apparent tension—measured by optical trapping experiments—validated the initial tension drop on de-adhesion. Furthermore, we not only reported changes in these parameters but also presented functional evidence of the altered physical state. Lowered tension state correlated with enhanced endocytosis (accumulation of Rab5 near the plasma membrane, Fig. 3 g) and reduced recycling (accumulation of Rab4 near the plasma membrane, Fig. 3 g). Importantly, we present straightforward evidence of the functional connection between higher fluctuations (and lower effective fluctuation-tension) leading to enhanced endocytosis which in turn leads to reduction of fluctuations. Active models of cell membrane undergoing endocytosis and exocytosis had been used in theoretical work [ 55 ] showing that active endo/exocytosis could give rise to an effective membrane tension. Their application to IRM data was not possible, but the pattern of changes of fluctuations (or effective fluctuation-tension) and Rab5/Tf punctas indicate a similar relationship. Future studies measuring the evolution of fluctuations and endocytosis in the same cells would further clarify their local dependence. Our data first highlight that the local (at the basal membrane) build-up of membrane fluctuations is actin-dependent and not solely dependent on de-adhesion/retraction. The changes in fluctuations were quick to resolve, and tension was negligibly altered in Cyto D-treated cells. While this aligns with recent work implicating the cytoskeleton–membrane connections in delaying tension flow and, therefore, its equilibration, further studies are needed to prove this conclusively. Identifying the stage when tension recovery was initiated was our primary target of investigations. We believe that the data strongly suggest that the formation of new invaginations starts the tension recovery. Even in the absence of de-adhesion, data in this manuscript (Fig. S7d) and reported earlier show that enhanced pit-formation (Dynasore treatment) could lead to a reduction of fluctuations and increase in tension (Fig. S7) while suppressing pit-formation by knocking down AP2 increases excess area, decreasing tension (Fig. 4 i). We have also observed that Shiga toxin created tubules and increased tension of ATP-depleted HeLa cells that contained GB3 but did not either create tubules or enhance tension in HeLa cells lacking GB3 (data not shown). Such tubules have also been reported to passively regulate the area of lipid bilayers on being strained [ 72 ]. The reverse, where flattening of invagination helps cells buffer a tension surge, has already been demonstrated. Hence, it is not far-fetched or physically impossible to use invaginations to perform this task. We believe it is possible that clustered lipids/proteins that initiate the pit formation can already start damping the fluctuations, although reports suggest so in simulations [ 52 ]. Clustering of resources, as reported for hotspots of CME [ 35 , 50 ] could also be ideal sites where pit-formation could remove the excess area. Pinching-off of endocytic buds or other ATP-dependent mechanisms, therefore, may not be critically essential for enhancing tension but required to prevent an excessive rise in tension, which could potentially inhibit many membrane processes or start unwanted processes. ATP-dependent machinery, we propose, acts as mechano-stats in the cell. They could be vital for keeping tension surges in check rather than being required only for enhancing tension. Our hypothesis is strengthened by the observation (Fig. 6 e) that the rate of tension-lowering is enhanced in cholesterol-depleted cells, which incidentally, also have a higher fraction of membrane covered by actin-membrane linker—Ezrin (Fig. S11). Whether other lipid-raft-dependent mechanisms are also at play [ 51 ] or cholesterol-dependent mechanisms downstream of membrane and actin-remodelling cannot be ruled out. Finally, some conclusions may be drawn about the relevance of the different pathways of endocytosis in tension regulation during de-adhesion. Assuming the three main pathways to be possibly the CG pathway, CME and caveolae-dependent pathway, cholesterol depletion is expected to block the CG and caveolae-dependent pathways even before pit formation. Cholesterol depletion, despite being reported to drastically (~ 80%) reduce internalization of CME, still is expected to allow ~ 50% of deep pits to form [ 73 ]. We observe that cholesterol depletion does not abolish tension recovery or its maintenance close to the Control's response (Fig. 6 e, C-P3). However, knocking down AP2, an adaptor protein required for the formation of CCPs during constitutive endocytosis, completely impairs the tension recovery. A lower colocalization with AP2 and more edge localization further confirm enhanced pit formation on de-adhesion. Together, this provides compelling evidence that pit-formation for CME significantly contributes to the mechano-regulation in de-adhering HeLa cells. The passive (ATP-independent mechanisms but cholesterol-dependent) pathways may also contribute by the induction of tubulated structures capturing excess membrane. We suggest that spontaneously pre-clustered platforms poised for other functions might tubulate on sudden tension reduction and contribute to the passive arm of the regulation. In conclusion, in this paper, we have presented the effect of de-adhesion on spatiotemporal fluctuations and effective cell membrane mechanics. We have demonstrated that an initial membrane slack is drastically reduced when the actin cortical network is weak. Pit formation on the plasma membrane has been shown to be a determining factor in regulating tension/fluctuations during the cellular process of de-adhesion. The restoration of the low effective tension used endocytic pit formation for increasing the tension, while the complete regulatory cycle utilized active as well as cholesterol-dependent passive forms of mechano-regulation (Fig. 8 ). Lead contact More detailed information and requests for the resources should be directed to and will be fulfilled by the lead contact, Bidisha Sinha ( [email protected] ). Material availability New materials and methods used in these studies will be available upon request to Bidisha Sinha. Data and code availability Data and codes used in this study for analysis purposes will be available upon request to the lead contact, Bidisha Sinha ( [email protected] ).
Adherent cells ensure membrane homeostasis during de-adhesion by various mechanisms, including endocytosis. Although mechano-chemical feedbacks involved in this process have been studied, the step-by-step build-up and resolution of the mechanical changes by endocytosis are poorly understood. To investigate this, we studied the de-adhesion of HeLa cells using a combination of interference reflection microscopy, optical trapping and fluorescence experiments. We found that de-adhesion enhanced membrane height fluctuations of the basal membrane in the presence of an intact cortex. A reduction in the tether force was also noted at the apical side. However, membrane fluctuations reveal phases of an initial drop in effective tension followed by saturation. The area fractions of early (Rab5-labelled) and recycling (Rab4-labelled) endosomes, as well as transferrin-labelled pits close to the basal plasma membrane, also transiently increased. On blocking dynamin-dependent scission of endocytic pits, the regulation of fluctuations was not blocked, but knocking down AP2-dependent pit formation stopped the tension recovery. Interestingly, the regulation could not be suppressed by ATP or cholesterol depletion individually but was arrested by depleting both. The data strongly supports Clathrin and AP2-dependent pit-formation to be central to the reduction in fluctuations confirmed by super-resolution microscopy. Furthermore, we propose that cholesterol-dependent pits spontaneously regulate tension under ATP-depleted conditions. Supplementary Information The online version contains supplementary material available at 10.1007/s00018-023-05072-4. Keywords
Supplementary Information Below is the link to the electronic supplementary material.
Acknowledgements BS acknowledges support from Wellcome Trust/DBT India Alliance fellowship (Grant number IA/I/13/1/500885), SERB (Grant number SERB_CRG_2458) and CEFIPRA (Grant number 6303-1). The authors are grateful to CSIR and IISER Kolkata for providing scholarships to TM and AB, respectively. TG thanks CEFIPRA for providing his fellowship. The authors are also thankful to the DBT Wellcome Imaging Facility (IA/I/16/1/502369) for confocal imaging, the Builder Imaging facility (BT/INF/22/SP45383/2022) for super-resolution STED imaging and the DIRAC supercomputing facility for tension mapping. Author contributions TM: investigation, formal analysis, data curation, writing (original draft), AB: investigation, methodology, software, formal analysis, writing (original draft), TG: methodology, software, investigation, formal analysis, SM: resources, AK: methodology, AB: methodology, resources, DM: resources, BS: conceptualization, methodology, writing (original draft), funding acquisition. All authors edited the manuscript. Funding Wellcome Trust/DBT India Alliance fellowship (grant number IA/I/13/1/500885), SERB (grant number SERB_CRG_2458) and CEFIPRA (grant number 6303-1). CSIR and IISER Kolkata for providing scholarships to TM and AB, respectively. TG fellowship is provided by CEFIPRA. Data availability The data that support the findings of this study are available from the corresponding author upon reasonable request. Codes used for this study are available at: https://github.com/BidishaSinha/Mechano-regulation-by-pit-formation-during-De-adhesion . Declarations Conflict of interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Ethical approval and consent to participate Not applicable. Consent for publication All authors agree to submit the manuscript for publications.
CC BY
no
2024-01-15 23:42:02
Cell Mol Life Sci. 2024 Jan 13; 81(1):43
oa_package/7f/66/PMC10787898.tar.gz
PMC10787899
38222211
Tuberculosis is one of the oldest known infective diseases. It continues to pose a major threat to global health. In low- and middle-income countries, drug-resistant strains of tuberculosis have made the disease increasingly dangerous. In high-burden countries like India, people with tuberculosis may not always have a comprehensive examination for severe illness (requiring hospitalization management) due to oversaturated outpatient departments, inadequate diagnostic capacity, and a lack of manpower. Oftentimes, these patients are initiated on treatment, and detailed assessments for severe illnesses are missed. This is particularly important in a country with the greatest tuberculosis burden, where two tuberculosis-related deaths occur every three minutes. The present article throws light on this grave issue, emphasizing the need for early triage at the time of diagnosis, which would ultimately impact overall mortality and treatment outcomes.
Editorial Tuberculosis (TB) is still an immense threat to health in endemic nations. Drug-resistant strains of mycobacteria have made matters worse, mandating creative approaches for efficient treatment [ 1 ]. At the national level, according to the National TB Prevalence Survey (NATBPS) 2019-2021, the estimated point prevalence of microbiologically confirmed pulmonary tuberculosis among individuals over 15 years of age was 316 per lakh population. It was predicted that there were 312 cases of all forms of tuberculosis per lakh people. Overall, in the year 2022, out of the total TB cases notified, 14,71,190 (61%) were male, 9,48,190 (39%) were female, and 1,023 (<1%) belonged to lesbian, gay, bisexual, transgender, queer, intersex, asexual, and other identities [ 2 ]. As per the World Health Organization's (WHO) recently released Global TB Report 2023, India contributes 27% of the total cases of tuberculosis, with 22% mortality worldwide. India seeks to eliminate tuberculosis by the year 2025. The goal of the National Strategic Plan 2017-2025 was to achieve 44 new cases of tuberculosis per lakh of citizens by the end of 2025. This figure, according to the 2023 report, is 199 cases per lakh. Hitting this goal will prove difficult because, by 2023, the plan called for an incidence of only 77 cases per lakh population. By 2025, the program also seeks to lower mortality to three deaths per lakh of people. The WHO has acknowledged the updated numbers for India; yet, this still comes out to 23 per lakh people [ 3 ]. With the current data on significant mortality in tuberculosis, it becomes important to assess the determinants for the same. One of the factors was an improper assessment of the severity of the disease at the time of diagnosis. In a study by Shewade et al., out of 3,010 cases of tuberculosis, 1,529 (50.8%) were screened at the time of diagnosis or notification, of whom 537 (35.1%) had a high risk of severe illness. Further, when comparing individuals without a high risk of severe illness (3.8%) to those with a high risk of severe illness (8.9%) at diagnosis, the incidence of early fatalities was considerably greater. Furthermore, early mortality was highest in the first two weeks of the disease and was highly correlated with a high risk of severe illness upon diagnosis or notification [ 4 ]. The burden of severe disease at notification or diagnosis and the viability of gathering these data in standard program settings are topics that have received little attention in the literature. The National Tuberculosis Elimination Program guidelines for 2021 in India, unequivocally advise conducting a severity evaluation as soon as feasible upon diagnosis and referring patients for inpatient care if they are very sick. However, this guideline needs clinical, laboratory, and radiographic examinations, which might be difficult in peripheral health institutions [ 4 ]. To address the issues in severity evaluation in the National Tuberculosis Elimination Program guidelines for 2021, a 'Differentiated TB Care Model' was introduced, which was basically a simplified approach with criteria for hospitalization [ 5 ]. As recommended from 16 districts of Karnataka in the findings of the study by Shewade et al., an easy and quick method for early assessment of the severity of illness based on vital signs, body mass index (BMI), and the inability to stand without assistance are simple ways to screen for the severity of the illness. These indicators are well-known risk factors for death that are straightforward to evaluate and understand. Patients who exhibit any of these signs may be directed to more advanced facilities, like a tertiary care hospital, for inpatient care and an extensive clinical evaluation [ 4 ]. A similar initiative is planned for Delhi, involving all 25 chest clinics or district tuberculosis centers. Beginning January 1, 2024, all adults (more than or equal to 15 years of age) who will be diagnosed with tuberculosis, both drug-sensitive and drug-resistant, notified by public health institutes will be triaged. This activity will be started as a pilot project on December 15, 2023. This triage will be done by assessing five indicators (Table 1 ). This triaging will be done as soon as possible at the diagnosing facility without even waiting for a formal notification of the patient or at the next earliest opportunity (at the home visit or start of the treatment or at the time of baseline investigations for tuberculosis). Two categories for severity of illness are defined based on the triage: either a triage positive or a triage negative. All the cases marked as triage positive will be registered in the notification registers at the health facility diagnosing the case and referred immediately to the designated tertiary care hospital in the national capital of India. Two hospitals, the National Institute of Tuberculosis and Respiratory Diseases and the Rajan Babu Institute of Pulmonary Medicine and Tuberculosis, are part of this initiative where these referrals will be made and a specific number of beds with trained staff designated to take care of these severe cases are planned. The five indicators mentioned in Table 1 would help in detecting three conditions at the time of diagnosis (Table 2 ). In India and other high-burden nations, there is a paucity of data on screening for severe illness on diagnosis or notification and the correlation between severe illness and early tuberculosis death in programmatic settings [ 4 ]. In a large study from Andhra Pradesh by Jonnalagada et al. on 8,240 tuberculosis patients, 50% of early deaths were reported within the first four weeks of treatment [ 6 ]. This data rose to 75% in the study from Karnataka [ 4 ]. Besides, another study from the Indian state of Tamil Nadu (during April-June 2022) implemented a differentiated care strategy called Tamil Nadu-Kasanoi Erappila Thittam (TN-KET) for all adults aged 15 years and older with drug-susceptible tuberculosis notified by public health facilities. Following notification of 14,961 TB patients, 11,599 (78%) underwent triage. Out of the people who were triaged, 1,509 (13%) had a high risk of developing a severe illness; of these, 1,128 (75%) underwent a thorough clinical evaluation in a nodal inpatient care health facility. In this study, 909 (92%) out of 993 confirmed as severely ill, were admitted, with 8% unfavorable admission outcomes, including 4% deaths [ 7 ]. To conclude, initial screening for severe sickness as an alternative strategy to reduce tuberculosis fatalities has existed since 2021, but a quicker and more targeted approach considering the issues at the grassroots level is warranted. This initiative in Delhi would be a remarkable step toward reducing mortality due to tuberculosis and achieving the desired aims of the National Strategic Plan 2017-2025. Additionally, this is the first such initiative where both drug-sensitive and drug-resistant tuberculosis cases will be triaged.
Acknowledge the Tamil Nadu Kasanoi Erappila Thittam (TN-KET) team.
CC BY
no
2024-01-15 23:42:02
Cureus.; 15(12):e50550
oa_package/4e/0e/PMC10787899.tar.gz
PMC10787900
38222222
Introduction A ureterocele is a cystic dilatation of the distal submucosal ureter, frequently located within the bladder [ 1 ]. The development of tumors within the ureterocele is very rare; there are few reports in the literature, and the majority are of urothelial carcinomas [ 2 ]. Urothelial cancer is developed countries' sixth most common cancer [ 3 ]. Bladder cancer (BC) accounts for 90-95%, upper tract urothelial cancer (UTUC) accounts for 5-10% [ 3 ], and urethral urothelial carcinoma is rare, less than 1% of all genitourinary malignancies [ 4 ]. We report a case of a man with urothelial carcinoma of a ureterocele and review the literature, especially concerning diagnosis and treatment.
Discussion A ureterocele is a congenital abnormality with an incidence as high as 1 in 500 in autopsy and a female preponderance four to six times greater than in men [ 5 ]. It can appear in single or duplex systems associated with the upper pole system, as in the Weigert-Meyer rule [ 1 ]. The insertion can be orthotopic in the bladder or ectopic, more frequently on the bladder neck or urethra, among others in the ectopic pathway [ 1 ]. In adults, it is usually asymptomatic. However, when symptomatic, it is associated with upper urinary tract obstruction. Other symptoms are bladder neck obstruction, calculus formation, and recurrent infections [ 6 ]. Their walls are made of two layers of urothelium (bladder and ureter) with muscle and collagen in the middle [ 5 ]. Perego et al. first described a tumor on a ureterocele in 1974 [ 7 ]. Few cases have been reported since, and Astigueta et a. in 2016 published 10 cases [ 2 ], and since then, only three more cases have been reported [ 8 - 10 ]. We did a table summarizing the cases reported in the literature, including ours (Table 1 ). The main clinical presentation is hematuria, followed by lower urinary tract symptoms (LUTS) and dysuria [ 2 ]. Other symptoms, such as lower back and supra-pubic pain, were sporadic [ 2 ]. Many imaging studies are used: intravenous urography (IVU), ultrasound (US), computed tomography (CT) urography, and MRI. The IVU is a historical exam with the typical alteration of the cobra-head sign in simple ureteroceles, with a distal ureter dilated in the bladder, surrounded by a thin and regular lucent line [ 18 ]. If there is a thickening or irregularity, it may suggest a pseudoureterocele [ 2 ]. A pseudoureterocele may be due to a tumor or edema from a stone in the ureterocele [ 2 ]. The ultrasound detects ureteroceles with the typical image of a "cyst within a cyst" located in the posterior lateral wall of the bladder. A bladder tumor is an echogenic, fixed mass in the bladder wall without acoustic shadow. A tumor inside should be excluded if these alterations appear in a ureterocele [ 6 ]. The CT urography detects the same alterations of the IVU with greater detail and can also show enhancement with contrast and exclude extravesical disease [ 2 ]. The MRI is usually not used, but it should detect the same findings as a CT urography [ 8 ]. Cystoscopy is used to confirm the diagnosis [ 2 , 10 ]. However, in some reports, the cystoscopic appearance is similar to a simple ureterocele if the tumor is completely inside the ureterocele [ 8 - 9 ], as in our case. The management varies, and there are no guidelines. Most reports used the same guidelines as in BC [ 10 ]. A TUR is useful for unroofing the ureterocele and removing the tumor for histological evaluation [ 2 , 5 - 8 ]. A more definitive and radical treatment was conducted in some reports. The most commonly used was a ureterocele resection and distal urethrectomy with ureteral reimplantation [ 2 , 7 ]. Other surgical options were RNU and even radical cystectomy [ 2 ]. Non-surgical options were adjuvant treatment with bladder instillations with bacillus Calmette-Guérin (BCG) [ 9 , 10 ], while other patients remained on surveillance only [ 2 , 8 ]. The histological evaluation was essential in most of these decisions, in which the tumor was non-invasive in the majority. In our report, since the tumor evolved into the distal ureter, we followed the guidelines for UTUC management with a single post-operative bladder instillation of MMC [ 19 ]. The follow-up scheme adopted was similar to UTUC with cystoscopy and URS [ 19 ], which is suggested by some authors [ 10 ]. Although the recurrence was a low-grade lesion on biopsy, the previous tumor was classified as a high-risk UTUC, and therefore, RNU was proposed. The RNU identified a single 15 mm low-grade pTa urothelial tumor with a clear surgical margin. There is no long-term follow-up data from the reports, so there are no recommendations concerning the best treatment option.
Conclusions Tumors in ureteroceles are very rare, and there are no guidelines for diagnosis, management, and follow-up. The diagnosis is challenging with the help of imaging studies and cystoscopy, although a normal exam may not exclude this diagnosis. Management varies according to histological results and imaging studies, from TUR and ureterocele resection to RNU. The follow-up data is unavailable; therefore, no evidence of the long-term outcomes is available in the literature.
Urothelial carcinoma on a ureterocele is extremely rare in the literature, and few case reports have been reported. There are no guidelines for diagnosis and management, and current practice is extrapolated from bladder and upper urothelial tract carcinoma. We present a case from a 61-year-old man with urothelial carcinoma on a ureterocele treated with ureterocele resection, distal urethrectomy, and reimplantation on the bladder. We also review the literature concerning diagnostic approaches and management.
Case presentation A 61-year-old caucasian man presented in the outpatient clinic with episodes of macroscopic hematuria for one year. The patient had no relevant past medical history and had no other complaints, and the physical examination was innocent. On blood analyses, there were no relevant alterations. On an excretory CT, the right collecting system had incomplete duplicity with hydronephrosis of the upper pole collecting system and merged in a single meatus with a ureterocele. The ureterocele presented a thickening of the wall protruding inside, extending to the distal ureter with 30/6 mm with contrast enhancement, compatible with a tumor inside the ureterocele (Figure 1 ). No other lesions were found. On cystoscopy, a single right meatus was found on a ureterocele; no papillary lesions came from that orifice, and clear urine was extracted during the exam; the left meatus and the bladder had no other alterations. The patient underwent resection of the ureterocele with distal urethrectomy of both duplicated systems and common-sheath reimplantation on the bladder dome with the Psoas-Hitch technique (Figure 2 ). The pathology report identified a high-grade urothelial carcinoma pT1 with clear surgical margins (Figure 3 ). An intravesical instillation of mitomycin C (MMC) was made post-operatively on the eighth day, extrapolating data from the UTUC management. A ureterorenoscopy (URS) was made at the three months of follow-up without any evidence of recurrence. At six months of follow-up, the patient was asymptomatic but presented on cystoscopy with bladder and ureteral recurrence (visible from the reimplanted ureter). The transurethral resection (TUR) of the bladder revealed a low-grade pTa urothelial bladder tumor, and a URS identified a ureteral tumor on the upper duplex system near the anastomosis, which was biopsied and identified as a low-grade ureteral tumor. The patient was submitted to a right radical nephroureterectomy (RNU) with the identification of a 15 mm low-grade pTa urothelial tumor with a clear surgical margin. An intravesical instillation of MMC was made post-operatively on the fifth day. The patient is on follow-up without any complaint or evidence of recurrence or progression.
CC BY
no
2024-01-15 23:42:02
Cureus.; 15(12):e50549
oa_package/d9/20/PMC10787900.tar.gz
PMC10787901
38222195
Introduction Iron accumulation in the brain, especially in the basal ganglia, frequently leads to neurodegeneration [ 1 ]. Mutations in C19orf12 have recently been identified in individuals with a diagnosis of neurodegeneration with brain iron accumulation (NBIA) [ 2 ]. The C19orf12 gene, which is localized in mitochondria and the endoplasmic reticulum, encodes a mitochondrial membrane protein. Mutations in this gene have been hypothesized to lead to the disruption of lipid homeostasis within the mitochondria, resulting in the onset of mitochondrial membrane protein-related neurodegeneration (MPAN) [ 1 - 3 ]. Patients diagnosed with MPAN typically experience gait disturbance as a result of lower limb spasticity and dystonia [ 4 ]. Other symptoms include parkinsonism, spastic paresis, dysarthria, cerebellar ataxia, behavioral problems, dementia, psychomotor delay, peripheral neuropathy, visual changes such as optic atrophy, and bowel/bladder incontinence [ 4 - 6 ]. MPAN cases exhibit more variability than other NBIA cases, however, in terms of age of onset (between three and 30 years) and rate of progression [ 7 ]. There is no rehabilitation protocol in the literature regarding exercise practices for MPAN syndrome. This study aims to reveal the effect of an 18-week therapeutic intervention on balance, coordination, and postural control in MPAN syndrome.
Discussion This is the first clinical case report to investigate the effects of a physiotherapy and rehabilitation program on body structure, function, and levels of activity and participation in a child with MPAN. The syndrome presents a variety of symptoms, ranging from gait disturbances to vision problems. Implementing a multidisciplinary approach to rehabilitation is therefore crucial in managing the syndrome. Prevalent physical disorders in cases of MPAN include spasticity, ataxia, muscle weakness, and associated gait issues. In our two cases, there was a crouch gait present due to the shortening of the hamstrings and gastro-soleus [ 4 - 6 ]. Initially, the disease appeared to have a spastic diplegic appearance. It was generally thought by pediatricians at the time that the children may have had cerebral palsy prior to the gene screening. However, based on the electroencephalogram reports and the bilateral involvement of the substantia nigra, red nucleus, and dentate nucleus in both siblings, as well as putamen involvement in the older sibling and globus pallidus in the younger sibling, the presence of brainstem involvement in our patients signifies the differentiation between MPAN disease and classical spastic diplegic cerebral palsy. One of the important features of the disease is that it can occur in a wide age range (from three to 30 years), as reported in the literature [ 7 ]. In our cases, diseases such as Duchenne muscular dystrophy, cerebral palsy, and hereditary spastic paraplegia were suspected until the older sibling reached the age of seven. However, the presence of the condition was confirmed through genetic screening. The rehabilitation program was developed to address the main issues of the cases. Both participants were attending school and aimed for academic success. Moreover, the families’ expectation was for their children to lead as independent a life as possible. Particularly in terms of joining friends at school, problems arose due to balance and coordination difficulties. In developing the rehabilitation program, we took into account other pediatric neurological studies and the typical symptoms experienced by our patients [ 16 - 18 ]. The significance of maintaining trunk control for balance is crucial for both adults and children [ 19 ]. Trunk control is an important milestone that supports all rehabilitation processes, particularly in children affected by neuromuscular disorders such as cerebral palsy [ 20 ]. Postural control in children with cerebral palsy worsens in comparison to their typically developing peers, potentially due to delayed and impaired development of neural motor control mechanisms and common secondary musculoskeletal abnormalities. After 18 weeks, we have made progress in parameters relating to balance and independence, a result of implementing a program focused on balance and coordination development. Although not evaluated in the study, the family reported a decrease in falls and improved running speed for both siblings following the treatment. We believe that this is closely linked to the use of balance exercises that involve trunk control. The study had several limitations. The rehabilitation facility was used by patients and physiotherapists concurrently. This made it challenging to hold the attention of patients during specific exercises. Additionally, varying motivations among the patients represented a significant limitation. For instance, the older brother was resistant to implementing the rehabilitation program, despite having more severe issues. Taking breaks during exercise is vital for this reason. This study is a case report, and further testing of the methods is necessary in evidence-based, randomized studies involving multiple participants.
Conclusions The purpose of the investigation was to evaluate the impact of an 18-week rehabilitation program, comprising balance and coordination exercises, on two siblings diagnosed with neurodegeneration caused by iron accumulation in the brain. Comprehensive understanding of these conditions is imperative to determine the most appropriate therapy. Hence, balance and coordination-based exercise regimes have become a vital element of clinical intervention for patients affected by MPAN syndrome. It is crucial that rehabilitation programs are tailored according to the fundamental requirements of the patients or their families. The 18-week rehabilitation program for MPAN syndrome in this study has beneficial effects on muscle strength, balance parameters, and independence.
This case report reports the effects of an 18-week physiotherapy program in children with mitochondrial membrane protein-associated neurodegeneration (MPAN). The study involved two brothers, aged 11 and 12, who had been diagnosed with MPAN. The physiotherapy program was divided into three phases and consisted of 18 weeks of training with a pediatric physiotherapist, including balance, coordination, and strengthening exercises. Muscle strength was assessed using pediatric manual muscle testing, functional balance using the Pediatric Berg Balance Test (PBBT), static balance using the Single-Leg Stance Test, dynamic balance using the Functional Reach Test, postural control using the 5-Time Sit-to-Stand Test, and independence using the Functional Independence Measure for Children (WeeFIM). Positive changes were observed in muscle strength, balance, and independence. After Phase I, PBBT scores (younger sibling +4, 8.1%; older +3, 6.8%) were higher than the minimal clinically important difference (MCID=3.66-5.83). After Phase III, although the PBBT scores improved (younger +2, 4.05%; older +1, 2.3%), the older sibling’s score was not higher than the MCID. Thus, the two children showed visible improvements in both body structure and function, as well as activity and participation levels.
Case presentation Two brothers with MPAN, aged 11 and 12, who were born at term with birth weights of 2,000 g (36 weeks) and 3,500 g (40 weeks), respectively, presented to the clinic with a complaint of falling. Muscle weakness, loss of coordination and balance, and gait disturbance were observed during the initial assessment (Table 1 ). The mother experienced her first pregnancy at the age of 25 (150 cm, 55 kg). Jaundice or trauma was not encountered in either child post-delivery. Apart from the premature birth of the younger sibling, no substantial birth difficulties were observed. The younger sibling did not demonstrate any motor developmental delays and acquired head control at three months and sitting at five months. The older sibling began walking at 14 months, whereas the younger sibling started walking at just nine months, without crawling first. At the age of five years, the older brother began experiencing falls, prompting his mother to seek medical attention. The younger brother underwent a check-up due to concerns raised by his teacher regarding vision and reading difficulties during his second year of primary school. When the older brother was seven years old, a physical therapist was consulted due to frequent falls and pes planus. Genetic screening was performed on the entire family, resulting in the detection of related gene damage in both siblings. Vitamin supplements were administered, and rehabilitation was recommended. Assessments Body Structures and Functions Assessments were based on the International Classification of Functioning, Disability and Health Child and Youth Version (ICF-CY). Hip extensor and abdominal muscle strength was evaluated by manual muscle testing (MMT) and scored from 0 to 5. Standard positions and procedures were used [ 8 ]. In pediatric research, instrumental maximum voluntary contraction should be preferred over MMT scoring [ 9 ]. Quantitative methods, however, have the disadvantage of requiring a specific tool rather than a clinician’s hands [ 10 ]. Functional balance was evaluated with the Pediatric Berg Balance Test (PBBT). Reliability testing for PBBT, performed with a sample of 20 children aged five to 15 years with mild to moderate motor impairments, showed good test-retest reliability (intraclass correlation coefficient [ICC] = 0.998) and good interrater reliability (ICC = 0.997) [ 11 ]. Validity testing for PBBT, performed with a sample of 30 children aged four to 10 years with spastic cerebral palsy in Gross Motor Function Classification System (GMFCS) Levels I-III, showed a strong correlation between the Pediatric Balance Scale and the self-care (r = 0.73, p < 0.001) and mobility (r = 0.82, p < 0.001) dimensions of the Pediatric Disability Evaluation Inventory [ 12 ]. After completing the 12-week program encompassing phases I and II, postural control and static and dynamic balance assessments were included. We assessed static balance using the single-leg stance test and found a standard error of measurement (SEM) of 8.7 s and a minimum detectable change (MDC) of 24.1 s at a 95% confidence level. We also assessed dynamic balance using the Functional Reach Test and postural control using five repetitions of the Sit-to-Stand Test. Activity and Participation Participation in activities of daily living (ADL) was evaluated with the Functional Independence Measure for Children (WeeFIM). WeeFIM is easy and efficient to administer and is a valuable predictor of function in children with disabilities. WeeFIM includes 18 items with a 20-minute administration time. As WeeFIM is shorter and quicker to administer, it lends itself to assessment of functional outcome in pediatric rehabilitation. Higher scores represent more independence in participation [ 13 ]. Test-retest for the six domains range from r = 0.83 to r = 0.99. Internal consistency (Cronbach’s alpha, or ICC) of the WeeFIM motor and cognitive scales was high (>0.90) and consistent for individual use. Interrater reliability was excellent, with ICC values of 0.98 and 0.93 for the motor and cognitive scales, respectively. Therapeutic management The initial phase of the rehabilitation program was structured according to the patients’ main problems. The primary concern was the impairment of balance during regular daily activities, followed by loss of muscle strength in the abdominals and hip extensors. It was identified that implementing exercises aimed at improving balance and coordination while walking was an important step forward [ 14 ]. Following these objectives, the Phase I scheme was developed in accordance with Table 2 . Results of Phase I Following the initial phase, which involved muscle strengthening and balance training, the children demonstrated improvement in several measures, including the PBBT, WeeFIM scores, and the strength of their abdominals and hip muscles. Notably, the increase in the PBBT score was significantly greater than the MCID reported in the literature for children with cerebral palsy [ 15 ]. In the older brother, there was an increase in the ability to stand on one leg (PBBT score from 1 to 2), complete a 360-degree turn (from 3 to 4), and reach forward while standing (from 3 to 4). The younger brother showed an improvement in unsupported standing with one foot forward position (from 0 to 4). The observed 2-point increase in WeeFIM scores for both siblings was attributed to an increase in stair-climbing activity. For detailed findings, please refer to Table 3 . Phase II (six to 12 weeks/at home) initiative was based on the caregiver's ability and the suitability of home conditions. Exercises were prescribed for home-based practice, as stiffness of the muscles around the feet and ankles in the morning and balance and coordination irregularities were observed. Following clinical conditions and legal regulations, a six-week hiatus was taken, and the objective was to preserve the previous progress through home-based exercises during this period. The home exercise regimen consists of straight leg lifts, side leg lifts, wrist flexor and gastro-soleus muscle stretches, standing up and sitting down repeatedly, and 20 minutes of cycling on an ergometer and light jogging every day. The younger sibling increased functional independence by 9 points, while the older sibling experienced an increase of 1 point before Phase III. Both siblings experienced a 2-point increase in balance; however, this improvement was below the MCID. As muscle strength was already gained to the desired extent during Phase I and there was no observed loss in relevant muscle strength before Phase III, strengthening exercises were not added to the rehabilitation program. Phase III was designed according to Table 4 . Results of Phase III The assessment conducted after Phase III showed that progress had been made in all measures except for the older sibling’s ability to stand on one leg and complete the 5-Time Sit-to-Stand Test. Among these achievements, only the dynamic balance metric for the younger sibling exceeded the minimum clinical significance threshold (Table 5 ).
CC BY
no
2024-01-15 23:42:02
Cureus.; 15(12):e50540
oa_package/c5/4c/PMC10787901.tar.gz
PMC10787902
38222127
Introduction and background Helicobacter pylori peptic ulcers may be treated with one of less than ten medications, including metronidazole, amoxicillin, clarithromycin, furazolidone, levofloxacin, and tetracycline. Antimicrobial resistance may be avoided and effective eradication is achieved by combining these antibiotics with bismuth salts and proton pump inhibitors (PPIs) [ 1 , 2 ]. The World Health Organisation has designated the study and development of novel medicines for H. pylori species that are resistant to clarithromycin as a "high priority'' [ 3 ]. Primary antibiotic resistance is the problem caused by poor adherence to medication, which may result in treatment failure and subsequent resistance. Clarithromycin and metronidazole resistance is a worldwide crisis that must be addressed because chronic H. pylori infection is associated with gastric adenocarcinoma and localized B-cell lymphoma of the stomach (formerly known as gastric MALT lymphomas) [ 4 , 5 , 6 , 7 ]. H. pylori infections have been linked to numerous extragastric symptoms, such as ischemic heart disease and refractory anemia, according to recent studies [ 8 , 9 ]. Despite a global drop in the percentage of H. pylori infections successfully treated, antibiotic resistance, particularly to highly effective broad-spectrum antibiotics, has been rapidly increasing [ 6 , 10 , 11 ]. Tetracycline and amoxicillin have lower rates of resistance [ 6 ]. The World Health Organisation published a study in 2018 on the worldwide spread of antibiotic resistance, and they discovered that resistance to levofloxacin, metronidazole, and clarithromycin had reached 15% worldwide. However, metronidazole and clarithromycin secondary resistance rates in certain areas were from 30 to 70%. Most often owing to initial clarithromycin resistance, eradication efforts with the most frequently used first-line regimens of clarithromycin, amoxicillin, and/or metronidazole failed more than 20-30% of the time [ 12 ]. Despite the bismuth quadruple regimen (PPI, bismuth salts, and two antibiotics) being successful as a first- or second-line therapy, adverse effects have been recorded in 16-33% of instances, resulting in treatment cessation and failure [ 13 ]. With high levels of resistance in H. pylori infections, both first and second-line treatment options have shown minimal elimination of the organism, making it a particularly challenging situation with limited alternative options. Primarily, combinations of antibiotics are being used for this purpose. Rifabutin, a rifamycin derivative, has emerged as an alternative to rifampicin in treating Mycobacterium tuberculosis. Its effect in the prophylaxis for Mycobacterium avium complex in immunocompromised patients is well known [ 14 , 15 ]. In recent years, novel regimens incorporating Rifabutin have been extensively studied for their efficacy in eliminating H. pylori, showing promising results even after multiple treatment failures [ 2 , 15 , 16 ]. Furthermore, rifabutin resistance in vivo is rare, and the risk can be significantly reduced when Rifabutin is administered in combination with other antibiotics [ 15 ]. Previous trials have used Rifabutin in various combinations, doses, and treatment durations. However, comprehensive reviews of multiple regimens still need to be included. This review aims to determine the most effective rifabutin-based regimen by considering its benefits and potential drawbacks for treating this infection.
Conclusions In conclusion, therapies that can successfully eliminate this infection in the first attempt must be considered the gold standard when prescribing medications. However, the rising global resistance to antibiotics like amoxicillin, metronidazole, clarithromycin, tetracycline, and levofloxacin raises concerns about their efficacy. Due to this, there is an increased need for developing efficient rescue regimens that can successfully eradicate this organism and prevent treatment failure. Our systematic review provides evidence showing that rifabutin regimens can be used to eliminate this infection successfully. Nearly all studies have shown that rifabutin-based regimens exhibited the lowest resistance rates compared to other drugs included in the therapy. It is noteworthy that rifabutin therapy has shown efficacy even in individuals with initial resistance to levofloxacin, clarithromycin, and metronidazole. Without relying on bacterial culture, it may be prudent to consider the empirical use of rifabutin as ''rescue therapy" in circumstances where these antibiotics have failed to work. Therefore, rifabutin-containing regimens should only be used as a fourth or subsequent therapy if early eradication regimens using other first-line antibiotics such as metronidazole, amoxicillin, clarithromycin, tetracycline, and levofloxacin fail. The microbial resistance, possible adverse effects, and the availability and efficacy of alternative drugs should all be carefully considered before choosing rifabutin as a first-line therapy choice for H. pylori infection.
Helicobacter pylori has been reported as a health problem worldwide, affecting a sizable portion of people. Peptic ulcers, gastric cancer, and various extra gastric conditions are associated with this bacterium. The rampant overprescribing of antibiotics has led to the emergence of H. pylori strains resistant to multiple antibiotics, causing a decline in the effectiveness of current treatments. Recently, there has been growing interest in researching alternative treatment options for H. pylori infections that do not respond to initial therapy. Rifabutin, a rifamycin derivative initially designed for tuberculosis treatment and preventing Mycobacterium avium complex infection, has gained attention as a potential rescue medication. It has shown efficacy against H. pylori and the potential to eradicate the bacterium when combined with other antibiotics. This systematic review article focuses on using rifabutin-based regimens as a treatment option after initial treatments have failed. The authors screened literature published in the last five years, between 2017 and 2022, across various search engines and closely examined relevant studies following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) criteria. The search covered a variety of electronic databases and focused on H. pylori gastritis, rifabutin-based treatment plans, and in vivo investigations in healthy individuals. The comprehensive review provides convincing evidence that rifabutin-based regimens are effective rescue treatments for H. pylori infections. Multiple studies in various areas consistently demonstrated high eradication rates, ranging from 70% to 90%, when rifabutin-containing regimens were used. The analysis found that only a tiny percentage of H. pylori strains (1%) were resistant to rifabutin therapy, further supporting the viability of Rifabutin as an alternative when other antibiotics failed to eradicate H. pylori. The cost of Rifabutin is a significant factor that may limit its accessibility, particularly in resource-constrained settings where H. pylori infection is common. Moreover, the potential side effects of Rifabutin, such as hematological problems, rashes, and digestive issues, need to be considered. However, these side effects are typically manageable and can be reduced by combining Rifabutin with other antibiotics. In conclusion, this systematic review provides evidence supporting the effectiveness of regimens derived from Rifabutin in eliminating H. pylori infections after initial therapy failure. Due to the observation that Rifabutin effectively eradicates resistant H. pylori infections, it can be considered a suitable choice for rescue therapy. Rifabutin-containing regimens should be reserved as fourth- or later-line therapy options, considering economic factors, the risk of microbial resistance, potential side effects, and the availability of alternative medications. Future research should focus on optimizing rifabutin-based regimens and investigating combination therapies that have better H. pylori eradication rates while also addressing the problem of resistant strains.
Review Methods The systematic review was done in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. A rigorous search strategy was developed to test this hypothesis and identify relevant literature. A comprehensive search of the various search engines and databases was done to include relevant literature. A PubMed search was conducted using the Mesh terminology ''Rifabutin or N'-acetyl rifabutin or Mycobutin or ansatipine or ansamycin or LM 427, H pylori or helicobacter pylori, Therapy or therapeutic* or regimen* or protocol*, ''Rifabutin/administration and dosage''[Mesh] OR ''Rifabutin/adverse effects''[Mesh] OR ''Rifabutin/metabolism''[Mesh] OR ''Rifabutin/pharmacokinetics''[Mesh] OR ''Rifabutin/therapeutic use''[Mesh] OR ''Rifabutin/toxicity''[Mesh]. The MeSH methods are listed in Table 1 . Data extracted via the PubMed search strategy is shown in Table 2 . The authors wrote inclusion criteria, and articles were then chosen. This criterion was laid down based on various factors, including those related to H. pylori gastritis, papers discussing the research subject, publications published in English, and other specific criteria. These criteria included selecting papers with full-text articles, studies involving in vivo investigations, papers focusing on patients without comorbidities, publications published within the last five years (2016-2022), papers specifically addressing the treatment of Helicobacter gastritis, and studies examining the use of rifabutin for treating this infection, among others. We thus included randomized control trials and observational studies in this systematic review. Articles published in languages other than English, those not related to the research topic, those without full-text availability, and those based solely on in vitro studies were not included in the analysis. Patients with other gastrointestinal (GIT) diseases and comorbidities, MALTomas, and gastric adenocarcinomas were also not included. PRISMA chart is given in Figure 1 . The authors independently reviewed the research design and the possibility of bias in selection or publication. Any disagreements in the reviews of the writers were settled by consensus. A comprehensive quality assessment of the retrieved publications was conducted using the appropriate methodologies for each type of study. The results of critical appraisal using the NewCastle-Ottawa Quality Assessment Tool evaluation for case-control and cohort studies and the Cochrane Risk of Bias Tool for RCT are given in Table 3 . Result Fifteen studies were finalized following multiple readings of the full text of the papers and discussions among co-authors. Study characteristics are given in Table 4 . Erick A. Argueta did research in the USA to see how new antibiotic resistance would affect efforts to eradicate H. pylori infection [ 17 ]. In a study including 101 patients, the bismuth quadruple (BQ) regimen was used. This treatment plan includes the use of bismuth, tetracycline, metronidazole, and PPIs. Eighty-six patients were given the triple regimen, which included PPIs, clarithromycin, and either amoxicillin, metronidazole, or rifabutin. The eradication rate for Cohort 2 (treated with the triple regimen) was 60.5%, whereas the eradication rate for Cohort 1 (treated with the BQ regimen) was 88.1%. The research indicated that of the antibiotics tested, metronidazole had the greatest resistance rate (33.3%), followed by clarithromycin (30%), levofloxacin (29.6%), amoxicillin (1.0%), rifabutin (0.5%), and tetracycline (0.5%). Not only that but multidrug resistance (MDR) was found in 65.6% of bacterial isolates. Youn I Choi conducted a study in Korea to identify potential candidates for a rescue regimen in cases of antibiotic-resistant H. pylori infections [ 18 ]. Nine patients showed no elimination when just one antibiotic was used, while 22 were resistant to several. Interestingly, none of the subjects showed resistance to rifabutin or furazolidone. However, this gastric infection was completely eradicated in every patient in both cohorts, showing a 100% success rate. The study also listed the rates of resistance for several antibiotics, with clarithromycin having the highest rate at 71.1%, followed by metronidazole at 67.7%, levofloxacin at 41.9%, and amoxicillin at 22.6%. Rifabutin and furazolidone showed no signs of resistance. In a Spanish study, susceptibility testing was used to identify treatment plans for 68 individuals with dual antibiotic resistance [ 19 ]. OAL (omeprazole, amoxicillin, levofloxacin) was administered to 43 patients, OAC (omeprazole, amoxicillin, clarithromycin) to 12 patients, and OAM (omeprazole, amoxicillin, metronidazole) to 13 patients for those with dual-resistant infections. The OAR (omeprazole, amoxicillin, rifabutin) regimen was used to treat 12 patients with triple antibiotic resistance. The study reported high eradication rates for the various regimens in both cohorts. Notably, dual antibiotic-resistant patients had a resistance rate of less than 10% following medication, indicating effective eradication. However, the resistance rate in the triple antibiotic-resistant group receiving the rifabutin-based regimen was 41.7%, indicating a lower level of effectiveness in this cohort. David Y. Graham looked into the Rifabutin-Based Triple Therapy (RHB-105) to see if it was beneficial in treating H. pylori infection [ 15 ]. Two hundred and twenty-eight patients were given the RHB-105 regimen (amoxicillin, omeprazole, and Rifabutin) while 227 patients were given an active comparator regimen (amoxicillin and omeprazole). The study used an intention-to-treat (ITT) methodology to evaluate the eradication rates. The RHB-105 regimen had an eradication percentage of 83.8%, while the active comparator had a rate of 57.7%. According to these results, RHB-105 significantly outperformed the active comparator in eliminating H. pylori. Rifabutin showed a lower resistance rate than the active comparator, highlighting the potential benefit of utilizing it to treat infections with H. pylori. The effects of the EAR treatment regimen (esomeprazole, amoxicillin, Rifabutin) were investigated by Kazumi Inokuchi using a historical control group of 12 patients [ 2 ]. Vonoprazan, amoxicillin, and rifabutin (VAR) were used to treat 57 different individuals. The research assessed the efficacy of two strategies for treating this infection, contrasting the ''intention-to-treat'' (ITT) approach with the ''per protocol'' (PP) approach. The case group had a greater rate of H. pylori eradication after receiving the triple therapy based on low-dose rifabutin. The 10-day regimen had a resistance rate of 16.7%, whereas the seven-day regimen had a resistance rate of 8.8%, according to the study's analysis of rifabutin regimen resistance rates. Triple rifabutin (RHB-105) was studied by Ira N. Klafus in the USA to determine its efficacy in treating this illness [ 21 ]. Seventy-seven people were given the RHB-105 regimen (which contained amoxicillin, omeprazole, and rifabutin), and 41 people were given a placebo. In both the ITT and PP analyses, the RHB-105 regimen showed statistically significant eradication. The resistance rate to the RHB-105 regimen was 10.6%, which could affect the treatment's efficacy. Chia Jung Kuo sought to understand better the clinical difficulties associated with multidrug-resistant infections. The trend of resistance being faced by the population was evaluated, with clarithromycin (92.7%) and levofloxacin (85.4%) having the highest resistance rates [ 22 ]. Rifabutin showed a resistance rate of 29.3%. Notably, the study disclosed neither the eradication rates attained nor the treatment regimens utilized. Instead, it focused on describing the MDR patterns and H. pylori's antibiotic susceptibility in the research population. A study by Chang Ming Lee had 323 participants divided into two cohorts [ 23 ]. The moxifloxacin-rifabutin triple therapy (MRT) treatment regimen was administered to 71 individuals who were unsuccessful with the first-line clarithromycin regimen (Cohort 1), while 252 patients with the same condition received the BQ treatment regimen (Cohort 2). The study found that eradication rates could differ depending on the treatment plan. MRT and BQ regimens showed effectiveness as second-line treatments for peptic ulcer cases unresponsive to the first-line clarithromycin regimen. Miftahussurur conducted research to evaluate different H. pylori infection treatment plans in Indonesian areas where metronidazole and levofloxacin resistance are common [ 24 ]. The study evaluated the susceptibility and resistance of five different antibiotics, furazolidone, sitafloxacin, garenoxacin, rifaximin, and rififabutin, to different strains of H. pylori bacteria. Among the participants, 61 out of 105 (58.1%) were responsive to each of the five antibiotics tested, indicating that these people may benefit from different eradication regimens that include these antibiotics. Notably, none of the organisms resisted sitafloxacin, furazolidone, or rifabutin. However, the study found various degrees of resistance to rifaximin and garenoxacin. Notably, 40 out of 105 subjects (38.9%) and seven out of 105 people (6.7%) were resistant to rifaximin and garenoxacin, respectively. This shows that these drugs are highly resistant in the Indonesian regions under study. The results do, however, indicate that adding furazolidone, rifabutin, and sitafloxacin to regimens can be successful in regions with great resistance to some antibiotics like levofloxacin and metronidazole. Another study took place in the Southeast Asian countries of Bangladesh and Nepal [ 25 ]. The study assessed the regions of Nepal where there was high antibiotic resistance. Garenoxacin resistance was seen with the tested antibiotics in 12 out of 42 people (28.6%), whereas sitafloxacin resistance was seen in two out of 42 participants (4.8%). Furazolidone and Rifabutin, however, did not show any signs of resistance. Another antibiotic, rifaximin, was tested, and 22 out of 42 people (52.4%) showed resistance. A similar evaluation was carried out in Bangladesh, demonstrating the antibiotic resistance patterns in these strains. Not more than one of the 56 subjects (1.8%) demonstrated resistance to sitafloxacin, compared to 29 out of 56 patients (51.8%) who showed resistance to garenoxacin. No rifabutin or furazolidone resistance was found, similar to Nepal. Resistance to rifaximin was seen in 36 out of 56 subjects (64.3%). Importantly, none of the tested strains in either nation showed rifabutin or furazolidone resistance. This shows that despite the significant resistance seen for other regularly used antibiotics such as garenoxacin, sitafloxacin, and rifaximin, these two antibiotics could be possible candidates for effective therapy regimens against H. pylori infection in Bangladesh and Nepal. Miftahussurur conducted a study in the Dominican Republic to find substitute antibiotics to deal with increased resistance to levofloxacin and metronidazole in treating this infection [ 26 ]. The distribution of strains of this organism in the Dominican Republic that were resistant to several antibiotics was evaluated. Rifaximin resistance was present in 52 out of 62 persons (82.5%) among the antibiotics tested, while garenoxacin resistance was seen in 22 out of 63 participants (34.9%). There was no resistance to sitafloxacin or furazolidone among the individuals. Rifabutin showed no resistance in any of the 63 participants. The study also found that 18.6% of the 63 patients exhibited double antibiotic resistance to rifaximin and garenoxacin. The study's conclusions are essential as they demonstrate how rare antibiotic resistance to rifabutin, sitafloxacin, and furazolidone is in the Dominican Republic. Therefore, these three medicines could be used to treat this infection in this area. To assess whether triple therapy based on rifabutin was efficient as a third- and fourth-line treatments for this infection, H. Mori conducted a study in Japan [ 27 ]. The 10-day and 14-day cohorts were each given a group of subjects. Twelve patients were administered esomeprazole, amoxicillin, and rifabutin four times daily in the 10-day group. The same drugs were given to 17 individuals in the 14-day group; however, rifabutin was only given once daily. In this study, we used ITT and PP analyses to evaluate the success rate at curing H. pylori infection as our major end measure. In the ITT analysis, the eradication rate for the 10-day group was 83.3%, while in the PP analysis, it was just 81.8%. Eighty percent of patients saw improvement after 10 days of treatment. ITT analysis showed a 94.1% eradication rate in the 14-day group, and PP analysis showed a 91.7% eradication rate. The 14-day treatment plan was successful in curing 90% of the illness. No one in the 10-day group showed resistance to rifabutin, whereas 5.9% of those in the 14-day group did. Rifabutin-based triple therapy showed promise in the study as a therapeutic option when first- and second-line treatments failed to eradicate this infection. In a sizable cohort of people suffering from this infection who had not responded to other treatment options, a trial by David et al. in Italy examined how effective this rifabutin-based rescue therapy was for treating this infection. For 14 days, all individuals received treatment with rifabutin, amoxicillin, and a PPI. ITT analysis was used to determine the primary objective: the percentage of these infections that had been eliminated completely. Male participants in the ITT study had an eradication rate of 70.6%, while female participants had an eradication rate of 72.1%. It is noted that 19 patients had adverse reactions, indicating side effects from the medication. The study stated that rifabutin resistance was minimal, but no information was given regarding its prevalence or extent. Overall, the study showed that rifabutin-based rescue therapy produced reasonably high eradication rates in patients with this infection who were challenging to treat. The research conducted in Italy by Saracino looked into antibiotic resistance and treatment results in individuals who had failed to eliminate this infection [ 28 ]. The participants then got categorized depending on the number of treatment regimen failures they had experienced. Among the participants with only one treatment regimen failure (n=415), the failed regimens included sequential therapy in 67 patients, levofloxacin-based therapy in 221 patients, rifabutin-based therapy in 109 patients, and pylera therapy in 18 patients. The second group consisted of patients with two treatment regimen failures (n=310), and the third group comprised patients who had experienced three or more treatment regimen failures (n=312). The specific failed regimens in these groups were similar to the first group, with varying frequencies. The trial gave the recruited subjects a 10- to 14-day therapy period. According to both resistance rates and minimum inhibitory concentration (MIC) values, the study found a considerable rise in secondary resistance to the tested antibiotics, including Rifabutin. This rise in resistance was closely related to how frequently the patients' treatments failed. The study brought attention to the fact that continued treatments that were eventually not able to eliminate the organism only led to an increase in building up resistance. Notably, the resistance to rifabutin grew with each succeeding treatment failure, reaching resistance rates of 26% after one unsuccessful treatment regimen, 41.6% after two unsuccessful regimens, and 69.2% after three unsuccessful regimens or more. YI Choi's study looked at possible rescue treatments for these infection strains that were resistant to antibiotics in China and Korea [ 29 ]. The esomeprazole, amoxicillin, and clarithromycin (EAC) group had 12 patients, the esomeprazole, amoxicillin, and metronidazole (EAM) group had 13 patients, the esomeprazole, amoxicillin, and levofloxacin (EAL) group had 31 patients, and the esomeprazole, bismuth, amoxicillin, and metronidazole (EBAM) group had 144 patients. These groups were divided based on the rescue regimens given. For a total of 14 days, the subjects underwent their rescue regimens. As per the analysis, the study discovered high eradication rates in all groups. For the EAC group, the eradication rate was 91.7%; for the EAM group, it was 92.3%; for the EAL group, it was 93.5%; and for the EBAM group, it was 95.1%. These findings show that susceptibility-guided therapy with the rescue regimens was very successful in individuals with several past H. pylori treatment failures. Among patients who had previously taken these antibiotics as part of their treatments, resistance levels for clarithromycin, metronidazole, and levofloxacin were 93.9%, 99.2%, and 98.3%, respectively. This emphasizes the significance of modifying treatment plans in accordance with profiles of antibiotic susceptibility to increase effectiveness. Overall, the research points to rifabutin and furazolidone as potential treatments for this infection that are resistant to antibiotics when used in rescue regimens. Research like this highlights the need to tailor antibiotic treatment plans to each patient's unique resistance profile in order to effectively eliminate H. pylori. Finally, the study and the resources we selected shed insight into the difficulties of antibiotic resistance in the treatment of H. pylori infections. Antibiotic resistance is a major problem that threatens the success of treatment plans. Rifabutin and furazolidone are two medicines that have shown promise as effective alternatives for treating H. pylori strains that are resistant to traditional antibiotics. These findings demonstrate the critical importance that individualized treatment regimens and susceptibility-guided therapy play in obtaining effective eradication rates. Healthcare providers may increase the likelihood of effective treatment results by choosing the most suitable antibiotics by analyzing the resistance patterns of specific patients. Further research and ongoing surveillance of antibiotic resistance patterns in this microorganism are essential to continuously refine treatment strategies and combat the growing problem of antibiotic resistance. Additionally, developing new antibiotics and innovative treatment approaches may be necessary to ensure effective management of these infections in the future. Discussion This 18-month study investigated the effect of treatment failure on 187 US patients diagnosed with H. pylori infection [ 17 ]. Next-generation sequencing was applied in this study to determine H. pylori's susceptibility to different types of antibiotics. Researchers believed this procedure was quite effective because it worked in 95% of the cases and yielded results in 72 hours. Sequencing led to the administration of two separate 14-day treatment regimens to groups of 101 and 86 patients, respectively; these were the BQ regimen and the triple regimen. After this intervention, it was observed that H. pylori exhibited antibiotic resistance rates ranging from 29.6% to 33.3% to levofloxacin, clarithromycin, and metronidazole. The rifabutin-PPI-amoxicillin triple regimen was administered to two individuals. With the BQ regimen, the overall success in eliminating was 88.1%, and with the triple regimen, it was 60.5%. Rifabutin, tetracycline, and amoxicillin resistance rates varied from 0.5% to 1.1%. These rates were lower than the other antimicrobial medications used in this trial. As a result, researchers concluded that early guidance on regimens reliant on these medications led to the efficient eradication of H. pylori. Youn I Choi et al. conducted a study on 31 patients in Korea to determine the optimal antibiotic treatment strategy for eradicating multidrug-resistant H. pylori [ 18 ]. The agar dilution method, as recommended by the Clinical and Laboratory Standards Institute (CLSI), was used to determine the lowest dilution concentration (MIC) for determining antibiotic resistance. The MIC of an antimicrobial agent is the lowest concentration at which the growth of bacterial colonies is inhibited. The CLSI guidelines were used to create the criteria for determining resistance to individual antimicrobials. In this case, the MIC for clarithromycin is higher than 1 g/mL. The resistance criteria were set at larger than 0.5 g/mL for amoxicillin, 8 mg/mL for metronidazole, 4 g/mL for tetracycline, 1 g/mL for levofloxacin antibiotics, 0.25 g/mL for rifabutin, and 4 g/mL for furazolidone. MDR is the presence of resistance to two or more antimicrobials. This traditional approach of gathering data on resistance opposes that of the study by Argueta et al. [ 17 ]. Of the 31 isolated strains, around 71.0% were resistant to more than two antibiotics. There were 13 that were resistant to two antibiotics, seven that were resistant to three, and two that were resistant to four. Only nine were immune to at least one antibiotic. The most prevalent combination of medication resistance was resistance to both clarithromycin and metronidazole (two strains, 0%). According to the in vitro results, the MIC for rifabutin ranged from 0.00098 g/mL to 0.0078 g/mL, while the MIC for furazolidone was higher. No rifabutin- or furazolidone-resistant isolates were detected, despite the fact that the concentration range on agar was lower than the CLSI-established threshold. Therefore, whether used together or individually, their regimen completely eliminates even the most resilient strains of multidrug-resistant H. pylori. To fully understand the risks associated with these drugs, however, further research is necessary. The aim of this 12-month study by Cosmo et al. in Spain was to assess the use of giving only those antibiotics that had shown susceptibility to treat the infection [ 19 ]. This approach involved the administration of a PPI and two other antibiotics for a specific period as a new treatment option in patients with significantly high resistance to more than one drug. In this study, all participants underwent stomach biopsy procedures to obtain microorganisms for bacterial culture and E-test (BioMerieux) susceptibility testing, which included 68 people with dual antibiotic resistance to clarithromycin, levofloxacin, and metronidazole who received triple regimens (OAL/OAC/OAM). A rifabutin regimen (OAR) was administered to 12 patients who had developed antibiotic resistance to all three medicines: clarithromycin, levofloxacin, and metronidazole. Following a 10-day course of medication with strict adherence to the prescribed regimens, it was discovered that 10% of patients with dual resistance and 41.7% of patients with triple resistance still had H. pylori, which had not been completely eradicated. The utilization of rifabutin in this trial specifically targeted cases with triple resistance that had been scientifically proven, resulting in different outcomes compared to studies conducted in the USA and Korea [ 17 , 18 ]. Consequently, rifabutin advice was not provided at an earlier stage than in previous studies. In this study, the success of eliminating the infection was 95.5% when a combination of PPI and two antibiotics was used in those who already had resistance to more than one drug. Another study by David Y. Graham on 455 patients with this infection in the USA aimed to assess how good the RHB-105 regimen was in eliminating this infection [ 15 ]. No treatment plans were recommended for the participant's symptoms. They were between the ages of 18 and 70 when they presented with dyspepsia that had lasted for at least two weeks. Upper endoscopy and 13C urea breath test (UBT) confirmed the diagnosis. Everyone with a favorable culture, histology, or rapid urease result was eligible for randomization. The RHB-105 treatment plan called for the use of omeprazole 120 mg, clavulanic acid 3 g, and rifabutin 150 mg. Two hundred and twenty-eight patients were given this regimen for 14 days, whereas another 227 patients were given an active comparator consisting of amoxicillin and omeprazole alone. In the ITT analysis, the RHB-105 regimen significantly outperformed the active comparator regarding the H. pylori eradication rate (83.8% vs. 57.7%). Susceptibility testing was performed for patients who experienced treatment failures; however, unlike in other studies from the USA, Korea, and Spain, the method employed in this study was not disclosed [ 17 , 18 ]. Based on the standard followed in this study, first-line treatment for any infection should have the highest probability of effectiveness. This is true for antimicrobial therapies. Unlike previous research, this study focused on the care of uninformed patients and concluded that the RHB-105 regimen could successfully eradicate H. pylori. This suggests that the problem of resistance to existing anti-H. pylori treatment may be handled by giving rifabutin and equivalent regimens early in the course of the illness. A new low-dose regimen of rifabutin with vonoprazan and amoxicillin was given for seven days, and effects studies by Inokuchi et al. on 69 patients with resistant gastric infections in Japan [ 20 ]. In contrast to a study conducted in the USA, the patients in this trial had treatment failure up to the sixth therapeutic regimen [ 15 ]. According to the MIC criteria established by the CSIL and utilized in the study by Youn I Choi et al., antimicrobial susceptibility was evaluated [ 18 ]. Twelve patients served as the historical control group and were given the 10-day course of the EAR therapy regimen (esomeprazole, amoxicillin, rifabutin). However, 57 patients were treated with the VAR regimen (lansoprazole, amoxicillin, and rifabutin) for seven days. ITT analysis showed that the eradication rate for VAR therapy was 91.2%, while it was only 83.3% for the EAR treatment group. The median rifabutin MIC for all strains was 0.00 0.01 g/mL. When comparing the two groups' safety profiles, those in the seven-day VAR group fared worst, with 31.6% (18/57) reporting side effects and two people dropping out of the study due to adverse reactions. Both patients stopped taking the medication on day four; one owing to stomach pain and the other due to headache and conjunctival hyperemia. There were no serious adverse events, and the symptoms of each patient diminished when the medicine was used up. Unlike previous studies conducted in the USA, South Korea, Japan, and Spain, this one really assessed potential risks. In this study, a seven-day low-dose rifabutin-based VAR treatment was shown to be a reliable and secure way to get rid of H. pylori. With the use of early advice for rifabutin regimens, this infection might be treated and symptoms reduced sooner rather than later. Isolates of H. pylori from individuals with refractory infections showed extremely high levels of antibiotic resistance, which greatly hampered effective therapy. The study discovered that resistance rates were quite high for well-known drugs such as metronidazole, levofloxacin, amoxicillin, and clarithromycin. Many people had dual resistance to clarithromycin and levofloxacin. The results highlighted the urgent need for alternative H. pylori treatment methods [ 21 ]. When clarithromycin failed to completely cure H. pylori infection, the efficacy of BQ treatment and MRT was examined. The results showed that BQ treatment, compared to MRT therapy, was more effective in curing the infection in patients with peptic ulcers. MRT should seldom be considered when BQ therapy is not viable [ 22 ]. In conclusion, the rising likelihood of antibiotic resistance among patients with refractory H. pylori infection highlights the urgent need for novel treatment modalities. According to the trial's findings, BQ therapy is superior to MRT as a second-line therapy for this condition. Even yet, further investigation is necessary to identify complementary treatments for those with resistant strains of H. pylori [ 21 , 22 ]. This retrospective research was carried out by Chang Ming Lee et al. [ 23 ]. Due to limited patient compliance and a convoluted dose strategy, finding a suitable third-line treatment when BQ therapy fails is difficult. Several Korean regimens failed to completely eradicate the illness. Therefore, the researchers looked at how well BQ treatment for peptic ulcers compared to the second-line MRT combination for curing the infection. Between January 2013 and December 2019, 665 H. pylori patients who had failed to respond to first-line clarithromycin were continually recruited at Gyeongsang National University Hospital. A total of 665 participants were included in the trial, with 71 receiving BQ and 252 receiving MRT. In order to conduct the peptic ulcer subgroup analysis, 132 patients from the BQ group and 51 patients from the MRT group were randomly selected. The study included 323 patients who met the inclusion criteria. Using an age and gender matching mechanism, 45 patients were examined, 15 from each group. Sixty-nine point zero percent of the cases were eradicated in the MRT group. These rates were much lower than the eradication rate seen in the BQ group, indicating that the treatment was far more successful. Analysis of patients who suffered from peptic ulcers revealed that those in the BQ group fared much better than those in the MRT group. Thirteen of the 14 patients who did not improve with MRT were cured after switching to a BQ regimen in the third line of treatment. BQ was effectively eradicated at a 90% rate after MRT had failed. From these findings, the researchers drew the conclusion that MRT treatment is not as successful as BQ therapy and should only be explored in cases when BQ is not possible. Miftahussurur et al. did research to determine effective therapy for this illness in areas of Indonesia where metronidazole and levofloxacin resistance are prevalent [ 24 ]. Clarithromycin and metronidazole resistance are serious problems in Indonesia. Resistance to levofloxacin and bismuth therapy is rising, which has resulted in a rise in the use of other antibiotic treatments. The dilution test was performed to determine how resistant H. pylori was to five different antibiotics. Additionally, next-generation sequencing was employed to identify mutations associated with antibiotic resistance. After being acquired from 1039 adult dyspeptic sufferers, 106 strains were tested, none showing furazolidone resistance. Furthermore, all the tested strains showed sensitivity to both rifabutin and sitafloxacin. In contrast, rifaximin and garenoxacin exhibited high resistance rates. Notably, garenoxacin-sensitive organisms have been found in places with high levels of clarithromycin resistance. However, rifaximin might not be a good antibiotic choice due to the prevalence of clarithromycin resistance. Finally, it was seen that rifabutin-based therapies were effective in eliminating this infection compared to the conventional protocols. Sitafloxacin should be explicitly considered to eradicate levofloxacin-resistant strains, while garenoxacin should be avoided. In the study conducted by Miftahussurur and colleagues, the researchers looked at successful treatment techniques for H. pylori infections in Bangladesh and Nepal, two countries where there was a considerable amount of resistance to the primary medicines [ 25 ]. Patients with adult dyspepsia who were undergoing endoscopies at the Tribhuvan University Teaching Hospital (TUTH) were included in this study. Patients came from both Nepal and Bangladesh. After homogenizing the antral biopsy tissues, 42 strains were successfully recovered from Nepal (146 patients), while 56 strains were successfully recovered from Bangladesh. These strains were then cultivated at 37°C for up to 10 days. Subcultures of H. pylori were grown in the same microaerophilic conditions on antibiotic-free Mueller-Hinton II agar medium with 7% horse blood. Brucella broth with 10% dimethyl sulfoxide and 10% horse serum was used to keep the H. pylori solution at an ideal temperature of 80°C. Two-fold agar dilution was employed to determine antibiotic susceptibility. The five antibiotics examined were rifaximin, sitafloxacin, furazolidone, garenoxacin, and rifabutin. Rifabutin and furazolidone were not resistant; however, sitafloxacin susceptibility was very high. In contrast, rifaximin was highly resistant. Garenoxacin resistance is also more prevalent in Bangladesh than in Nepal because of its association with levofloxacin resistance. As a remedy for this infection, Rifabutin should be taken cautiously, according to the authors, because of the drug's interaction with Bangladesh's widespread tuberculosis. The high susceptibility to furazolidone and sitafloxacin increased the prospects for their future use in Bangladesh and Nepal. In a study, Muhammad Miftahussurur et al. sought to identify alternative H. pylori antibiotics for the Dominican Republic to mitigate resistance, which is widespread throughout the entire nation [ 26 ]. The MICs of five different antibiotics were determined using a two-fold agar dilution method against a panel of 63 different strains from the Dominican Republic. We employed next-generation sequencing to analyze genomic changes associated with antibiotic resistance. Rifabutin, furazolidone, and sitafloxacin were active against all 63 strains. On the other hand, resistance to rifaximin and garenoxacin was widespread (82.5% and 34.9%, respectively). Garenoxacin resistance (8/9, 88.9%, OR=45.33, P=0.002) and dual resistance to antibiotics (7/9, 77.8%, OR=31.5, P=0.009) were more likely in individuals older than 60 years old (P=0.004, r=0.357). More than three changes in rpoB were present in most rifaximin-resistant strains, with the most common new variants being S352L, I2726L, and V2465A. A strong correlation between the vacA genotype and rifaximin resistance was found (P=0.042). The lowest inhibitory concentration was significantly higher among the 23 levofloxacin-resistant organisms (39.1%, 9/23) and was positively correlated with resistance (P=0.001, r=0.84). Additionally, the vast majority of these organisms were resistant to garenoxacin (82.6%, 19/23, P=0.001). In a study with 10 strains, 16 (or 84.2% of the total) exhibited gyrA mutations at D91 and N87, making them resistant to garenoxacin. The study exhibited several limitations, outlined as follows: first, due to the limited sample size and the exclusive focus on samples collected solely from the capital of the Dominican Republic, the generalizability of the results to the rest of Latin America was compromised. Second, the study examined only a subset of the numerous H. pylori genes that could potentially be associated with antibiotic resistance, thus offering a limited understanding of the topic. Third, the resistance rates of the five different antibiotics were solely estimated through in vitro experimentation. Alternatives to levofloxacin and metronidazole that could be incorporated into the eradication strategy include sitafloxacin, rifabutin, and furazolidone. Such regions include the Dominican Republic, which has a high rate of resistance to these medications. The purpose of the prospective randomized investigation by Hideki Mori et al. is to determine whether rifabutin, used as part of a triple therapy regimen, is effective as a third- or fourth-line rescue medication [ 27 ]. Participants in the trial were those who had not improved after receiving both primary and secondary eradication treatments. A stomach biopsy was carried out to ascertain whether the rpoB mutation was present. For 10 or 14 days, patients took eradication medication with rifabutin (300 mg once daily), amoxicillin (500 mg four times daily), and esomeprazole (20 mg four times daily). Eighty percent of the sample reported that they did not take their medication as prescribed. A 13C UBT or a stool antigen test was used to establish if H. pylori had been successfully eradicated from the body after therapy had concluded. There were a total of 22 people who took part in the experiment, 12 of whom were placed in the 10-day group and 17 in the 14-day group. In the 14-day group, the success rate of eradicating the infection was 94.1% while in the 10-day group it was 81.8% (depending on the technique of analysis). Additionally, between 8% and fewer than 30% of patients in the two cohorts stopped taking their prescriptions as a result of side effects including diarrhea. However, more patients in the 10-day group finished the fourth eradication medication, suggesting that the 14-day group's success rate may have been overestimated. In addition, there was no reliable way to assess eradication rates, rpoB mutations, or rifabutin MICs. Finally, we can say that both the 10-day and 14-day rescue regimens worked, with the latter reaching an eradication rate of over 90%. However, when therapeutic tolerance is taken into account, it is possible that a 10-day course of treatment is all that is necessary for complete eradication. A few issues with this study should be noted. The self-reported approach for assessing therapy compliance was the first. This approach may be limited by its dubious precision and subpar assessment accuracy. Second, there needs to be more resistance data due to the exclusive reliance on clinical assessment without including susceptibility testing. As a result, the reasons for earlier treatment failures in these patients (lack of drug adherence or a rise in antibiotic resistance) are unknown. Furthermore, the study population exhibited a lack of complete homogeneity due to various characteristics of the study design. In summary, our study demonstrated that, when repeated attempts to eradicate H. pylori using standard antibiotics failed, rifabutin-based rescue therapy is a practical and secure option. A sizable group of people who were very difficult to treat participated in the trial. In addition, Saracino et al. carried out a retrospective analysis to examine how antibiotic resistance emerged, how therapy progressed, and how MIC values changed over time in patients who had been unable to eradicate H. pylori with at least one medication [ 28 ]. The effectiveness of levofloxacin, metronidazole, and clarithromycin against H. pylori bacteria was examined in a study including individuals who had UGIE after prior medication. Pylera® or a susceptibility-guided treatment plan was provided to the patients. Antibiotic susceptibility data was provided for 1037 of the 1223 patients who tested positive for H. pylori. Antibiotic resistance was found to be at substantially higher rates, MIC values were much lower than expected, and the number of failed treatment efforts was much higher than expected. Except for when sequential therapy is utilized as a third-line therapy, the eradication rates of antibiogram-tailored drugs remained unchanged. It was found that when Pylera® therapy was utilized as a last resort, it resulted in cure rates comparable to those of other culture-guided treatments. Unfortunately, the Pylera® regimen is difficult to administer and more frequently leads to adverse effects than alternative treatments. Up to 14 pills taken four times over the course may be necessary, and more than 30% of patients experience side effects, which increases the likelihood that therapy will be stopped early. In conclusion, the number of therapeutic failures was inversely linked to the rate and MIC values of secondary resistance to the three tested antibiotics. The eradication rates from Pylera® therapy were comparable to those from treatments tailored to a particular culture. Rifabutin, furazolidone, and other antibacterial medications were tested for their antimicrobial efficacy in a second prospective trial by Youn I Choi et al. with 200 participants to investigate their potential efficacy as therapies for multidrug-resistant H. pylori along with H. pylori strains, which are resistant to other antibiotics [ 29 ]. After conducting a retrospective analysis of the information available at the Helicobacter Registry of two famous medical centers, along with four reference strains, a total of 31 single- or multidrug-resistant H. pylori bacteria were selected for testing. The strains were subjected to the broth microdilution technique to evaluate their resistance to a number of different antibiotics. No rifabutin- or furazolidone-resistant H. pylori isolates were discovered among the 31 antibiotic-resistant strains. Only one of the strains tested positive for tetracycline resistance. Rifabutin and furazolidone are more effective at curing the infection than amoxicillin and levofloxacin, which have resistance rates of 22.6% and 1.9%, respectively. Only 3.2% of H. pylori bacteria were resistant to the antibiotic tetracycline. Since the eradication rate was not demonstrated by treating actual patients, it is possible that the in vitro results do not reflect what happens in vivo. It is also important to remember that there may be a disparity between the results of the study and the actual prevalence of antibiotic-resistant H. pylori in Korea due to selection bias. Clinicians should exercise caution before extrapolating the findings to the general public due to ethical considerations, and the research's lack of information on antibiotic resistance in H. pylori is another drawback. In conclusion, rifabutin, furazolidone, and tetracycline can be used to eliminate this infection successfully. They can be used as solitary treatment options or together as a complete regimen in regions where species of this organism have developed very high rates of resistance. However, further research and careful consideration are warranted before implementing these findings on a broader scale. As H. pylori becomes increasingly resistant to medications, new approaches are required to treat this persistent infection. The review of key data presented in this research sheds light on the challenges posed by antibiotic-resistant H. pylori and the need to develop new therapeutic techniques. In view of the proliferation of multidrug-resistant strains of H. pylori, the development of new drugs and treatment methods is essential for the successful elimination of this pathogen. Rifabutin, furazolidone, and sitafloxacin have shown promise as potential therapeutic options, especially in regions where resistance to conventional antibiotics is prevalent. Healthcare providers must keep abreast of the most recent findings on H. pylori resistance patterns and treatment choices in order to give patients with H. pylori infection effective and individualized therapeutic interventions. Furthermore, larger-scale studies with a more representative sample of the general population are required to demonstrate the efficacy and safety of these supplementary drugs. In conclusion, combating H. pylori infection and the ensuing rise in antibiotic resistance requires an all-encompassing strategy that includes continual monitoring, the cautious use of drugs, and the exploration of potential innovative treatments. The medical community has to make considerable efforts worldwide to improve H. pylori infection treatment and reduce the prevalence of related illnesses. Limitations There were no in vitro clinical trials included in this systematic review. We might have overlooked a few crucial publications because we could not obtain a few articles as the complete text was unavailable, and these were not included in our study.
We would like to express our gratitude to all those who contributed to the successful completion of this systematic review article. It is through the collective efforts and dedication of each author that this work has come to fruition.
CC BY
no
2024-01-15 23:42:02
Cureus.; 15(12):e50541
oa_package/28/bf/PMC10787902.tar.gz
PMC10787903
38217706
Introduction Successful treatment of urethral stricture disease requires not only adequate surgical experience but also appropriate preoperative diagnosis. The basic tools widely used for the initial evaluation of patients with suspicion of urethral stricture (US) are uroflowmetry, supplemented with the IPSS (International Prostate Symptom Score) questionnaire. However, these non-invasive tests remain only supplements to the available imaging methods. Currently, the standard imaging of the urethra includes urethroscopy, cystourethrography (CUG) with voiding cystourethrography (VCUG), and increasingly used sonourethrography (SUG) and magnetic resonance urethrography (MRU) [ 1 – 3 ]. Comprehensive data collection is of utmost importance prior to an operation, because factors such as stricture length, location, and extent of periurethral pathology have a key impact on the choice of surgical approach, reconstruction technique, and the final outcome. The implementation of SUG has been already described more than 30 years ago, yet importantly, this method is still evolving. Compared to the first data provided by McAninch in 1988, who was the first to describe implementation of SUG in US diagnosis, the currently widely available high-quality ultrasound devices offer incomparable image quality and detail in the assessment of pathological tissue [ 4 ]. The main limitation of the ultrasound technique includes operator dependence and lower sensitivity for evaluation of posterior urethra—a limitation that according to some authors can partly be overcome by the use of transrectal ultrasound [ 5 ]. Sonourethrography has shown significant value in several studies and in the light of the growing interest in the application of this method, this narrative review provides a summary of the available literature on the diagnostic role of SUG in the management of urethral strictures. The aim of this review is a thorough analysis of the SUG including technical aspects of the procedure, operator dependency, advantages, and limitations. Pathophysiology of urethral stricture disease The pathophysiology of urethral stenosis is linked to excessive fibrotic growth at the level of the corpus spongiosum. The result of this pathological process is known as “spongiofibrosis”. In contrast to the normal urethral wall, the epithelial layer at the site of stricture is much thicker. Dense packing of elastin fibers around the narrowed urethra causes the loss of natural elasticity of the urethra until they finally prevent proper urination [ 6 ]. Fluid’s irritative effect at the site of urethral damage may theoretically intensify the process, but this mechanism has not been practically explored in human studies [ 7 ]. It is yet noteworthy, that the first murine model for urinary extravasation revealed that mesenchymal spongiofibrosis can be induced by urethral injury with subsequent extravasation. Understanding of this cause-and-effect sequence explains the need to look for more accurate diagnostic methods that provide information on pathology beyond the urethral lumen [ 8 ]. Conventional urethral imaging techniques: urethrocystography and urethroscopy Cystourethrography and voiding cystourethrography have been the oldest and most used imaging modalities for patients with US, still being the “gold standard”. The examination is widely accessible and the location and length of the stricture can be evaluated instantly and at a relatively low cost. A great advantage of this method is the ability to assess the entire length of the urethra including the posterior urethra. In the case of complete obliteration of the urethra, in patients who are already on a suprapubic catheter—the proximal segment can be visualized by performing the antegrade urethrogram. Furthermore, CUG/VCUG also detects presence of diverticula, stones, fistula or false path. The main limitation is lack of information about the tissue beyond the lumen of the urethra; thus, information on spongiofibrosis cannot be obtained. Some comparable studies suggest that CUG/VCUG underestimates stricture length [ 9 – 14 ]. Moreover, both the patient and physician may be exposed to ionizing radiation during the procedure, unless an infusion line is used to fill the urethra and bladder. The impact of radiation can be especially significant when repeated examinations are necessary. On the other hand, urethroscopy enables a real-time endoscopic visualization of the urethral lumen without exposure to harmful radiation. Noteworthy, urethrocystoscopy and CUG/VCUG have been considered the preferred tools in post-urethroplasty follow-up protocols to detect a recurrent stricture [ 15 , 16 ]. Yet, urethroscopy rarely enables assessment of the stricture length as the caliber of symptomatic strictures is usually narrower than the standard cystoscopes used [ 17 ]. Moreover, urethroscopy is limited in providing a clear diagnosis in complex cases such as multiple strictures, or complete urethral obliteration. Novel urethral imaging technique: magnetic resonance urethrography Magnetic resonance urethrography stands out among the methods used in the diagnosis of urethral stricture, because it provides three-dimensional images of urethral stricture disease, including data on the tissue surrounding the urethra. One of the major differences, which also determines the choice of one of these methods, is the range of urethra evaluation. Magnetic resonance urethrography was found to be accurate in assessment of both anterior and posterior urethra. The value of MRU is particularly emphasized for the evaluation of the posterior urethra, because preoperative assessment of these strictures correlated more closely with operative findings compared to RUG/VCUG [ 18 – 20 ].
Materials and methods A comprehensive literature search was performed using the Medline and Cochrane databases in October 2022. Studies that evaluated the use of SUG in the diagnosis of urethral stricture disease were included in the analysis. Prospective studies were selected for this review to obtain the most informative data possible. This exclusion of case reports, editorials, and commentaries, while potentially limiting the scope of the review, was deemed necessary to ensure the highest quality and clinical relevance of the findings. Articles were screened by two reviewers (M.F. and K.M.) who followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement. The selection process is presented in the PRISMA flowchart (Fig. 1 ) [ 21 , 22 ]. The articles were screened using the keywords “sonourethrography”, “urethral ultrasound”, “urethral stricture”, and “SUG”. Only human studies and articles in English were included. Case reports, conference abstracts, editorials, and comments were excluded from detailed analysis.
Results Seventeen papers were selected as a result of our literature review on the use of SUG in assessing urethral strictures. Final analysis is based on prospective studies, the majority of which are limited by a small patient population (number of patients varied from 28 to 113). Nine studies included patients with urethral stricture located in anterior urethra and eight studies included patients regardless of the stricture location. As most of the available literature assesses the value of ultrasound based on comparing this method to other modalities and/or surgical findings, the collected data are presented in the Table 1 . As shown in the table, the diagnostic accuracy of SUG was compared to RUG/VCUG, MRU, and sonoelastography. The accuracy of SUG was generally high, with most studies reporting a sensitivity of over 80% and a specificity of over 90%. However, there was some variability between studies, with the accuracy of SUG being lower for strictures in the posterior urethra. Sonourethrography technique The technique of the procedure itself has not changed much since its introduction and most of the authors follow the same steps. The ultrasound transducer should be positioned on the perineal area, and high-frequency ultrasound waves are directed into the urethral tissue. The ultrasound frequencies should be adjusted in different parts of the urethra—15–18 MHz for the penile urethra (from meatus to the distal bulbar urethra) and 9–12 MHz for the bulbar urethra (up to the urethral external sphincter). Special attention should be paid to the impact of the examiner’s pressure created with the transducer against the skin, as too much pressure may generate an impression of a false stricture. Moreover, to avoid artefacts and evaluate the dynamic view of the intraurethral flow, the urethra should be filled during the examination. The following is a general description of the steps involved in injecting contrast for sonourethrography: Patient preparation: Patient is positioned in a lithotomy position, with legs supported and separated. Perineal area is cleansed with an antiseptic solution. Filling the urethra: A tip of a thin catheter is inserted into the urethral meatus, and the saline is prepared in a syringe. In case of a distal urethra stricture, a blunt plastic cannula can be used. Saline is slowly administered into the urethral lumen, typically in small portions. Any discomfort or adverse reactions should always be noted. Ultrasound imaging: The ultrasound transducer is positioned and moved from the urethral meatus toward the perineal area, and the saline-filled urethra is imaged in real-time. The physician should carefully assess the images for any abnormalities or areas of narrowing. The direction of examination is not relevant; however, the entire length of the urethra available for examination should always be assessed. Post-procedure: Once the imaging is completed, the catheter or cannula is removed, and the patient is instructed to void. Anatomy of male urethra on sonourethrography Normal urethra as seen on Fig. 2 presents as an anechoic tubular area, with smooth outline, usually of 8–10 mm in diameter [ 23 ]. If saline is introduced, small hyperechoic echoes may be visible within the urethral lumen (Fig. 3 ). Alterations in course of spongiofibrosis present as hyperechogenic areas in comparison to the normal echogenicity of corpus spongiosum (Fig. 4 ). Calcifications may be encountered. Ultrasound also allows the evaluation of mucosa and its abnormalities, lumen abnormalities such as diverticula, Cowper glands, paraurethral soft tissues and/or perineal masses, posttraumatic changes, etc., as well as imaging of the bladder (which may show a thickened trabeculated bladder wall in case of high-pressure voiding due to the presence of stricture). Additional ultrasound techniques Sonoelastography also known as virtual or electronic palpation is a novel technique used for measurement of tissue stiffness. Talreja et al., in a study on 77 patients with clinical features of anterior urethral stricture disease concluded that sonoelastography estimates stricture site and length better in comparison with RUG/VCUG and SUG. It estimates the degree of spongiofibrosis which serves as an important prognostic factor for stricture recurrence more accurately than SUG. Despite several subsequent studies, it is not widely used [ 24 – 29 ]. Bosio described contrast-enhanced voiding urosonography (CE-VSUG) via the transperineal approach in a pediatric population after catheter filling of the bladder with ultrasound contrast diluted in serum, and its use for assessing posterior urethral anomalies and the degree of vesicoureteral reflux in children has become widespread [ 30 ]. Sonourethrography vs. other imaging methods Diagnostic accuracy of sonourethrography compared to other methods and surgical findings Most of the studies compared SUG findings with that of RUG/VCUG in the diagnosis of urethral stricture. In two studies SUG was found to be more accurate at diagnosing stricture presence and estimating the stricture length compared to RUG [ 9 , 10 ]. Yet, the sensitivity in detecting the stricture and estimating its length using the SUG largely depends on the part of the urethra where the stricture is located. In six studies, SUG has been found to be superior to RUG for anterior urethral strictures [ 9 , 10 , 26 , 31 – 33 ] The highest correlation for stricture length at operation was for strictures located in the penile urethra [ 6 ]. Another early study comparing SUG to conventional RUG found that RUG tended to underestimate actual stricture length as compared to SUG [ 32 , 34 ]. Tembhekar and colleagues evaluated the role of SUG in 70 male patients referred to the urology department for symptoms suggestive of urethral stricture disease. This study diagnosed 39 strictures in 33 patients. RUG/VCUG and SUG were equally efficacious in diagnosing anterior urethral strictures; however, only one of three (33.3%) posterior urethral strictures were adequately visualized on SUG. The group also concluded that SUG was superior in evaluating spongiofibrosis; however, this appeared to be subjective, based on authors’ opinion. Interestingly, 61 of the 70 (87%) of patients involved in this study preferred SUG over conventional RUG, as it was felt to be less invasive and caused less discomfort [ 35 , 36 ]. Only in one study, SUG was the least accurate method compared with RUG/VCUG and MRU with average overestimation of 2 mm as related to the operative measure [ 18 ]. Despite high accuracy of SUG in most patients, the authors of this study experienced some notable outliers in the SUG measurements. None of these problems occurred in the penile urethra; instead, they were all exclusive to the bulbar or membranous urethra. This accurately depicts the technical challenges of performing SUG in the posterior urethra, which is nearly impossible despite optimal patient placement and considerable operator expertise [ 18 , 37 ]. Also, it was discovered that in 44 out of 232 (19%) patients undergoing anterior urethral reconstruction included in the study, the results of the intraoperative SUG changed the planned reconstructive technique (based on the preoperative RUG). The authors of this study described criteria to perform an anastomotic urethroplasty based on the intraoperative urethral ultrasonogram findings demonstrating a bulbar urethral stricture length of < 25 mm on aggressive urethral distension [ 38 ]. Sonourethrography for the assessment of spongiofibrosis Most authors concluded that SUG enables the evaluation of spongiofibrosis in the anterior urethra and provides similar accuracy as compared to MRU. More anatomical detail is MRU’s principal benefit, which is offset by the cost of the modality and the difficulty of image interpretation. A qualitative and quantitative evaluation of spongiofibrosis may also be provided by SUG incorporating real-time elastography [ 26 , 30 , 37 ]. It is yet unknown whether determining the exact extent of spongiofibrosis before the surgery has significant clinical value and is still to be investigated in further research. However, most authors agree that it has an influence on the choice of surgical technique as excision of the fibrotic fragment and end-to-end anastomosis is preferred in the case of extensive spongiofibrosis [ 38 ]. In a study by Ravikumur et al. [ 31 ], SUG appeared to more accurately depict stricture length, stricture diameter, and degree of spongiofibrosis when correlated with cystoscopic and intraoperative findings. Sonourethrography as a sole imaging technique Most of the articles that have been published demonstrate the value of SUG as an auxiliary modality in addition to the standard methods of diagnosing urethral strictures such as RUG or urethroscopy. However, in a recent study, Bryk and colleagues evaluated the viability of using SUG as the sole imaging technique for diagnosing urethral strictures prior to surgical treatment. This study demonstrates that, in a high-volume center with an experienced team, SUG may be the sole imaging modality needed to plan a definitive urethral reconstruction. It should be highlighted that this study only included patients with anterior urethral strictures. In comparison to RUG, which was 90% accurate in this study of 30 men who underwent both procedures, SUG was 100% correct for anterior urethral strictures, but only 60% accurate for posterior urethral strictures. Hence, as the authors concluded, it is not recommended to extend these findings to the posterior urethra. In the light of available data on SUG, because of its limited value in detecting posterior urethral strictures, the standard urethrography should remain the basic ‘road-map’ prior to surgery, particularly in patients with suspected urethral stricture undergoing initial diagnosis [ 39 ].
Conclusion Sonourethrography assessment of the male anterior urethra in patients with anterior urethral strictures is a safe, well-tolerated, minimally invasive and cost-effective diagnostic modality. For the posterior urethra, this technique cannot be recommended, based on the available published evidence. While more studies are needed to better characterize SUG, it could be proposed as an additional diagnostic modality, especially in severe and recurrent cases. More evidence on SUG and more data from studies with larger patients' groups need to be collected in the next future, as so far no randomized clinical trials have been published. Although in the future, SUG might replace CUG/VCUG as the investigation of choice in the diagnosis of anterior urethra strictures, at present, combining RUG/VCUG still remains the gold standard in evaluating urethral stricture disease.
Purpose To synthetize the current scientific knowledge on the use of ultrasound of the male urethra for evaluation of urethral stricture disease. This review aims to provide a detailed description of the technical aspects of ultrasonography, and provides some indications on clinical applications of it, based on the evidence available from the selected prospective studies. Advantages and limitations of the technique are also provided. Methods A comprehensive literature search was performed using the Medline and Cochrane databases on October 2022. The articles were searched using the keywords “sonourethrography”, “urethral ultrasound”, “urethral stricture” and “SUG”. Only human studies and articles in English were included. Articles were screened by two reviewers (M.F. and K.M.). Results Our literature search reporting on the role of sonourethrography in evaluating urethral strictures resulted in selection of 17 studies, all prospective, even if of limited quality due to the small patients’ number (varied from 28 to 113). Nine studies included patients with urethral stricture located in anterior urethra and eight studies included patients regardless of the stricture location. Final analysis was based on selected prospective studies, whose power was limited by the small patients’ groups. Conclusion Sonourethrography is a cost-effective and safe technique allowing for a dynamic and three-dimensional urethra assessment. Yet, because of its limited value in detecting posterior urethral strictures, the standard urethrography should remain the basic ‘road-map’ prior to surgery. It is an operator-dependent technique, which can provide detailed information on the length, location, and extent of spongiofibrosis without risks of exposure to ionizing radiation. Keywords Open Access funding enabled and organized by Projekt DEAL.
Highlights and clinical indications Retrograde urethrography has historically been the gold standard for identifying urethral strictures; however, because of its certain drawbacks, novel imaging techniques have been investigated and evaluated. Before deciding on surgical intervention, it is crucial to thoroughly consider the length, location, number of the strictures and their morphology since each may affect the choice of the treatment method. Modern high-resolution ultrasound is widely available; thus, the quality of data provided by this diagnostic method has improved significantly since the first description several decades ago. Sonourethrography has nowadays become a viable supplement to the standard modalities and provides additional valuable information. Fibrous scarring of the corpus spongiosum leading to a decrease in the urethral lumen is the fundamental theory explaining the pathogenesis of urethral stricture disease. Sonourethrography provides data on spongiofibrosis with satisfactory accuracy making this method widely used mostly in specialized reconstructive urology centers. As a high-resolution, multi-planar, and cost-effective technique that can be performed in an outpatient setting, SUG has found its place in the new standards of diagnostics of anterior urethral strictures. It is safe for both the patient and the physician because neither are exposed to radiation. Moreover, the possibility of using saline instead of iodine contrast makes it applicable also for allergic patients. However, knowing in which clinical situations SUG is of the greatest value is crucial. As proven in numerous publications, the satisfactory accuracy of the SUG refers primarily to the penile urethra. Some authors question the value of the radiological assessment of strictures of the distal urethra and its impact on the choice of surgical technique. These strictures are often extensive or multiple, rather than single as mostly observed in iatrogenic bulbar strictures. Thus, regardless of length and extent of spongiofibrosis, these strictures often require onlay urethroplasty with opening the urethral lumen when most accurate assessment of the pathology may be achieved during the surgery. On the other hand, in these cases, SUG seems to be the best method to show the periurethral pathology up to the urethral opening with high accuracy, allows discussion of the surgical plan with the patient before the surgery. Moreover, SUG can be of particular use to calculate the flap width in the pendulous urethra, where fasciocutaneous flaps are frequently used for reconstruction. For this purpose, Morey and McAninch proposed a straightforward formula 26–3 D (where D is the urethral diameter in mm) [ 40 ]. The lumen diameter can be measured with satisfactory accuracy with ultrasonography. This prevents excessive flap width from causing urine pooling and enables the fasciocutaneous flap to be harvested before the urethra is opened. Furthermore, SUG can be particularly valuable in cases when conventional ascending urethrography is challenging or impossible due to the anomalous anatomy of the distal urethra. This is particularly the case in patients with hypospadias, when both the native and reconstructed urethra are often extremely difficult to evaluate. While descending SUG avoids the need to inject a contrast agent, micturating SUG, although challenging, is feasible even in very complex cases and does not require catheterization of the urethra. The use of SUG in these patients should also be particularly considered as a follow-up tool—without exposing the patient to radiation. Thus, future research should investigate the accuracy of sonourethrography in the follow-up of patients after urethral stricture surgery. This could be a way to detect early recurrence of the stricture. In addition, research is ongoing to develop new ultrasound techniques that can improve the accuracy and clinical utility of sonourethrography. For example, researchers are exploring the use of three-dimensional ultrasound and contrast-enhanced ultrasound. One of the significant limitations of SUG is operator dependency and although the statistical analysis on the issue is scarce or non-existent, nearly all papers stress that despite wide availability of ultrasound and inclusion of the technique in both urethral stricture diagnostic algorithms and guidelines, it has not yet been entirely incorporated into urological everyday practice [ 41 – 43 ]. Also the issues of long learning curve, limitation in evaluating the posterior urethra, technical aspects of the examination, such as patient preparation and the length of the examination itself are being raised [ 43 , 44 ].
Author’s contribution MF: protocol/project development, data Collection, manuscript writing/editing. MV: protocol/project development, data Collection, manuscript writing/editing. KM: data Collection, manuscript writing/editing. JA: data Collection, manuscript writing/editing. FC-J: protocol/project development, data collection. AC: protocol/project development, data collection. CMR: data collection, manuscript writing/editing. WV: protocol/project development, data collection, manuscript writing/editing. MW: protocol/project development, data collection. GM: protocol/project development, data collection, manuscript writing/editing. MM: protocol/project development, data collection, manuscript writing/editing. Funding Open Access funding enabled and organized by Projekt DEAL. Data availability This article is a narrative review that synthesizes findings from existing literature. It does not contain any new data generated by the authors. Data supporting the findings of this review are available within the cited articles. Readers are referred to these original publications for access to the specific datasets analyzed. Declarations Conflict of interest There are no potential conflicts of interest. Research involving human participants and/or animals Not applicable. Informed consent Informed consent was not required for this study.
CC BY
no
2024-01-15 23:42:02
World J Urol. 2024 Jan 13; 42(1):32
oa_package/ec/0f/PMC10787903.tar.gz
PMC10787904
38112900
Introduction Urinary tract infections (UTIs) are the most prevalent outpatient illnesses and affect about 50% of the population at some point in their lives [ 1 ]. The incidence of UTIs is increasing with age, and close to 10% of postmenopausal women indicate that they had a UTI in the previous year [ 2 ]. Moreover, it is a frequent emergency department (ED) diagnosis with reportedly high diagnostic inaccuracy [ 3 ]. According to clinical criteria alone, the diagnosis of UTI has a diagnostic error rate of approximately 33% [ 4 ]. Different classification systems for UTI exist. Despite this diversity, defining UTI is reduced to the presence of bacteria in the urinary tract accompanied by related symptoms and dividing UTI into noncomplicated and complicated groups, with the latter leading to severe consequences, such as urosepsis, if untreated [ 5 ••]. The algorithm for diagnosing patients with suspected UTIs consists of several stages; each of the subsequent ones allows for more reasoned further diagnostics to make the correct diagnosis. Figure 1 shows the artificial intelligence (AI)–based treatment and diagnosis of UTIs. Liquid-based laboratories, such as urine analyses with microscopy and culturing, represent the standard for the initial diagnosis of UTI to suspect pathologies of the urinary system and to specify indications for an instrumental approach [ 6 •, 7 ]. Currently, it is well known that UTI-associated urine changes could be present in non-infection urinary tract pathologies, leading to decreased urine microscopy accuracy [ 8 ]. Moreover, urine culture suffers from several shortcomings, such as being time-consuming and highly susceptible to contamination, leading to incorrect antibiotic prescription, overutilization, antibiotic resistance, and postponed treatment [ 9 ]. The use of AI in the medical industry has grown and expanded over time. Among them, the development of intelligent decision-making for clinical medicine is the fastest [ 10 ]. Currently, it has already been stated that AI-based models can significantly improve physicians’ workflow when examining patients with UTI [ 11 ••]. However, most contemporary reviews focus on examining AI usage with a restricted quantity of data, analyzing only a subset of AI algorithms, or performing narrative work without analyzing all dedicated studies. Given the preceding, the goal of this work was to conduct a mini-review to determine the current state of AI-based systems as a support in UTI diagnosis.
Material and Methods Search Strategy In July 2023, the systematic publication search was done in several databases, including ACM Digital Library, CINAHL, IEEE Xplore, PubMed, and Google Scholar via Boolean operators with the use of the following terms: “AI,””artificial intelligence,” “UTI,” “urinary tract infection,” “cystitis,” “pyelonephritis,” “prostatitis,” “orchitis,” “epididymitis,” “urine,” “urinalysis,” and “urine culture.” Inclusion criteria : description of the development and validation of AI-based approaches for UTI diagnosis based on clinical and/or laboratory and/or instrumental data, description of the AI model used, presence of performance metrics; publication date within 5 years from the search time; English-written papers; accessibility of full papers. Exclusion criteria : papers not in the English language; papers published more than 5 years ago. Also, papers describing solely the technological aspects of the proposed method without its clinical implementation were excluded. Studies Process Two reviewers (A. T. and N. N.) independently identified all papers. All studies fitting the inclusion criteria were selected for full review. If there was disagreement or discrepancy, the senior author (B. K. S.) made the final decision. Data Extraction and Analysis We reviewed studies and extracted information related to the objective, dataset volume, data used for the training, AI approach with precise classification or networks used, performance metrics, outcomes, and validation type. In papers comparing several AI-based models, the most accurate was included in the table. After investigation of the included papers, we divided the described AI models into basic clinical scenarios where they are supposed to be used. This study was reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines.
Results Out of the 782 papers that were considered, only 14 studies on AI models in the area of UTIs met the criteria for inclusion (Fig. 1 ). These can be grouped according to the scenarios the AI models were developed for, namely, (1) diagnosis of uncomplicated UTI and symptoms checkers, (2) diagnosis of complicated UTI, and (3) diagnosis of UTIs in specific population groups. Among models, 12 and two papers described machine and deep learning approaches, respectively. The most popular machine learning model was the artificial neural network (ANN) described in six studies, followed by extreme gradient boosting (XGBoost) ( n = 3), support vector machine (SVM) ( n = 1), CatBoost ( n = 1), and ensemble learning model (ELM) ( n = 1). Among predictive inputs, demographic parameters were used in 10/14 (71.4%) studies and mostly in the view of age ( n = 9), gender (9), race, and weight ( n = 1). Notably, the latter are included in papers with pediatric patients. Anamnesis was considered in 7/14 (50%) papers, namely, history of previous UTIs ( n = 4), history of previous antibiotic treatment failure ( n = 1), history of previous urine culture results ( n = 1), and invasive urethral procedures ( n = 1). Comorbidities were used in 3/14 (21.4%) studies: diabetes ( n = 2), pneumonia ( n = 2), classification of stroke ( n = 1), and the presence of mixed cerebrovascular disease ( n = 1). Logically, the last two were used in developing AI decision support for stroke patients. UTI-associated symptoms were included in 7/14 (50%) papers: dysuria ( n = 4), fever ( n = 3), suprapubic pain ( n = 3), frequency and urgency ( n = 1), pollakiuria ( n = 1), and urine incontinence ( n = 1). Urinalysis was used, and prognostic input was provided in 7/14 (50%) papers; two of them included dipstick tests only. When urine microscopy was used, red blood cells (RBC), white blood cells (WBC), bacteria’s presence, nitrites, epithelial cells, and glucose were analyzed in 2, 3, 2, 1, 1, and 1 studies, respectively. The study used urine cloudiness as one of its features. Imaging data were used in 3/14 (21.4%): one study analyzed the cystoscopic appearance of the lower urinary tract, and two papers described ultrasound imaging usage (for estimation of hydronephrosis and vesicoureteral reflux grades, respectively). Also, there were other inputs not related to the abovementioned groups: length of stay (LOS) ( n = 3), length of urethral catheterization ( n = 2), immunological urine markers ( n = 1), ward ( n = 1), serum creatinine and albumin ( n = 1 and n = 1), glucocorticosteroid use ( n = 1), and duration of immobility ( n = 1). Performance metrics, validation type as well, and the abovementioned data arranged to include studies are discussed and presented in the review. Uncomplicated UTI AI-Based Diagnosis and Symptom Checkers Research on AI-based models for uncomplicated UTI diagnosis and symptom checking is listed in Table 1 . The study by Ozkan et al. [ 12 ••] sought to determine the accuracy of several artificial intelligence models in predicting the likelihood of cystitis and non-specific urethritis disorders, given similar symptoms from the urinary system. Anamnesis, urinalysis, and ultrasound results from 59 individuals were gathered as a training and validation dataset for the study. Four distinct artificial intelligence techniques were applied: decision trees (DT), random forests (RF), support vector machines (SVM), and artificial neural networks (ANN). When these models were compared, it became evident that ANN had the greatest accuracy for UTI detection, with a result of 98.3%. This ANN model only requires the variables pollakiuria, erythrocyturia, and suprapubic pain to acquire a diagnosis with comparable accuracy to a clinical-based diagnosis (Fig. 2 ). It was demonstrated that the ANN-based model structure could categorize UTIs without the requirement for expensive laboratory testing, ultrasounds, or invasive methods. Hence, it results in a cheaper diagnostic cost and a quicker decision-making process. The motivation behind Gadalla et al.’s [ 13 ] paper is that women with uncomplicated UTI symptoms are frequently treated with empirical antibiotics, leading to antibiotic misuse and the development of antimicrobial resistance. The authors looked into 17 clinical and 42 immunological potential predictors for bacterial culture using a random forest or support vector machine (SVM) paired with recursive feature removal (RFE). The most effective clinical predictor to rule in and rule out UTI was urine cloudiness. Interestingly, adding the selected immunological biomarkers to the model with clinical features (including cloudiness or turbidity) did not improve the predictive properties. Dhanda et al. [ 14 ] described the NoMicro model, which does not take into account urine microscopy. Instead, the results of the urine dipstick test are used. Moreover, the authors generated NoMicro models based on several machine learning classificators, namely XGBoost, RF, and ANN, and compared their efficiency. The primary outcome was a pathogenic urine culture growing ≥ 100,000 colony-forming units. Predictor variables included age; gender; dipstick urinalysis nitrites, leukocytes, clarity, glucose, protein, and blood; dysuria; abdominal pain; and history of UTI. According to the results, the AUC of the NoMicro approach reached 0.85 in external validation and did not statistically differ from the version considering urine microscopy results. Arches et al. [ 15 ] described an application providing an analysis of the urine test strip using smartphones. According to the results, among the 65 participants, the confirmed UTI AI model achieved an overall accuracy rate of 96.03% and an overall reliability rate of ≥ 0.9, which is interpreted as excellent. Complicated UTI AI-Based Diagnosis Research describing AI-based models for complicated UTI diagnosis is listed in Table 2 . Møller et al. [ 16 ] aimed to develop two predictive models, using data from the index admission as well as historic data on a patient, to predict the development of UTI at the time of entry to the hospital and after 48 h of admission (HA-UTI). The ultimate goal was to assess the individual patient’s risk. The methodology included developing five machine learning models using features such as demographic information, laboratory results, past medical history, and clinical data. The unstructured features, such as the narrative text in electronic medical records, were preprocessed and converted to structured form by natural language processing. The area under the curve ranged from 0.82 to 0.84 for the entry model ( t = 0 h) and 0.71 to 0.77 for the model predicting HA-UTI. Taylor et al. [ 17 ] performed a single-center, multi-site, retrospective cohort analysis of adults who visited the emergency department based on urine culture results, clinical symptoms, and blood tests. Using both laboratory and clinical data, models for UTI prediction were created using six machine learning algorithms: RF, XGBoost, SVM, adaptive boosting, elastic net, and ANN. A full set of 211 variables and a reduced set of 10 variables (age, gender, history of UTI, dysuria, the presence of nitrites in urine, white blood cells (WBC), red blood cells (RBC), bacteria, and epithelial cells) were both used to develop the models. Comparisons between the UTI predictions and previously recorded UTI diagnoses were made. XGBoost, which has an area under the curve of 0.904, was found to be the best-performing method. It was also shown to have greater sensitivity when compared to the documentation of the UTI diagnosis. According to the results obtained, in practical application, approximately 1 in 4 patients will be re-classified from false positive to true negative, and 1 in 11 patients will be re-categorized from false negative to true positive on account of implementing the algorithm. Mancini et al. [ 18 ] created a machine learning model that can forecast a patient’s likelihood of developing a multidrug-resistant (MDR) UTI after being admitted to the hospital. The paper added a user-friendly cloud platform called DSaaS (Data Science as a Service), which is ideal for hospital organizations where healthcare operators might not have specialized programming language skills but need to analyze data, via machine learning techniques including CatBoost, SVM, and ANN. The paper employed DSaaS on a real antibiotic stewardship dataset. The development of an MDR UTI was predicted using data related to 1486 hospitalized patients, namely, sex, age, age class, ward, and time period. According to the results obtained, CatBoost exhibited the best predictive results, with the highest value in every metric used. Cai et al. [ 19 ] described two models based on ANN for predicting fluoroquinolone-based therapy failure (model 1) and fosfomycin-based therapy failure (model 2) among patients with recurrent UTI. Input data mostly consisted of previous urine culture profiles as well as types of antibiotic therapy failures. After the completion of the ANN learning and prediction processes, our neural network showed a sensitivity of 87.8% and a specificity of 97.3% in predicting the clinical efficacy of empirical therapy. Interestingly, the previous use of a specific class of antibiotic was not a risk factor for developing bacterial resistance to the same class (except for the fluoroquinolones), but instead, the most important risk factor for predicting resistance is the use of other classes of antibiotics. Chen et al. [ 20 •] compared models based on LR and ANN in defining UTI risk after cystoscopy to reduce antibiotic overuse. As input data, previous UTI history as well as cystoscopic findings such as benign prostatic hyperplasia (BPH), diverticulum, trabeculation, blood clot, cystocele, stone, and tumor was selected. The neural network model had a high accuracy of 85%, sensitivity of 80%, and specificity of 88%. Hong et al. [ 21 ] constructed a prediction model for urosepsis risk for patients with upper urinary tract calculi with the use of a machine learning ANN model. Several clinical and laboratory features, as well as a hydronephrosis degree based in the USA, were taken as predictive inputs. The area under the receiver operating curve in the validation set was 0.95. According to the results, the proposed model could provide risk assessments for urosepsis in patients with upper urinary tract calculi. UTI AI-Based Diagnosis in Susceptible Subgroups Papers describing AI-based models for uncomplicated UTI diagnosis in susceptible subgroups are listed in Table 3 . Pregnant women and children represent a separate subgroup of patients more susceptible to UTIs and requiring specific diagnostic flow and treatment. Pregnancy immunologic and urinary tract alterations predispose women to UTIs. Progesterone-induced smooth muscle relaxation and gravid uterine compression cause ureter and renal calyces dilatation. Also, vesicoureteral reflux may occur. These modifications exacerbate urinary tract infections [ 22 ]. In turn, UTIs are among the most prevalent bacterial pediatric infections. They are equally prevalent in males and girls during the first year of life but become more prevalent in girls following the first year [ 23 ]. This high susceptibility makes the development of decision support models based on AI even more relevant. Bertsimas et al. [ 24 ] developed a machine learning model to better stratify pediatric patients with vesicoureteral reflux complicated by UTI according to the effect of continuous antibiotic prophylaxis. The authors used the following data as input: vesicoureteral reflux grade, serum creatinine, race/gender, fever, dysuria, and weight, and achieved an AUC of 0.82. The described model allows better identification of patients for whom continuous antibiotic prophylaxis will be more effective, thereby providing a personalized approach, while minimizing use in those with the least need. A study by Burton et al. [ 25 ] aimed at introducing a way to increase the efficiency of urine culture results among pregnant women and children by reducing the number of query samples to be cultured and enabling diagnostic services to concentrate on those in which there are true microbial infections. This research discussed two methods of classification to test: one is a heuristic approach using a combination of features such as urine WBC and bacterial counts, and the second is testing typical machine learning models such as random forest, neural network, and XGBoost using independent features such as demographics, previous urine culture results, and clinical details as well. The most optimal solution found was three separate XGBoost algorithms trained separately for pregnant patients, children, and the rest of the categories. Combining the three models yielded a workload reduction of 41% and a sensitivity of 95% for each patient group. The work shows the possibility of using supervised machine learning models to improve service efficiency in situations where demand exceeds the number of resources available to public healthcare providers. Immobile stroke patients also represent a highly susceptible patient subgroup. The prevalence of urinary tract infections is approximately 19%. In addition, the occurrence of an infection can exacerbate the physical harm caused by a stroke, forming a vicious circle with the stroke [ 26 ]. Zhu et al. [ 27 ] aimed to develop a prognostic model to define the risk of UTI among immobile stroke patients. Six machine learning models and an ensemble learning model were derived and evaluated. The latter achieved the best performance metrics both in internal and external validation sets, with an AUC of up to 0.82. Xu et al. [ 28 ] created an effective prediction model for identifying UTI risk in immobile stroke patients and compared its prediction performance to establish machine learning algorithms. They addressed this issue by developing a Siamese network that employed commonly used clinical criteria to identify patients at risk of UTIs. The model was developed and validated using a countrywide dataset of 3982 Chinese patients. A Siamese network is a deep neural network architecture with two or more identical subnetworks that are commonly employed in object detection. With an AUC of 0.83, the Siamese deep learning network did better than all the other machine learning–based models at predicting UTIs in stroke patients who were unable to move.
Conclusion AI-driven UTI detection is a promising new area of healthcare research; however, it is still in the exploratory rather than implementation phase. To fully realize AI’s potential for enhancing UTI diagnosis, more study is needed, ideally guided by larger, more diversified datasets and rigorous validation techniques. By resolving these issues, we can bring AI to bear on this important aspect of healthcare, which will improve patient care, cut costs, and slow the spread of antibiotic resistance. Further studies utilizing large, heterogeneous, prospectively collected datasets, as well as external validations, are required to define the actual clinical workflow value of artificial intelligence.
Purpose of Review Artificial intelligence (AI) can significantly improve physicians’ workflow when examining patients with UTI. However, most contemporary reviews are focused on examining the usage of AI with a restricted quantity of data, analyzing only a subset of AI algorithms, or performing narrative work without analyzing all dedicated studies. Given the preceding, the goal of this work was to conduct a mini-review to determine the current state of AI-based systems as a support in UTI diagnosis. Recent Findings There are sufficient publications to comprehend the potential applications of artificial intelligence in the diagnosis of UTIs. Existing research in this field, in general, publishes performance metrics that are exemplary. However, upon closer inspection, many of the available publications are burdened with flaws associated with the improper use of artificial intelligence, such as the use of a small number of samples, their lack of heterogeneity, and the absence of external validation. AI-based models cannot be classified as full-fledged physician assistants in diagnosing UTIs due to the fact that these limitations and flaws represent only a portion of all potential obstacles. Instead, such studies should be evaluated as exploratory, with a focus on the importance of future work that complies with all rules governing the use of AI. Summary AI algorithms have demonstrated their potential for UTI diagnosis. However, further studies utilizing large, heterogeneous, prospectively collected datasets, as well as external validations, are required to define the actual clinical workflow value of artificial intelligence. Keywords Open access funding provided by Manipal Academy of Higher Education, Manipal
Limitations and Future Directions AI algorithms can identify unique correlations between symptoms, urinalysis results, and inflammatory processes in the urinary tract, as well as concise variable sets that are accurate in predicting urinary tract infections. Unquestionably, artificial intelligence is a highly precise and reliable instrument for predicting various events in healthcare [ 29 ]. In contrast to conventional statistics, artificial intelligence forecasts events by identifying distinct patterns. Sadly, along with new opportunities, associated difficulties with their application have emerged, necessitating a reduction in general optimism in order to comprehend the actual state of this technology, especially in the UTI field. A sufficient amount of data is required for training neural networks to attain optimal performance metrics. In addition, limited dataset sizes may lead to estimation instability and overfitting [ 13 ]. According to our review, 10 of the 14 studies included more than 1000 cases, which, at first sight, may be an argument in favor of the utility of AI in the context of UTIs. In addition to quantity, however, the dataset must also be of sufficient quality. To be generalizable, the data should ideally be multicenter and prospectively collected, as well as span multiple geographic regions [ 14 ]. Furthermore, validation is an essential aspect of the reliability of the results. To obtain as objective and unbiased performance metrics as feasible, validation should be performed externally with samples that AI has never seen before [ 30 ]. Only two works provided external validation results in our review. On the other hand, the disparity in laboratory thresholds between medical centers and guidelines further complicates the collection of multicenter datasets and the routine application of the resulting AI-based models. For instance, there is currently no accepted level for a positive urine culture, with published values ranging from 10^2 to 10^5 cfu/mL. Conceivably different thresholds would result in different test performances [ 17 ]. The limitations outlined above represent only a small portion of the issues associated with the application of artificial intelligence and the interpretation of the results obtained. Despite this, the results of the studies included in this review demonstrate the potential utility of AI-based models for diagnosing UTIs. Clarifying the issues associated with the use of such technologies is an integral part of comprehending how the urological community should advance their sophistication. To facilitate the training of models, it is essential that as many medical centers around the world as possible converge on a common terminology for UTI, threshold values for various indicators, and research quality. To ensure generalizability, future studies should be prospective and multicenter to transition AI-based models from a stage of experimental development to a stage where they can be utilized in the clinical practice of urologists. Lastly, the advancement of AI in the field of UTIs is directly related to the general enhancement of diagnostic techniques. As new markers, new modalities, and improved interpretation become available, studies should be conducted to ascertain their utility in predicting UTIs using AI, thereby enhancing our knowledge of the future development of this technology. There is great potential in using AI algorithms for the detection of urinary tract infections (UTIs), but there are also an array of challenges that need to be addressed. The investigations discussed demonstrate that machine and deep learning models have the potential to significantly improve UTI diagnosis, leading to faster, more precise diagnoses. The limitations and unknowns of their clinical influence, however, must be noticed. Although AI models have shown remarkable precision in specific settings, a wider variety of data is necessary to guarantee their consistency and generalizability. Larger datasets that comprise patients from many different backgrounds, ages, and locations fall under this category. To minimize errors and verify the efficacy of AI algorithms in actual clinical situations, prospective data collecting is essential. In addition, it is essential to acknowledge the value of external validation in making AI models robust and useful in various healthcare settings. Moving forward, healthcare facilities should work together to develop standardized diagnostic criteria and terminology for UTIs. This will help alleviate problems caused by disparities in laboratory methods and terminology. This will allow AI algorithms to be more seamlessly integrated into the diagnostic workflow, minimizing the need for intrusive and expensive laboratory testing and imaging.
Author Contribution NN—conception, data collection, data analysis, writing the main manuscript; AT- conception, data collection, data analysis, writing the main manuscript; DRS—data analysis, writing, and editing; BZH—conception, data collection, data analysis, writing main manuscript; RZ—data analysis, writing, and editing; BKS—conception, data collection, and analysis, writing, and editing, supervision. Funding Open access funding provided by Manipal Academy of Higher Education, Manipal Data Availability Data is available on request. Declarations Ethical Approval Not applicable. Conflict of Interest Nil. Human and Animal Rights and Informed Consent This article does not contain any studies with human or animal subjects performed by any of the authors.
CC BY
no
2024-01-15 23:42:02
Curr Urol Rep. 2024 Dec 19; 25(1):37-47
oa_package/c0/67/PMC10787904.tar.gz
PMC10787905
38217694
Introduction Schwannomas are peripheral nerve sheath tumors (PNST) found in cranial and spinal nerves throughout the body. Schwannomas of the orbit are rare, and there are less than 500 cases reported in the literature [ 16 ]. It is estimated that these tumors constitute only about 1% of orbital tumors [ 9 ]. They are typically benign, slow growing, and encapsulated tumors that occur without any predilection for sex or age [ 16 ]. The growth of these tumors has been associated to functional and/or aesthetic morbidity and malignant transformation seems to be exceedingly rare [ 8 ]. Symptoms and clinical signs vary depending on the size and location of the tumor and may include a palpable mass, bulb dislocation, ptosis, optic neuropathy, and diplopia. Management of these tumors is challenging and often requires a multidisciplinary collaboration that involves ophthalmologists, neurosurgeons, otolaryngologists, and maxillofacial surgeons. When possible, surgery with gross total resection (GTR) is the treatment of choice [ 5 ]. The orbit is a confined space delimited by bone, containing delicate neuro-ophthalmic structures and with a limited capacity to expand [ 15 ]. In the context of orbital schwannomas, curative approaches are often pursued to prevent tumor growth and subsequent compression and injury to intra-orbital structures [ 19 ]. The majority of cases in the literature have been offered surgery and most of the remaining have been treated with radiation therapy or stereotactic radiosurgery [ 12 , 16 ]. Nevertheless, the surgical risks inherent to the complexity of the neurovascular structures of the orbit are not negligible [ 2 , 6 , 10 , 18 ] and warrant further investigation of the outcomes of conservative management. Recently, three separate reports have questioned the validity of an aggressive treatment strategy and advocate watchful waiting in selected cases [ 3 , 7 , 16 ]. In addition, increased use, availability, and sensitivity of diagnostic imaging makes incidental findings of orbital tumors more common, which also indicates a need for conservative approaches. In brief, the natural course of OS has been poorly studied and support for conservative management is consequently lacking. The aim of this study was to evaluate the national Swedish experience of surgical and conservative management of OS.
Methods Admission routine The study center is the primary Swedish referral center for multidisciplinary management of orbital tumors, including schwannomas. During the period of 2005 to 2021, 16 patients with a new diagnosis of OS were managed. Other neurosurgical and neuro-ophthalmological centers in Sweden were contacted to identify cases that may have been overlooked. One neurosurgical center reported having managed up to five cases during the study period. Unfortunately, no data on patients treated outside the authors’ institution could be retrieved. Once a patient with a lesion suspicious of schwannoma is referred to the study center, a complete neuroophthalmological examination is performed. This includes additional imaging, automated perimetry, Optical Coherence Tomography (OCT) and fine-needle aspiration biopsy if deemed necessary. Management is then discussed at a multidisciplinary conference of ophthalmologists, neuro-ophthalmologists, radiologists, and in some cases neurosurgeons. Typically, a conservative management is considered for all patients with OS if the indications for early surgery are not met. Surgery and postoperative follow-up Indications for early surgery are deformation of the globe or compression of the optic nerve leading to optic neuropathy. Relative indications for surgery may include impaired motility with double vision and cosmetic implications including proptosis, hyper- or hypoglobus, or ptosis with a palpable mass. The surgical approach is determined by the surgeon’s preference and the tumor location, size, and extension. All procedures are performed with microsurgical techniques and under general anesthesia. Tumors extending into the orbital apex, or through the orbital fissure, or those causing severe thinning of adjacent bony structures are often operated jointly with a neurosurgeon. OS located in the superior part of the orbit can be accessed by an anterior orbitotomy through an eyelid crease incision. OS located posteriorly in the orbital apex, sometimes with skull base extensions, are often accessed transcranially via craniotomy. In GTR, the nerve harboring the tumor is cut to allow complete removal of the tumor. In subtotal resection (STR), the tumor capsule is incised along the longitudinal axis of the nerve, and the schwannoma is excised leaving the capsule to retain the integrity of the nerve. This approach offers surgical plane separated from delicate structures surrounding the nerve, allowing safe tumor removal deep in the apex of the orbit. A neuro-ophthalmic examination and MRI are performed at one-week post-surgery. If radicality is confirmed on MRI and the pathology report, no further follow-ups are required. However, in the case of STR, MRI is recommended at 3 months post-surgery and annually afterwards. In these cases, the follow-up is generally carried out at the primary referring clinic. At follow-up, tumor growth was defined as the radiological growth of a tumor remnant following SRT, while tumor recurrence was defined as the reappearance of a tumor following GTR. Conservative management and follow-up Conservative management is offered to patients with mild symptoms, no signs of compressive optic neuropathy and low-risk of neuro-ophthalmic impairment. This encompasses individuals incidentally discovered, and those reporting minimal subjective discomfort, proptosis, globe displacement, or ocular motility issues that do not interfere with everyday tasks, such as medical requirements for driving. Low-risk neuro-ophthalmic impairment is defined as cases where the tumor is not in the apex or direct contact with the optic nerve. For these patients, radiological and ophthalmological examinations are performed at regular intervals. In patients where the tumor is in the orbital apex or located in the vicinity of the optic nerve automated perimetry and, in later years, OCT with measurements of the peripapillary retinal nerve fiber layer, macular ganglion cell layer thickness and macular ganglion cell inner plexiform layer is added. OCT with these measurements is a useful tool as they provide early signs of compressive optic neuropathy often preceding changes on automated perimetry [ 1 , 17 ]. Additionally, patients are encouraged to seek care upon onset or worsening of symptoms. There is no consensus on the intervals or the optimal length of follow-up. In our practice, patients with stable tumors are examined twice a year during the first 2 years after the diagnosis and then once a year, until 5 years. However, longer follow-ups may be indicated in selected cases, including children. In patients with slowly growing tumors, MRI is performed twice a year until cessation of tumor growth or surgery. After the follow-up period patients are encouraged to seek care for any eye-related symptoms. Study setting Patients included in this study were divided into 3 groups: patients with surgery as the primary management (group 1), patients converted to surgery after an initially conservative management (group 2), and patients conservatively treated throughout the entire follow-up period (group 3).
Results Incidence Considering the 16 documented cases of OS between the years of 2005 and 2021, the apparent incidence of OS in Sweden was estimated at 0.1 per million and year. However, the true incidence may be higher. Baseline characteristics A total of 16 patients with OS, diagnosed in Sweden between 2005 and 2021, were included in this study (Table 1 ). Patients were aged between 8 and 74 years (median 52) at the time of diagnosis and 50% ( n = 8) were female. Six patients (37%) were referred due to the incidental detection of an orbital mass during imaging for unrelated causes, such as dementia, stroke, or hydrocephalus. The remainder of the patients actively sought care for ophthalmic symptoms. Patients were referred to the study center under preliminary diagnoses, including OS, cavernous malformation, or dermoid cysts. A final diagnosis of OS was established in all cases, based on either histology and imaging ( n = 11; 69%), or imaging alone ( n = 5; 31%). Initial imaging revealed well-demarcated tumors of varying sizes, often heterogeneous ( n = 11; 69%) and cystic ( n = 11; 69%) on MRI. Half of the tumors exhibited lobular growth patterns ( n = 8). The location of the tumor was extraconal in eight cases (50%), intraconal in six cases (38%), and mixed intra- and extraconal in 2 cases (12%). Extension of the tumor into the skull base was found in five cases (31%), where four passed through the orbital fissure and one through the orbital roof. Thinning of adjacent bone in conjunction with the slow growth of the tumor was seen in 10 patients (63%), involving the superior orbital fissure, the orbital roof, the medial, and lateral walls of the orbit, or the frontal bone (Table 1 ). Surgical management was chosen in 4 patients (25%) and conservative management in 12 patients (75%). Among the 12 conservatively managed, later follow-ups resulted in delayed surgery in 3 patients (25%, Table 1 ). Presenting signs and symptoms Ten patients had symptoms on presentation (63%), the rest of the cases were discovered incidentally. Symptomatic patients reported an average symptom duration of 18 months prior to presentation and diagnosis. Proptosis was the most common finding ( n = 12; 75%), followed by vertical displacement of the eye globe ( n = 7; 44%) with all but one exhibiting hypoglobus rather than hyperglobus. Six of the patients (37%) presented with diplopia, five (31%) with pain or discomfort localized to the orbital region, four (25%) with impaired ocular motility and visual impairment, and three (19%) with ptosis. Presenting signs and symptoms were more common among patients who underwent surgery. More than half of the patients (56%) managed conservatively had been incidentally diagnosed with OS, as opposed to only one patient (14%) among those surgically managed (Table 2 ). Outcomes Four patients underwent early surgery (group 1), three with GTR (75%) and one with STR (25%). Three patients, initially treated conservatively, were later operated on with STR, (group 2). One of these three was initially offered surgery due to advanced tumor size and extent of symptoms but had declined. After 15 years of observation, the patient deteriorated due to tumor growth and underwent emergency surgery (Patient 8 in Table 1 ; and Fig. 1 ). The two other patients did not experience clinical worsening but were both offered surgery after 1.5 years of observation due to rapid tumor growth, increasing pressure on the eye globe, or thinning of surrounding bony structures (Patients 6 and 13 in Table 2 ). For groups 1 and 2, the mean postoperative radiological follow-up was 96 and 24 months, while the mean clinical follow-up was 81 and 44 months, respectively. In both groups, no tumor growth or recurrence were detected (two patients lost to radiological follow-up) and all patients had favorable postoperative outcomes characterized by complete symptom resolution (Table 3 ). One patient operated for an incidentally discovered tumor remained asymptomatic postoperatively. Among the patients conservatively managed (group 3), none experienced any worsening of symptoms at an average of 30 months of clinical follow-up. Moreover, only one of these patients (Patient 12, 11%) experienced tumor growth after 17 months, while the rest had no evidence of further growth over a radiographic follow-up period of 24 months (Table 3 , Fig. 2 ).
Discussion This study reports on a population-based cohort of orbital schwannomas. Of the 16 patients included, four (25%) underwent early surgery at the time of diagnosis, while three were surgically treated later, after initial conservative management (19%), and nine (56%) did not require any surgical intervention during the study period. Although no intraoperative complications occurred in this series, orbital surgery carries the risk of nerve, vessel, and extraocular muscle damage, which may result in both functional and cosmetic sequelae [ 10 , 18 ]. For patients with no or mild symptoms or who are poor candidates for surgery, the risks of surgical treatment may outweigh the benefits. Conservative management with regular clinical and radiological follow-ups was therefore adopted. Additionally, patients were encouraged to immediately seek care in the advent of any new symptoms or worsening of previous symptoms. Currently, it is a matter of debate whether GTR should be the gold standard for all patients diagnosed with OS [ 3 ]. To date, there are only three cases of conservative management reported in the literature. One case was observed over a 4-year period due to the risk for cosmetic disfigurement in case of surgery. There was no deterioration or increase in tumor size during the observation period [ 16 ]. In the second case, surgery was not offered given lack of symptoms and high risk of injury to orbital neurovascular structures. The patient’s neuro-ophthalmic status remained stable at 6 months of follow-up [ 7 ]. The third case was an 8-year-old with an orbital schwannoma involving the extraocular muscles. However, outcomes at follow-up were poorly disclosed [ 4 ] Most schwannomas are benign and slow growing WHO grade I tumors. However, intraorbital space is limited and the contained structures are sensitive. To minimize the risk of damage to these structures, surgery is often performed promptly. Surgical tumor removal may come at the cost of injury to the intraorbital structures. Consequently, a conservative management strategy may provide a safer course in selected cases. Of the 12 conservatively managed patients, only three required delayed surgery due to tumor growth ( n = 3) or worsening of symptoms ( n = 1). The patient who experienced both tumor growth and worsening of symptoms was initially offered surgery but declined. The rest ( n = 9) were conservatively managed without requiring further intervention during the study period. Among these patients, only one experienced tumor growth. However, the growth was mild (from 2.4×1.7×1.8 cm to 2.6×1.6×2.1cm) and the patient had no associated symptoms. Patients presenting with signs of neuro-ophthalmic compromise were offered early surgery. At the study center, STR was often considered when the tumor reached deep into the apex or was adjacent to major neurovascular structures. Of the seven surgically treated patients, four underwent STR (57%) with no reported intraoperative complications, contrary to previous reports [ 2 , 6 , 21 ]. Despite STR, none of these patients experienced tumor growth at 65 months of radiological follow-up. In fact, only two cases of tumor recurrence have been reported in the literature, lending further support to the benign behavior of these tumors.[ 13 ] Taken together, this indicates that STR may be sufficient to secure tumor control when GTR may not be safely achievable. Delaying surgery may arguably do harm through the prolonged compression of intra-orbital structures. However, conservative management allows for surgical intervention when indicated by changes in symptoms or imaging. Nonetheless, this strategy is best suited for tumors that do not compromise the optic nerve or the globe. The group that underwent late surgery had similar outcomes as the early surgical group, demonstrating the success of this strategy. Radiosurgery, predominantly Gamma Knife surgery (GKS), is increasingly used for benign orbital tumors including schwannomas [ 20 ]. However, doses greater than 12 Gy are considered unsafe due to the risk of optic neuropathy and visual acuity impairments are seen with doses ranging from 6 to 16 Gy [ 14 ]. Several studies report good tumor control after GKS, however, there are no guidelines for when to pursue GKS over surgical resection [ 11 , 20 ]. None of the patients in this study were considered for radio surgery, reflecting our own institutional practices. However, radiosurgery may be considered as a viable alternative to surgery or conservative treatment and should incorporated in future treatment guidelines. In summary, the results of this study suggest that surgery may be avoided or delayed in a selected number of patients presenting with mild or no symptoms. Conservative management, with clinical and radiological follow-ups with gradually increased intervals are therefore advocated in these cases. In cases requiring surgery, STR offers tumor control with less surgical risks compared to GTR. Postoperative yearly radiological controls, for a period of 5 years, are suggested to detect rare cases of tumor growth or recurrence (Fig. 3 ). Yearly follow-ups are motivated by the rarity of the disease and the risk of permanent visual impairment. Strengths and limitations This population-based study is one of the largest series of reported OS with more than 5 years of follow-up. The retrospective study design and the small sample size, however, hamper the level of evidence. Since patients were referred from different regions, radiological and clinical follow-ups were not always consistent. The diagnosis of schwannoma was not histologically confirmed in four of the patients.
Conclusion There were no differences in long term outcome between patients who had been managed with early surgery and those operated later after an initially conservative management. Conservatively treated patients had minimal to no symptoms and remained clinically stable throughout the follow-up period. Based on these findings, conservative management may successfully be adopted in cases with mild symptoms and low risk of neuro-ophthalmic impairment. Conversion to surgical management is indicated upon clinical deterioration or tumor growth.
Abstract Introduction Orbital schwannomas (OS) are rare occurrences with no more than 500 cases reported in the literature. The tumor’s potential to compromise the delicate neuro-ophthalmic structures within the orbit prompts surgical removal. Tumor removal is performed by ophthalmologists, often requiring a multidisciplinary surgical approach. The literature contains a very limited number of cases managed non-surgically. However, the inherent risks of orbital surgery warrant a comparison of the outcomes of conservative and surgical management strategies. Aims To review the national Swedish experience with the management of orbital schwannomas. Methods The study center is the primary Swedish referral center for the multidisciplinary management of orbital tumors, including schwannomas. During the period of 2005 to 2021, 16 patients with an OS diagnosis were managed at the center. Results Four patients initially underwent surgery where gross total resection (GTR) was achieved in three (75%) and subtotal resection (STR) in one (25%) case. The remaining 12 patients, who had a low risk of neuro-ophthalmic impairment, were managed conservatively with radiological and clinical examinations at regular intervals. After an average follow-up of 17 months, surgery was performed in three of these cases (25%). No recurrences or tumor growths were detected on radiological follow-ups (mean 50 months), and all patients experienced postoperative improvement at clinical follow-up (mean 65 months). The remainder of the conservatively treated patients ( n =9) experienced no clinical progression (mean 30 months). A slight radiological tumor progression was detected in one patient after 17 months. Conclusion There were no differences in long-term outcome between patients who had been managed with early surgery and those operated later after an initially conservative management. Conservatively treated patients had minimal to no symptoms and remained clinically stable throughout the follow-up period. Based on these findings, conservative management may successfully be adopted in cases with mild symptoms, no signs of compressive optic neuropathy and low risk of neuro-ophthalmic impairment. Conversion to surgical management is indicated upon clinical deterioration or tumor growth. Based on the findings of this study a decision tree for the management of orbital schwannomas is suggested. Supplementary Information The online version contains supplementary material available at 10.1007/s00701-024-05899-1. Keywords Open access funding provided by Karolinska Institute.
Supplementary information
Author contribution All authors approved of the submitted manuscript. All authors qualify for authorship according to the ICJME guidelines, as all have participated in the conceptualization, design, data extraction, writing, and drafting, as well as reviewing the manuscript. AET, EE, and EB supervised the work. Funding Open access funding provided by Karolinska Institute. Declarations Conflict of interest The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:02
Acta Neurochir (Wien). 2024 Jan 13; 166(1):9
oa_package/a7/d3/PMC10787905.tar.gz
PMC10787907
38217801
Introduction Periodontal disease is a chronic, multifactorial, and infectious disease caused by bacteria. It is characterized by the formation of an inflammatory response in the supporting bone and connective tissue against microbial dental plaque, and the nature of the resulting inflammatory response determines the course of periodontal disease [ 1 , 2 ]. Although periodontitis is a disease that develops in response to bacteria and their products, the course of the disease is regulated by the host tissue response. Periodontal tissue may act as a source of endocrine-like inflammatory mediators (such as IL-1β, TNF-α, and IL-6) that are important for treating periodontal inflammation and can affect glucose and lipid metabolism [ 3 ]. According to current study, Body Mass Index (BMI) was correlated with clinical attachment loss (CAL), pocket depth (PD), plaque index (PI), stage and grade of periodontitis, and number of remaining teeth [ 4 ]. Adipose tissue, particularly white adipose tissue, is an endocrine organ in which a number of pro- or anti-inflammatory cytokines known as adipokines are produced, including adiponectin, visfatin, leptin, and resistin [ 5 , 6 ]. Additionally, periodontal diseases may have an effect on adipokines or on the periodontal response [ 7 ]. Asprosin, a recently discovered glucogenic adipokine, is produced by the C-terminal cleavage product of profibrillin coded by profibrillin gene (FBN1), is mainly synthesized by white adipose tissue and released during fasting [ 8 ]. Asprosin circulates in the blood at nanomolar levels and is taken to the liver, where it triggers the G protein-cAMP-PKA pathway, causing rapid glucose delivery into circulation [ 9 ]. Asprosin is associated with diseases such as diabetes mellitus, obesity, polycystic ovary syndrome, and cardiovascular diseases [ 10 ]. Recent research has revealed that asprosin is implicated in many cell types’ inflammatory responses. According to the TLR4-JNK pathway, asprosin causes pancreatic islet cells to become inflamed and dysfunctional [ 11 ]. Moreover, asprosin stimulates endoplasmic reticulum stress-induced skeletal muscle inflammation in a PKC-dependent way [ 12 ]. We thought that periodontitis might have an impact on the level of circulating asprosin because of its systemic effects (similarly to other chronic inflammatory diseases). As far as the study’s authors are aware, no study could be found that investigated the relationship between periodontitis and asprosin, a hormone that may be helpful in the approach to periodontology as an additional biochemical marker to monitor the periodontal status and BMI of patients. This study aimed to (I) investigate levels of serum and saliva asprosin in patients with periodontitis (II) determine whether asprosin levels were related to clinical periodontal parameters, and (III) investigate the relationship between serum and saliva asprosin levels and periodontitis by grouping it according to BMI.
Methods Sixty-five systemically healthy individuals were recruited from the Clinic of Periodontology, Faculty of Dentistry, Atatürk University, between October 2021 and July 2022. The study population was designated into two groups: 1. Thirty-five patients with periodontitis (periodontitis group), 2. Thirty periodontally healthy patients (control group). This study was approved by the Institutional Internal Review and Ethics Board (AU-IIREB reference code: 511) and conducted according to the 2008 Declaration of Helsinki and its later amendments. Registration number (NCT05449821) was taken from ClinicalTrials.gov website. A signed consent form was obtained from all participants prior to the commencement of the study. The inclusion criteria of the study were as follows: Participants were aged between 18 and 60 years old, were generally healthy, non-smoking, and none had undergone periodontal therapy and/or antibiotic therapy in the past 6 months [ 13 ]. Patients with systemic diseases, such as diabetes mellitus, rheumatoid arthritis, and cancer, pregnant or breastfeeding women were excluded [ 14 ]. Clinical examination Complete clinical periodontal examination was performed by a single-blinded examiner (DÖE) using Williams probes (Williams, Hu-Friedy, Chicago, IL, USA). The periodontal parameters probing depth (PD), clinical attachment level (CAL), plaque index (PI)[ 15 ], gingival index (GI) [ 16 ] and bleeding on probing (BOP) were all measured during a full-mouth clinical examination of all patients. The measurements were recorded from 6 sites (disto-buccal, mesio-buccal, mid-buccal, and disto-palatal, mesio-palatal, mid-palatal). A sample group of 10 people with periodontitis who rejected to take part in the study served as the intra-examiner calibration sample prior to the study’s start. Within 24 h, each patient’s CAL and PD were measured twice at six different sites on each tooth. After the assessments achieved the established success criteria of CAL and PD (the percentage of agreement within 1 mm between repeated measurements should be at least 90%), they were considered to be reproducible. The BMI values of the participants were used using an electronic scale (Charder MS-3400-1, Taiwan) that can give height, weight and BMI values, and the method of dividing the weight by the square of the height was used [ 17 ]. Study groups Using the following criteria, the cases were allocated to the study groups: Control group: The sites presented BOP < 10% and PD ≤ 3 mm, no sites had attachment loss. There was no evidence of alveolar bone loss on radiographs or a history of periodontitis. Periodontitis group: At least two non-adjacent teeth showed signs of interdental CAL or buccal or oral CAL ≥ 3 mm and PD >3 mm. In line with the periodontal parameters obtained, it was divided into stages according to the classification conducted by the 2018 EFP and AAP [ 17 ]. Sample Collection During the collection of the serum samples, blood samples were taken from the antecubital fossa while the patients remained in a sitting position for the purpose of standardization. Blood samples taken for biochemistry tests were kept for 20–30 min for coagulation. The tube was centrifuged at +4°C and 4000 rpm for 15 min while remaining upright [ 18 ]. The collected serum samples were aliquoted, stored until analysis day at -80° in a deep freezer. Between the hours of 9:00 AM and 10:00 AM, whole saliva samples were taken from the patients and controls and placed into weighted 5-ml sterile polypropylene tubes for 10 min. Ambient conditions were provided for all participants to sit comfortably in a resting position. For two hours prior to collection, no oral stimulation was allowed in order to rule out the influence of mastication or eating. During this time, the sitting patients collected their saliva, gathered it at the back of the mouth, and occasionally emptied it into a collection tube. For 5 minutes, the samples were centrifuged at 10,000 rpm [ 19 ]. The final supernatant was kept at -80°C until it was time to use it. Biochemical analysis As directed by the manufacturer, the serum samples were examined using a “Human Asprosin ELISA Kit.” (BT LAB, Cat. No. E4095 Hu, China). This kit has a detection range of 0.5–100 ng/mL. This test has a sensitivity of 0.23 ng/mL. The kit manufacturer specifies that the asprosin measurement’s interassay and intraassay coefficients of variation are < 10% and <8%, respectively. In a summary, each sample and standard (128, 64, 32, 16, 8, and 4 ng/mL) were introduced to the appropriate well that had already been coated with human asprosin antibodies. The antibodies that were coated on the wells bound the asprosin that was present in the sample. After that, human asprosin antibodies that had been biotinylated were added and bound to the asprosin in the sample. The biotinylated asprosin antibodies were then combined with streptavidin-HRP and bound to them. Unbound streptavidin-HRP was removed following incubation. The substrate solution was then added, and the color of the mixture changed in direct proportion to the level of human asprosin present. By adding an acidic stop solution, the process was stopped, and absorbance at 450 nm was measured. The optical density (OD) of the asprosin samples was compared to the standard curve to estimate their levels [ 20 ]. Statistical analyses The results were described as the mean ± standard deviation (SD). The normal distribution suitability of the parameters was determined using Kolmogorov-Smirnov tests. Since the asprosin values of the healthy and test groups were normally distributed, Student’s t- test was used to compare the asprosin levels of the two groups. BMI comparisons of the control and periodontitis groups were performed on an independent sample using Student’s t -test. The stage of the periodontitis group was analyzed statistically using One-way ANOVA. Also, analysis of covariance (ANCOVA) was used to investigate the effect of the body mass index as covariates. Pearson’s correlation test was also performed. ROC curve analysis was used to determine the discriminating power of asprosin in the diagnosis of periodontitis. The value of p < 0.05 was statistically significant. Statistical analyses were performed by using the SPSS 20.0 statistical software program (SPSS Inc., Chicago, IL, USA).
Results Sixty-five patients were included in this study, and there was no statistically significant difference between the groups in terms of gender ( p =0.97). Asprosin levels and other factors were compared between participants with periodontitis and those with control (Table 1 ). Both the serum and saliva asprosin levels were statistically significantly higher in the periodontitis group than in the control group ( p <0.001). Saliva asprosin and serum asprosin levels were statistically significantly different in the periodontitis group ( p <0.001). Saliva asprosin and serum asprosin levels were statistically significantly different in the control group ( p =0.04). In the periodontitis group, when classified as stage I, II, III and IV according to the 2018 EFP/ AAP classification, serum asprosin levels are respectively; 47.48 ± 2.19, 57.27 ± 8.21, 82.30 ± 24.67, 124.18 ± 38.43 and saliva asprosin levels are respectively; 34.91±5.23, 46.01±3.23, 54.50±2.07, 64.41±14.27. A statistically significant difference was found between both serum and saliva asprosin levels in the Stage IV periodontitis group compared to the Stage I and II periodontitis groups ( p <0.001) (Table 2 ). The distribution of asprosin levels according to BMI in the periodontitis group are shown in Table 3 . BMI was significantly higher in the periodontitis group than in the control group ( p =0.017). A strong positive and significant correlation was found between the asprosin levels and BMI index ( r =0.77, p <0.001). Also, in an ANCOVA taking BMI as covariance the difference between periodontitis and control group level remained significant for saliva and serum asprosin level, (F: 161;61, respectively, p <0.001). The correlation of saliva and serum asprosin levels with other variables are shown in Table 4 . There was a significant and positive correlation between the serum and saliva asprosin levels and PI, GI, BOP, CAL, and PD ( p <0.001). The correlation between the asprosin levels with CAL and PD was the strongest. There was a significant relationship between the serum and saliva asprosin levels r: 0.663 , p <0.001. When the serum cut-off value was 25.36, the sensitivity was 94% and the specificity was 83% (AUC = 0.94, p < 0.001) (95% confidence interval, 0.88–1.0). When the saliva cut-off value was 19.81, the sensitivity was 89% and the specificity was 80% (AUC = 0.92, p < 0.001) (95% confidence interval, 0.85–0.99). The ROC curve is shown in Fig. 1 .
Discussion Asprosin was first identified by Romere et al., who also verified that patients with neonatal progeroid syndrome have mutations at the 3' terminus of the fibrillin-1 (FBN-1) gene that cause asprosin to be absent near the C-terminal cleavage site [ 8 ]. As a result of this study, it was found that both serum and saliva asprosin levels were higher in the periodontitis group compared to the control group. Based on the results of a literature search, it was concluded that this study is one of the first of its kind to investigate the serum and saliva asprosin levels of periodontitis patients. Asprosin has been demonstrated to bind to hepatocyte membranes and stimulate glucose synthesis through the cyclic AMP (cAMP)-protein kinase A (PKA) pathway [ 8 ]. According to a different study, asprosin affects mice through the olfactory receptor OLFR734 that is a G-protein coupled receptor [ 21 ]. Li et al. demonstrated that asprosin therapy significantly decreased circulating glucose levels in OLFR734-knockdown mice compared to their wild-type counterparts. Additionally, asprosin enhanced circulation in response to fasting raised hepatic cAMP levels and glucose synthesis, however these effects were reportedly less pronounced in the OLRF734-knockdown mice [ 22 ]. The role of asprosin in the inflammatory response in different cell types has been detailed in a number of recent research. According to research, asprosin activates the TLR4-JNK pathway, which results in inflammation and malfunction of pancreatic islet cells [ 11 ]. Additionally, in vivo studies showed that asprosin increased endoplasmic reticulum stress, glucose intolerance, insulin resistance, and the flow of pro-inflammatory cytokines (monocyte chemoattractant protein-1, IL-6, and TNF) [ 23 ]. Huang et al. reported that asprosin has a proinflammatory effect on the vascular endothelium, and also alleviates vascular endothelial inflammation caused by high-fat diet by neutralizing it with asprosin antibody. Asprosin has been reported to induce and exacerbate vascular endothelial inflammation through the IKKβ-NF-κBp65 pathway [ 24 ].Asprosin induced a pro-inflammatory response in THP-1 macrophages, as shown by Shabir et al, by greatly increasing the production and secretion of important pro-inflammatory mediators such TNF, IL-1, IL-8, and IL-12 [ 12 ]. Periodontitis is a chronic inflammatory condition that produces an inflammatory and immune response through the interaction of bacterial products and multiple cell populations [ 25 ]. Since periodontitis is a chronic inflammatory disease, asprosin levels in both serum and saliva were higher in the periodontitis group compared to the control group. We considered that periodontitis might have an impact on the level of circulating asprosin because of its systemic effects (similarly to other chronic inflammatory diseases). Accordingly, asprosin may be a biomarker in detecting periodontitis and linking periodontitis with BMI status. In this study, periodontitis was staged, and it was investigated whether asprosin levels were affected according to the severity of the disease. When the serum asprosin levels were analyzed, a statistically significant difference was found between stages II and IV and between stages I and IV ( p <0.001). Accordingly, as the severity of periodontitis increased, the serum asprosin level increased. The increase in serum asprosin level according to the severity of periodontitis revealed the relationship between the severity of inflammation and asprosin levels. Additionally, asprosin levels had a significant and positive correlation with PI, GI, BOP, CAL, and PD. The correlation between asprosin levels with CAL and PD was the strongest. Periodontal disease severity and degree of damage are related to CAL and PD. The positive correlation of asprosin with clinical indicators suggests that this marker can be used to determine the severity of periodontal disease. Asprosin and BMI index showed a substantial positive and significant association in this investigation. Uğur and Aydın reported that asprosin levels in serum and saliva increased with BMI, which is consistent with our study [ 26 ]. However, in addition to this study, it was determined by the help of ANCOVA analysis that there was no confounding factor of BMI with high asprosin levels in the periodontitis group in our study and that this disease was a result of periodontitis. Additionally, a number of clinical trials revealed that people who were overweight or obese had significantly higher serum asprosin levels [ 27 – 29 ]. Circulating asprosin was found to cross the blood-brain barrier and directly activate orexigenic AgRP+ neurons through a cAMP-dependent pathway in a study. It was also reported to trigger appetite and boost body weight by inhibiting anorexigenic proopiomelanocortin (POMC)-positive neurons in a downstream manner in a GABA-dependent manner [ 9 ]. In humans, a genetic deficiency in asprosin causes NPS, which is characterized by anorexia and extreme weakness [ 8 ]. However, there is conflicting information on the levels of circulating asprosin in obese kids. Children with obesity had considerably lower serum levels of asprosin than children of normal weight, according to studies by Long et al. and Corica et al. [ 30 , 31 ]. Additionally, Silistre and Hatipoglu discovered no gender-related changes in the asprosin levels in the blood of children with obesity compared to the normal weight group [ 32 ]. Many studies have reported that obesity and overweight are associated with the risk of periodontitis [ 33 , 34 ]. The reason for this relationship may be the involvement of proinflammatory cytokines secreted from adipose tissue in the formation of excessive inflammatory response in periodontitis [ 35 ]. It has been shown that the relationship between obesity and periodontitis is bidirectional, and the presence of inflammation in periodontal tissues may be a predisposing factor for obesity [ 36 ]. As with obesity, adipokines and systemic low-grade inflammation may explain another mechanism by which periodontitis is associated with obesity. It has been suggested that obesity may contribute to periodontitis through altered adipokine levels [ 37 ]. Since there is no study of asprosin, a recently discovered adipokine, with periodontitis, there is no study in which we can compare the relationship of serum and saliva with BMI in individuals with periodontitis. One of the limitations of this study was that the gender and age factors were not limited. The levels of adipokine hormones may differ between different genders and age groups. Age and gender standardized-controlled trials need to obtain precise data. Another limitation of the study was examining the serum and saliva asprosin levels using a limited sample size. For the objective of evaluating activity, it may be helpful to compare the levels of asprosin in the serum and saliva with those in GCF and gingival tissue (as well as research at how periodontal treatment affects asprosin levels) in patients with various types and degrees of periodontitis.
Conclusion As a result of our study, asprosin may be a useful parameter as a biomarker in the course of periodontal diseases. However; BMI status should be considered when evaluating asprosin levels in patients with periodontitis. It is necessary to conduct molecular research to determine the function of asprosin in the inflammatory pathway in light of the pathophysiology of periodontal tissues.
Objectives A newly discovered adipokine known asprosin in serum and saliva in patients with periodontitis has not been explored. The aim of this study was to determine the relationship between serum and saliva asprosin levels and periodontitis by grouping it according to body mass index (BMI). Materials and methods The study was conducted on 65 systemically healthy patients (35 patients with periodontitis (periodontitis group), 30 periodontally healthy patients (control group)). In each patient, age, BMI, and clinical periodontal parameters (plaque index (PI), gingival index (GI), probing depth (PD), and clinical attachment level (CAL)) were evaluated. Statistical analyses were conducted utilizing the Student t -test, ANOVA, and Pearson correlation analysis. For the significance level of the tests, p <0.05 were accepted. Results The serum and saliva were collected to assess asprosin levels. Both the serum and saliva asprosin levels were statistically significantly higher in the periodontitis group than in the control group ( p <0.001). Saliva and serum asprosin levels were directly proportional to the severity of the periodontal disease ( p<0.05 ). Asprosin levels were higher in patients with a higher BMI ( p<0.05 ). Conclusion Asprosin levels were increased in periodontitis, and even a high BMI status apparently affected the levels of this hormone. It is thought that asprosin may be a useful biomarker in evaluating the relationship between periodontal status and BMI. Clinical relevance Asprosin may be a useful parameter as a biomarker of periodontal disease progression. However, BMI status should be considered when evaluating asprosin levels in patients with periodontitis. Keywords Open access funding provided by the Scientific and Technological Research Council of Türkiye (TÜBİTAK).
Author contributions SNSG worked on the conception and design of the study, interpretation of data, and drafted the manuscript. AD worked on the conception and design of the study and critically revised the manuscript. EL worked on the execution of biochemical analysis, interpretation of data, and critically revised manuscript. DÖE worked on sampling and interpretation of data and critically revised the manuscript. TA worked on the conception and design of the study and critically revised the manuscript. All authors contributed significantly to the drafting and revision of the manuscript and gave final approval of the version to be published. Funding Open access funding provided by the Scientific and Technological Research Council of Türkiye (TÜBİTAK). Declarations Ethics approval This study was approved by the Institutional Internal Review and Ethics Board (AU-IIREB reference code: 511). Informed consent Informed consent was obtained from all subjects involved in the study. Written informed consent has been obtained from the patient(s) to publish this paper. Conflict of interest The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:02
Clin Oral Investig. 2024 Jan 13; 28(1):91
oa_package/dc/a1/PMC10787907.tar.gz
PMC10787908
38217553
Introduction Eating disorders (ED) are defined, according to the current edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5-TR) [ 1 ], by specific disturbances in eating behaviors, and by a persistent and undue influence of body weight or shape on the self-evaluation of the individual [ 1 ]. Rather than completely demarcated clinical entities, Anorexia Nervosa (AN), Bulimia Nervosa (BN), and Binge Eating Disorder (BED) may share a common psychopathological core [ 2 – 4 ]. In fact, potential transitions between diagnoses have been shown to occur in patient during their lifetime [ 5 – 7 ]. Psychological treatments currently adopted for EDs—such as Cognitive Behavioral Therapy—posit the existence of a specific core of pathological beliefs shared across AN, BN and BED. This psychopathological core is represented by a primary low self-esteem, an over-evaluation of achievements, and a clinically relevant intolerance for adverse mood states, all driving the persistent and undue influence of body weight or shape on self-evaluation [ 2 , 3 ]. Phenomenological research has offered a novel perspective on this point, focusing on the role played by embodiment in shaping eating psychopathology [ 8 – 10 ]. Traditionally, phenomenology has developed a distinction between “lived body” ( Leib ) 1 and “physical body” ( Koerper ) 2 . In this perspective, the “lived body” has been defined as the subjective preconscious, coenesthetic 3 experience of one’s own body, while the “physical body” represents its material dimension [ 11 , 12 ]. Recently, Stanghellini and colleagues [ 9 ] attempted to conceptualize EDs psychopathology in terms of the “lived-body-for-others”—a concept which was first introduced by Sartre [ 13 ]. In addition to the previously described dimensions of corporeality, Sartre described that one can apprehend one’s own body from another point of view, as one’s own body when it is looked at by another person [ 13 ]. When we are looked by another person, the “lived body” is no longer a direct, first-personal experiential evidence, but it is an entity that exists as viewed from an external perspective. This third-person perspective of oneself is defined as the “gaze of the Other” (Fig. 1 ). According to several observations [ 19 – 21 ], individuals with EDs report a particular difficulty in experiencing their own body from within, or, in other words, from the coenesthetic perspective. The dialectic integration between the inputs arising from the “lived body” and those coming from the “physical body” is impaired. Stanghellini and colleagues [ 9 ] have hypothesized that people with ED tend to experience their own body first and foremost as an object looked by another person (the above-mentioned “lived-body-for-others”), rather than from a first-person perspective [ 14 ]. This core psychopathological feature would explain the main characteristics of EDs, which are represented by the adoption of external, objective measures to define one’s self and for the clinically relevant preoccupation with controlling one’s own body shape and/or weight [ 9 , 14 , 15 ]. In this perspective, symptoms such as severe dieting or obsessive weight-control might represent a dysfunctional coping strategy to manage the feelings of alienation and extraneousness towards one’s own corporeality. Individuals may report a divergence in the degree to which dieting attempts and eating restriction can be applied or followed [ 4 ]. Nonetheless, even when impulsiveness and loss of control over eating represent the main features of the clinical presentation—such as in the case of BED—individuals frequently report a reduced possibility to feel their body from within [ 9 ], as well as a perceived similarity between their lived experiences and the lived experience of patients with AN or BN [ 16 ]. According to Stanghellini et al. [ 17 ], people with EDs define themselves reaching external measures. Thus, they perceive their body as an objectified entity to which aesthetic and moral judgments can be applied. Further discussion should also be reserved for the process of perception itself. Contemporary experimental research, and previous theoretical studies, have questioned the hierarchy of psychological processes shaping individual perception. Sensory stimuli, rather than constituting the object of perception, can represent external constraints for internal representations [ 18 – 20 ]. In fact, according to recent theoretical frameworks (e.g., active inference), sensory stimuli are first integrated at a preconscious level, and only later cognitively appraised [ 18 ]. Therefore, a perceptual and preconscious representation of one’s own body relies at the basis of a conscious experience of it [ 21 ]. This novel perspective on perception and consciousness has since triggered a wave of innovations in psychological sciences [ 14 , 22 , 23 ]—while open questions remain in the field of EDs [ 23 ]. For instance, as previously mentioned, an altered optical, visible and aesthetic self-appraisal of one’s own body has been postulated for ED 4 [ 17 , 24 ]. To individuals with ED, their body would principally be given as an object “to be seen”. The “gaze of the Other” would serve as an optical prosthesis to cope with an altered coenesthesia 3 and as a device through which persons with ED can define themselves. This hypothesis has been empirically supported [ 14 , 23 , 25 – 28 ], and embodiment disorders have been shown to play a role as maintain factors in longitudinal studies focusing on ED symptomatology [ 29 ]. Nonetheless, this hypothesis does not yet address the significant gender gap observed within EDs and within clinical practice [ 1 ]. In addition, the role of socio-cultural factors in influencing both embodiment and psychopathological symptoms has not been fully previously embedded within this conceptual structure. The present study thus aims at offering a novel perspective on the strong female preponderance for EDs [ 1 ], in light of the role played by the “gaze of Other” in eating psychopathology. For these reasons, a brief review is first offered on the role of socio-cultural factors as informing the lived experience of individuals. Subsequently, the role of gender identity in embodiment is discussed. Finally, the complex interaction between embodiment, gender identity, and psychopathology is presented.
Conclusions As long as the “feminine” is lived and conceptualized as the “Other”, the feminine gender will be associated with a higher risk to adopt a dysfunctional identification of the self through the “gaze of the Other”, especially in an ocular-centric society. Current theoretical models positing a role for embodiment in shaping eating psychopathology can be updated, appreciating the interplay between the feminine gender and the risk to engage in maladaptive self-objectification. This maladaptive self-objectification may be attempted through the subjugation to the visual representation of oneself through the “gaze of the Other”. The interplay between gender and EDs needs to be considered as embedded within a world of values—both at the individual and sociocultural level.
Background Phenomenological research has enriched the scientific and clinical understanding of Eating Disorders (ED), describing the significant role played by disorders of embodiment in shaping the lived experience of patients with ED. According to the phenomenological perspective, disorders of embodiment in ED are associated with feelings of alienation from one’s own body, determining an excessive concern for external appearance as a form of dysfunctional coping. The purpose of the present narrative review is to address the role of gender identity as a risk factor for EDs in the light of phenomenological approaches. Methods Narrative review. Results The current study discusses the interplay between perception, gender identity, and embodiment, all posited to influence eating psychopathology. Internalized concerns for body appearance are described as potentially associated with self-objectification. Furthermore, concerns on body appearance are discussed in relation to gendered social expectations. The current review also explores how societal norms and gender stereotypes can contribute to dysfunctional self-identification with external appearances, particularly through an excessive focus on the optical dimension. The socio-cultural perspective on gender identity was considered as a further explanation of the lived experience of individuals with ED. Conclusions By acknowledging the interplay between these factors, clinicians and researchers can gain a deeper understanding of these disorders and develop more effective interventions for affected individuals. Level of evidence Level V narrative review. Keywords Open access funding provided by Università degli Studi di Firenze within the CRUI-CARE Agreement.
Lived experience and sociocultural factors The present narrative review suggests a possible integration of the phenomenological approach with some elements of other available theoretical perspectives (e.g., cognitive–behavioral, psycho-dynamic, psychophysiological findings). Phenomenological explorations typically target the lived experience of individuals. The result is a rich and detailed collection of qualitative self-descriptions from patients. In an attempt to better grasp the pathogenesis of ED, it is crucial to shift the attention from abnormal eating behaviors to more complex and subtler psycho(patho)logical features, especially experiential ones. For instance, caloric restriction and rapid weight loss are both frequently observed among athletes competing in a specific weight-class [ 30 – 32 ]. These behaviors may even reach clinical significance—possibly inducing side-effects, such as amenorrhea—while still not been conceptualized as constituting a psychopathological disorder [ 30 – 32 ], although potentially satisfying all criteria to be diagnosed with AN or BN. By contrast, a core difference in EDs is that eating behaviors transcend their original significance of nutrition, entertainment, pleasure, and social function. Thus, thinness becomes crucial for self-worth, dietary restraint for the need for control, and binge-eating to manage emotions [ 10 ]. The personal meanings behind these behaviors might be analyzed in terms of individual history, and in terms of a dialectical interaction with sociocultural factors, which may vary across cultures and history. Sociocultural factors and EDs Some of the pathological connotations attributed to weight/food control by persons with ED are a product of the fashion industry or media [ 33 – 37 ]. Almost two centuries ago (after the industrial revolution) a sober and controlled alimentation became a spread and largely shared value, and a thin body was considered a symbol of efficiency [ 38 , 39 ]. Across time the image of the ideal body changed, but it constantly mirrored social position and individual worth, with a number of studies documenting the trend of increasing thinness between the 1950s and the 1990s [ 40 , 41 ]. Contemporary estimates report a prevalence of 0.7% for EDs in Europe, and a rise of around 15% from 1990 to 2019 for this group of diagnoses in the same region [ 42 ]. Common risk factors have been identified for AN, BN and BED [ 43 , 44 ]. One of the strongest factors were observed in relation to gender [ 43 ] and cultural acculturation [ 43 , 45 ]. Sociocultural influences for EDs have also been noted to interest certain professional sectors more specifically. In particular, those exposed to ideals of beauty and self-control, such as ballet dancers [ 46 – 48 ] and fashion models [ 49 ]. Moreover, the importance given to thinness, as an expression of power and control over one’s self [ 50 ], as well as a means to reach higher social desirability [ 46 , 51 ], has been observed as influencing the risk to develop EDs during the lifetime. Interestingly, preliminary evidence has also shown that this risk is influenced by a negative assessment of the position reserved to women in family or society [ 50 ]. An effective appraisal of lived experiences along and not in contrast to physical and social determinants is, therefore, of primary importance to the advancement of psychiatry, and to our understanding of EDs in particular. Gender identity and eating disorders More than 70 years ago, Simone de Beauvoir published “Le Deuxième Sexe” (The Second Sex; [ 52 ]). Its second volume (“L'Expérience Vécue”, The Lived Experience) is a seminal book, which is arguably at the basis of contemporary thought on gender, gender identity and what a feminine gender entails in general [ 53 – 55 ]. The main focus of this second volume is to ponder “What is woman?” [ 52 ]. Beauvoir argues that a woman is, by definition, the “Other”: Then, how can the “Other” define itself if not through a third-person view? Positing the feminine as the essential “Other”, Beauvoir implies that a woman is objectified, or, in other words, that a woman becomes connoted by passivity and thus alienated from her true self. Agency, and the active possibility for women to autonomously obtain self-representation and self-definition, is undermined [ 56 ]. The essential quality of “Other”-ness can also be internalized [ 57 ], with distinct expectations for what concerns gender roles within society and with an explicit focus on the visual representations of the self [ 58 ]. The higher prevalence of EDs in Western countries [ 42 ], where gender equality is higher [ 59 ], can then be interpreted in light of the influence exerted by gender stereotypes on social roles [ 59 ]. In fact, occupational segregation between genders may be more readily appraised in more egalitarian and developed countries [ 60 – 62 ]. This cultural and social representation of gender stereotypes, in the occupational or educational sector, can partly explain the equality paradox 5 [ 59 ]. A common theoretical framework to explain this paradox is to posit that virtue-signaling or group-affiliation may shape personal identities, driving social roles to become increasingly more divergent between genders [ 63 – 71 ]. A commonly employed example would be relegating professions involving care to the feminine, and technical oriented careers to the masculine [ 61 ]. According to the bio-psycho-social model for EDs [ 43 , 44 ], biological sex has been recognized as a risk factor for development of these disorders, as demonstrated in both clinical and non-clinical samples. A strong female to male ratio for EDs is estimated from population-based studies [ 42 ]. In parallel, a higher risk for ED has consistently been observed at the epidemiological level for transgender women in comparison with transgender men [ 72 ], irrespective of gender-affirming hormonal therapy or surgical interventions [ 73 ]. Transgender women appear to be at a higher risk of being diagnosed with an ED also when accounting eating restraints aimed at modulating hormonal effects on body weight and shape [ 73 ], that is when eating restraints are not secondary solely to gender-distress. Therefore, gender, and not solely sex, should also be recognized as influencing eating behaviors. For this reason, a perspective on lived experiences for women should move beyond mainly characterizing biological characteristics in relation to sex as informing the risk to develop an ED during the lifetime. While genetic and hormonal factors have a role in eating psychopathology [ 74 ], the particular onset of most EDs around puberty [ 1 ] may also be appreciated beyond mechanistic or reductionist claims of hormonal influences on mental disorders [ 75 ]. In fact, a child going through puberty may feel that their body is escaping them, that their body is no longer the clear expression of their individuality [ 76 , 77 ]. This experience of alienation from one’s own body is also a function of social expectations for what concerns gender identity [ 78 – 81 ]. In other words, female and male adolescents who do not fully conform to gendered expectations for what concerns primary or secondary sexual characteristics, or who do not fully conform to social expectations for what concerns gender roles or gender expression, may be at a higher risk of experiencing feelings of alienation from their body [ 82 – 85 ]. In this instance, the body becomes foreign, and, at the same moment, it is grasped by others as an “object”. If the child is a woman, she may also more frequently be objectified or sexualized [ 86 ]. The visual, or ocular characteristics, are those more readily grasped by the “gaze of the Other”, and may thus become a primary target for body modification goals, or concealment [ 73 , 87 ]. Gender identity and embodiment Since ancient times, the optical dimension has been specific to the feminine, and the mirror is the feminine utensil par excellence—at least in the stereotypical and common-sense meaning [ 88 , 89 ]. It evokes the radiance of beauty, the charm of the gaze, seduction. To reflect oneself in the mirror is to project one's image before oneself, to split oneself into a self that is looked at and one that is looked at. The mirror is used to see, know, modify, and disguise oneself. The face in Greek is called prosopon—the figure that offers itself to the eyes of the “gaze of Other” as a seal of its own identity [ 90 ]. Female identity has thus always been linked to the optical dimension—both to one's own appearance reflected in the mirror and to one’s own appearance offered to the “gaze of the Other”. This dependence of identity on gaze (not female only), and especially on the “gaze of the Other”, has not diminished in the course of history, but on the contrary has been further strengthened in the “society of the spectacle” [ 91 ] whose distinctive trait is, precisely, “ocular-centrism” [ 92 ]. This mode of access to oneself mediated by visual representations can turn out to be alienating, since images convey individual ghosts and cultural aspects, social prejudices, gender stereotypes. At the same time, the attempt to experience and define one’s own self through the “gaze of the Other” may be captivating or socially rewarding [ 93 ]. However, defining one’s self in this manner exposes to the risk of being enthralled into an alienated representation of the self, fully enmeshed and intertwined within social expectations [ 94 ], in complete opposition to an authentic definition of the self. Theoretically, any individual can ultimately succeed in grasping itself only by alienating itself, positing oneself both as a subject and, vis-à-vis oneself, as an object [ 13 ]. In fact, the act of defining oneself is not solipsistic in nature, but requires another being to compare oneself, and to which to be compared [ 95 ]. Moreover, the “Other”, as an existent being itself, can form a representation of us in its mind, and we, in turn, can define ourselves as a function of this representation [ 96 ]. Women have not been equally supported in the maturation of an autonomous definition of their identity by the presence of widespread, culturally and socially relevant gender models, by which to define themselves [ 97 ]. Similarly, gender minorities and non-stereotypical males may experience psychological distress secondary to social expectations in relation to their body weight or shape [ 98 – 102 ], as well as their gender roles or their gender expression [ 103 ]. Social expectations for what concerns “feminine” can strongly influence the lived experience of any individual [ 52 , 104 ], and the essential quality of the “Other” internalized in relation to gender identity (Fig. 2 ). Gender identity, embodiment, and psychopathology The interplay between gender identity, embodiment and psychopathology has not yet been fully elucidated. Here, the authors posit that the “gaze of the Other” can exert a dual effect on the individual. On one hand, the “gaze of the Other” offers an external reality, capable to offer us self-recognition and validation [ 105 ]. On the other, it may be a source of distress, defining us, possibly in contrasts with our personal values [ 13 ]. The individual may react to the challenge posited by the “gaze of the Other” symmetrically and in a dual manner: either rejecting this external “definition” of itself, or, vice-versa, seeking it, to possibly grasp itself through a well-defined, stable and “objectified” representation [ 106 – 108 ]. A healthy well-being may result from the balance of these two processes. On the contrary, a maladaptive self-identification can be attempted in persons with ED [ 23 ], fueled by a disproportion in the preconscious optical–coenesthetic experience of oneself. Nonetheless, the attempt to nullify oneself to resist an external definition may be observed within EDs as well. As the visible body is the channel by which oneself is subjugated by the “Other”, coercing it can diminish the possibility of being grasped. This process is more readily apparent in victims of abuse or sexual violence [ 27 , 28 , 109 ]. Nullifying one’s own body can represent a form of self-injury and self-preservation at the same time [ 110 ]. Repetition of trauma and hypersexuality may also be observed, even in individuals with a history of adverse experiences in the sexual domain [ 111 – 114 ]. In fact, while previous theoretical contributions focused on the role of trauma in predisposing individuals for emotional dysregulation, and thus, potentially, hypersexuality, sexual activity can also represent an attempt to alienate and subjugate oneself in a perceptual manner [ 111 ]. In addition, it may reflect an effort to experientially, affectively, seek an alternative encounter with one’s own body, thus escaping a dysfunctional optico-coenaesthesis 6 [ 115 ]. Strengths and limits The strength of the current study lies in its phenomenological approach. The subjective experience of individuals with ED was considered, which allowed for a deeper and nuanced understanding of their lived corporeality. Previous theoretical contributions have been integrated with a sociocultural perspective, in light of self-objectification theory. A novel perspective on the gender gap observed within AN, BN and BED was reached. In summary, the current study proposes the interplay between perception, gender identity and embodiment as a potential target for future empirical research and clinical interventions. The limits of the study, on the other hand, are represented by its narrative nature, relying on the existing literature. Furthermore, while the study discusses important theoretical implications of gender identity and perception on embodiment, it may lack full empirical evidence to support these claims, and its exploratory intent should be taken into consideration. What is already known on this subject? Some key features of embodiment disturbances in ED have been previously described, such as the experience of a distorted body image, or the lack of effective integration between interoceptive or visual stimuli within an appropriate cognitive appraisal of body weight and shape. In addition, individuals with ED report feelings of alienation from their own body, feeling their material self to be extraneous from themselves. For this reason, a subjective experience of ‘estrangement’ from their physical self is known to fuel distress in these individuals, and thus contribute to disordered eating patterns. As a response, individuals with ED frequently report relying on external measures to reach an effective definition of their corporeality. What this study adds? The present narrative review attempts to integrate phenomenological accounts with the psychosocial model for EDs, remarking the role of the broader social context in which these disorders develop and are experienced. Accordingly, the role of the feminine as the “Other” was discussed as shaping self-objectification among women, and as a partial explanation of gender discrepancies in the epidemiology of EDs. The current review also emphasizes the need to move beyond solely describing biological sex as contributing to the female/male disparity in EDs, advocating for a socio-cultural perspective on gender identity.
Author contributions All authors equally contributed. L.T. wrote the first draft of the manuscript, under the supervision of G.S., V.R. and G.C.; all authors reviewed the manuscript and agree with its content. Funding Open access funding provided by Università degli Studi di Firenze within the CRUI-CARE Agreement. Open access funding provided by Università degli Studi di Firenze within the CRUI–CARE Agreement. This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Data availability Not applicable. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no potential conflict of interest.
CC BY
no
2024-01-15 23:42:02
Eat Weight Disord. 2024 Jan 13; 29(1):8
oa_package/b1/93/PMC10787908.tar.gz
PMC10787909
38222998
Introduction Constipation can notably affect children's health and their parents' quality of life. Functional constipation (FC) in children is defined as irregular or complex bowel movements without underlying systemic or anatomical causes [ 1 ]. Although the prevalence of constipation varies among societies based on many factors, including sociocultural and dietary habits, constipation is a common pediatric health problem, affecting between 0.7% and 29.6% of children [ 2 ]. One study reported inadequate nutrition as a primary (58%) risk factor for constipation, with mental instability (21.2%) and genetic factors (3.5%) also identified [ 3 ]. Worldwide, many diagnostic criteria have been approved for diagnosing functional constipation. However, the Rome III criteria and the North American Society for Pediatric Gastroenterology, Hepatology, and Nutrition (NASPGHAN) guidelines are the clinical criteria most often utilized in Saudi Arabia [ 4 ]. Recently, the Rome IV criteria have been widely used to diagnose FC. According to Rome IV, the diagnosis of FC in children under the age of four requires the existence of at least two of the following symptoms for at least one month in the absence of an organic cause and with insufficient criteria to diagnose irritable bowel syndrome: maximum of two attempts to defecate per week, history of constipation, history of uncomfortable, painful bowel movements, history of hard or large stools, having a large fecal mass in the rectum. Rome IV also notes the following criteria for diagnosing FC in toilet-trained children over four years of age: one or more incontinence incidences per week and a history of having large, thick stools that could block the toilet [ 5 ]. Fecal incontinence, frequent abdominal pain, bleeding of the rectum, enuresis, and urinary retention and infection are all well-recognized complications of FC in children [ 6 ]. Fecal occlusion, intestinal/ileus blockage, toxic megacolon, and bowel perforation are the more catastrophic potential effects of FC [ 7 ]. The North American and European Societies for Pediatric Gastroenterology, Hepatology, and Nutrition (NASPEGHAN/ESPGHAN) released updated clinical recommendations for diagnosing and managing FC in children. These include "a normal intake of fibers and liquids, normal physical activity, and an extra pharmacological treatment for rectal fecal impaction followed by a pharmacological maintenance therapy." The societies also recommend polyethylene glycol (PEG) with or without electrolytes (0.2-0.8 g/kg) for maintenance therapy [ 8 ]. Providentially, FC is estimated to account for 90% of cases seen in healthcare centers, with the remaining 5-10% having organic causes [ 9 ]. Because FC is a relatively benign disorder, it often goes undiagnosed and untreated, which can lead to a variety of medical and psychosocial issues for children and their parents, cause concern for healthcare budgets, and, presumably, place a significant socioeconomic burden [ 1 , 10 , 11 ]. A 2022 study in Saudi Arabia found that parents' knowledge of childhood constipation was linked to their practices in dealing with it. The study involved 568 participants and concluded that poor practices were due to inadequate knowledge [ 12 ]. A 2019 study in kindergartens in Jatinangor found that in 111 parents there was a correlation between parental knowledge and their children's toilet training practice and behavior. The study highlights the importance of educating parents about fecal continence in children [ 13 ]. In addition, a 2020 systematic review of 13 studies attempted to locate, organize, and synthesize the current data on the experiences of parents caring for children with FC and the information needed by such parents to facilitate successful treatment. The study revealed that most parents have inadequate knowledge of childhood constipation, with a relatively small percentage having sufficient and correct information about the condition [ 14 ]. In this study, pediatricians were also found to have inadequate knowledge of FC in children, with notable differences in the pediatricians' knowledge and patterns of practice regarding childhood constipation [ 3 , 4 , 14 ]. To the best of our knowledge, no studies in the Western region of Saudi Arabia assess the level of parents' knowledge regarding childhood FC. Thus, the study aimed to evaluate parents' knowledge, attitudes, and practices regarding childhood FC in Makkah, Saudi Arabia to identify knowledge gaps that can be filled to reduce morbidity through early detection, appropriate access to medical care, and increased public awareness.
Materials and methods Study design and participants This web-based descriptive cross-sectional study evaluated parents' knowledge, attitude, and practice toward childhood constipation. The data were obtained through an online questionnaire directed to parents in Makkah, Saudi Arabia. Parents working in the medical field were excluded. Ethical considerations and sampling technique The researchers were committed to all ethical considerations required to conduct the research. The Biomedical Ethics Committee at Umm Al-Qura University provided ethical approval no. HAPO-02-K-012-2023-04-1539 to conduct this study. Participants’ privacy was maintained, and the responses remained confidential. Using the Raosoft sample size calculator (Seattle, WA: Raosoft, Inc.), the sample size was determined with a 95% confidence level, a 5% margin of error, and a 50% response distribution, and 385 participants were considered the minimal sample size [ 2 ]. The overall sample size increased to a maximum of 824 participants in case of potential data loss and to generalize the study results more efficiently. Study tool and data collection A previously validated questionnaire has been utilized [ 12 ]. An online questionnaire was sent electronically through Google Forms to 824 parents; two opted out of the study and 26 were related to the health sector. After the exclusion of these 28 parents, a total of 796 parents satisfying the inclusion criteria were included in the analysis. The questionnaire is composed of three sections. The first section comprised parents' sociodemographic characteristics, such as age, gender, employment status, residence, and educational status. The second part included seven questions to evaluate parents' knowledge, causes, and symptoms. The third section assessed the response, treatment, and source of knowledge. The questionnaire has been closed-ended with predefined choices. The questionnaire was already prepared and translated into Arabic, the local language. Google Forms have been used for the design. Statistical analysis Descriptive and inferential statistical analyses of the data were carried out. Simple descriptive statistics of the sociodemographic characteristics and other categorical variables in the form of frequencies and percentages were calculated and tabulated. Continuous variables medians and interquartile ranges (IQRs) were reported as measures of central tendency and dispersion, respectively, owing to the relatively non-normal distribution of variables as determined by the Kolmogorov-Smirnov test (p<0.001). Seven questions assessed the parents' knowledge and awareness of childhood constipation, and one point was given for each correct response. These were summed up to obtain the total knowledge score of each participant. Because some questions involved multiple responses, the total possible knowledge score of a participant ranged from 0 to 18. Similarly, the practice scores of the participants ranged from 0 to 4. The scores of participants with different sociodemographic characteristics were compared using the non-parametric Mann-Whitney U test and the Kruskal-Wallis test. Additionally, non-parametric Spearman's rank correlation was applied to assess the correlation between knowledge and practice scores. Significance was established at a p-value of 0.05, indicating a 95% confidence interval. All statistical calculations were performed using SPSS version 27.0.1 (Armonk, NY: IBM Corp.).
Results Among 796 participants, a notable majority were female, accounting for 74.2% of the sample, while the remaining 25.8% were male. Each participant willingly agreed to partake in the survey. The age distribution varied, with the largest group falling in the 35-60 years category at 62.8%, followed by 25 to 34-year-olds at 22.0%; a smaller representation of individuals aged 15-24 years (6.7%) and those above 60 years (8.5%). The educational backgrounds of the participants were diverse, with a significant majority having a university education (67.5%), while 12.1% possessing postgraduate qualifications. The participants also came from various occupational backgrounds, with government employees making up the largest group (34.3%) and unemployed individuals accounting for 31.0%. Most participants were married (92.5%), while 7.5% were divorced. Additionally, 80.7% of the participants reported that their children had previously experienced constipation (Table 1 ). Table 2 presents the participants’ responses to knowledge questions regarding childhood constipation, highlighting the percentages of correct answers. Only 11.1% of participants correctly identified the definition of constipation as “less than three bowel movements per week,” while 63.6% correctly recognized it as a “symptom” rather than an illness. In terms of common causes, a significant 88.4% recognized “organic constipation” and 81.3% identified “functional constipation” as common causes. Regarding the causes of functional constipation in children, 84.3% correctly attributed it to a “low-fiber diet.” In terms of symptoms, substantial percentages correctly identified “hard, dry stool” (71.0%) and “painful defecation” (63.0%) as common symptoms. However, fewer participants recognized the symptom of “large diameter stool that may obstruct the toilet” (19.3%). Concerning the necessity of complete tests for constipated children, 38.7% correctly answered “no,” while 30.2% believed “yes” and 31.2% responded with “I don't know.” Finally, for complications, the majority correctly identified “painful anal fissures” (75.1%) and “fecal impaction” (60.6%) as potential complications, while fewer recognized “intestinal perforation” (7.3%) and “fecal incontinence” (8.3%). The participants’ total knowledge score had a median of 8.0, indicating a moderate overall understanding of childhood constipation. This summary emphasizes the varying levels of correct answers among the participants and underscores areas where educational efforts may be beneficial (Table 2 ). Table 3 provides insights into the participants’ responses regarding their practices related to childhood constipation, with a focus on the percentages of correct practices. When asked about the initial home treatment for their child’s constipation, 27.9% correctly suggested “giving him high-fiber food,” while 23.5% recognized the importance of “giving him plenty of fluids.” However, a notable 23.7% mentioned “giving him laxatives” as an initial treatment, which is not recommended without medical guidance. Furthermore, only a small percentage (10.6%) suggested “give him bananas or honey.” Regarding the treatment for fecal impaction and intestinal blockage, 42.8% correctly identified “do an enema to expel stool” as the most successful treatment, indicating a good understanding of this critical aspect. In terms of recommending fiber-rich foods for children with constipation, 71.7% rightly suggested “fruits like watermelon, apple, and banana” and 68.8% recognized “vegetables” as a suitable option. This indicates a relatively strong awareness of dietary recommendations for managing constipation. The participants’ total practice score had a median of 2.0 (IQR: 1.0-3.0), suggesting a moderate level of adherence to correct practices related to childhood constipation (Table 3 ). When asked about their biggest concern regarding chronic childhood constipation, the majority (65.1%) expressed the fear of it “continuing into adulthood.” A significant portion also expressed concerns related to severe medical conditions, with 30.9% fearing “internal abdominal tumors” and 17.3% fearing “congenital abnormalities of the colon (stricture).” These responses highlight the substantial apprehension parents may have about the long-term implications of childhood constipation on their children’s health (Table 4 ). Regarding sources of information about constipation, the internet emerged as the most commonly cited source, with 25.5% of the participants relying on online resources. Friends and relatives also played a significant role, with 23.6% turning to them for information. Doctors and medical staff were mentioned by 14.6% of the participants, while frequent medical practice was cited by 13.8%. Magazines, books, TV, and other sources made up the remainder of the information channels. This indicates a diverse range of sources from which parents seek information about childhood constipation, with the internet being a prominent choice. This underscores the importance of ensuring accurate and reliable online resources for parents seeking guidance on this health issue (Figure 1 ). The findings indicated a significant difference in knowledge scores based on gender (p<0.001). Female participants had a higher median knowledge score of 8.0 (IQR: 7.0-10.0) compared to males, with a median score of 8.0 (IQR: 5.0-10.0). Regarding educational level, there was a significant difference in knowledge scores (p=0.001). Postgraduate participants exhibited the highest median knowledge score of 9.0 (IQR: 7.0-11.0), while those with a middle-level education had the lowest median score of 7.0 (IQR: 5.0-9.0). However, no significant associations were observed between knowledge scores and age, occupation, marital status, or whether their children had experienced constipation before. In summary, parents and educational level were associated with variations in knowledge about childhood constipation among the participants. Female participants and those with postgraduate education tended to have higher knowledge scores (Table 5 ). The analysis revealed some significant findings. First, there was a significant difference in practice scores based on age (p=0.007). The participants in the 25-34 years age group had a median practice score of 2.0 (IQR: 1.0-3.0; mean=2.0), and those in the 35-60 years group (mean=2.2) and more than 60 years age group (mean=2.4) also had a median score of 2.0 (IQR: 2.0-3.0). However, participants in the 15-24 years age group had a comparatively lower score of 2.0 (IQR: 1.0-3.0) (mean=1.9). Other sociodemographic characteristics, including gender, agreement to participate in the survey, educational level, occupation, marital status, and whether their children had experienced constipation before, did not show significant associations with practice scores (Table 6 ). The analysis, performed using Spearman’s rank correlation coefficient, revealed a statistically significant positive correlation (ρ=0.328, p<0.001) between knowledge and practice scores among the 796 participants (Table 7 ). This finding indicates that the participants who demonstrated higher levels of knowledge about childhood constipation tended to exhibit better practices related to its management. In other words, as the participants’ knowledge scores increased, their practice scores also showed a positive trend. This suggests that improving knowledge about childhood constipation can lead to more effective and appropriate practices in its management and prevention (Figure 2 ).
Discussion Our study aimed to determine parents’ knowledge, attitudes, and practices toward childhood constipation in a major city in Saudi Arabia. The dominance of female participants, constituting 74.2% of the sample, was a noteworthy observation. Moreover, the high percentage of participants with a university education (67.5%) and postgraduate qualifications (12.1%) reflects a relatively well-educated sample. Educated parents are more likely to actively seek information and engage in health-related discussions [ 15 ]. However, the educational diversity within the sample may lead to varying levels of health literacy. This diversity highlights the importance of tailoring health education programs to different educational backgrounds to ensure effective communication and understanding. This study provides valuable insights into the participants’ understanding of childhood constipation. While many respondents correctly identified common causes of constipation, such as “organic constipation” and “functional constipation,” there were noticeable gaps in their knowledge [ 1 ]. This indicates that constipation is perceived differently by individuals in the Makkah community. For instance, only 11.1% of persons correctly identified the definition of constipation as “less than three bowel movements per week,” consistent with the findings of Gray [ 16 ]. This lack of awareness regarding a fundamental aspect of constipation could have led to misconceptions and delayed recognition of the condition, which may have implications for children’s health. Thus, there is a lack of standardized definitions and a shared understanding of constipation within the Makkah community [ 17 ]. This lack of consensus could result in challenges in healthcare settings, as healthcare providers may need to clarify and educate patients and their families about what constipation entails. Additionally, the relatively low recognition of specific symptoms, such as “large diameter stool that may obstruct the toilet,” and severe complications, such as “intestinal perforation,” underscores areas where parents may not be fully aware of the potential seriousness of childhood constipation, echoing concerns raised by the study conducted in Nigeria [ 18 ]. This highlights the need for comprehensive education that not only addresses common aspects of the condition but also delves into less prevalent but critical facets to ensure a holistic understanding among parents in Makkah. The practice assessment reflects the actions that parents take when their child experiences constipation. A significant percentage suggested giving high-fiber foods and plenty of fluids as initial home treatments, which aligns with evidence-based recommendations for managing childhood constipation [ 19 , 20 ]. However, the finding that 23.7% of the participants mentioned “give him laxatives” as an initial treatment is concerning. This finding aligns with concerns about laxative misuse in similar populations, as identified by Roerig et al. [ 21 ]. The inappropriate use of laxatives can lead to dependence, electrolyte imbalances, and other adverse effects, highlighting the need for caution in their administration. It is crucial to highlight the importance of clear and accessible guidance on appropriate home remedies to avoid potential harm to children. Education campaigns can play a pivotal role in promoting safe and effective home treatments. The expression of concern by parents regarding the long-term consequences of childhood constipation is a significant finding. This reflects parental anxieties about their children’s health, which has been documented by Rajindrajith et al. [ 22 ]. The fear of constipation continuing into adulthood and concerns about severe medical conditions indicate that parents in Makkah are deeply invested in their children’s well-being. This reflects a forward-looking approach to healthcare decisions. Healthcare providers should recognize and address these fears during consultations, offering reassurance and guidance to alleviate parental anxiety [ 23 ]. This personalized approach to addressing parental concerns can improve overall healthcare experiences for families in Makkah. The analysis of associations among knowledge, practice, and sociodemographic factors revealed intriguing patterns. There was a significant difference in the knowledge scores based on gender and educational level. The female participants and those with postgraduate education tended to have higher knowledge scores. This suggests that tailored educational interventions can be particularly effective for these groups. Additionally, it highlights the importance of gender-inclusive health education efforts to bridge the knowledge gap among parents. The positive correlation between knowledge and practice scores is a crucial finding. It suggests that when parents are better informed and have a deeper understanding of constipation, they are more likely to implement effective strategies for caring for their children with constipation. It emphasizes that improving parents’ knowledge about childhood constipation can lead directly to better practices in managing the condition, as indicated by Thompson et al. [ 14 ]. This underlines the potential impact of educational interventions in enhancing the care and well-being of children with constipation in Makkah. The limitations of this study should be acknowledged to provide a comprehensive assessment of its findings and implications. The study’s reliance on a self-report survey may introduce response bias, as the participants may provide socially desirable answers or unintentionally misrepresent their knowledge, attitudes, and practices related to childhood constipation. The research was conducted in Makkah, Saudi Arabia, and may not be fully representative of the broader Saudi population. The predominantly female sample may also introduce gender bias and limit the generalizability of the results. Third, the study’s cross-sectional design restricts the ability to establish causal relationships between sociodemographic factors and knowledge or practice scores. Additionally, the survey’s reliance on closed-ended questions may not capture the depth of the participants’ attitudes and practices, and qualitative research methods could complement future investigations in this regard.
Conclusions This study highlights the variability in knowledge levels among parents, with a moderate overall understanding of childhood constipation. It emphasizes a moderate level of adherence to recommended practices related to childhood constipation, with some room for improvement in certain areas. These findings underscore the importance of targeted educational efforts to improve parents’ understanding and behavior concerning childhood constipation in Makkah, Saudi Arabia. Additionally, the study emphasizes the role of the internet and interpersonal sources in disseminating information, highlighting the need for reliable online resources and healthcare professionals to provide accurate guidance to concerned parents.
Introduction Functional constipation in children is described as irregular or difficult bowel motions without underlying systemic or anatomical causes. Although constipation can have a serious negative impact on a child's health and the lives of their parents. This study aimed to assess the knowledge of parents about childhood constipation, intending to reduce morbidity and mortality through increased public health education in Makkah, Saudi Arabia. Methods The current study was a web-based, descriptive cross-sectional study. The data were obtained from May 2023 to November 2023 through an online questionnaire directed to parents in Makkah, Saudi Arabia, and analyzed using SPSS version 27.0.1 (Armonk, NY: IBM Corp.). Results A total of 796 participants were included in the present study, of which 205 (25.8%) were males and 591 (74.2%) were females. The knowledge levels among them varied, with 11.1% correctly defining constipation and 63.6% recognizing it as a symptom. Common causes like organic and functional constipation were acknowledged by 88.4% and 81.3% of participants, respectively. Regarding practices, 27.9% recommended high-fiber foods for initial home treatment, and 42.8% acknowledged that an enema is effective for fecal impaction. In the dietary recommendations, 71.7% suggested fruits and 68.8% mentioned vegetables. Concerning attitudes, 65.1% expressed fear of childhood constipation continuing into adulthood, while 30.9% feared severe medical conditions. The internet (25.5%) and friends/relatives (23.6%) were the primary sources of information. Knowledge was significantly higher among females and those with postgraduate education. Conclusion This study highlights the variability in knowledge levels among parents, with an overall moderate understanding of childhood constipation. It emphasizes a moderate level of adherence to recommended practices related to childhood constipation, with some room for improvement in Makkah, Saudi Arabia.
CC BY
no
2024-01-15 23:42:02
Cureus.; 16(1):e52236
oa_package/56/1d/PMC10787909.tar.gz
PMC10787910
38217744
Background Colorectal cancer (CRC) is more common among men than women, and as of 2020, CRC had a 5-year prevalence rate of 51.9 per 100,000 people in Asia and an incidence rate of 68 per 100,000 in Hong Kong [ 1 , 2 ]. Surgical methods are commonly applied to remove tumors from the colon or rectum, including low anterior resection (LAR), laparoscopy, abdominoperineal resection, and transanal microsurgery. However, fecal incontinence (FI), defined as the “involuntary loss of feces when feces are solid and/or liquid,” commonly develops following surgical management of CRC and chemoradiation [ 3 ]. Following surgery, FI can develop due to damage to muscular, fascial, or neural tissues during surgery [ 4 ]. Seventy to 90% of the patients who undergo sphincter preserving surgery experience FI in addition to other symptoms such as incontinence for flatus, increased intestinal gas, and rectal urgency (commonly referred to known as LAR syndrome) [ 5 ]. The reported incidence of FI in patients treated with pelvic chemoradiation ranges between 3 and 53% [ 6 , 7 ]. FI has a significant negative impact on the quality of life (QoL) among CRC survivors [ 8 ]. The inability to control the involuntary leakage of stool can cause embarrassment and fear of such episodes may hinder social participation or physical activity, adversely affecting mental health [ 9 ]. Therefore, the post-surgical management of FI is necessary. Treatments for FI include both pharmaceutical and non-pharmaceutical interventions. Non-pharmaceutical treatments may include diet adjustments, anal plugs, or physiotherapy [ 10 , 11 ]. Physiotherapy interventions for the treatment of FI include pelvic floor muscle training (PFMT) with or without biofeedback, neuromuscular electrical stimulation (NMES), and acupuncture [ 12 ]. A recent review by Kim and Oh [ 13 ] evaluated the effectiveness of PFMT on bowel function and health-related QoL among patients who have undergone LAR and included studies published until 2019. Since then, further studies have been published on physiotherapy interventions in the management of FI, and therefore, the review requires updating. The objective of this review is to investigate the effectiveness of physiotherapy interventions on FI and the QoL following colorectal surgery.
Methods This systematic review and meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [ 14 ] and was registered in the International Prospective Register of Systematic Reviews (PROSPERO, CRD42022337084). Search strategy Electronic databases of English-language articles, including the Allied and Complementary Medicine Database (AMED), The Cochrane Central Register of Controlled Trials (CENTRAL), the Cumulative Index of Nursing and Allied Health (CINAHL), Embase, MEDLINE, Ovid, the Physiotherapy Evidence Database (PEDro), Scopus, and Web of Science, and electronic databases of Chinese-language articles, including Chinese National Knowledge Infrastructure (CNKI) and Wanfang Data, were searched to identify potentially relevant articles. Searches were conducted from database inception to December 2021 and updated in November 2022. Searches were conducted by two independent review authors (P.C. and C.T.C). The specific search strategy applied to the Medline database is presented in Supplementary Appendix 1 . Four themes, “physiotherapy intervention,” “colorectal cancer and surgery,” “bowel incontinence,” and “randomized controlled trials,” were utilized to identify potentially relevant articles. Related terms associated with each theme were combined using the Boolean operator “OR.” All four themes were combined using the Boolean operator “AND.” EndNote 20 citation management software was used to archive and organize the search results and remove duplicates. Manual searches were also performed by examining the reference lists for each included trial and relevant systematic reviews to identify additional candidate trials. Study eligibility criteria Trials were included if they (1) were randomized controlled trials (RCTs, including pilot, cluster, or crossover trials) comparing physiotherapy interventions (acupuncture, biofeedback, electrical stimulation, aerobic exercises, resistance exercises, stretching exercises, manual therapy, PFMT, or yoga) with a control condition (no treatment, usual care [UC], placebo, or active control) on FI and QoL outcomes following colorectal surgery; (2) measured FI using the Cleveland Clinic Incontinence Score (CCIS, also known as the Wexner score), the low anterior resection syndrome (LARS) score, or anorectal manometry (ARM) or measured QoL using the Fecal Incontinence Quality of Life (FIQL) scale, the EuroQoL 5-Dimension (EQ-5D) scale, the European Organization for Research and Treatment of Cancer (EORTC), Quality of Life Questionnaire (QLQ) module for colorectal cancer (QLQ-CR29), the EORTC QLQ module for cancer (QLQ-C30), or the 36-Item Short Form Survey (SF-36); and (3) were available in full-text format in either English or Chinese (traditional or simplified). Unpublished theses were also included in the review if they met the review criteria described above. Studies were excluded if they were systematic reviews, RCT/systematic review protocols, or case reports. Trials that included subjects with FI associated with causes other than colorectal surgery or radiotherapy for CRC (such as labor, prostate cancer, malabsorption, spinal surgery, or congenital disorders) were excluded. Trials reporting data as median and interquartile range, trials that included subjects with FI secondary to other medical conditions, such as neurological diseases, and trials that evaluated the effectiveness of non-physiotherapy treatments (such as anal plugs, Chinese medication, herb medication, injection, acupressure, moxibustion) were also excluded. Article screening Articles identified through the electronic searches underwent a three-stage screening process, including title, abstract, and full-text screening. Studies were screened for inclusion by two reviewers (P.M.Y. and L.P.H.). Disagreements between reviewers were resolved by discussion, and a third reviewer (P.K.) was consulted if the disagreement remained unresolved after discussion. Data extraction Data extraction was conducted independently by three authors (P.C., P.K., and C.K.H). The following data were extracted from each included study: last name of the first author, publication year, country of origin, sample size, mean age/age range of study subjects, intervention and control, outcome measures, and pre- and post-treatment data (mean and standard deviation) of ARM measured by anal resting pressure (ARP) and maximum squeeze pressure (MSP) for FI, and lifestyle, coping behavior, depression, and embarrassment components of QoL. For post-treatment data, only the data associated with the longest follow-up period was extracted. Quality assessment Five authors (P.C., P.K., C.K.H., P.M.Y., and L.P.H) independently assessed the methodological quality of the included trials using the Revised Cochrane Risk-of-Bias (RoB 2) tool for randomized trials [ 15 ]. Disagreements between reviewers were resolved by discussion. An additional reviewer (PK) was consulted for any unresolved disagreements. The RoB 2 categorizes the overall RoB level as high, moderate, or low, based on the RoB identified in five domains: (1) randomization process, (2) deviation from intended interventions, (3) missing outcome data, (4) outcome measurements, and (5) the selection of the reported results [ 15 ]. The overall RoB evaluation was guided by signaling questions in the five domains [ 15 ]. Data synthesis and statistical analysis All meta-analyses were performed using Comprehensive Meta-Analysis software (CMA version 3.3.070, Biostat Inc., Englewood, NJ, USA). Statistical heterogeneity was measured using the I 2 test. A random-effect model was applied for high heterogeneity ( I 2 > 50%); otherwise, a fixed-effect model was utilized [ 16 ]. Trials evaluating similar physiotherapy interventions, control conditions, and outcome measures were grouped together for meta-analysis. For continuous data, the weighted mean difference (WMD) and 95% confidence interval (CI) were calculated. WMD was chosen because similar units of measurement were used for outcome measures in the included trials [ 17 ]. In the meta-analyses and throughout the “ Results ” section, all data for ARP and MSP reported by Chen [ 18 ] were converted from mmHg to kPa (by multiplying by 0.133) [ 19 ] to standardize the unit for the calculation of WMD. Significance was defined as a p -value ≤ 0.05.
Results The study selection process, which followed the PRISMA approach, is summarized in Fig. 1 . The studies that were excluded at the full-text screening stage and the reasons for exclusion are listed in Supplementary Appendix 2 . A total of 4413 articles were identified via electronic and manual searches. After all screening stages were applied, only 10 trials met the inclusion criteria for the meta-analytic review. Characteristics of the included trials Table 1 summarizes the characteristics of the included trials. Across all 10 trials, data were extracted for 608 subjects. The sample sizes of the included trials ranged from 12 to 100. The mean age of participants in the included trials ranged from 45.6 to 66.8 years. All subjects in the included studies had fecal incontinence following surgery for CRC. The interventions assessed in the included trials were PFMT ( n = 2), biofeedback ( n = 4), and biofeedback combined with PFMT ( n = 4). Among the 10 included trials, three trials reported QoL outcomes, and eight trials reported FI outcomes. Risk of bias in the included trials Figure 2 A shows the distributions of RoB levels across each domain, and Fig. 2 B shows the overall RoB levels assessed for all included trials. Of the 10 included trials, three (30%) reported adequate randomization processes, one (10%) had few to no deviations from the intended intervention, six (60%) were free of missing outcome data, seven (70%) blinded the outcome assessors, and eight (80%) had reported all planned outcomes. Of the 10 included trials, four had moderate RoB, and six had high RoB. Effects of interventions on QoL among individuals with FI after CRC surgery PFMT versus UC Two trials compared the effects of PFMT with those of UC on QoL [ 20 , 21 ]. The RoB for these two trials was high. Both trials [ 20 , 21 ] measured QoL using the FIQL and reported the lifestyle, coping behavior, depression, and embarrassment components. PFMT was performed daily in both trials, but the protocols varied. In Hung, Lin (20), participants performed four PMFT sessions per day, with each session consisting of 20 contractions. In Xia et al. [ 21 ], participants performed three PMFT sessions per day, with each session consisting of 20 contractions. The duration of PFMT ranged from 3 to 9 months. Meta-analysis of data from the two trials [ 20 , 21 ] ( n = 112) revealed significant effects of the intervention compared with UC for the lifestyle (WMD 0.54; 95% CI 0.03, 1.05; p = 0.04; Fig. 3 A), coping behavior (WMD 1.14; 95% CI 0.24, 2.04; p = 0.01; Fig. 3 B), and embarrassment (WMD 0.417; 95% CI 0.14, 0.70; p = 0.00; Fig. 3 C) components of the FIQL; however, no significant effect was observed for PFMT compared with UC on the depression component (WMD 0.424; 95% CI − 0.24, 1.09; p = 0.21; Fig. 3 D). Effect of interventions on FI Biofeedback versus UC Three trials compared the effects of biofeedback with those of UC on FI, measured by ARM [ 22 – 24 ]. The RoB for these three trials ranged from moderate to high. ARP and MSP were reported in all three trials, but only two trials reported RRP [ 22 , 24 ]. In all three trials [ 22 – 24 ], electromyography (EMG) biofeedback therapy was implemented after colorectal surgery. One [ 22 ] of the three trials used an anal electrode inserted into the lower rectum, with adhesive electrodes placed on the external oblique muscles, forming a circuit to enable the detection of muscle activities during bowel movements. In this trial, daily biofeedback therapy, provided for 45–60 min per session, was performed for 15 days. Two trials [ 23 , 24 ] provided insufficient descriptions of the method used to detect electrical activity during bowel movements or the treatment parameters. The meta-analysis of all three trials [ 23 , 24 ] ( n = 226) demonstrated a significant effect for biofeedback compared with UC for improving ARP (WMD 9.55; 95% CI 2.59, 16.51; p = 0.01; Fig. 4 A) and MSP (WMD 25.29; 95% CI 4.08, 48.50; p = 0.02; Fig. 4 B). The meta-analysis of the two trials reporting RRP [ 22 , 24 ] ( n = 152) found a significant effect for biofeedback compared with UC on RRP improvement (WMD 0.51; 95% CI 0.10, 0.92; p = 0.02; Fig. 4 C). PFMT plus biofeedback versus PFMT The effects of PFMT alone compared with the effects of PFMT combined with biofeedback on anorectal dynamics were examined in three trials [ 18 , 25 , 26 ]. The RoB of these three trials varied from moderate to high. All three trials [ 25 , 26 ] reported ARP and MSP as outcomes, but only two trials [ 25 , 26 ] reported RRP. Chen (18) and Zheng, Wu (26) included EMG biofeedback therapy in addition to PFMT. The electrical activities of the pelvic floor muscles were measured by anal electrodes in all three trials. Zheng, Wu (26) also used adhesive electrodes to detect the electrical activity of external oblique muscles. Biofeedback therapy was performed two to three times per week for 20–30 min each time, with the total training period ranging from 3 to 13 months. In the trials by Chen (18) and Zheng, Wu (26), PFMT was performed daily, consisting of five sets of 10 repetitions consisting of contractions lasting 5–10 s per repetition, with 10 s of rest between repetitions. Subjects in the study by Yang, Wang (25) performed PFMT by contracting pelvic floor muscles for 10 s. The total PMFT training period varied from 16 months to the time subjects needed to be discharged [ 18 , 25 , 26 ]. The meta-analysis of the three trials [ 18 , 25 , 26 ] ( n = 211) showed a significant effect for biofeedback combined with PFMT compared with PMFT alone on ARP (WMD 3.00; 95% CI 0.40, 5.59; p = 0.02, Fig. 5 A) and MSP (WMD 9.35; 95% CI 0.17, 18.53; p = 0.05, Fig. 5 B). The meta-analysis of the two trials reporting RRP [ 25 , 26 ] ( n = 135) revealed a significant effect for biofeedback combined with PMFT compared with PMFT alone on RRP (WMD 1.54; 95% CI 0.60, 2.48; p = 0.00, Fig. 5 C). Biofeedback versus PFMT The effects of biofeedback therapy alone were compared with the effects of PFMT alone on FI using the CCIS in two trials [ 27 , 28 ]. The RoB for the two trials varied from moderate to high. Both trials provided biofeedback therapy for the experimental group, whereas the control group performed Kegel exercises. Cho, Kim (28) provided the experimental group with biofeedback training designed to strengthen their external anal sphincter. Subjects were educated to slowly contract and relax the pelvic floor muscles and were presented with visual or audible signals proportional to their anal squeezing pressure. The training was provided one to two times per week for 6 months. Kim, Jeon (27) provided the experimental group with biofeedback training in which an anorectal probe was used to train subjects to achieve adequate squeeze pressure using a visual feedback display. Each training session lasted 10 to 30 min, and subjects were encouraged to repeat the exercises five times each day. The PFMT dosage used for the control groups was not specified in either trial [ 27 , 28 ]. The meta-analysis of data from both trials [ 27 , 28 ] ( n = 168) showed a non-significant effect for biofeedback alone compared with PFMT alone on the CCIS (WMD 0.49; 95% CI − 1.68, 2.66; p = 0.66; Fig. 6 ).
Discussion This systematic review and meta-analysis investigated the effectiveness of physiotherapy interventions on FI and QoL following colorectal surgery. The literature searches identified 4413 potentially relevant articles indexed in both English- and Chinese-language databases; however, only 10 trials met the pre-defined inclusion criteria for the meta-analysis. The interventions examined in the included trials were PFMT alone, biofeedback therapy alone, and the combination of PFMT and biofeedback therapy. The RoB of the included trials ranged from moderate to high. Meta-analysis of data from two trials [ 20 , 21 ] comparing PFMT alone with UC revealed a significant effect of the intervention on QoL components, including lifestyle, coping behavior, and embarrassment, as measured using the FIQL. The minimal clinically important difference (MCID) values reported for lifestyle, coping behavior, and embarrassment are 0.2, 0.3, and 0.2, respectively, among the noncancerous population [ 29 ]. The mean estimated effects obtained in the studies included in the current review surpassed the MCID for lifestyle (0.54), coping behavior (1.14), and embarrassment (0.43), indicating that these effects might be clinically meaningful. Despite the significant results obtained for QoL in the current review, the findings are limited by the high RoB, the limited number of pooled trials ( n = 2), and the varying PFMT protocols used by each included study. Nevertheless, considering the effect size and safety of PFMT [ 30 ], this intervention should be considered a potential treatment option for improving QoL among individuals who experience FI following CRC surgery. Meta-analysis of data from trials with moderate to high RoB comparing biofeedback alone with UC revealed significant effects of the intervention on the ARP, MSP [ 22 – 24 ], and RRP [ 22 , 24 ] measures of ARM. Meta-analysis of data from trials [ 18 , 25 , 26 ] with moderate to high RoB identified similar significant effects on FI parameters when PFMT combined with biofeedback was compared with PFMT alone. ARM is a non-invasive procedure used to objectively quantify anorectal function and has been found to be clinically relevant for assessing the severity of FI in both children and adults [ 31 – 33 ]. No MCID has been established for ARM, preventing interpretation of the estimated effect size. However, the 95% CI values for both interventions (PFMT plus biofeedback and biofeedback alone) were below the MCID, indicating the potential for clinically trivial effects. Additional data examining the effects of these interventions would narrow the 95% CI and provide more precise estimates of the average effects of PFMT plus biofeedback and biofeedback alone for the treatment of FI following CRC surgery. The results obtained in the current study for PFMT plus biofeedback agree with results reported in previous systematic reviews [ 34 , 35 ] examining the effects of pelvic floor rehabilitation, including PFMT plus biofeedback, on improving anorectal function following rectal resection surgery. However, these prior systematic reviews [ 34 , 35 ] did not include quantitative analyses, and both reviews included non-RCTs. By contrast, the current review quantitatively evaluated the efficacy of PFMT plus biofeedback and only included RCTs, offering a higher level of evidence [ 36 ]. Although the current review found significant effects for biofeedback alone and PFMT plus biofeedback on FI as measured by ARM, these findings are limited by a considerably high RoB, large variations in the PFMT and biofeedback protocols described in the included trials, and the small number of trials included in the meta-analysis. However, considering the non-invasive nature of PFMT and the minimally invasive nature of biofeedback, both PFMT combined with biofeedback and biofeedback alone should be considered potential interventions for improving FI following CRC surgery. Future studies should investigate additional ARM other than ARP, MSP, and RRP, such as urge volume and volume of first sensation [ 37 ], to obtain a more holistic understanding of the effects of these interventions on FI. Meta-analysis of data from two trials [ 27 , 28 ] comparing biofeedback alone with a PMFT alone revealed a non-significant effect of the intervention on FI following colorectal surgery. Based on these results, no recommendations can be made regarding the effectiveness of biofeedback alone compared with PFMT alone. Future studies of high methodological rigor are required to confirm the results obtained in this review for biofeedback alone compared with PFMT alone. Strengths and limitations of the review The current review has several strengths, including being the first review to include meta-analyses evaluating the effectiveness of various physiotherapy interventions on FI and QoL following CRC surgery. A comprehensive search strategy was applied to identify RCTs evaluating the effectiveness of various physiotherapy interventions for the treatment of FI following CRC surgery. Meta-analyses revealed significant effects for PFMT alone, biofeedback alone, and the combination of PFMT with biofeedback for improving QoL and FI following CRC surgery. Our review also has some limitations. More than half of the included studies (6 of 10) were identified as having a high RoB. Other limitations are the inclusion of unpublished theses, which might hinder the study quality because unpublished studies may be of lower methodological quality than published studies [ 38 ], heterogeneity in terms of PFMT and biofeedback protocols utilized in the included trials, which minimizes the applicability of these findings to clinical settings; a high degree of statistical heterogeneity was evident across the pooled estimates as indicated by large I 2 values, small sample sizes in some of the included trial;, and the small number of trials included in meta-analyses.
Conclusion The current systematic review and meta-analysis identified significant improvements in the lifestyle, coping behavior, and embarrassment components of the FIQL among patients who received PFMT compared with those who received UC. Considering the non-invasive nature of PFMT and the sizes of the effects obtained for this intervention on different QoL components, PFMT should be considered an intervention that may improve QoL among individuals who experience FI following CRC surgery. Meta-analysis revealed that biofeedback alone was superior to UC and that PFMT plus biofeedback was superior to PFMT alone, with both superior interventions resulting in significant improvements in ARP, MSP, and RRP when assessed using ARM. Biofeedback is a minimally invasive intervention that can be applied alone or in combination with PFMT to treat FI following CRC surgery. However, the efficacy of biofeedback alone compared with PFMT alone remains inconclusive. Future high-quality RCTs remain necessary to standardize and optimize PFMT and biofeedback parameters for FI rehabilitation following CRC surgery and to confirm the results obtained in this review.
Purpose To investigate the effectiveness of physiotherapy interventions compared to control conditions on fecal incontinence (FI) and quality of life (QoL) following colorectal surgery. Methods Electronic searches in English-language (Scopus, Web of Science, Embase, AMED, CENTRAL, CINAHL, MEDLINE, Ovid, and PEDro) and Chinese-language (CNKI, Wanfang Data) databases were conducted. Trials comparing physiotherapy interventions against control conditions and assessing FI and QoL outcomes were included in the review. Results Ten trials were included. Meta-analysis revealed statistically significant improvements in lifestyle (0.54; 95% CI 0.03, 1.05; p = 0.04), coping behavior (MD 1.136; 95% CI 0.24, 2.04; p = 0.01), and embarrassment (0.417; 95% CI 0.14, 0.70; p = 0.00) components of QoL among individuals receiving pelvic floor muscle training (PFMT) compared with those receiving usual care (UC). Meta-analysis showed biofeedback to be significantly more effective than UC in enhancing anal resting pressure (ARP; 9.551; 95% CI 2.60, 16.51; p = 0.007), maximum squeeze pressure (MSP; 25.29; 95% CI 4.08, 48.50; p = 0.02), and rectal resting pressure (RRP; 0.51; 95% CI 0.10, 0.9; p = 0.02). Meta-analysis also found PFMT combined with biofeedback to be significantly more effective than PFMT alone for ARP (3.00; 95% CI 0.40, 5.60; p = 0.02), MSP (9.35, 95% CI 0.17, 18.53; p = 0.05), and RRP (1.54; 95% CI 0.60, 2.47; p = 0.00). Conclusions PFMT combined with biofeedback was more effective than PFMT alone, but both interventions delivered alone were superior to UC. Future studies remain necessary to optimize and standardize the PFMT parameters for improving QoL among individuals who experience FI following CRC surgery. Review registration This systematic review is registered in the PROSPERO registry (Ref: CRD42022337084). Supplementary Information The online version contains supplementary material available at 10.1007/s00520-023-08294-1. Keywords Open access funding provided by The Hong Kong Polytechnic University.
Supplementary Information Below is the link to the electronic supplementary material.
Acknowledgements The research team acknowledges Mr. Muhammad Usman Ali for his assistance and support. Author contributions P.K. contributed to the study’s conception and design. P.C. and C.T.C performed electronic and hand searches. P.M.Y. and L.P.H. completed the study screening. P.C., P.K., and C.K.H performed data extraction. All authors conducted the risk of bias assessment and read and approved the final manuscript. Funding Open access funding provided by The Hong Kong Polytechnic University. Declarations Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:42:02
Support Care Cancer. 2024 Jan 13; 32(2):103
oa_package/88/1d/PMC10787910.tar.gz
PMC10787911
37405589
Method Data come from a 100-day intensive longitudinal study of adolescent daily experiences. The study protocol was approved by University of Michigan Institutional Review Board for Health Sciences and Behavioral Sciences (IRB-HSBS). No data have been previously reported. Participants The final sample included 106 adolescents (57.5% girls, 38.7% boys, 3.8% other gender identity 1 ). Adolescents were between 9 and 18 years old ( M = 13.34, SD = 1.92), and 75% were at least 12 years old. Most were White (75.5%) and not Latin(o/a) (91.5%), with others identifying as more than one race (17.9%), Black/African American (5.7%), or American Indian/Alaskan Native (0.9%). Families were recruited through social media, virtual flyers, and university-affiliated databases. Families were defined as one parent or legal guardian and two children (of any degree of genetic relatedness) between 8 and 21 years old, with at least one child between age 8 and 17. After enrollment, some adolescents did not complete the study or meet inclusion criteria for this paper (see below). Thus, the final sample reported here consists of 63 families with 43 sibling pairs and 20 singletons. The average daily response rate of included adolescents was 94%. The full study sample consisted of 174 adolescents (54.9% girls, 42.9% boys, 2.2% other gender identity) aged 8 to 20 years old ( M = 13.37, SD = 2.19). Of these adolescents, 6.3% withdrew from the study, 13.3% were dropped from the study because they failed to complete at least 50% of the daily diaries after the first 30 days, and 19.5% completed the study but were excluded from this paper because their response rate was less than 80%. This follows procedures and simulation results from past research on data fidelity and missingness (Rankin & Marsh, 1985 ; Wright et al., 2019 ). Included participants did not differ from excluded participants demographically (i.e., gender, age, or race/ethnicity, all p s ≥ 0.05), however, they did report fewer impulsive behaviors as well as greater working memory and attentional control (all p s ≤ 0.01); they did not report differences in perceptual sensitivity ( p > .05). Procedure The study was conducted virtually between March 2021 and August 2022. Families first completed a baseline session in which parents provided electronically signed informed consent for themselves and for their children under 18 years old, who also provided informed assent. Adolescents aged 18 years or older provided informed consent. Parents and adolescents then independently completed 90-minute online surveys (using any Internet-capable device) via Qualtrics. Surveys contained questions about their identities, feelings, and behaviors as well as cognitive tests. Then, every night for the next 100 days, adolescents completed a 20-minute online survey. Around 5:00PM, unique survey links were sent to the parent’s email address 2 , who distributed them to their children. Adolescents were asked to take the survey at 8:00PM or after that day’s activities; survey links expired the next day at noon. Among other measures, the daily surveys included questions about externalizing behaviors and social experiences as well as the novel IC task. Each family member received $15 for completing the baseline survey. Adolescents received $1 for each completed daily survey, which doubled to $2 if they completed at least 80% of the surveys; they received a $35 bonus if they completed at least 90% of the daily surveys. Measures The focus of this paper is on the novel intensive longitudinal measure of IC. Baseline measures of cognition (i.e., task-based working memory and self-reported attentional control) and age were used to assess convergent validity; measures of perceptual sensitivity and gender were used to assess discriminant validity. To examine how IC’s daily average and fluctuations were linked to externalizing behaviors, baseline impulsive behaviors were used. Daily impulsive behaviors and social experiences were also used in illustrative person-specific analyses. Baseline Measures Task-Based Working Memory. The Symmetry Span task (Foster et al., 2015 ) was used to assess working memory. Adolescents were shown a series of highlighted red squares in a 4 × 4 black and white matrix. Interspersed was a symmetry task in which adolescents had to judge whether a pattern composed of black and white squares was symmetrical along its vertical axis. Set lengths ranged from 2 to 5 symmetry-matrix combinations (12 trials total). At the end of each set, adolescents were asked to recall the location of each red square in the correct order. Adolescents received two practice trials and were required to maintain at least 85% accuracy on the symmetry trials. The absolute span score (i.e., sum of all perfectly recalled sets) was used in the subsequent analyses (Shipstead et al., 2012 ). Self-Reported Attentional Control. Attentional control was assessed using a subscale from the Early Adolescent Temperament Questionnaire-Revised (EATQ-R; Ellis & Rothbart 1999 ). It contained six items on capacity to focus and shift attention (e.g., “It is easy for me to really concentrate on homework problems”; α = 0.75). Items were rated on a 5-point scale (1 = Almost always untrue to 5 = Almost always true ) with higher scores reflecting greater control. Self-Reported Perceptual Sensitivity. Perceptual sensitivity was assessed using another subscale from the EATQ-R (Ellis & Rothbart, 1999 ). It contained four items on the detection or awareness of slight, low-intensity stimulation in the environment (e.g., “I am very aware of noises.”; α = 0.81). Higher scores reflect greater sensitivity. Self-Reported Impulsive Behaviors. Impulsive behaviors were assessed with the 40-item UPPS-P Impulsive Behavior scale (Lynam et al., 2006 ). It has five subscales, measuring a lack of premeditation (e.g., “I am one of those people who blurt out things without thinking”; α = 0.76), lack of perseverance (e.g., “I tend to give up easily”; α = 0.76), sensation seeking (e.g., “I quite enjoy taking risks”; α = 0.87), positive urgency (e.g., “I tend to act without thinking when I am really excited”; α = 0.93), and negative urgency (e.g., “When I feel bad, I will often do things I later regret in order to make myself feel better now”; α = 0.89). Items were rated on a 4-point scale (1 = Not at all like me to 4 = Very much like me ) with higher scores reflecting greater impulsivity. Daily Measures IC. The Stroop Color Word Test (Golden, 1975 ) was adapted for the 100 daily assessments. In the classic measure, participants received 3 pages with 100 color words each (“red”, “green”, “blue”). On the first page, all color words were printed in black ink. On the second page, all color words were printed in congruent ink (e.g., the word “red” printed in red ink), and on the third page, all color words were printed in incongruent ink (e.g., “red” printed in green ink). Answer choices included each color word printed in black ink. Participants were given 45 seconds per page to circle as many answers corresponding to the ink color as possible. It was expected that more items would be correctly circled on congruent than incongruent pages; the number of correctly answered incongruent items was the IC score. Since 1975, subtraction methods (e.g., comparing reaction time on incongruent and congruent trials) have gained popularity, but they have significant limitations for accurately detecting individual differences (Draheim et al., 2019 ; Weigard et al., 2021 ), and thus, intraindividual variation. Therefore, in this study, adolescents were presented with a randomized set of 100 color words (“red”, “green”, blue”, or “yellow”) in incongruent colors only (e.g., “red” in green font). Consistent with other work (Heitz & Engle, 2007 ), there were no neutral or congruent conditions, and all combinations of words and font colors were presented the same number of times per day. As in the classic measure, adolescents indicated the font color of as many words as they could in 45 seconds by selecting the correct color word presented in black ink (see supplemental materials for task images) 3 . The number of correct responses indexed each day’s IC score; this is consistent with the original measure and has been suggested as a viable alternative to reaction time-based indices (Khng & Lee, 2014 ). On the first day, adolescents also completed five practice trials with feedback. The task is openly available ( https://osf.io/9yabr/ ). To ensure data fidelity, some trials were excluded for some adolescents. Days adolescents completed one or fewer trials correctly were excluded (<0.5%), as they likely reflected low effort or technical issues. Also, days adolescents completed nearly all 100 trials were winsorized to three standard deviations above that day’s average (<0.5%), as they likely reflected technical issues (e.g., screen freezing) and internal testing suggested it would be difficult to complete more than 75 trials. Daily Impulsive and Social Experiences. Daily positive urgency and daily negative urgency (i.e., tendencies to act impulsively when experiencing positive and negative emotions, respectively) were assessed via the Short UPPS-P (Cyders et al., 2014 ); each scale contained four items adapted to reflect adolescents’ impulsive behaviors that day (e.g., positive urgency: “ Today , I tended to act without thinking when I was really excited”) and were rated on a 4-point scale (1 = Not at all like me to 4 = Very much like me ). Similar measures have been adapted and used in intensive longitudinal studies (Sperry et al., 2016 ; Tomko et al., 2014 ). Reliabilities were good, according to multilevel confirmatory factor analysis (Schuurman & Hamaker, 2019 ). For daily positive urgency, between-person ω = 0.80 and within-person ω = 0.79. For daily negative urgency, between-person ω = 0.81 and within-person ω = 0.88. Daily social experiences were assessed using a modified activity questionnaire (McHale et al., 1999 ). Adolescents completed one item indicating how much time (in minutes) they spent visiting or hanging out that day on a sliding scale from 0 to 100. Responses were binned: 0 = Did not visit or hang out today ; 2 (1–49 min) = Visited or hung out a little ; 3 (50–99 min) = Visited or hung out a moderate amount ; 4 (100 + minutes) = Visited or hung out a lot . Further details for the daily measures are provided in the supplemental materials. Analytic Plan Three sets of analyses were conducted. First, the reliability and validity of the novel IC measure were assessed. Second, average daily IC scores and fluctuations in those scores were associated with baseline impulsive behaviors. Finally, personalized network analyses (using GIMME; Gates & Molenaar 2012 ) were conducted for a subset of individuals to illustrate the utility of intensive longitudinal data for person-specific inferences. Analyses were conducted in SPSS (version 26) and R (v4.1.2; R Core Team, 2022 ). Reliability and Validity of Daily IC. Parallel forms reliability was assessed by comparing the interindividual IC means and standard deviations ( SD s) across all 100 days. Daily means and SD s across participants for each day were expected to be approximately equal and normally distributed, suggesting that random differences in stimuli order did not systematically impact IC assessment. Intraclass correlation coefficients (ICCs) were also calculated, with values greater than 0.50 indicating moderate reliability, 0.75 good reliability, and 0.90 excellent reliability (Bartko, 1966 ). Moderate to good reliability across days was expected (see Kelly & Beltz 2021 ). Convergent validity was assessed by correlating each day’s IC score with standard baseline measures of cognition (i.e., working memory and attentional control) across participants. Low to moderate correlations were expected because they assessed similar, but distinct domains of cognition, and in the case of attentional control, in a different modality (Toplak et al., 2013 ). Convergent validity was also assessed by correlating each day’s IC score with age. Older youth were expected to have higher IC. Discriminant validity was assessed by correlating each day’s IC score with baseline perceptual sensitivity and via gender differences. No relations were expected given weak, null, and inconsistent findings in the extant literature (Weafer, 2020 ). All reliability and validity analyses used listwise deletion for daily missing data, but there were at least 92 participants (87%) included in each day’s analysis. Effect sizes for independent analyses were evaluated using r and Cohen’s d , with small effect sizes corresponding to r = 0.1 and d = 0.2, medium to r = 0.3 and d = 0.5, and large to r = 0.5 and d = 0.8 (Cohen, 1988 ). Nested analyses (accounting for family dependencies) showed the same pattern of results as the independent analyses described in the main text; they are available in the supplemental materials. Fluctuations in Daily IC and Links to Impulsive Behaviors. Each adolescent’s average or mean IC (i M ) across the 100 days was calculated. Fluctuations were calculated using intraindividual standard deviations (i SD ); smaller i SD s reflect fewer and/or smaller deviations from an adolescent’s own average, and larger i SD s reflect more and/or larger deviations from that average. For instance, an adolescent with i M = 34 and i SD = 2.7 demonstrates relatively consistent performance across days compared to an adolescent with i M = 34 and i SD = 7.5 who has the same level of performance but with larger variability, sometimes scoring well below or above their average. A one-sample t -test was used to determine whether the sample showed fluctuations (i.e., i SD s significantly different from zero). Gender and age effects were also explored. Multilevel models (nesting individuals within families) were then used to assess associations between baseline impulsive behaviors (i.e., lack of premeditation, lack of perseverance, sensation seeking, positive urgency, and negative urgency) and daily IC (i M , i SD ). Each outcome was assessed in a separate model and all models included age and gender (0 = boys; 1 = girls) 4 . Illustrative Adolescent-Specific Network Analyses . Finally, illustrative adolescent-specific network analyses were conducted via GIMME to highlight the utility of intensive longitudinal methods for future work on externalizing behaviors. Specifically, 12 adolescents who reported any substance use (e.g., tobacco, marijuana, or alcohol) during the 100-day study were matched (see supplemental materials) with 12 adolescents who reported no substance use. This extreme groups comparison is ideal for highlighting heterogeneity among adolescents who use substances (and demographically similar youth who do not). Their daily IC, positive urgency, negative urgency, and social time were linearly detrended by day (as many time series approaches assume stationarity; Beltz & Gates 2017 ), and then submitted to confirmatory subgrouping-GIMME (CS-GIMME; Henry et al., 2019 ). CS-GIMME is a variant of GIMME which uses unified structural equation models (or uSEMs) in combination with a grouping algorithm to derive sparse, person-specific networks of directed relations among intensively measured variables (Gates & Molenaar, 2012 ). It is unique among network approaches because it estimates contemporaneous (i.e., same day) and first order lagged (i.e., next-day) relations, providing some temporal indexing for relations, and because it provides both nomothetic and idiographic inferences via relations that can apply to the whole sample or to a single adolescent. As described in Fig. 1 , GIMME derives person-specific networks through a multistep, data-driven process. The analysis begins with a null model. Then, a directed relation is added between two variables if it would significantly improve model fit for at least 75% of the sample (as determined by Lagrange Multiplier tests; Sörbom 1989 ); these group-level relations are iteratively added until none meets the 75% criterion. In this analysis, autoregressive relations (i.e., variables predicting themselves from one day to the next) are specified at the group-level to facilitate model fitting (a common procedure; Lane et al., 2019 ). Individual-level relations are then added if an individual’s model does not fit well with only group-level relations. After each level is fit, models are pruned of relations that no longer meet criteria, and final models are evaluated using standard fit indices. In this study, CS-GIMME was used to permit a priori comparisons between adolescents who did and did not use substances over the 100 days. In CS-GIMME (Henry et al., 2019), subgroup-level relations are iteratively estimated after group-level relations and before individual-level relations (with a 51% criterion). This means that relations common among the majority of youth who used substances have the opportunity to be estimated separately from youth who did not use substances. Thus, each resulting network reflects a personalized set of relations with unique weights (some of which apply to everyone, some of which apply only to an adolescent’s subgroup, and some of which are unique to an adolescent). GIMME uses full information maximum likelihood, and only includes individuals with 80% or more of the daily dairies. GIMME, and its extensions, have been widely used and are well-supported by largescale simulations (see Gates & Molenaar 2012 ; Henry et al., 2019 ; Lane et al., 2019 ). After GIMME, network node centrality was calculated and compared across subgroups. Node centrality is the number of relations for each variable divided by the total number of network relations. It reflects the relative influence of each behavior in an adolescent’s person-specific model (Beltz & Gates, 2017 ). Independent sample t -tests were conducted to determine whether node centrality differed across adolescents who used substances and those who did not.
Results Reliability and Validity of Daily IC The interindividual means and SD s of IC across 100 days are shown in Fig. 2 . Each bar represents that day’s IC averaged across all adolescents. The thick black line represents the overall IC score ( M = 34.19), aggregated across adolescents and days, and the thin dashed line represents daily variation across adolescents’ scores (overall SD = 1.82). Overall skewness was 2.23 and kurtosis was 8.84, suggesting that the distribution was negatively skewed and heavy-tailed. This is easy to visualize in Fig. 2 ; IC notably increases over the first week (indicated by light gray bars). Excluding this week, the mean for the remaining 93 days was similar, and SD expectedly decreased ( M = 34.53; SD = 1.14); the skewness and kurtosis also decreased to close to zero (skewness: 0.18, kurtosis: − 0.70), approximating a normal distribution. The ICC was 0.53 (95% CI = [0.47, 0.61]), suggesting relatively good reliability across days, and it was 0.56 (95% CI = [0.50, 0.63]) across the last 93 days. Convergent and divergent validity were assessed via correlations of daily IC with standard baseline measures. These correlations are shown in Fig. 3 . Regarding convergent validity, there were small-to-moderate, positive correlations. IC’s average correlation with working memory (Fig. 3 , dark gray) was r = .16 ( SD = 0.10; range: 0.03 - 0.37), and 37% of the correlations were significant at p < .05. IC’s average correlation with attentional control (Fig. 3 , light gray) was r = .20 ( SD = 0.07; range: 0.01 - 0.37), and 54% were significant. IC’s average correlation with age was r = .20 ( SD = 0.07; range: 0.01 - 0.38), and 52% were significant; thus, older adolescents had higher IC. Regarding divergent validity, perceptual sensitivity (Fig. 3 , black) was not significantly correlated with any day’s IC performance (average r = -.03, all p s > 0.05). On average, girls ( M = 35.03; SD = 11.72) had better daily IC than boys ( M = 33.43; SD = 11.30), but this difference was not significant ( p > .05), and girls only significantly outperformed boys on a single day, which likely reflects Type I error. The average effect size (Cohen’s d ) across all 100 days was small ( d = 0.15; SD = 0.11; range: 0.002 - 0.56). Each daily correlation and multilevel models of the same relations adjusting for family dependencies are provided in supplemental materials.
Discussion IC is often conceptualized as a relatively stable construct, but growing evidence suggests that cognition varies across contexts and time (Brose et al., 2014 ; Sliwinski et al., 2006 ). Although this has significant implications for adolescent externalizing behaviors, the relevant literature has been stagnated by a lack of suitable intensive longitudinal cognitive assessments. Thus, the present study considered the extent to which adolescent IC – captured by 100 novel daily measurements – fluctuates from day-to-day in ways that are meaningfully associated with impulsivity and differ by the daily experiences of adolescent who do and do not use substances. IC Can Be Measured Daily Findings suggest that the novel IC task is a reliable and valid method for assessing daily cognitive processes related to inhibition. Parallel forms reliability was determined by comparing M s and SD s across days and calculating ICCs. The 100-day distribution of scores was non-normal: Specifically, IC performance increased over the first week before stabilizing, which may be due to initial task-related learning. Indeed, when the first seven days were excluded, IC scores from the remaining 93 days approximated a normal distribution. Moreover, variation across participants increased over the course of the study until about day 70, indicating increasing then plateauing between-person differences in IC. This is consistent with the notion that contextual demands can reduce individual differences (Miller et al., 2006 ), with the novelty of the study being an initially constraining context. ICCs also indicated that individuals’ scores displayed moderate reliability across days (Bartko, 1966 ). This is aligned with expectations for daily studies of cognition - prior work has suggested moderate ICCs are likely well-suited to intensive longitudinal research, as they reflect both between-person stability of assessments and potentially true variation in those assessments within a person over time (Bolger & Laurenceau, 2013 ). Notably, most other daily cognition validation studies have not shown similar first-week effects, although there was evidence for learning in a mental rotation task across a 75-day study (Kelly & Beltz, 2021 ), and there was a short-term practice effect for executive function over the first 4 days of a recent 2-week study (Wang et al., 2021 ). Thus, the current findings may reflect something unique about executive function (or repeated assessments of it), as IC is typically thought to be a subdomain. Findings could also reflect something unique about adolescence. Younger participants may need more time to become accustomed to intensive assessments (in which case, the first week could reflect noise), or adolescents may be more flexible and malleable thinkers compared to young and older adults (Laube et al., 2020 ), who have been the participants in most intensive longitudinal studies (in which case, the first week could reflect learning). This is an exciting area for future research. Clearly, this finding has significant implications for the design and implementation of future intensive longitudinal research as well. Quite a few ‘intensive’ studies are 14 days long (McNeish et al., 2021 ), which means that their first half could reflect task-related noise or learning effects rather than reflections of daily experiences. Extended training opportunities may be necessary in future studies, and data analytics that explicitly consider stationarity should be used. For instance, daily data detrending was used in this paper’s network analyses and other studies have excluded the first day (or several days) of data collection (e.g., Kelly & Beltz 2021 ), but these approaches are debatable, and no consensus exists. If more than 14 days of data are available, then continuous time and other models of non-stationarity could be employed, which is an area ripe for future applications (Zhu et al., 2021 ). Validity was assessed via correlations of daily IC with standard baseline measures (across participants, separately for each day). As expected, correlations with cognitive measures (i.e., working memory and attentional control) and age were low to moderate, but were always positive and were replicated in multilevel analyses adjusting for sibling dependencies; this provides some evidence of convergent validity. Also as expected, perceptual sensitivity and gender were not associated with IC, providing some evidence of divergent validity. As higher correlations between daily IC and baseline cognition might have been expected, the reported moderate correlations could reflect some degree of unreliability or some influence of daily experiences. IC and working memory are distinct executive functioning skills that continue to differentiate during adolescence (Miyake & Friedman, 2012 ), and low correlations ( r = 0.20 - 0.30) are common in nomothetic and population-level studies (Chaku et al., 2022 ; Ferguson et al., 2021 ). Attentional control was also measured via self-report and low correlations are expected with cross-modality measures (Snyder et al., 2021 ). In task-based measures, participants provide ‘objective’ data with narrowly defined goals, but in self-report measures, participants provide ‘subjective’ data about how they broadly utilize cognitive skills across situations and time (Dang et al., 2020 ). One exciting direction for future intensive longitudinal research involves assessing multiple inhibitory control tasks, or even other executive functioning tasks. Indeed, a recent study assessing multiple daily executive function tasks over two weeks found that about 40% of daily variation in the latent construct could be attributed to within-person variability (Wang et al., 2021 ). Thus, future work could consider additional psychometric properties of daily IC or even individual differences in the structure of executive function via multilevel or person-specific factor analyses, respectively (Nesselroade & Ford, 1985 ; Vogelsmeier et al., 2019 ). Alternatively, moderate correlations between daily and baseline measures could reflect true variation. Indeed, unreliability and true daily fluctuation would both appear as occasion-to-occasion variation in daily data. The IC task was developed specifically to capture adolescents’ daily cognition. It was expected to fluctuate and change potentially in concert with adolescent experiences (Molenaar, 2004 ; Nesselroade, 1991 ), whereas existing, standard cognitive measures assume stability across contexts and that variation over short periods of time is merely noise (Bielak et al., 2017 ). Given these conceptual differences, it is not surprising that daily IC exhibited low-to-moderate correlations with baseline cognition. They shared some degree of common variance, but daily IC ultimately provided distinct, complementary information. As this measure is openly available ( https://osf.io/9yabr/ ), there is ample opportunity for future work to examine the extent to which IC fluctuations are truly explained by daily experiences. Daily Fluctuations in IC Matter for Impulsive Behaviors IC exhibited significant fluctuations across days. Gender was not associated with fluctuations, but older adolescents demonstrated fewer fluctuations than younger adolescents. This aligns with an extant literature showing that intraindividual variability decreases from childhood to adolescence, as cognitive skills become more efficient and stable (Dykiert et al., 2012 ; Hultsch et al., 2011 ). Thus, intraindividual variability may contain valuable information about development (Lövden et al., 2013 ; Tamnes et al., 2012 ) and IC fluctuations (i.e., i SD ) may be a developmental marker – a construct independent of an individual’s level of IC (i.e., i M ). Links between daily IC and baseline impulsivity support this notion. Notably, fluctuations in IC were uniquely associated with greater lack of perseverance, suggesting that adolescents whose IC levels varied more across days were also more likely to give up during difficult tasks. This was true even when controlling for average IC, indicating that fluctuations in IC provide distinct information about constructs related to impulsivity. Adolescents who experience more fluctuations in IC may live in more chaotic environments or may be more susceptible to environmental changes (e.g., less sleep, more stress); thus, variability may be a developmental marker of stress reactivity or vulnerability (Nesselroade, 1991 ). Understanding how these fluctuations are linked to everyday experiences is just beginning to be realized and future time-indexed, multivariate studies would be well-poised to investigate these questions. Beyond fluctuations and perseverance, higher average IC was also associated with less positive and negative urgency, suggesting that average IC – assessed in the context of adolescents’ everyday lives – may be particularly relevant for impulsive behaviors that occur in emotional or motivational situations. This nomothetic inference is consistent with the larger literature suggesting that IC is particularly salient for externalizing behaviors (Heffer & Willoughby, 2021 ). Adolescents Are Unique Research has largely focused on characterizing the ‘average’ adolescent, but averages cannot characterize the biopsychosocial experiences that are unique to a single adolescent (Molenaar, 2004 ), so a person-specific approach (i.e., GIMME) was used to accurately model developmental heterogeneity among the daily data of adolescents who used substances and those who did not (Gates & Molenaar, 2012 ; Henry et al., 2019 ). Even though GIMME prioritized commonalities across all adolescents and within subgroups during model building, none were found; in fact, no two individuals shared the exact same pattern of associations. This highlights the significant heterogeneity among adolescents who use substances and how links between IC and externalizing behaviors may unfold in unique ways for unique youth. As with all approaches, though, some caution is warranted because several features of the dataset (e.g., power or measurement intervals) could impact estimation of individual-level network features (see Weigard et al., 2021 ). Despite this, indices that quantified the importance of each variable in the adolescent-specific networks (i.e., centrality) suggested that IC played a more important role in the networks of adolescents who used substances than those who did not. Thus, IC may represent a hub that integrates and distributes information within and between externalizing and social behaviors for adolescents who use substances. Indeed, nomothetic studies have found that IC explains more variance (i.e., plays a central role) in externalizing behaviors than working memory or other cognitive skills (Young et al., 2009 ). Future intensive longitudinal work could explore the specific influence of IC on externalizing behaviors, for example, by leveraging the direction of relations with IC to determine whether it is more likely to influence (i.e., out-degree) or be influenced by (i.e., in-degree) other network behaviors. Such directional inferences require specialized network analyses, such as the multiple solutions version of GIMME (Beltz & Molenaar, 2016 ). Although this GIMME analysis only focused on a subset of adolescents who used substances in order to emphasize insights into externalizing behaviors, it nonetheless illustrates one way in which intensive longitudinal data can be used to inform future personalized studies of externalizing behaviors. Clinical science already has a rich history of examining externalizing behaviors in context via EMA (reviewed in Votaw & Witkiewitz 2021 ), and future research could benefit from intensively measuring individuals during interventions, paving the way forward for individualized, tailored, and even adaptive treatments (Ginexi et al., 2014 ). For example, a randomized clinical trial used 2-week pre-treatment EMA data to develop individualized cognitive behavioral therapy plans for individuals struggling with substance use, and those strategies were utilized during temptation, according to post-treatment EMA data (Litt et al., 2009 ). As EMA typically has fewer and shorter measures than the 100-day intensive longitudinal study described here, it is easy to envision the extensive opportunities for personalization in both basic and applied science. There are also new and exciting questions that could be answered with personalized analyses of intensive longitudinal data. For instance, although relations between cognition and other behaviors (e.g., sleep, affect, and mood) are well-established at the between-person level, there is limited and somewhat contradictory, evidence at the within-person level (Hawks et al., 2022 ; Neubauer et al., 2019 ). As reliable and valid intensive cognitive assessments (like the IC task used here) become increasingly available, though, within-person insights may become clear. Another important direction concerns how intraindividual variation in IC (e.g., its stability or behavioral network centrality) over development is related to interindividual differences in sensation seeking and other externalizing behaviors (e.g., substance use, antisocial behavior, or conduct disorders) known to normatively shift across adolescence. IC’s daily fluctuation or centrality in a network may be a marker of development or maturation, which could be revealed in measurement burst designs that incorporate intensive measurement (e.g., over days or weeks) with traditional longitudinal measurement (e.g., annually; Sliwinski 2008 ). Study Considerations This study had several limitations that should be considered. First, data collection was completed during the novel coronavirus-19 (COVID-19) pandemic; inferences may have been impacted by this period of heightened instability. All testing was conducted during the pandemic, though, so effects were similarly distributed across adolescents, and intensive longitudinal studies emphasize intraindividual variation (not necessarily levels ). Second, because all testing was conducted virtually, a small subset of days (representing < 0.5 of all data) may have been affected by technical difficulties (e.g., screen freezing or spotty connectivity). Further, it was difficult to examine potential impacts of the timing of daily survey completion (e.g., whether adolescents completed a daily survey the next morning) due to software limitations. This is somewhat common in intensive longitudinal studies (Keijsers et al., 2022 ). Future research could use local application-based assessments (that do not require Internet connectivity) or ask participants to report technical difficulties, but these alternatives may be costly or burdensome for participants. Third, the sample ( N = 106) was smaller than in other validation studies (Collins et al., 2016 ), including some that included intensive longitudinal cognitive measures (e.g., Kelly & Beltz 2021 ), but the 100-day time series was longer than other intensive longitudinal studies, and retention was excellent. Of recruited adolescents, 61% completed over 80% of the daily diaries (with an average response rate of 94%). This completion rate is like other intensive longitudinal studies using adult samples (Teague et al., 2018 ; Wright et al., 2019 ) and better than others using adolescent samples (Keijsers et al., 2022 ). Fourth, although the sample was generally reflective of the surrounding area, there was limited race/ethnic diversity, with most adolescents endorsing White race/ethnicities (75.5%); thus, findings may not be generalizable to all adolescents and must be replicated in larger and more representative samples of uniquely diverse youth. Further, although excluded participants were similar in age, gender, and race/ethnicity to included participants, they did report more impulsive behaviors and lower cognitive skills, suggesting that study findings may also have limited generalizability to participants with high externalizing problems. This is unfortunately a common finding in traditional cross-sectional and longitudinal research even though other intensive longitudinal studies do not systematically report associations between psychological characteristics and participant retention (see Ewing et al., 2018 ). Although not without trade-offs, ways to facilitate retention include using shorter daily surveys, collecting data for fewer days, implementing planned missingness or related designs (Yuan et al., 2020 ), and leveraging passive sensing data (Ponnada et al., 2022 ). The present study was designed to be 100 days because it provided enough within-person power to estimate person-specific models with 20% missing data (Rankin & Marsh, 1985 ); with a 30-day study, fewer data would likely be missing, but person-specific analyses could not be conducted. Fifth, identical measures were not used at baseline and in the daily assessments, limiting direct comparisons, but this is true in almost all intensive longitudinal studies, as measures must be adapted for repeated use (Chaku & Beltz, 2022 ). Finally, only adolescents (e.g., not parents) reported on daily externalizing behaviors; there was low endorsement of externalizing behaviors generally, and clinically-significant externalizing disorders were not directly assessed. Only 12 adolescents endorsed using any type of substance during the study and rates of impulsive behaviors were generally low. This could be related to adolescent ability to access substances during the COVID-19 pandemic. Nonetheless, these are important avenues to explore in future research. Parent reports would be especially relevant for younger samples and replicating these analyses in clinical samples could shed light on patterns unique to those on the externalizing spectrum, as daily IC may represent different constructs in those with, for example, substance use disorders versus antisocial behavior.
Conclusions Externalizing behaviors normatively increase during adolescence in ways that accentuate individual differences and implicate IC (Young et al., 2009 ). Yet, IC – and its expression across different contexts and times – is thought to be stable even though growing evidence suggests it may vary within individuals (Brose et al., 2014 ; Sliwinski et al., 2006 ). Thus, single assessments that describe ‘average’ adolescents may not accurately capture the lived experiences of individuals, including their daily highs and lows. In an attempt to fill this knowledge gap, this intensive longitudinal study followed over 100 adolescents for 100 days while they completed a novel, openly available, IC task. The task was generally found to be reliable and valid, with daily levels of IC and fluctuations in IC associated with individual differences in impulsive behaviors. Moreover, an adolescent-specific network analysis comparing adolescents who used substances over the 100 days to those who did not revealed that IC was central to the daily interplay among impulsivity and adolescent social experiences, but only for substance users. Thus, the new IC measure – and the illustration of its use in adolescent intensive longitudinal research – encourages future innovation in the investigation of adolescent-specific externalizing behaviors.
Inhibitory control is a transdiagnostic risk factor for externalizing behaviors, particularly during adolescence. Despite advances in understanding links between inhibitory control and externalizing behaviors across youth on average , significant questions remain about how these links play out in the day-to-day lives of individual adolescents. The goals of the current study were to: (1) validate a novel 100-occasion measure of inhibitory control; (2) assess links between day-to-day fluctuations in inhibitory control and individual differences in externalizing behaviors; and (3) illustrate the potential of intensive longitudinal studies for person-specific analyses of adolescent externalizing behaviors. Participants were 106 youth (57.5% female, M age = 13.34 years; SD age = 1.92) who completed a virtual baseline session followed by 100 daily surveys, including an adapted Stroop Color Word task designed to assess inhibitory control. Results suggested that the novel task was generally reliable and valid, and that inhibitory control fluctuated across days in ways that were meaningfully associated with individual differences in baseline impulsive behaviors. Results of illustrative personalized analyses suggested that inhibitory control had more influence in the daily networks of adolescents who used substances during the 100 days than in a matched set of adolescents who did not. This work marks a path forward in intensive longitudinal research by validating a novel inhibitory control measure, revealing that daily fluctuations in inhibitory control may be a unique construct broadly relevant to adolescent externalizing problems, and at the same time, highlighting that links between daily inhibitory control and impulsive behaviors are adolescent-specific. Supplementary Information The online version contains supplementary material available at 10.1007/s10802-023-01071-y. Keywords
Behaviors on the externalizing spectrum, such as impulsivity, substance use, and conduct problems, increase during adolescence. Although this increase is generally considered normative, individual differences in externalizing behaviors can result in social difficulties, school failure, and persistent deviance (Odgers et al., 2008 ). Inhibitory control (IC; the ability to suppress prepotent responses) may play a key role in regulating externalizing behaviors (Young et al., 2009 ). This field of research, however, has overwhelmingly relied on cross-sectional or a few longitudinal assessments, and typically represents IC as a trait or locally stable construct. Yet, growing evidence suggests that cognition fluctuates at granular timescales and is meaningfully linked to personal characteristics and wellbeing (Brose et al., 2014 ; Sliwinski et al., 2006 ). Thus, it is likely that IC is associated with externalizing behaviors in unique ways for individual adolescents, but that the research methods employed to-date could not empirically detect this. Intensive longitudinal designs can be used to understand cognitive fluctuations and their psychological significance. These methods involve many repeated assessments of the same variables and participants over relatively short periods of time: from moment-to-moment (e.g., neural activity; Demidenko et al., 2022 ), hour-to-hour (e.g., affect; Brose et al., 2014 ), or day-to-day (e.g., personality; Kelly et al., 2020 ), revealing how behavior, including cognition, unfolds within an individual. Intensive longitudinal designs can also support a range of inferences, including generalizable nomothetic inferences made from examinations of interindividual (i.e., between-person) variation, idiographic inferences about unique individuals made from examinations of intraindividual (i.e., within-person) variation, or a combination of the two (Molenaar, 2004 ). Unfortunately, there is a lack of cognitive measures that are well-suited for intensive longitudinal assessment, potentially because highly controlled, lab-based measurement is at odds with mobile, ecologically-valid assessments (Roche et al., 2016 ). Thus, the psychometric validation of intensive longitudinal cognitive assessments is a challenging, nascent area of research (but see Kelly & Beltz 2021 ; Sliwinski et al., 2018 ). The goals of the current study were to: (1) develop and validate a novel 100-occasion measure of IC; (2) assess how daily IC fluctuations were related to baseline impulsive behaviors; and (3) explicate the potential of idiographic analyses for revealing adolescent-specific links among daily IC and externalizing behavior in social contexts. IC is a Risk Factor for Adolescent Externalizing Behaviors IC is one of the most salient cognitive factors associated with externalizing behaviors (Young et al., 2009 ). Although most work has focused on children (see Schoemaker et al., 2013 ), IC plays an important role in adolescence, too. Deficits in IC reliably mark adolescent populations with clinically relevant problem behaviors. For instance, low IC predicts early onset of alcohol use-related problems (López-Caneda et al., 2014 ), illicit drug use (Nigg et al., 2006 ), and high likelihood of antisocial or oppositional defiant disorder diagnoses (Bonham et al., 2021 ). Even in non-clinical samples, low IC is consistently linked with increased externalizing behaviors in both cross-sectional and longitudinal research (Kim-Spoon et al., 2019 ). IC may be particularly important for understanding adolescent behaviors because it continues to mature into early adulthood (Luna et al., 2015 ) and occurs in the context of early adolescent increases in neural reactivity to socioemotional stimuli (Foulkes & Blakemore, 2018 ), which may contribute to normative rises in externalizing behaviors (Lydon-Staley & Bassett, 2018 ). Most studies on this topic to-date are cross-sectional and limited to nomothetic inferences even though they assume cognitive control and socioemotional reactivity change within individuals over time (see Beltz 2018 ). Thus, there is theoretical acknowledgement of the importance of intra individual variation for adolescent externalizing behaviors, but there is little empirical evidence for it, as the extant literature primarily concerns inter individual variation (Chaku & Beltz, 2022 ). Intensive longitudinal research is needed to fill this critical knowledge gap. From Nomothetic to Idiographic Assessments of IC IC is often conceptualized as a relatively stable trait that can be assessed once and generalized across time. Converging evidence increasingly shows, however, that short-term fluctuations in cognition have significance for psychological wellbeing (Castellanos et al., 2005 ; Hultsch et al., 2011 ). Numerous factors, such as interpersonal experiences, contextual demands, and emotional processes, can influence or be influenced by cognition at any given moment. For example, increased working memory fluctuations (measured across four weeks with four assessments a day) have been associated with poor sleep quality in early adolescents (Galeano-Keiner et al., 2022 ). Similarly, better working memory (measured daily over two weeks) has been associated with less negative mood and greater school belonging in late adolescence (Wang et al., 2021 ). Unfortunately, work in this area is challenged by a lack of sufficient measurement tools. Current cognitive assessments (e.g., Stroop Color Word, Stop-Signal, and Go/No Go tasks) have primarily been used to capture normative IC-linked processes, but they have limitations (see Snyder et al., 2015 ). Specifically, recent work suggests that the subtraction methods often used to analyze recorded task data (e.g., comparing reaction times on incongruent versus congruent trials) are unreliable and do not index individual differences well (Salthouse & Hedden, 2002 ; Weigard et al., 2021 ), which limits their suitability for studies of intraindividual variation. Nonetheless, understanding short-term cognitive fluctuations is essential because patterns of intraindividual variability may reliably differ across individuals, potentially reflecting an enduring personal characteristic (Nesselroade, 1991 ). Although the extent to which fluctuations may reflect vulnerability versus resilience, and in turn adjustment, likely depends on the specific cognitive construct, context, or sample, the general importance of cognitive variation is being increasingly realized. For instance, greater response variability on a lab-based cognitive task was associated with lower perceptions of real-world risks (e.g., substance use) in 14-to-16-year-olds (Goldenberg et al., 2017 ). Yet, most studies only capture trial-to-trial fluctuations during a single, highly controlled task, so little is known about how longer-term fluctuations in IC are associated with externalizing behaviors in everyday life. This is unfortunate because daily fluctuations are likely related to adolescents’ unique but recurring biopsychosocial experiences. Beyond understanding fluctuations in cognition, daily data (if abundant) can also be used to make person-specific inferences. Adolescence is characterized by increased, variable opportunities for learning, adaptation, and exploration (Knoll et al., 2016 ). These experiences do not just differ in degree, but also in nature and biopsychosocial context across individual adolescents (Chaku & Beltz, 2022 ). To capture this heterogeneity, averages (and even average fluctuations) must be eschewed in favor of person-specific models (Molenaar, 2004 ). For instance, group iterative multiple model estimation (GIMME; Gates & Molenaar 2012 ) is a person-specific analytic technique that maps variation among pre-selected variables within a person over many relatively short intervals. Applied to daily adolescent behavior, GIMME considers each adolescent as if they are a sample of N =1, allowing for unique inferences about the timing of relations among variables (i.e., same-day or next-day) that potentially reflect adolescents’ real-world contexts. GIMME is unique among person-specific analytic approaches because it also affords some generalizable inferences. It does so without averaging by prioritizing relations that are common across a sample. Thus, when combined with daily data, GIMME and other person-specific techniques can be used to make potentially important insights into when , for whom , and in what situations links between IC and externalizing behaviors matter. Intensive Longitudinal Measures of IC The feasibility of intensive longitudinal studies is increasing due to wide accessibility of internet-capable devices and sophisticated analytic techniques for dependent data (Lydon-Staley & Bassett, 2018 ). Still, there is a paucity of cognitive assessments available for inclusion in these studies. Beyond their traditionally restricted use in laboratories, this may be because cognitive measures cannot be intensively repeated due to trial counts, length, or practice effects, or because they require timing mechanisms that are difficult to apply in situ (Ladouce et al., 2016 ). Nonetheless, a few intensive cognitive assessments have been validated. For example, Sliwinski and colleagues ( 2018 ) validated working memory (i.e., dot memory and n-Back) and perceptual speed (i.e., symbol search) tasks across 14-day ecological momentary assessment (EMA) bursts. Although links to other behaviors were not examined, the working memory task exhibited lower intraindividual reliability than the perceptual speed task. Recently, Kelly and Beltz ( 2021 ) validated novel, openly available 75-occasion measures of delayed verbal recall ( https://osf.io/vhr7u/ ) and three-dimensional (3D) mental rotations ( https://osf.io/m6ae8/ ). This involved creating 75 unique sets of stimuli and establishing procedures for parallel forms reliability across sets. Overall, the 3D mental rotations measure had better psychometrics than the delayed verbal recall measure, however, there are significant challenges in distinguishing between unreliability and meaningful fluctuation in intensive longitudinal measurement (Keijsers et al., 2022 ). Thus, the authors posed that verbal recall may be more influenced by daily experiences compared to mental rotations but did not empirically examine this possibility. This small literature suggests that intensive longitudinal measurement of cognition is feasible, but it has not yet been widely employed to study adolescent IC in everyday contexts. The Current Study IC likely varies from day-to-day in ways that have significance for externalizing behaviors in unique adolescents, but measures are not yet available to examine this. Thus, the overarching study goal was to reveal adolescents’ daily associations between IC and externalizing behaviors through three aims. First, a 100-occasion measure of IC was developed and psychometrically validated. Second, daily IC (both 100-day averages and fluctuations) were related to baseline impulsive behaviors; low averages and high fluctuations were expected to predict impulsivity. Third, the potential of person-specific analyses for revealing links between IC and externalizing behaviors was illustrated by comparing the daily networks of adolescents who used substances to the networks of a matched subsample who did not. Fluctuations in Daily IC and Links to Impulsive Behaviors Daily IC scores for 12 illustrative adolescents are shown in Fig. 4 along with their i M s and i SD s. Adolescents demonstrated remarkably different trajectories. For example, ID 36 exhibited large day-to-day variation (i SD = 10.44), whereas the ID 73 exhibited far less variability (i SD = 3.16). Further, some adolescents were characterized by a relatively flat trajectory (e.g., ID 11) while others were better characterized by positive (e.g., ID 77), negative (e.g., ID 41), or non-linear (e.g., ID 29) trajectories. Across adolescents, the i M ranged from 12.79 to 55.58, and the i SD ranged from 3.16 to 15.44. A one-sample t -test indicated that the i SDs were significantly different from zero, t (105) = 31.53, p < .001, providing statistical evidence for IC fluctuations. The i M and i SD were not significantly correlated ( r = -.01, p = .91). Although the i M s and i SD s did not vary by gender ( p s > 0.05), they did covary with age, as older adolescents had higher scores (i M : r = .27, p = .01) and fewer fluctuations (i SD : r = -.19, p = .04). Multilevel models (nesting siblings within families) were then used to examine how i M s and i SD s were associated with impulsive behavior, accounting for gender and age. Higher i M s were associated with less positive urgency, b = -0.02(0.01), p = .03, and less negative urgency, b = -0.02(0.01), p = .04, suggesting that adolescents who had higher average daily IC tended to make less impulsive choices during extreme emotions. Higher i SD s were associated with lack of perseverance, b = 0.03(0.01), p = .03, even after controlling for i M s; this suggests that adolescents who had more variable IC were more likely to give up during difficult tasks. No significant links were found between daily IC and lack of premeditation or sensation seeking. Analyses were repeated excluding the first week of skewed assessments, and inferences were the same; see the supplemental materials. Illustrative Adolescent-Specific Network Analyses All adolescent-specific GIMME models fit the data well according to average indices: CFI = 0.99, NNFI = 1.00, RMSEA = 0.02, SRMR = 0.06. Surprisingly, there were no group- or subgroup-level relations (besides the specified autoregressions), indicating substantial heterogeneity among adolescents who used and did not use substances during the 100 days. Figure 5 A shows the network of an adolescent (11.42-year-old male) who used alcohol. Circles represent variables and directed arrows represent relations between those variables; solid lines reflect same-day relations, dashed lines reflect next-day relations, red lines reflect positive relations, and blue lines reflect negative relations. Thus, this adolescent’s network contains eight relations, and only four were not specified a priori (i.e., the autoregressives). There is a positive, same-day relation between social time and positive urgency, indicating more impulsive behaviors due to positive emotions on days he was more social. There is also a pair of relations between positive and negative urgency, suggesting that negative urgency is related to same-day increases in positive urgency that propagate across time. Finally, there is an inverse, lagged relation between negative urgency and IC suggesting that increases in negative urgency are associated with IC declines the next day. This could reflect a downstream cognitive consequence of actions taken to alleviate unwanted emotions. Figure 5 B shows the network of the matched adolescent (11.83-year-old male) who did not use substances. His network is less dense, with only two relations that are not autoregressives. Like the adolescent in Fig. 5 A, there is a positive, same-day relation between negative and positive urgency. There is an additional positive, lagged relation between social time and negative urgency, though, potentially suggesting that his social interactions are related to next-day increases in negative urgency. Interestingly, however, his daily IC was not linked to externalizing or social experiences. It is important to emphasize that the adolescents in Fig. 5 do not characterize the all the adolescents in the sample nor do they characterize all the adolescents who used substances or those who did not in the sample; otherwise, the relations would have appeared at the group- or subgroup-level. Nonetheless, follow-up analyses suggested that adolescents who reported using substances tended to have more relations involving IC (i.e., greater IC centrality) compared to adolescents who did not use substances, t (22) = -2.24, p = .05, d = 0.74. Most (60%) relations with IC were negative (i.e., worse IC was associated with more negative urgency), suggesting that IC was more salient for the daily experiences of substance users. Electronic Supplementary Material Below is the link to the electronic supplementary material.
Acknowledgements The authors thank members of the Methods, Sex differences, and Development – M(SD) – Lab at the University of Michigan, especially Amy Loviska who assisted with study implementation, Kaitlyn Zhao and Emma Donnelly-Sironen who assisted with participant testing, Jennifer Cleary and Heidi Westerman who assisted with clinical assessments, and Gwyn Reece who assisted with data management. The authors also thank Anthony Provenzola for providing programming expertise related to the novel inhibitory control task. Finally, the authors thank the adolescents and their families for graciously sharing their daily experiences. Funding This research and A.M.B. were supported by the Jacobs Foundation. This work was also supported in part by a James S. McDonnell Foundation Understanding Human Cognition Opportunity Award to A. Beltz (doi: 10.37717/2020-1145). N.C. was supported by NICHD T32 HD007109 (to Chris Monk, Vonnie McLoyd, and Susan Gelman). A.S.W was supported by K23 DA051561. The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Declarations Conflict of interest The authors have no competing interests to declare.
CC BY
no
2024-01-15 23:42:02
Res Child Adolesc Psychopathol. 2024 Jul 5; 52(1):93-110
oa_package/6c/cb/PMC10787911.tar.gz
PMC10787912
37878150
Background Breast cancer affects almost 56,000 women every year in the United Kingdom (UK) [ 1 ] and despite improvements in treatment, approximately 40% of these will require mastectomy [ 2 ]. Seroma formation following mastectomy and/or axillary clearance is common, with reported incidence in the literature varying between 10 and 85% [ 3 ]. Although rarely a serious complication of breast surgery, seroma can cause delayed wound healing, require repeated aspiration with the risk of infection, and may delay the start of adjuvant treatments [ 4 , 5 ]. Strategies to reduce the formation of seroma include the use of surgical drains and flap fixation methods such as quilting or tissue glue and external compression which all act by minimising the surgical dead-space and evidence to support the effectiveness of different approaches has been summarised in several systematic reviews [ 6 – 10 ]. Many of these reviews, however, have highlighted the lack of high-quality research to support practice and the need for future well-designed studies in this area. For future research to be meaningful, it is vital that the study design should reflect current practice and address key uncertainties that are important to patients and the clinical community. We aimed to survey breast surgeons to determine current approaches to the management of seroma in the UK; particularly the use of drains after simple breast cancer surgery to inform the feasibility and design of a future randomised controlled trial (RCT).
Methods An online national practice survey was developed in REDCap ® to capture current UK practice regarding the use of strategies to reduce seroma formation following mastectomy and axillary surgery including the use of drains and flap fixation techniques, and details of the patient pathway for the management of seroma post-operatively. Preliminary work suggested that drains were the most commonly used method of reducing seroma in the UK, so questions focussed on evaluating the feasibility of a future trial comparing the use of drains versus no drains and key elements of trial design including inclusion/exclusion criteria and selection of the primary outcome (Appendix 1). All consultant breast surgeons and senior breast surgery trainees/fellows, defined as being within their final two years of training, were invited to complete the survey through the professional associations (Association of Breast Surgery (ABS) and the Mammary Fold (UK trainee breast surgery group) and via social media networks. The survey was open for a 4-month period (December 2021–March 2022) and regular reminders were sent to optimise participation. Simple descriptive summary statistics were calculated for each survey item and free text responses were analysed using content analysis [ 11 ].
Results Respondent demographics A total of 147 responses were received of which 97 were completed with data that could be included in the analysis. The partially complete responses were excluded from further analysis where they had very limited, or no data entered. Respondent demographics are summarised in Table 1 . Most respondents were consultant breast surgeons ( n = 75, 77%) with responses received from surgeons practicing across the UK (Table 1 ). Use of peri-operative interventions to reduce seroma formation Of the 97 surgeons who completed in the survey, the majority ( n = 82, 85%) used drains either routinely ( n = 38, 39%) or in certain circumstances ( n = 44, 45%). This was most frequently a single drain, although when mastectomy was combined with an axillary node clearance (ANC), two drains were used by a proportion of surgeons (Table 2 ). The indications for selective drain use included patient (e.g. age, high body mass index or large breasts) and treatment factors (e.g. extent of surgery or after neo-adjuvant chemotherapy). The most common reasons for not using a drain included non-compliance and patient preference (Table 2 ). Whilst further details on how patient preference influences drain use were not collected by this survey, drain placement is part of the informed consent process and some patients may therefore decline drain insertion despite the reasons for use being explained. Few surgeons ( n = 26/97, 27%) reported routinely using flap fixation methods to reduce seroma. The most commonly used methods were quilting ( n = 11, 11.5%) and glue sealants ( n = 8, 8.5%). These methods were more frequently used by surgeons who did not routinely use drains ( n = 20, 34% vs n = 6, 16%) (Table 2 ). Post-operative patient pathway Significant variability in the post-operative management of drains and seromas was reported. Of the 79 respondents using drains, the majority ( n = 59, 62%) removed them based on the volume of seroma drained per day, most commonly < 50 mls/24 h (Table 3 ). Drains were most frequently removed by nursing staff ( n = 63/106, 60%) in breast clinic ( n = 44, 56%). Breast care nurses were most frequently responsible for assessing patients; deciding when seromas should be drained (77/199, 39%) and performing the procedure (72/250, 29%). Factors influencing decision-making regarding seroma drainage included patient symptoms (88/240, 37%) and assessment of actual (e.g. infection, 64/250, 26%) or impending (e.g. concerns re skin viability, 79/250, 33%) complications. Very few respondents reported draining all seromas (Table 3 ). Feasibility and design of a future RCT Of the 91 surgeons completing this section of the survey, just under half ( n = 37, 41%) expressed uncertainty regarding the use of drains after routine breast cancer surgery. Of those indicating some uncertainty, this was mostly regarding the use of a drain following mastectomy and sentinel lymph node biopsy (SLNB) ( n = 33, 70%) and isolated ANC ( n = 29, 62%). Two-thirds of surgeons felt a trial comparing the use of drains vs no drains after simple breast cancer surgery was needed (n = 59, 65%). Almost half of the surveyed surgeons ( n = 45, 49%) indicated they would be willing to randomise all patients undergoing mastectomy ± axillary surgery in an RCT comparing the use of drains versus no drains. The remaining half expressed reluctance randomising specific groups of patients. These groups primarily included patients in whom the surgical dead-space was anticipated to be large, for example following mastectomy and ANC (29/83, 35%); in women with Body Mass Index (BMI) > 30 (16/83, 19%); in those with large breasts (12/83, 15%) or; in those perceived to be at higher risk of post-operative complications (13/83, 16% e.g. those with high risk of bleeding; post neo-adjuvant chemotherapy etc.) (Table 4 ). There was a lack of consensus regarding the most suitable primary outcome for a future trial. Respondents were asked to select all outcomes of importance and/or interest in order to try and elucidate whether a single or co-primary endpoint might be necessary for a future trial. The most commonly selected outcome was the number of symptomatic seromas drained (67/302, 22%), followed by the number of hospital/healthcare provider visits (49/302, 16%) and patient reported outcomes such as patient satisfaction ( n = 41/302, 14%) (Table 4 ). Almost three-quarters ( n = 67, 73.5%) of respondents felt that a future trial was feasible and almost 80% would be definitely ( n = 54) or possibly ( n = 18) interested in recruiting patients to a future RCT. Of 91 respondents, 52 (57%) provided free text comments relating to the feasibility and design of a future trial. Three key themes emerged (Table 5 ) from the content analysis. These were: 1. The need for high-quality evidence to inform, change and address variation in practice (specifically no longer using drains); 2. Clinical (personal and community) equipoise and; 3. Trial design, outcome selection and feasibility. Overall, respondents felt that as there was significant variability in UK practice and no clear evidence to support the use of drains, a trial was necessary.
Discussion This survey has demonstrated considerable variation in the management of seromas following mastectomy and axillary surgery in the UK and the need for and potential feasibility of a large-scale pragmatic RCT to establish best practice. It is likely that a future trial would compare the use of drains vs no drains as this is currently the main strategy for reducing seroma development in the UK. Surgeons’ attitudes to a potential trial in this area reflects the lack of high-quality evidence to support the use of drains following breast cancer surgery. A systematic review [ 6 ] considered seroma formation in 1347 women following mastectomy (± axillary lymph node clearance) with and without suction drainage. The review included two RCTs [ 12 , 13 ] and six non-randomised studies [ 14 – 19 ]. The data were found to be at a high risk of bias, heterogeneous with variable use of flap fixation methods and with an inconsistently defined primary outcome of seroma formation. The authors concluded that there was some evidence that drainage following mastectomy and axillary surgery could be safely omitted without increasing seroma formation or complications but highlighted the need for further high-quality research to determine the role of surgical drains following breast cancer surgery including outcomes of importance to patients. These findings were consistent with previous reviews [ 9 , 20 , 21 ] suggesting that drainage does appear to reduce seroma rates but may be associated with longer hospital stays. However, it should be noted that drain use increasingly may not affect hospital stay as significantly as it has done in the past. The 2021 Getting It Right First Time (GIRFT) report (Using Hospital Episode Statistics, HES Data April 2015-March 2018) [ 22 ], demonstrated that only 20% of mastectomies without reconstruction, were conducted as a day case and that rates vary widely across trusts from 0% to 78.28%. The report recommended that day case mastectomy rates should be increased to 75%. Increasing day case mastectomy has perhaps recently been driven by the COVID-19 pandemic, and the consequent need to avoid hospital stay and risk of infection. In this survey, 77% (58/75) of respondents reported that patients went home on the day of surgery more than 75% of the time (Table 2 ). Several recent or ongoing European RCTs, as well as comprehensive literature reviews [ 23 ] have considered techniques to reduce seroma formation and the need for drains after mastectomy. The Dutch SAM trial [ 24 , 25 ] (NCT03305757), was a multicentre three arm RCT of flap fixation with sutures or tissue glue and conventional closure, with closed suction drains in all arms. This showed a significant reduction in clinically significant seroma in both flap fixation arms with the greatest reduction in the sutured flap fixation group [ 24 ]. Ongoing RCTs include the single-centre Dutch SARA [ 26 ] (NCT04035590) trial which will compare flap fixation with and without suction drainage; the multicentre Dutch QUILT (NCT05272904—not yet recruiting) trial comparing quilting without a drain and conventional closure; and the multicentre French QUISERMAS trial [ 27 ] (NCT02263651) comparing conventional closure with a drain and flap fixation without a drain. This study completed in 2018 but has yet to report. None of these trials, however, reflect current UK practice or include outcomes of importance of patients. Quilting is not standard practice in the UK, perhaps due to the increased costs associated with the time quilting takes and the use of more expensive self-locking sutures to perform the procedure. In addition, there are perceived concerns regarding compromising skin flap viability, particularly where the skin flaps are thin or in those already deemed to be high risk for complications such as smokers. Whilst the most common outcome for the trial suggested by surgeons in this survey was the number of symptomatic seromas drained (20% of respondents), this was closely followed by the number of hospital/healthcare provider visits (16% of respondents). Work with our patient and public involvement (PPI) group highlighted that hospital visits were perceived as a major burden to patients. This outcome would comprehensively evaluate drain-related issues, symptomatic seromas; wound complications and patient concerns which may require clinical evaluation while being objective and easy to measure. As such, hospital visits would pragmatically be the most appropriate primary outcome for a future trial. This is a national practice survey with limitations that require consideration. Firstly, it only includes the views of a relatively small group of UK breast surgeons. From the Getting It Right First Time (GIRFT) report in 2021 [ 22 ], there were 130 breast surgery units in England, but this number varies year to year depending on service mergers and closures. The surgeons who responded, may be more engaged in research and thus may not be representative of the breast surgical community more broadly. Whilst this is possible, the survey has included surgeons from across the UK, in all major geographical areas, with various degrees of experience. Furthermore, this engaged group of surgeons is likely to include those who will open and recruit to any future study. It could therefore be argued that their views are the most relevant as they will determine whether a future study would be successful. It is, however, possible that willing to participate in a future RCT in principle, does not always translate into actual participation in practice. Despite limitations, this work demonstrates there is a need for a high-quality RCT to determine if, when and in whom closed suction drains are necessary following mastectomy and axillary surgery in the UK. Perhaps more importantly, this is a question that is also meaningful to patients as in the recent James Lind Priority Setting Partnership (PSP) in breast cancer surgery [ 28 ], one in three patient respondents submitted questions related to seroma and the benefits of drains after breast cancer surgery. Overall, this question was ranked as the 11 th most important research priority to patients completing the survey and although it narrowly missed being considered for the top 10 research priorities [ 28 ], it is clearly an area where more research is needed. Work to design and gain funding for a future trial is now underway. Given the large volume of procedures performed, it is likely that that such a trial would recruit quickly and easily and utilisation of the breast trainee collaborative research network may represent a cost-effective means of delivering the study in a timely fashion [ 29 – 31 ]. If an RCT proves that drains are unnecessary in all or most patients undergoing mastectomy and/or axillary surgery, it will provide the necessary high-quality evidence to change practice. This will reduce NHS costs and the burden on scarce resources, but more importantly, improve patient experiences of breast cancer treatment.
Purpose Up to 40% of the 56,000 women diagnosed with breast cancer each year in the UK undergo mastectomy. Seroma formation following surgery is common, may delay wound healing, and be uncomfortable or delay the start of adjuvant treatment. Multiple strategies to reduce seroma formation include surgical drains, flap fixation and external compression exist but evidence to support best practice is lacking. We aimed to survey UK breast surgeons to determine current practice to inform the feasibility of undertaking a future trial. Methods An online survey was developed and circulated to UK breast surgeons via professional and trainee associations and social media to explore current attitudes to drain use and management of post-operative seroma. Simple descriptive statistics were used to summarise the results. Results The majority of surgeons (82/97, 85%) reported using drains either routinely (38, 39%) or in certain circumstances (44, 45%). Other methods for reducing seroma such as flap fixation were less commonly used. Wide variation was reported in the assessment and management of post-operative seromas. Over half (47/91, 52%) of respondents felt there was some uncertainty about drain use after mastectomy and axillary surgery and two-thirds (59/91, 65%) felt that a trial evaluating the use of drains vs no drains after simple breast cancer surgery was needed. Conclusions There is a need for a large-scale UK-based RCT to determine if, when and in whom drains are necessary following mastectomy and axillary surgery. This work will inform the design and conduct of a future trial. Supplementary Information The online version contains supplementary material available at 10.1007/s10549-023-07042-7. Keywords
Supplementary Information Below is the link to the electronic supplementary material.
Acknowledgements List of PUBMed citable collaborators (names as given in the online survey; some who completed did not leave a name). Nick Abbott, Raj Achuthan, Goran Ahmed, Rachel Ainsworth, Laura Arthur, Salena Bains, Zoe Barber, Jeremy Batt, Ashleigh Bell, Jane Carter, Alice Chambers, Anna Conway, Carol-Ann Courtney, Ian Daltrey, Raouf Daoud, Isabella Dash, Rajiv Dave, Julia Dicks, Urszula Donigiewicz, Hiba Fatayer, Daniel Glassman, Nikki Green, Eleanor Gutteridge, Ahmed Hamad, Anita Hargreaves, James Harvey, Shaziya Hassan Ali, Sophie Helme, Julia Henderson, Susan Hignett, Fiona Hoar, Jonathan Horsnell, Thomas Hubbard, Alex Humphreys, Javeria Iqbal, Omotayo Johnson, Meera Joshi, Charlotte Kallaway, Isabella Karat, Baek Kim, Eleftheria Kleidi, Manish Kothari, Chrissie Laban, Kelly Lambert, Siobhan Laws, Alexander Leeper, Serena Ledwidge, Valentina Lefemine, Jonathan Lund, E Jane Macaskill, Mariam Malik, James Mansell, Loaie Maraqa, Yazan Masannat, Julia Massey, Ross McLean, Jennifer McIlhenny, Colin McIlmunn, Louise Merker, Geraldine Mitchell, Jo Mondani, Elizabeth Morrow, Nabila Nasir, Olubunmi Odofin, Caroline Osborne, Polly Partlett, Anna Powell-Chandler, Sreekumar Sundara Rajan, Clare Rogers, Chandeena Roshanlall, Matthew Philip Rowland, Walid Abou Samra, Lucy Satherley, Brendan Skelly, Richard Sutton, Anne Tansley, Marios Konstantinos Tasoulis, Simon Timbrel, Nader Touqan, Alison Waterworth, Lisa Whisker, Kate Williams, Nihal Gonen Yildirim, Charles Zammit. Previous communication Abstract (Poster at Association of Breast Surgery Conference): Fairhurst, Katherine et al. Are drains necessary following mastectomy and axillary surgery? Feasibility work for the Diamond Study. European Journal of Surgical Oncology, Volume 48, Issue 5, e207. Author contributions All authors contributed to study conception and design. Material preparation and data collection were performed by Katherine Fairhurst. Data analysis was completed by Kirsty Roberts and Katherine Fairhurst. The first draft of the manuscript was written by Katherine Fairhurst and all authors read and approved the final manuscript. Funding This work was supported by an NIHR Academic Clinical Lectureship (CL-2020-25-002) for Katherine Fairhurst. Shelley Potter is an NIHR Clinician Scientist (CS-2016-16-019). The views expressed are those of the authors and not necessarily those of the UK National Health Service or National Institute for Health and Care Research. Data availability The datasets generated and analysed during this study are stored under the provisions of the National Data Protection Act and the University of Bristol requirements. Data may be made available to bona fida researchers only, on reasonable request to the corresponding author, after their host institution has signed a Data Access Agreement. Declarations Ethical approval Not required. Consent to participate/publish All participants voluntarily participated and were made aware of potential publication. All data is presented anonymously, and no patient participants were involved. Competing interests The authors have no relevant financial or non-financial interests to disclose.
CC BY
no
2024-01-15 23:43:45
Breast Cancer Res Treat. 2024 Oct 25; 203(2):187-196
oa_package/1d/dd/PMC10787912.tar.gz
PMC10787939
38222147
Introduction Stress urinary incontinence (SUI) is a prevalent (4-35% prevalence rate) and often distressing condition that affects individuals, primarily women, across various age groups [ 1 ]. It is characterized by the involuntary leakage of urine during activities that increase intra-abdominal pressure, such as sneezing, coughing, running, or lifting heavy objects [ 2 ]. The impact of SUI extends beyond the physical symptoms, as it can significantly disrupt an individual's social life and personal well-being, ultimately compromising their overall quality of life [ 3 ]. Intriguingly, the prevalence of SUI appears to transcend age boundaries, affecting women not only as they age but also in their younger years [ 4 ]. This widespread occurrence underscores the importance of addressing this condition comprehensively. However, in many societies, particularly in countries like India, women often endure urinary incontinence silently, with reluctance to discuss their health issues openly [ 4 ]. Consequently, the reported prevalence rates might not accurately reflect the true scope of this problem as the reported statistics revealed that 20% of women consulted healthcare professionals for urinary incontinence, while 72% of those affected had been silently battling it for over a year, citing reasons such as unawareness and societal taboos as barriers to seeking help [ 4 ]. Understanding the causes of SUI reveals a complex interplay of factors. Vaginal deliveries, for instance, are associated with a heightened risk due to potential alterations in pelvic floor innervation, injury to the levator ani muscle, and damage to the endopelvic fascia from stretching or tearing. Moreover, vaginal births may lead to reduced mobility of the bladder neck, further exacerbating the risk of SUI [ 5 ]. The concept of the integrated continence system, encompasses deficits in the intrinsic urethral closure system, the urethral support system, and the lumbopelvic stability system [ 5 ]. These systems are interconnected through neural and endopelvic fascia connections and thus play a critical role in maintaining urinary continence [ 5 ]. The lumbopelvic region's control, in turn, relies on the coordination of various muscle groups, including the diaphragm, transversus abdominis, pelvic floor musculature, and lumbar multifidus [ 5 ]. These muscles regulate intra-abdominal pressure and the tension in the thoracolumbar fascia, influencing postural control [ 6 ]. However, addressing SUI is not as straightforward as concentrating solely on pelvic floor muscle strengthening. Emerging evidence suggests that the abdominal muscles also play a crucial role in achieving optimal results [ 7 , 8 ]. It has been asserted that pelvic floor muscle rehabilitation reaches its full potential when addressing the abdominal muscles in tandem [ 7 ]. It has been reported that pelvic floor muscle contraction led to increased activity in the abdominal muscles and vice versa, highlighting the potential benefits of abdominal muscle (transverse abdominis) training in SUI rehabilitation [ 7 ]. To tackle the multifaceted nature of SUI, a promising approach known as dynamic neuromuscular stabilization (DNS) may be introduced, which works on synergistic action of the entire core musculature [ 8 - 11 ]. DNS relies on developmental kinesiology, comparing the stabilizing patterns of individuals to those of healthy infants [ 12 ]. It targets the integrated spinal stabilization system, involving deep cervical flexors, diaphragm, transversus abdominis, multifidus, and the pelvic floor [ 8 ]. This system provides a stable foundation for purposeful activities by maintaining axial spine extension and centered extremity joint positions, thereby increasing intra-abdominal pressure sensed by the central nervous system [ 8 ]. DNS exercises aim to activate the spinal stabilizing system effectively through repetition, helping individuals regain control during various tasks [ 8 ]. Furthermore, dynamic core stability is crucial for optimal athletic performance and hinges on the precise coordination of the integrated spinal stabilization system and the regulation of intra-abdominal pressure [ 8 - 13 ]. This coordination is far more vital than simply building strength in isolated muscle groups as the muscles do not work in isolation [ 8 - 11 , 14 ]. The DNS method acknowledges that the core musculature functions as a cohesive unit in which the various components work together synergistically [ 15 ]. When one part of this system is weak or dysfunctional, it can have repercussions on the functioning of other core muscles [ 15 , 16 ]. DNS exercises are designed to address this interconnection by fostering comprehensive activation and strengthening of the entire core musculature, rather than targeting individual muscles in isolation [ 15 , 16 ]. By engaging in DNS exercises, individuals can train their core muscles to operate in harmony, providing stability and support to the spine, pelvis, and surrounding structures. As a result, this approach improves posture, enhances control over movements, and bolsters overall body stability [ 15 ]. DNS exercises contribute to the enhancement of core muscle strength by promoting synchronized activation, improving stability, and optimizing the performance of the entire core unit. By taking a holistic approach to core training, DNS exercises offer a comprehensive strategy for boosting core strength and stability, ultimately leading to improved overall physical function and performance. While DNS has been effectively utilized in various conditions, such as sports injuries, cerebral palsy, hemiplegia, and other diverse ailments [ 16 - 22 ], there is currently no research investigating the impact of DNS exercises on SUI. Considering the intricate connections between the pelvic floor and core musculature, it becomes evident that a comprehensive approach to treating SUI is warranted. Thus, this study seeks to evaluate the efficacy of a DNS exercise program in comparison to a traditional pelvic floor exercise program for women dealing with SUI.
Materials and methods The study commenced after obtaining approval from the Institutional Review Board of Amity Institute of Health Allied Sciences (AUUP/IEC/2021-Jan/03) and was registered in the Clinical Trial Registry, India under the reference CTRI/2021/09/036247. It was designed as a single-blinded (participants were blinded), single-center randomized controlled trial. We recruited a convenient sample of 24 female participants (12 in each group) [ 23 ]. Participants were eligible if they were females aged 18-40 years, married, diagnosed with mild to moderate SUI, at least one-year post delivery, and medically and physically fit for assessment and physiotherapy. Exclusion criteria included continuous urinary leakage, current urinary incontinence drug therapy, pelvic prolapse > stage I, pregnancy, vaginal/urinary tract infections, menstruation during the examination, presence of tumors/fractures/acute inflammatory diseases, current estrogen treatment, and use of anticholinergics, antidepressants, or serotonin-affecting substances [ 24 ]. All the participants were recruited by convenience sampling after being referred and diagnosed by a gynecologist. Randomization and allocation of participants Participants who met the inclusion criteria and were screened for exclusion were assigned to one of the treatment groups in a 1:1 ratio. A random number table, generated by a statistician, was used to allocate participants to their respective groups. Subsequently, an allocation plan was meticulously documented in sequential order and securely sealed within opaque envelopes. During the allocation process, an independent individual, unrelated to the study, was responsible for opening the envelope to disclose the assigned group. This unbiased procedure determined whether participants were allotted to either the DNS or Kegel exercise group. Procedure Baseline measurements for outcome measures included pelvic floor muscle (PFM) strength (perineometer), electromyography (EMG) of PFM, and quality of life (Urogenital Distress Inventory-6 (UDI-6)). These measurements were repeated after a 12-week intervention period. The Consolidated Standards of Reporting Trials (CONSORT) flowchart, presented in Figure 1 , provides a visual representation of the participant enrollment and progression throughout the study, illustrating their inclusion, allocation, follow-up, and analysis stages. The experimental group followed a four-phase exercise protocol (detailed in Table 1 ), while the control group performed PFM contractions, holding each contraction for 10 seconds, with 10 repetitions per set and three sets in total, with one minute of rest between sets [ 24 ]. According to the DNS principles, for correct activation of the integrated spinal stabilization system, the abdominal muscles should not only expand in the caudal direction but should expand in all directions, i.e., the posterior and lateral as well. Thus, the investigator should palpate the medial to anterior superior iliac spine anteriorly and the posterolateral aspect of the abdominal wall below the lower ribs from behind to assess the correct expansion of the entire abdominal wall [ 16 ]. The exercises for both groups were carried out under the supervision of the investigator in the clinical setting. They were instructed not to perform the exercises at home. Outcome measures In the completed study, the outcome measures encompassed changes in perineometer values, average, peak, and maximum voluntary contraction (MVC) of PFM as measured by EMG, and the impact of urinary incontinence on quality of life by using UDI-6 after the 12-week treatment period. Perineometer The device (Bionics Perineometer Analogue, Mumbai, India) consisted of a conical vaginal insert linked to a handheld microprocessor for measuring pressure in mmHg upon compression of the insert. Participants assumed a crook-lying position, and the perineometer was inserted into the vaginal canal until the compressible portion extended beyond the hymeneal ring. Baseline pressure readings were recorded, and participants were instructed to exert maximum effort to contract their PFM for two to three seconds, completing three consecutive squeezes with a one-minute rest interval. The peak of these three contractions represented their maximum perceived strength. Careful observations and a pressure biofeedback device were employed to ensure a neutral spine position and prevent elevated perineometer readings due to excessive intra-abdominal pressure. The vaginal insert was covered with a condom to ensure safe and hygienic use by multiple participants. The perineometer demonstrated high reliability, indicated by an intraclass correlation coefficient (ICC) of 0.95 [ 25 ]. EMG EMG data were gathered using the NeuroTrac MyoPlus2A version 11.1 device (Hampshire, UK). A pear-shaped intravaginal sensor with stainless steel electrodes was gently inserted into the vagina while patients were in a supine lithotomy position. A reference electrode was placed on the right anterior superior iliac spine. Prior to examination, participants received training to ensure accurate contraction of their PFM. PFM strength was evaluated by recording the average scores from three maximum contractions. The EMG values were documented in terms of average (EMG average), peak (EMG peak), and maximum voluntary contractions (EMG MVC), with measurements in microvolts for EMG average and EMG peak (μV) and in percentage (%) for EMG MVC. An automated protocol software offered on-screen instructions and voice guidance, indicating when to contract and relax the PFM, following a pattern of five-second work and rest intervals [ 26 ]. The data were displayed after getting filtered by the inbuilt software. Urogenital Distress Inventory-6 (UDI-6) This inventory featured six questions, with responses categorized as "not at all" (0), "a little bit" (1), "moderately" (2), and "greatly" (3). Scores were obtained by summing the responses, dividing by six, and multiplying by 25 to yield the Urogenital Distress Inventory (UDI) score. Scores exceeding 33.33 indicated higher distress [ 27 ]. The kappa statistics vary from 0.699 to 0.350 for each question [ 27 ]. The sensitivity for UDI-6 is 97% [ 28 ]. Statistical analysis The analysis was carried out after the completion of the 12-week treatment period for all participants. We adhered to the Enhancing the Quality and Transparency of Health Research (EQUATOR) network guidelines for reporting descriptive statistics and tests. This approach accounted for the possibility that dropouts might have returned to their initial readings, and any missing data were replaced with the readings and responses from the first day of assessment. To enhance sensitivity, data available after the 12-week period were also analyzed separately. However, there were no dropouts in the current study. To explore the differences between the two groups, the Mann-Whitney U-test was utilized. Mean differences at a 95% confidence interval were reported to convey the data. However, for within-group comparison, the Wilcoxon sign rank test was used. All data analyses were conducted by a biostatistician who was blinded to the treatment group assignments. The analysis was performed using IBM SPSS Statistics for Windows, version 23.0 (IBM Corp., Armonk, NY). Additionally, we calculated the rank-biserial correlation coefficient (r) as a measure of effect size in our study. “r" quantifies the strength and direction of the relationship between groups in a non-parametric context. The interpretation of r is as follows: 0 indicates no relationship, 1 indicates a perfect positive relationship, and -1 indicates a perfect negative relationship. Traditionally, the magnitude of r is considered small if around 0.1, medium if around 0.3, and large if around 0.5 or above. This provides insight into the practical significance of the observed differences between groups. A higher absolute value of r suggests a more substantial effect, and it complements the p-value by quantifying the magnitude of the observed differences [ 29 ]. The formula used for the calculation of “r” was r = Z / √N, where, r represents the rank biserial correlation coefficient, and Z is the Z-score obtained from the Mann-Whitney U test. The Z-score is a measure of how many standard deviations an observation or data point is from the mean of a distribution. √N is the square root of the total sample size [ 29 ].
Results Demographic data analysis conducted with Mann-Whitney U-tests revealed no statistically significant differences between the DNS and Kegel exercise groups in terms of age, height, weight, BMI, number of children, and duration of symptoms (as presented in Table 2 ). Strength of the pelvic floor by using a perineometer Initially, there was no significant difference between the two groups at baseline (p = 0.075). Median ± SD values for DNS and Kegel groups at baseline were 10 ± 2.71 and 10 ± 4.58, respectively. However, at 12 weeks, a statistically significant difference was observed (p = 0.005). Notably, the DNS group exhibited a significant increase in pelvic floor muscle (PFM) strength compared to the Kegel exercise group, i.e., median ± SD values were found to be 24 ± 4.30 and 14 ± 4.60, respectively (Table 3 ). When comparing each group individually from baseline to 12 weeks, significant improvements were observed in both groups (DNS group: p = 0.005; Kegel exercise group: p = 0.020), as presented in Table 4 . Electromyography of the pelvic floor muscles All EMG data were automatically filtered by the inbuilt software of the NeuroTrac MyoPlus2A version 11.1. At baseline, no statistically significant difference was found between the groups for EMG average, EMG peak, and EMG MVC with p-values of 0.545, 0.696, and 0.807, respectively. Median ± SD values for EMG average, EMG peak, and EMG MVC for DNS and Kegel exercise groups at baseline were 41.1 ± 19.08, 78.1 ± 43.11, 100.30 ± 90.7 and 37.4 ± 16.06, 60.2 ± 17.2, 45.2 ± 21.72, respectively. However, a statistically significant difference was observed at 12 weeks for EMG average, EMG peak, and EMG MVC (average p = 0.005; peak p = 0.001; MVC p = 0.009). Significant improvements were noted in both groups for all three components (EMG average, EMG peak, and MVC) when compared from baseline to 12 weeks (DNS group: average p = 0.001; peak p = 0.001; MVC p = 0.030; Kegel exercise group: average p = 0.005; peak p = 0.001; MVC p = 0.005), as presented in Tables 3 , 4 for between and within-group analyses, respectively. UDI-6 No statistically significant difference was found between the two groups at baseline (p = 0.481). Median ± SD values for DNS and Kegel exercise group at baseline were 42 ± 4.95 and 41 ± 4.48, respectively. However, a statistically significant difference was observed between the two groups at 12 weeks (p = 0.001) with median ± SD values for DNS and Kegel exercise groups at 12 weeks as 14 ± 5.45 and 21 ± 7.21, respectively. When evaluating improvements in both groups individually from baseline to 12 weeks, statistically significant improvements were seen in both groups (DNS group: p = 0.032; Kegel exercise group: p = 0.046), as detailed in Tables 3 , 4 for between and within-group analyses. Calculation of “r” effect size In our study, the "r" values were as follows: 0.72 for perineometer, 0.70 for EMG average, 0.74 for EMG peak, 0.8 for EMG MVC, and 0.89 for UDI-6. These values indicate that all variables had "r" values greater than 0.7. An effect size of "r" value above 0.7 signifies a medium to large effect size [ 28 ]. This suggests that women in the DNS group experienced a more effective treatment compared to the Kegel exercise group for all the variables (Table 5 ).
Discussion In this study, we conducted a thorough evaluation to assess the relative effectiveness of DNS exercises in comparison to conventional Kegel exercises for the management of SUI in women. Our study is a significant addition to the existing body of knowledge regarding the array of treatment approaches available for addressing SUI, shedding light on potentially more efficient interventions for this widespread condition. Our investigation revealed substantial improvements across various key parameters following a 12-week intervention with DNS exercises. These improvements included enhanced PFM strength, increased electrical activity in the PFM, and lower scores on the UDI-6. Conversely, the group that underwent traditional Kegel exercises did not exhibit similar significant improvements when compared to the DNS group, particularly in terms of PFM strength, pelvic floor EMG activity, and UDI-6 scores. These findings suggest that DNS exercises may hold promise as a more advantageous intervention, especially for women dealing with SUI. One of the intriguing aspects of our study is the direct comparison of DNS exercises and Kegel exercises. This comparison fills a gap in the literature, as there has been limited research directly comparing these two approaches in the context of SUI. Our study provides valuable insights into whether DNS exercises, designed to target the integrated spinal stabilization system, including the pelvic floor, present a more effective alternative to the commonly recommended Kegel exercises. The rationale behind the superior outcomes observed in the DNS exercise group likely stems from the biomechanical link between the diaphragm, abdominals, multifidus, and the pelvic floor [ 5 - 8 ]. DNS exercises are structured to activate and coordinate these muscle groups, which constitute an integrated spinal stabilization system [ 9 - 12 ]. This holistic approach emphasizes the importance of these muscles working together efficiently. Coordination of the diaphragm, abdominal muscles (including the transverse abdominis), multifidus, and pelvic floor is central to the success of DNS exercises [ 9 - 12 ]. These exercises emphasize the coordinated activation of this integrated system to maintain the continence mechanism under stress. Additionally, it has been reported that if one part of the stabilizing system is affected, it impacts the other parts as well [ 8 ]. In contrast to previous approaches that predominantly focused on strengthening individual muscles, DNS recognizes the importance of these muscles acting in unison. The literature also supports the idea that treating the lumbopelvic system as a whole is crucial for effective SUI rehabilitation [ 4 , 6 , 7 ]. Another study illustrated that the training of the diaphragm and abdominal muscles along with the pelvic floor was superior to the isolated contraction of PFM for urinary incontinence [ 24 ]. The DNS approach, which targets this entire system, has been shown to lead to more significant improvements in our study. It is worth noting that while some studies have focused on different conditions, such as enhancing core stability in stroke patients, they have highlighted the potential of DNS exercises to enhance core stability [ 9 - 11 ]. Core stability, in turn, plays a pivotal role in providing support to the pelvic floor [ 5 - 7 ]. The timeline for observing improvements in our study was 12 weeks, which aligns with the existing consensus in muscle physiology [ 24 ]. Previous research has demonstrated that improvements in pelvic floor strength can be observed within a similar timeframe, validating our findings and underscoring the potential for relatively short-term interventions to yield meaningful results [ 24 ]. Limitations It is important to acknowledge the limitations of our study. We did not include a follow-up to assess the long-term effectiveness of the treatment. Our study was single-blinded, and it is essential to consider that the age group of participants was limited to females between 18 and 40 years old, which may affect the generalizability of our results to other age groups. Additionally, participants with severe SUI were not included in the study, restricting the use of DNS exercises to mild to moderate SUI cases. Beyond these considerations, our study did not incorporate an examination of various pertinent factors, including socioeconomic status, the presence of endometriosis, and dietary habits. These omissions represent a potential source of confounding that could influence the observed outcomes. Consequently, the broader applicability of our findings to diverse demographic and health contexts is constrained. Addressing these limitations necessitates future research endeavors incorporating a follow-up mechanism, diverse age groups, and a more inclusive participant selection process. Additionally, a comprehensive exploration of confounding factors and their potential impact on treatment outcomes will contribute to a more nuanced understanding of the intervention's efficacy.
Conclusions In conclusion, our study's results suggest that DNS exercises, which emphasize the coordinated activation of the diaphragm, abdominals, multifidus, and pelvic floor, may offer a more effective approach for managing stress urinary incontinence in women compared to traditional Kegel exercises. The biomechanical synergy of these muscle groups appears to be a critical factor in the observed improvements. However, further research is warranted to explore the long-term effectiveness and specificity of this approach for different populations by incontinence type, severity, age, and other important factors. Nonetheless, our study opens new avenues for the management of SUI, offering individuals and healthcare professionals a potentially more effective intervention to consider.
Background and objective Stress urinary incontinence (SUI) is a prevalent condition affecting women of various age groups, significantly impacting their quality of life. To address this multifaceted issue, a comprehensive approach that goes beyond traditional pelvic floor exercises is needed. Dynamic neuromuscular stabilization (DNS) exercises, targeting the integrated spinal stabilization system, offer a promising alternative. Thus, this study aimed to compare the effectiveness of DNS exercises and Kegel exercises in managing SUI among women. Methods This single-blinded, pilot study involved 24 women aged 18-40 years with mild to moderate SUI. Participants were divided into DNS and Kegel exercise groups. Outcome measures included perineometer readings, electromyography (EMG) data, and the Urogenital Distress Inventory-6 (UDI-6). Statistical analysis compared baseline and 12-week data within and between groups, and rank-biserial correlation coefficient (r) as a measure of effect size in our study was calculated. Results At 12 weeks, the DNS group showed significant improvement in pelvic floor muscle strength compared to Kegel exercises (p = 0.005). Both groups had significantly enhanced pelvic floor muscle strength (p < 0.05). A significant change occurred for EMG average, EMG peak, and EMG maximum voluntary contraction (MVC) at 12 weeks (average p = 0.005; peak p = 0.001; MVC p = 0.009), with significant improvements in both groups (p < 0.05). For UDI-6, a significant difference emerged between the two groups at 12 weeks (p < 0.05), with significant improvements in both groups individually from baseline to 12 weeks (p < 0.05). The effect size "r" for all variables indicated a medium to large effect size, underscoring the substantial and significant impact of DNS exercises in managing SUI among women compared to Kegel exercises. Conclusion This study suggests that DNS exercises, emphasizing the coordinated activation of the diaphragm, abdominals, multifidus, and pelvic floor, may provide a more effective approach for managing SUI in women compared to traditional Kegel exercises.
CC BY
no
2024-01-15 23:43:45
Cureus.; 15(12):e50551
oa_package/aa/f9/PMC10787939.tar.gz
PMC10787940
38222225
Introduction Hemobilia is an infrequent cause of upper gastrointestinal bleeding, which is most often secondary to intrabiliary procedures, hepatic trauma, inflammatory diseases, biliary tumors, and vascular malformations [ 1 , 2 ]. Although common in the gastrointestinal tract, angiodysplasia is extremely rare in the biliary ducts, with few cases described thus far. The resolution of these pathologies has become quicker and more precise with the advancement of cholangioscopy as a diagnostic and therapeutic tool [ 3 ]. Until now, the diagnosis of these conditions depended on radiological assistance and empirical treatments. We report a case of a patient with hemobilia due to angiodysplasia of the major biliary duct, diagnosed by cholangioscopy, and the condition was resolved after the placement of a biliary stent.
Discussion Hemobilia remains a challenging entity to diagnose and treat, requiring the use of various technologies and a multidisciplinary approach. Common causes include intrabiliary procedures (endoscopic or percutaneous), hepatic trauma, cholangiopathies secondary to cirrhosis, tumors, and vascular malformations. These, however, account for a minority of cases, usually occurring secondary to portobiliary fistulas, varices, hereditary telangiectasias, and rarely angiodysplasias [ 4 , 5 ]. Angiodysplasia is a common finding in upper and lower gastrointestinal endoscopies, primarily located in the stomach, duodenum, and colon. Its occurrence in the biliary tract is extremely rare, making diagnosis challenging and requiring specific resources and trained professionals, with few cases reported to date. It is characterized by the malformation of vessels in the mucosa and submucosa, with no hereditary or racial connection, but with a higher incidence in elderly patients with aortic stenosis, chronic kidney disease, lung diseases, and von Willebrand disease. Its pathophysiology remains uncertain but may be related to common bile duct contractions causing intermittent vessel obstruction, leading to focal dilations, tortuosities, and collateral vessel formation [ 6 , 7 ]. The advancement of endoscopic devices and technologies now allows for the determination of the causes of hemobilia. Once a diagnostic tool, ERCP has become an important therapeutic tool, evolving toward digital cholangioscopy. Adverse events, such as biliary tract perforation, air embolism, and bacteremia, are rare, though slightly higher than ERCP. Risk factors include patient age, stent placement, and, primarily, lithotripsy for biliary stones [ 8 , 9 ]. In our case, cholangioscopy played a crucial role in diagnosis and treatment. After an unsuccessful attempt at endoscopic treatment and cholangiography showing no major abnormalities, cholangioscopy revealed the source of bleeding as an angiodysplasia in the distal major biliary duct. The few cases reported in the literature have had varying approaches to this finding. One patient with a history of hemobilia, but no active bleeding at the time of diagnosis, maintained stable hemoglobin levels without intervention. Another used laser therapy for both angiodysplasia and actively bleeding neoplasms, and a third underwent radiointerventional embolization [ 3 , 10 , 11 ]. Factors influencing our treatment decision were active bleeding, a distal location near the papilla, and the unavailability of hemostasis accessories with SpyScope DS® at the time of the examination. Possible complications related to biliary stent placement include acute pancreatitis, cholangitis, migration, and acute cholecystitis. In patients eligible for laparoscopic cholecystectomy, the procedure can be performed as described in our case. Percutaneous or endoscopic ultrasound-guided gallbladder drainage is an alternative for gallbladder decompression in patients who are not surgical candidates [ 8 , 12 ].
Conclusions In conclusion, digital cholangioscopy played a crucial role in the diagnosis and treatment decision for hemobilia due to angiodysplasia of the major biliary duct. There is still a lack of data in the literature to define the gold-standard therapy, but the use of a fully covered metallic stent proved effective in achieving hemostasis.
Hemobilia is described as bleeding from the intra- or extrahepatic biliary tree expressed through the major duodenal papilla into the duodenum, with angiodysplasia of the major biliary duct as a rare etiological factor with few cases reported in the literature. Cholangioscopy plays a pivotal role in diagnosing and making therapeutic decisions regarding biliary tract lesions. We report a case of the diagnosis and treatment of hemobilia secondary to bleeding from angiodysplasia of the major biliary duct, which was resolved after the placement of a fully covered metallic stent, with a review of the literature.
Case presentation An 87-year-old male patient with atrial fibrillation who was on apixaban 5 mg/day presented to the emergency department with exertional dyspnea and lower limb edema. He had no history of external bleeding but complained of mild abdominal pain upon examination. Laboratory tests revealed acute anemia with a hemoglobin level of 5.7 g/dL. Due to the patient's age and comorbidities, the attending team opted for a capsule endoscopy to avoid a more invasive procedure like a traditional endoscopy after an inconclusive computed tomography. Capsule endoscopy revealed blood residues in the stomach and small intestine, with recent clots predominantly in the duodenal region. Upper gastrointestinal endoscopy (EGD) identified active bleeding at the major duodenal papilla with an unsuccessful attempt at hemostasis through sclerosis (Figure 1 ). Following an investigation with magnetic resonance imaging of the biliary tract, which identified a punctate enhancement focus in the distal common bile duct (Figure 2 ), the patient underwent endoscopic retrograde cholangiopancreatography (ERCP). During the procedure, bleeding at the papillary orifice persisted, and cholangiography showed no abnormalities. Cholangioscopy with SpyScope DS® (Boston Scientific, Massachusetts) immediately above the papilla revealed anomalous vascular formations, suggestive of angiectasias with active bleeding (Figures 3 , 4 ). A 10 mm x 60 mm fully covered self-expanding metallic stent (WallFlex® Biliary RX Stent, Boston Scientific, Massachusetts) was placed for mechanical hemostasis, resulting in a favorable outcome. The patient remained hospitalized, with stabilization of hematimetric values but developed acute cholecystitis due to obstruction of the cystic duct by the stent. He underwent laparoscopic cholecystectomy and was discharged on the fourth postoperative day. The stent was removed by EGD after 90 days, with resolution of the bleeding, and anticoagulation was resumed.
CC BY
no
2024-01-15 23:43:45
Cureus.; 15(12):e50552
oa_package/14/dd/PMC10787940.tar.gz
PMC10787941
38222993
Introduction Primary ciliary dyskinesia (PCD), an autosomal recessive condition characterized by motile cilia malfunction, displays clinical and genetic variation. Some of the clinical symptoms of PCD include left-right lateralization, infertility, and chronic upper and lower respiratory illness [ 1 - 5 ]. There is no one best way to diagnose PCD; instead, a variety of methods can be used, such as a combination of nasal nitric oxide (nNO), high-speed genetic analysis, immunological fluorescence of ciliated cells, transmission electron microscopy (TEM), high-speed video microscopy analysis, immune fluorescence of ciliated cells, and genetic analysis (gene panel analysis or extensive genetic analysis) [ 6 , 7 ]. There is a significant correlation between phenotype and specific genetic changes. Reduced generation of multiple motile cilia (RGMC) has been linked to CCNO gene mutations, which are also more likely to have a more severe respiratory disease phenotype with pulmonary failure at a younger age [ 8 ]. For the first time, we describe the CCNO NM 021147.4 (c.258 262dup.p, Gln88argfs*8 homozygous) gene mutation that caused severe PCD in a consanguineous Indian family. This article was previously presented as a meeting abstract at the 2022 American Thoracic Society International Conference Meeting on May 13-18, 2022.
Discussion We described two Indian sisters who experienced early-onset respiratory symptoms, obstructive ventilatory dysfunction, and situs solitus. They were identified as having PCD based on the full exon of the CCNO gene and low nNO levels. Patients with ciliogenesis-related mutations in the genes CCNO and MCIDAS have been reported to have a worse phenotype than those with other kinds of PCD, which is consistent with the current cases [ 8 ]. There are currently 50 known PCD genes that cause a variety of functional defects ranging from abnormal beat patterns through impairment of dynein arms to a complete absence of cilia [ 9 , 10 ]. Although nNO showed good diagnostic accuracy as a PCD diagnostic test when compared to the extended reference standard of TEM and/or genetic testing [ 11 ], it is crucial to confirm the genetic diagnosis, given the present evidence of links between genotype and phenotype [ 12 ]. Patients with PCD, which is secondary to CCNO mutation, have early-onset respiratory symptoms, recurrent lower respiratory tract infections, and progressive loss of lung function, which is consistent with our cases. Situs solitus, like in the cases presented, has only been documented in people with CCNO mutations [ 8 ]. The other documented symptoms are recurrent pneumonia, sinusitis, and otitis media (Table 1 ). There are no reports of genetic PCD in India, and only one publication stated that patients with recurrent sinusitis, otitis, and pneumonia in India who also had a fractional exhaled NO level under 10 ppb and who had not undergone genetic testing were likely to have PCD [ 13 ]. To our knowledge, this is India’s first report on PCD with CCNO which accounts for two out of 318 cases that have been published in the literature (Table 1 ). In 16 people who had the first CCNO mutation in 2014, a malfunction in the mother centriole formation and migration at a late stage of multiple motile cilia (MMC) differentiation led to a significantly decreased number of MMCs. Congenital mucociliary clearance disorder with RGMC is used to denote this hereditary condition [ 8 ]. Following the study, there have since been several CCNO reports (Table 1 ). In a study of five PCD patients from three different Irish traveler families, it was discovered that a sibling pair in Irish family B had the CCNO gene [ 14 ]. In another investigation, researchers looked for CCNO mutations in 170 Israeli families with mucociliary clearance disorders and identified two novel variations (p.Gly56Alafs38; c.165delC, c.638T>C, p.Leu213Pro), and two known mutations were found in 15 individuals from 10 families (6% prevalence) [ 15 ]. In Lisbon, Portugal, 12 patients underwent PCD genetic testing confirming the diagnosis, with three presenting CCNO mutations [ 16 ]. In Turkey, out of a total of 265 patients with PCD during a five-year period, 46 had genetically determined PCD using whole-exome sequencing at a single facility, and four had CCNO [ 17 ]. There have been multiple publications from China. One of these described 58 individuals with PCD, 51 with hereditary PCD, and three with CCNO [ 18 ]. This presenting case study is not a meta-analysis, and its limitation arises from the small number of patients from India, which has not been previously explored in the current literature.
Conclusions PCD with CCNO mutations is a relatively rare disease. Our findings highlight the significance of considering PCD based on CCNO mutations in people with situs solitus to have a more severe respiratory disease phenotype with lung failure at earlier ages. It is also critical to include individuals from various racial and ethnic origins in PCD-associated genetic CCNO mutation.
Primary ciliary dyskinesia (PCD) is a heterogeneous autosomal recessive disease marked by organ lateralization in 50% of patients, chronic sinopulmonary disease, infertility in men, and neonatal respiratory distress. Respiratory control cells contain CCNO in their apical cytoplasm, which is necessary for the development of multiciliate cells, basal body amplification, and migration. Reduced generation of multiple motile cilia, a rare form of PCD, has been linked to CCNO gene abnormalities . Individuals with CCNO mutations have been reported to suffer from severe lower respiratory infections that cause progressive impairment of lung function. For the first time, we describe the CCNO NM 021147.4 (c.258 262dup.p, Gln88argfs*8 Homozygous) gene mutation in an Indian consanguineous family that resulted in severe PCD.
Case presentation Our case study focuses on two consanguineous sisters from India who visited the Pediatric Pulmonary Department at Sidra Hospital in Qatar when they were 17 and 15 years old. They presented with chronic wet cough and were found to have progressive loss of lung function that eventually led to end-stage lung disease and the requirement for lung transplantation in the case of the elder sister. Since infancy, both siblings had suffered significant lower respiratory infections, chronic rhinorrhea, and recurrent ear infections. Bronchiectasis was detected by computed tomography (CT) of the chest when the elder sister was 14 years old (Figure 1 ) and the younger sister was seven years old (Figures 2 , 3 ). Physical examination was pertinent for bilateral crackles and clubbing in both sisters. The elder sister was hypoxic requiring 1‐2 L of oxygen per minute via a nasal cannula. Expiratory flow volume (spirometry) revealed significant mixed restrictive and obstructive airway disease, which was worse in the elder sibling. The forced expiratory volume in one second (FEV1) and forced vital capacity (FVC) in the elder sister were 25% and 40% predicted, respectively (Figure 4 ). Whereas the FEV1 and FVC in the younger sister were 39% and 57% predicted, respectively (Figure 5 ). Immune deficiency and cystic fibrosis were ruled out by blood tests and sweat chloride measurements. Both sisters had low nNO levels. The elder sister’s nNO was 63.3 ppb (predicted 200‐1,000) and the younger sister’s nNO was 88.3 ppb. PCD was diagnosed based on clinical phenotype and low nNO levels, which were further confirmed through genetic sequence analysis and deletion/duplication for PCD, which revealed the pathogenic variant CCNO NM 021147.4 (c.258 262dup.p, Gln88argfs*8 homozygous) gene mutation.
CC BY
no
2024-01-15 23:43:45
Cureus.; 16(1):e52237
oa_package/e0/e3/PMC10787941.tar.gz
PMC10787942
38222226
Introduction Acute appendicitis is one of the most prevalent causes of presentation to the emergency department (ED) [ 1 ]. With the treatment being resection, whether open or laparoscopic, it is not uncommon for patients to present with complications of appendectomy. One of the delayed complications of this procedure is stump appendicitis, a once rare entity with a now increasing incidence [ 2 ]. The estimated figure in literature is one in 50,000, although this is likely underestimated and the condition underreported [ 3 ]. Stump appendicitis can occur as soon as two months and as late as 50 years post-appendectomy [ 4 ]. It is often diagnosed radiologically on an abdominal CT scan performed to rule out other diagnoses [ 5 ]. A lack of awareness by the ED physician can lead to delayed diagnosis, which inadvertently leads to delayed treatment and an increased complication rate [ 6 ]. We report a case of a young male presenting with acute abdominal pain and a history of appendectomy. Stump appendicitis was diagnosed incidentally on CT and the patient was managed conservatively.
Discussion Stump appendicitis (SA) is defined as interval inflammation of residual appendicular tissue after appendectomy [ 5 ]. First described in 1945 by Rose, its incidence has since been slowly rising [ 2 ]. This has been attributed to laparoscopic surgery by some authors but refuted by others; an alternative explanation is the increasing awareness to this clinical entity in recent years [ 8 ]. However, a low index of suspicion remains one of the principal reasons behind delayed diagnosis and treatment [ 9 ]. A 2020 literature review by Enzerra et al. counted 160 cases reported in published literature; another review in the same year by Burbano et al. concluded that “the incidence of stump appendicitis seems to be higher than the one reported of 1 in 50,000” [ 10 , 11 ]. Several anatomic and surgical risk factors have been suggested to increase the risk of SA; the most consistent seem to be inappropriate identification of the base of the appendix at the time of surgery and length of the residual stump [ 4 ]. Although there is no clinical consensus on the exact measurement, the general recommendation is that the stump be less than 5 mm long [ 12 ]. A disputed risk factor is undergoing laparoscopic as opposed to open appendectomy. Theoretically, “the lack of a three-dimensional approach and the absence of a tactile return” increase the risk of SA post-laparoscopic appendectomy by increasing the length of tissue left behind [ 2 ]. Interestingly, one review found the incidence post-laparoscopy to be less than half of that post open appendectomy [ 13 ]. Ultimately, surgical technique and operator experience play a role in the complication rate of any procedure. Other risk factors include a retro-colic position of the appendix, poor blood supply, and appendicolith formation [ 5 ]. Clinical presentation of SA is similar to that of acute appendicitis, further confounding the diagnostic process. Although no one symptom or sign is specific, the most common findings are RLQ pain, leukocytosis, peritonism, fever, nausea and vomiting [ 11 ]. Epidemiologically, the diagnosis is made in patients ranging from eight to 80 years, with a median age of 33 years, and a male-to-female ratio of 1.1:1 [ 14 ]. Abdominal imaging is often needed to confirm the diagnosis, with computerized tomography (CT) being the gold standard [ 15 ]. Compared to ultrasound scans, CT is more sensitive and specific and can simultaneously exclude other abdominopelvic disease entities [ 8 ]. Findings are similar to those of acute appendicitis and can include cecal wall thickening, free fluid, abdominal collections or abscesses- as well as identification of the appendicular stump [ 8 ]. Less frequently, barium enema and colonoscopy can aid in the diagnosis [ 16 ]. In cases of ambiguity or diagnostic uncertainty, a decision to proceed to diagnostic laparoscopy can be made [ 17 ]. The vast majority of reported cases have cited a completion appendectomy as the treatment of choice for SA, with the open approach being preferred to laparoscopic surgery [ 6 ]. This may be due to the fact that a delayed diagnosis leads to more complicated presentations, such as stump gangrene, perforation, and peritonitis [ 5 ]. One review cited a perforation rate as high as 70% [ 14 ]. Conservative management with parenteral antibiotics and analgesia with or without colonoscopy may be appropriate for some patients, such as the case reported above; although it is less commonly described in the literature [ 4 ]. Potential uses of colonoscopy are washout, pus drainage and clearance, and appendicolith removal with a snare [ 8 ]. An interval appendectomy has also been suggested by some authors after the acute inflammatory phase subsides with medical management to improve intra-operative visualization of the stump and prevent recurrence [ 4 ].
Conclusions Right lower quadrant abdominal pain comes with a wide range of differential diagnoses. In patients with a clinical picture of acute appendicitis and history of appendectomy, it is the emergency physician’s responsibility to maintain a high index of suspicion for stump appendicitis. The gold standard for diagnosis is an abdominal CT. Prompt identification and treatment- primarily with a completion appendectomy- is key to avoid complications such as perforation.
Acute appendicitis is one of the most common diagnoses in the emergency department. As with other surgical procedures, post-appendectomy complications are numerous and can be either immediate or delayed. Stump appendicitis is an underreported and underrecognized complication that is often diagnosed radiologically while ruling out other diagnoses. We report a case of a 26-year-old male presenting with acute right lower quadrant abdominal pain. Although he initially denied any surgical history, a focused abdominal exam revealed an incisional scar which turned out to be the result of an appendectomy nine years ago. The patient was worked up for alternate causes of right lower quadrant pain. Investigations revealed high inflammatory markers and hematuria. We proceeded with a non-contrast CT scan to rule out vesicoureteric junction stone. Instead, the scan was suggestive of stump appendicitis. The patient was admitted and treated conservatively. Maintaining a high index of suspicion for stump appendicitis, especially in patients with a clinical picture typical of appendicitis but a history of appendectomy, is key to making an early diagnosis and avoiding further complications.
Case presentation A 26-year-old male presented to ED with a one-day history of right lower abdominal pain associated with nausea and one episode of vomiting. The pain was of gradual onset, dull achy character, mild to moderate intensity, and intermittent course. He denied fever, altered bowel habit, or dysuria; review of systems was otherwise unremarkable. During the initial interview, he denied any past medical or surgical history. His only medications were simple analgesics, he was a non-smoker and did not consume alcohol, and family history was non-contributory. Vitally, the patient was stable and afebrile. Focused abdominal examination revealed an incisional appendectomy scar; when asked about this finding, the patient then added that he underwent an open appendectomy nine years ago. He also had moderate right lower quadrant (RLQ) tenderness with rebound and mild guarding. No other peritoneal signs were present, and the systemic examination was otherwise normal. At this point, appendicitis was ruled out and the work-up was aimed at identifying alternate causes for his RLQ pain. Investigations Laboratory investigations showed mild leukocytosis (12.6 x 109/L) with neutrophilia (8.6 x 109/L) and an elevated C-reactive protein (CRP) (140.4mg/L). Urinalysis was positive for erythrocytes. The high inflammatory markers combined with hematuria led to the consideration of vesicoureteric junction stone as a possible diagnosis; thus, we proceeded to a non-contrast CT scan of the abdomen and pelvis, which revealed diffuse wall thickening of a blind-ended, tubular structure arising from the cecal pole, suggestive of inflamed appendicular stump (Figure 1 ). A repeat CT with contrast further delineated the inflammatory changes to include the cecum and proximal ascending colon with surrounding fat stranding and multiple prominent reactive lymph nodes. Differential diagnosis Although the clinical picture of RLQ pain and tenderness allows for a wide range of differential diagnoses, it is confounded by the history of appendectomy. One literature review concluded that the most common misdiagnoses were constipation, gastroenteritis and right-sided diverticulitis [ 7 ]. Treatment General surgery was consulted, and no acute surgical intervention was deemed necessary. The patient was then admitted as a case of stump appendicitis and managed conservatively. He was kept nothing by mouth (NPO) and started on intravenous analgesia, anti-emetics and antibiotics (cefuroxime, metronidazole). During admission, the gastroenterology service was consulted and the patient underwent colonoscopy which was grossly normal. Random biopsies were taken for pathology examination. Outcome and follow-up The patient was discharged three days later on an oral course of the same antibiotics for a total of seven days. On a follow-up appointment with gastroenterology, he reported improved symptoms and pathology reports showed no significant histological abnormality. Repeat labs showed a downward trend of inflammatory markers. The final diagnosis was stump appendicitis.
CC BY
no
2024-01-15 23:43:45
Cureus.; 15(12):e50557
oa_package/4a/d6/PMC10787942.tar.gz
PMC10787943
38222158
Introduction Common variable immunodeficiency (CVID) is one of the prevalent primary immunodeficiencies, characterized by a wide range of symptoms and recurrent bacterial infections. Patients with CVID are at increased risk of infections, including gastrointestinal infections. Chronic diarrhea is one of the presentations of CVID, and it is often the initial symptom that leads to diagnosis. Recognizing the link between CVID and chronic diarrhea is vital. Early diagnosis can guide appropriate treatments, including immunoglobulin therapy, improving patients' quality of life and preventing complications. This case report underscores the significance of considering CVID in patients with chronic diarrhea, promoting prompt recognition and intervention.
Discussion The patient had chronic diarrhea with recurrent sinopulmonary infections, hypogammaglobinemia, hypoproteinemia, and vitamin deficiency, along with lymphoid nodular hyperplasia of the duodenum, suggesting the diagnosis of CVID (primary immunodeficiency disorder). Other differential diagnoses that were kept were: HIV-associated enteropathy (acquired immunodeficiency disorder) as the patient had fever, recurrent respiratory tract infections, chronic diarrhea, weight loss, and a history of blood transfusion, however, HIV antibodies were negative; microscopic colitis as the patient had chronic watery diarrhea and weight loss favored the diagnosis, but the normal findings of a colon biopsy favored against it; bile salt malabsorption as the patient had chronic diarrhea, weight loss, and vitamin deficiencies favored the diagnosis; however, there was no history of abdominal pain, cramping, bloating, urgency, difficulty controlling bowel movements, or steatorrhea, which was against the diagnosis of bile salt malabsorption. Inflammatory bowel disease (IBD) was our fifth differential as the patient had chronic diarrhea, anemia, and hypoalbuminemia favored the diagnosis; however, other clinical features of IBD (blood in stools, joint pain, or other extraintestinal manifestations), histopathological evidence on biopsy, or biochemical evidence (raised ESR, CRP, and fecal calprotectin levels) were absent. Lastly, our sixth differential diagnosis was disseminated tuberculosis (pulmonary and gastrointestinal tuberculosis) as the patient had a fever with an evening rise in temperature and a chronic cough. The patient was residing in a tuberculosis-endemic region, and the patient had significant weight loss. However, there was no evidence of tuberculosis based on histopathological examination of abdominal lymph nodes and sputum, and there was no history of pain in the abdomen or alteration of bowel habits or ascites. Chronic diarrhea is defined as the passage of abnormally liquid or unformed stools at an increased frequency with a duration of more than 4 weeks [ 1 - 2 ]. The approach to chronic diarrhea is different from acute diarrhea and requires complex evaluation [ 1 ]. Primary clinical assessment is done to distinguish common possibilities with further diagnostic approach requires differentiating osmotic diarrhea from secretory diarrhea by stool osmotic gap and based on this approach the study patient had secretory diarrhea [ 3 ]. Microscopic colitis presents with similar clinical features but was ruled out due to lack of histopathological evidence, similarly, bile salt diarrhea also has similar presentation but due to lack of availability of its diagnostic tests (75SeHCAT scan) in our setup and other clinical features it was not approached further. Since the patient was asymptomatic before the onset of her symptoms 9 years back, reduced immunoglobulin levels observed were likely due to acquired hypogammaglobinemia. Various causes of acquired hypogammaglobinemia are defined in the literature, namely drug-induced, infections, malignancy, and excess losses of immunoglobulins [ 4 ]. CVID is one of the primary immunodeficiency syndromes which is defined by The European Society for Immunodeficiencies (ESID) as hypogammaglobulinemia with IgG levels two standard deviations below the mean; poor vaccination responses or no isohemagglutinins; and ruling out alternative causes of hypogammaglobulinemia [ 5 ]. There is no single definitive test for CVID, and diagnosis can sometimes be challenging. The study patient was diagnosed with CVID due to her recurrent respiratory tract infection, chronic diarrhea, reduced serum IgA and IgG levels, reduced anti-HBS antibody titers, no other causes of hypogammaglobinemia were found and the disease started in the patient's second decade of life. Whole exome sequencing (WES) was performed to detect known phenotypic gene variants causing immunodeficiency and no pathognomonic or likely pathognomonic variants causative of the phenotype were detected. Prominent respiratory symptoms were also observed in study patients which could be due to recurrent respiratory tract infection or in rare cases granulomatous-lymphocytic interstitial lung disease (GLILD), which is reported in around 8-20% of cases [ 6 ]. In the large single-center prospective study done by Resnick et al. conducted in 2012, 94% of their study patients diagnosed with CVID had a history of infections; 68% of the study patients developed non-infectious complications, 29% of study patients had chronic lung disease and 15% of study patients had gastrointestinal inflammatory disease [ 7 ]. The most common infections reported in patients with CVID are bacterial infections causing sinopulmonary infections and gastrointestinal infections [ 8 ]. Apart from infections, 10-20% of patients with CVID are reported to have gastrointestinal manifestations with diarrhea being the most common symptom. These manifestations include inflammatory bowel-like disease, nodular lymphoid hyperplasia, bacterial overgrowth, nonspecific malabsorption, and gastrointestinal lymphoma [ 9 - 12 ]. Our patient had diffuse nodular lymphoid hyperplasia which can occur in up to 20% of CVID patients. It occurs due to chronic antigenic stimulation, these can be asymptomatic or present with pain in the abdomen, chronic diarrhea, intestinal obstruction, and, very rarely, a massive GI bleed. Immune globulin replacement, which is the cornerstone of therapy, has significantly changed the clinical course of CVID by lowering the burden of recurrent infections and subsequent complications [ 13 ]. Early recognition and appropriate treatment not only alleviate symptoms but also enhance the overall well-being of affected individuals.
Conclusions Evaluation of chronic diarrhea requires thorough clinical evaluation and should be evaluated with an open mind. A review of a patient’s investigations should always be considered when clinical judgment regarding a case is not satisfied. Although common diseases should be considered first in differential diagnosis points favoring the diagnosis and points against diagnosis should be compared. CVID is a primary immunodeficiency disorder that can cause a wide range of clinical manifestations, including chronic diarrhea. Early diagnosis and treatment with immunoglobulins are essential to improve the quality of life and reduce the risk of complications in patients with CVID.
Chronic diarrhea poses a diagnostic challenge due to its diverse etiology, encompassing various gastrointestinal disorders. This case report emphasizes the clinical significance of considering common variable immunodeficiency (CVID) as a potential underlying cause in a patient presenting with chronic diarrhea. In this case study, we describe a 36-year-old female with a 9-year history of chronic diarrhea, recurrent sinopulmonary infections, and weight loss for 3 years, where previous evaluations failed to yield a diagnosis. This case underscores the diagnostic hurdles faced by healthcare professionals, often causing a delay in identifying fewer common conditions like immunodeficiency syndromes. Early recognition of CVID is crucial, enabling timely intervention with immunoglobulin replacement therapy, markedly enhancing patients' quality of life and averting complications. This report highlights the necessity for a comprehensive evaluation of non-responsive chronic diarrhea cases and raises awareness about CVID as an essential consideration, facilitating precise diagnoses and tailored treatments.
Case presentation A 36-year-old female presented to our emergency department with complaints of watery stools for 9 years, recurrent respiratory tract infections for 9 years, and weight loss for 3 years. Initially, the patient developed loose motions that were watery in consistency, with a stool output of about 150 mL per episode and a frequency of four to five episodes of loose motions per day, which were associated with nocturnal diarrhea. The patient also complained of coughing and nasal stuffiness for 9 years, which was associated with expectoration for the past 4 years. The patient had also complained of fever for 9 years. The frequency of the fever varied, ranging from occurring three to four times a week to sometimes once a month. The fever was mostly associated with coughing, and there was no notable association between the fever and diarrhea. The maximum temperature recorded by the patient was 100°F sublingually. The fever was associated with an evening rise in temperature, which subsided after taking over-the-counter medications. For the past 4 to 5 years, the patient has also complained of fatigue and unintentional weight loss of 8 kg. There was a history of blood transfusions in the past 10 years due to blood loss during delivery, and she was vaccinated for hepatitis B and COVID-19 in the past. There was no significant family history, or history suggestive of pulmonary tuberculosis in any family members. There was no significant obstetrical or gynecological history. At presentation, the patient’s blood pressure was 106/60 mmHg, pulse rate was 92 per minute, respiratory rate was 14 per minute, and the temperature recorded was 98.6°F sublingually. On a general physical examination, mild pallor was observed. There were no signs of icterus, cyanosis, clubbing, pedal edema, or lymphadenopathy. The patient’s systemic examination there was nonremarkable. On anthropometric examination, her body mass index (BMI) was 18.5 kg/m 2 , and mid-upper arm circumference was 23 cm. Initial investigations, which were performed and are shown in Table 1 , revealed mild macrocytic anemia, thrombocytopenia, hypoproteinemia with hypogammaglobinemia, vitamin B12 deficiency, folate deficiency, and vitamin D deficiency. The stool examinations were also performed with negative results for stool microscopy (for ova, cysts, and parasites), stool culture and sensitivity for pathogenic organisms, and fecal occult blood test. The patient’s calculated stool osmotic gap suggested secretory diarrhea, which corresponded with a normal 72-hour fecal fat estimation test that ruled out malabsorption syndrome and functional diarrhea (Table 2 ). Upper gastrointestinal endoscopy was performed, which revealed atrophic gastric mucosa and nodular lymphoid hyperplasia of the D2 part of the duodenum (Figures 1a , 1b ) (Table 3 ). Histopathological examination of these lesions revealed chronic active duodenitis with nodular lymphoid hyperplasia with no evidence of dysplasia or malignancy. Similarly, a lower gastrointestinal endoscopy was also performed, but it was unremarkable (Table 4 ). Biopsies were taken from the transverse and descending colons and were also unremarkable(Table 5 ). An x-ray of the chest was also done and showed bilateral calcified and fibrotic lesions (Figure 2 ), and the sputum examination (for biochemical tests, microscopic examination for gram stain and acid-fast stain, and culture for pathogenic organisms) was nonremarkable. A Triple-Phase Contrast-Enhanced CT A whole abdomen was performed to rule out any structural disease and showed mild splenomegaly with variable-sized, homogenously enhancing mesenteric lymph nodes. A CT-guided abdominal lymph node biopsy was then performed, which showed reactive lymphocytosis with no evidence of granulomatous changes or malignancy (Table 5 ). A secondary clinical assessment was done to assess for chronic diarrhea associated with immunodeficiency. A serological test was conducted to detect viral markers for human immunodeficiency virus (HIV), and the result was negative. Since hypogammaglobinemia was already documented during the initial investigations, the patient was then evaluated for a poor response to immunization. The patient's anti-HBs titer was ordered, and despite receiving a booster dose of the hepatitis B vaccine in the past, low levels of anti-HBs antibodies were observed. (Table 6 ). Initially, the patient was given a course of injectable antibiotics for 5 days. Ceftriaxone (Gram-positive and Gram-negative coverage) and metronidazole (anaerobic coverage) were given along with supportive treatment of antisecretory agents and encouraged oral fluid intake. The patient was relieved for a few days, but her symptoms began again. After the diagnosis of CVID was considered, the patient was started on intravenous immunoglobulin (IVIG) at 500 mg/kg body weight every four weeks with premedication under supervision. The patient was discharged in satisfactory condition and was in follow-up. Serum immunoglobulin levels were done in follow-up, and they showed rising titers of serum IgA, IgG, and IgM. During follow-up, the patient was asked about her general well-being, appetite, symptoms of cough, water stools, fever, and any complications related to IVIG administration. Two weeks after receiving the third dose of IVIG, the patient’s watery stools improved and her fever subsided. After the sixth dose, the patient's bowel movements returned to normal, her cough subsided, her appetite increased, and her BMI increased from an initial 18.5 to 21.2. The patient did not suffer from any complications from IVIG, and after the eighth dose of IVIG, the patient became asymptomatic. One year after her first dose of IVIG, the patient is still asymptomatic.
CC BY
no
2024-01-15 23:43:45
Cureus.; 15(12):e50556
oa_package/e1/87/PMC10787943.tar.gz
PMC10787944
38222183
Introduction The occurrence of instrument separation during endodontic therapy poses a challenging situation, with reported incidences ranging from 2% to 6% in investigated cases [ 1 ]. The presence of a distinct instrument within the root canal obstructs access to the root apex during nonsurgical root canal therapy. These instruments commonly encompass a variety of types, including files, reamers, peeso reamers, Gates-Glidden drills, thermomechanical compactors for gutta-percha compaction, Lentulo spirals, or the tips of specific hand instruments such as gutta-percha spreaders or explorers [ 2 ]. Common etiologies of file separation include incorrect use, restrictions in its physical properties, insufficient access, aberrant anatomy of the root canal, and possible manufacturing flaws [ 2 ]. If the broken piece protrudes from the root apex, it may irritate the periapex or obstruct comprehensive root canal shaping and cleaning treatments apical to the point of its separation. This is significant in endodontic therapy, impacting the final treatment outcome [ 3 ]. Therefore, it is crucial to attempt instrument retrieval or bypass before considering obturation to the level of separation or resorting to surgical intervention. The Masserann approach is unique among the many ways to remove foreign items from the root canal [ 4 ]. With success rates as high as 55%, this method is very useful for extracting intracanal silver points, broken fragments of files, and posts [ 5 ]. The armamentarium employed includes long, crown-cutting diamonds (Shofu Preparation Kit, Kyoto, Japan), Gates-Glidden drills (Mani Inc., Tochigi, Japan), a slow-speed, contra-angle handpiece (NSK, Japan), and the Masserann kit (Micro Mega, Besançon, France) [ 6 ]. By severing the surrounding radicular dentin, the Masserann kit trephine burs, which are end-cutting and color-coded, progressing in size, rotate anticlockwise to liberate space in the periphery of the separated instrument’s coronal end. The extractor resembles a tube and has a stylet or plunger rod. It seals the fragment's exposed coronal end just short of the extractor's end against an internal dent when screwed within. After that, the fragment can be eliminated by rotating anticlockwise [ 6 ]. This case report describes the successful extraction of a separated file firmly lodged in the root canal dentin of a first premolar on the right mandible.
Discussion A significant obstacle to root canal cleaning and shaping is the separation of instruments within the canal, which prevents access to the apex. As a result, there is a danger to the endodontic treatment outcome and a decreased likelihood of successful retreatment [ 5 , 7 ]. Some of the factors that influence the prognosis in these kinds of cases are the state of the root canal (vital or nonvital), the tooth's status (symptomatic or asymptomatic, with or without periapical pathology), the level of cleaning and shaping at the time of separation, and the location of the separation within the canal. Generally speaking, the prognosis is worse than with conventional endodontic therapy [ 1 ]. As a result, every attempt should be taken to retrieve or avoid using the detached instrument. The root canal’s length, curvature, and diameter of its cross-section; thickness of dentin and root morphology; the instrument's content and cutting action (counterclockwise or clockwise); and the location, length, and degree of binding of the instrument within the canal are among the variables that affect orthograde retrieval [ 5 ]. Three instrument retrieval strategies exist, including chemical, mechanical, and surgical techniques [ 7 ]. Since surgery does not necessitate a crucial amount of dentin removal, it should be considered first when the separated fragment is mostly or entirely outside the root canal [ 7 ]. Chemical methods that corrode the fractured metallic instrument with solvents such as nitric acid, sulfuric acid, iodine trichloride, hydrochloric acid, and iodine crystals [ 5 ] or dissolve the instrument electrochemically using electrolyzed solutions of sodium chloride or fluoride [ 8 ] are inefficient for retrieving instruments because they take a significant amount of time to dissolve the metallic instrument completely. Furthermore, because these chemical solvents are limited to the shattered instrument surface in the canal, they are regarded as unpredictable and may harm the nearby soft and hard tissues [ 7 ]. There are two steps in every mechanical technique for retrieving instruments. The initial step in preparing the root canal is using ultrasonic or rotary instruments to release the broken instrument. The next step is to try to retrieve the broken instrument using ultrasonics or special equipment [ 7 ]. Mechanical methods for retrieving instruments can be broadly divided into two categories: those that use trephine bursts to penetrate the separated instrument's periphery during the preparation phase, followed by attempts to remove the instrument using equipment, and those that use ultrasonics or particular files to create a tiny space only on the fractured instrument's side during the preparation phase, followed by attempts to remove the instrument using devices or ultrasonics. Particular files and loops are included with these devices to remove the broken instruments [ 7 ]. The Canal Finder system, the EndoPuls system (EndoTechnic, San Diego, CA), and small-diameter ultrasonic tips, such as ET25 and TFRK-S, are examples of systems employing ultrasonics or specific files [ 7 ]. These devices provide a vertical movement using a handpiece and specific files, which helps bypass the separated instrument [ 9 ]. The needle-sleeve technique, Masserann kit, Endo Extractor (Brasseler Inc., Savannah, GA), Cancellier Extractor Kit (SybronEndo, Orange, CA), and Micro-Retrieve and Repair System (Superline NIC Dental, China) are mechanical devices using trephine burs in endodontic procedures. In the preparation process, these technologies uncover the coronal region of the separated instrument using a hollow tube with a cutting end and a diameter of 0.7-2.4 mm [ 7 ]. As in our case the separated instrument was present in the straight portion of the posterior teeth and most of it was present within the canal, we employed the nonsurgical mechanical method and used the Masserann kit for its retrieval. With over 30 years of experience, the Masserann kit has been used to remove damaged tools from teeth. Success rates for anterior and posterior teeth are 73% and 44%, respectively [ 10 ]. However, because using relatively large and rigid trephines can result in removing considerable quantities of root dentin, weakening the tooth, or increasing the risk of perforation, it requires regular radiographic monitoring. It may be less effective in teeth having thin or curved roots or in apically fractured fragments [ 3 ]. Notwithstanding these drawbacks, the Masserann kit works incredibly well to remove metal fillings from front teeth with robust, straight roots. The extractor's locking mechanism provides significant retention for grasping and removing firmly wedged impediments in the canal. A direct path to the fragment makes it easier for the trephine to center over it, releasing the coronal end circumferentially and safely removing the surrounding dentin. This makes it easier to grasp the fragment firmly and makes it easier to retrieve it along the root's long axis, allowing for regular retreatment [ 6 ]. Our successful attempts to remove detached files in posterior teeth using the Masserann kit challenge the literature's suggestion that employing the Masserann approach for posterior teeth can be challenging [ 11 ]. Every operation was carried out under rubber dam isolation. In one instance, the clamp's wing obscured the separated segment's appearance on the radiograph, requiring the clamp to be removed to obtain a clean image. In another instance, where the tooth was severely damaged, wedges were used to hold the rubber dam in place rather than clamps. The optimum course of action is prevention, and in situations where instruments separate, the significance of safe retrieval or bypassing is emphasized [ 12 ]. Although the Masserann procedure is time-consuming and technique-sensitive [ 13 ], detached files from maxillary lateral incisors and maxillary and mandibular molars were successfully recovered through strategic use within clinical constraints and operator expertise. On the other hand, in some situations, using ultrasonics and a dental operating microscope can increase efficacy [ 14 ].
Conclusions Instrument separation during endodontic therapy poses a significant challenge, affecting the success of the treatment. The presented case report demonstrates the successful removal of a tightly wedged file in a mandibular first premolar using this technique. Despite its limitations, such as the need for frequent radiographic monitoring and restricted application in certain tooth types, the Masserann kit proves effective in many cases. The importance of instrument retrieval before surgery is emphasized, underlining the technique's clinical relevance and the positive outcomes in challenging cases.
Instrument separation during endodontic therapy is a complication occurring in 2% to 6% of cases. Focusing on the Masserann technique, the study presents a success rate of 55% in retrieving separated instruments. The technique's effectiveness is demonstrated through a case involving retrieving an instrument from the mandibular first premolar. The technique utilizes various tools, including trephine burs and an extractor, providing a reliable means to dislodge tightly wedged fragments. Despite limitations in specific tooth types and the necessity for frequent radiographic monitoring, the Masserann kit proves effective and underscores the importance of attempting retrieval before considering surgical interventions. The presented case exemplifies the technique's clinical applicability and positive outcomes in intricate scenarios, emphasizing its significance in endodontic practice.
Case presentation A 50-year-old male patient visited the department with a complaint of pain in the lower right back region of his jaw that had been there for three days. The pain was localized, of moderate intensity, characterized as a dull ache, and continuous. It exacerbated while lying down and during mastication. Upon clinical examination, temporary restorations were observed on teeth 44 and 45. Radiographic examination revealed a white radiopaque shadow in the root canal, indicating the presence of a separated instrument. The fragment was situated in the coronal third of the root of tooth 44, extending approximately 5 mm beyond the apex, with a length of 13.6 mm. Periapical radiolucency was evident around the separated fragment, as shown in Figure 1 . Considering the single-rooted nature of the tooth, the absence of root canal curvature, and the fragment's location in the coronal third of the root, attempts to bypass the fragment proved unsuccessful. Therefore, the decision was made to use the Masserann technique for fragment removal. Gates-Glidden drills were used one after the other to straighten the root canal to allow radicular access to the fragment's coronal end. A contra-angle handpiece was used to attach a pre-selected 1.2 mm trephine, which was then revolved in a counter-clockwise orientation to form a trench surrounding the fragment's coronal end and remove dentin. The trephine was centered correctly over the fragment, and radiographic confirmation was obtained. The piece was then sleeved by sliding an extractor tube with a 1.2 mm diameter into the trough. The extraction tube's plunger rod was manually rotated clockwise to grasp the fragment against its wall upon radiological confirmation of its placement, as shown in Figure 2 . The entire assemblage was spun counterclockwise to release the instrument from the dentin and enable removal as soon as the tightest hold was detected, as shown in Figure 3 . The working length was determined using an apex locator, with tooth 44 having a length of 14 mm. Biomechanical preparation was carried out on tooth 44 using Dentsply hand protaper files until reaching file F3. Calcium hydroxide intracanal medicament was placed, and a temporary dressing was applied to tooth 44. The patient was recalled after seven days. During the subsequent visit, the patient was asymptomatic. The temporary dressing was removed, and the canal was thoroughly irrigated with normal saline and 5.25% sodium hypochlorite. Mastercone fit was evaluated, and the canal was obturated, followed by post-endodontic restoration with teeth 44 and 45, as shown in Figure 4 . The patient was then scheduled for a follow-up after one month, during which he remained completely asymptomatic.
CC BY
no
2024-01-15 23:43:45
Cureus.; 15(12):e50559
oa_package/68/da/PMC10787944.tar.gz
PMC10787948
38222244
Introduction Novel oncological therapies are a major area of new drug development and comprise a substantial portion of new drug approvals in the United States and Europe [ 1 , 2 ], driven by modern advances in targeted molecular therapies [ 3 ]. However, newer anti-cancer therapies are increasingly expensive, and the therapeutic benefit provided is often modest in comparison to prices [ 4 ]. National regulatory agencies such as the Medicines and Healthcare Products Regulatory Agency (MHRA) in the United Kingdom (UK) assess whether medicines provide an appropriate benefit to the risk profile of patients to licence them for marketing but do not take economic aspects into account. Independent organisations such as the National Institute for Health and Care Excellence (NICE) in England and the Scottish Medicines Consortium (SMC) in Scotland are responsible for pharmacoeconomic appraisals to assess the cost-effectiveness of medications. These organisations publish guidance on whether the National Health Service (NHS), the public healthcare provider in the UK, should reimburse a treatment after assessing the clinical benefits of the treatment against the financial costs of the treatment. Cost-effectiveness is commonly defined through the quality-adjusted life year (QALY), equal to one year of life in perfect health [ 5 ], as a measure incorporating both lifespan and quality of life. The incremental cost per additional QALY added by a novel therapy compared to existing therapies is assessed as a metric of the health utility provided by a medicine. The SMC does not have a formal QALY threshold for funding, whilst NICE commonly utilises an incremental threshold of £20,000 to £30,000 per QALY added to guide funding decisions [ 6 - 8 ]. Although both organisations share the use of the QALY as a cost-effectiveness measure, they use different criteria, methodologies and implementation processes for appraising medicines. Despite this, the two organisations generally align on appraisal outcomes, although the SMC is noted to take longer to publish appraisals for oncology indications [ 9 ]. Whilst the health technology assessment (HTA) process provides another protection for the health service and taxpayers, the process can be lengthy and delay public access to medications. Common criticisms include ambiguity about decision-making, inappropriate funding decisions, lack of consistency, delays and lack of access for patients to treatments, particularly for cancer and orphan indications [ 10 - 14 ]. Healthcare providers in England have a statutory responsibility to make funding available for a medicine within 90 calendar days after guidance recommending its use is published by NICE. In Scotland, NHS providers are expected to reach a decision on the provision of an SMC-accepted medicine within 90 days of the SMC issuing advice to the health board that the medicine is recommended. Thus, the guidance of bodies such as NICE and SMC is integral to ensuring that novel oncological therapies are cost-effective and appropriate for funding, and their decisions and timelines can have significant impacts on the availability of therapeutic options for cancer patients in the UK.
Materials and methods Single health technology assessments (HTA) of oncological therapies published by NICE from January 1, 2017, to December 31, 2022, were identified on the NICE website. All appraisals of single technologies for oncology indications published by SMC between January 1, 2017, and December 31, 2022, were identified from published documents on the SMC website. The time from marketing authorisation (MA) until publication of HTA guidance was the primary outcome measure. The UK MA approval dates for the relevant indication or dates of label extension were obtained from the European Medicines Agency (EMA) or MHRA websites or the UK Summary of Product Characteristics for the product. If this data was not publicly available using the aforementioned sources, then it was obtained via an information request from the medical information department of the MA holder. Re-appraisals of technologies where HTA guidance was previously published were excluded. Assessments that were terminated due to non-submission by the MA holder were excluded. Differences in time from the date of the UK MA to publication of HTA between SMC and NICE were compared using a two-sided Mann-Whitney U test. P-values of <0.05 were considered statistically significant, and there was no control for multiplicity. Data collation and analysis were carried out in Microsoft Excel (Microsoft Corporation, Washington, USA) and Minitab version 21 (Minitab, LLC., Pennsylvania, USA).
Results Two hundred and six single HTAs were published by the SMC, and 253 were published by NICE between January 1, 2017, and December 31, 2022, for oncology therapies. Following exclusions, we included 148 HTAs for SMC and 161 HTAs for NICE, of which 111 technologies had HTA guidance published by both agencies during the study period. Figure 1 shows a flow diagram of the selection process. Overall median time from MA to publication of guidance was not significantly different between organisations: 291 days (IQR 222-406) for SMC and 257 days (IQR 167-448) for NICE (p=0.054, not significant) (Table 1 ). A similar amount of technologies were recommended in guidance by SMC and NICE (90.5% and 89.4%, respectively), with a similar proportion recommended with restrictions on their use (29.7% SMC and 26.1% NICE). The majority of HTAs were for solid organ cancers (70% for both organisations) (Table 1 ). The median time from MA to publication of HTA guidance for solid organ cancers was significantly lower for NICE at 231.5 days (IQR 148-392.25) compared to SMC at 273 days (IQR 202-378) (p=0.039) (Table 1 ). There was no significant difference in time from MA to publication of HTA guidance for haematological malignancy between SMC and NICE (p=0.597) (Table 1 ). The most common tumour types for technologies were lung and breast for both organisations (SMC: breast 23 and lung 23 appraisals, NICE: breast 20 and lung 32 appraisals) (Table 2 ). The median time to publication of HTA guidance after MA was significantly shorter for solid organ cancer therapies than haematological malignancy for NICE (231.5 days (IQR 148-392.25) vs. 339 days (IQR 206-623), p=0.010) and SMC (273 days (IQR 202-378) vs. 327 days (IQR 258.5-780), p=0.012). There were 111 technologies with published guidance by both agencies between January 1, 2017, and December 31, 2022 (Table 3 ). For these technologies, the median time from MA to HTA publication date was significantly longer for SMC with 287 days (IQR 217-362) than NICE with 233 days (IQR 144-358) (p=0.005) (Table 3 ). A similar number of technologies were recommended in guidance for SMC and NICE (90.1% and 91%, respectively) (Table 3 ), and similar proportions were recommended with restrictions (28.8% SMC and 26.1% NICE). All technologies not recommended were for solid organ cancer. The median time from MA to publication of HTA guidance was significantly shorter for NICE than SMC for solid organ cancer (NICE 225 days (IQR 135-313), SMC 272 days (IQR 220.5-354.75) (p=0.006)) but not for haematological malignancy (NICE 293 days (IQR 179-489), SMC 318 days (266-400), p=0.338). There were 14 technologies assessed by both agencies where there was discordance in the final guidance recommendation (Table 4 ). Six technologies were rejected by NICE and approved by SMC, and eight technologies were rejected by SMC and approved by NICE (Table 4 ).
Discussion In this study, we assessed 161 single HTAs published by NICE and 148 by SMC for oncological therapies. A similar proportion of technologies were recommended for use by both agencies (SMC 90.5%, NICE 89.4%). The number of technologies that were recommended with restrictions on their use was also similar between agencies. An overall recommendation rate of 85% is reported by NICE for data from 2000 to 2023 [ 15 ], although this also covers appraisals for non-oncology therapies. A 2012 study by Ford et al. reported an 80% recommendation rate for NICE and a 71.4% recommendation for SMC for cancer therapies [ 9 ]. This study reported a median time from MA to publication of single HTA guidance for cancer drugs of 25.2 months for NICE and 8.0 months for SMC. That study looked at data from different decades (the 2000s vs. 2010s/2020s), where NICE and SMC processes were different and the method of data extraction for SMC was different (from annual appraisal summary documents rather than single published guidance documents), which may explain the discrepancies seen. More recent studies looking at HTA final guidance documents between 2014 and 2016 for oncology therapies found recommendation rates of 79% for NICE and 75% for SMC [ 16 , 17 ]. In our study, there was disagreement between SMC and NICE on recommendation or non-recommendation for a minority of technologies (12.6%). This is consistent with previous work that has shown a high degree of agreement between SMC and NICE for cancer therapies compared to other European countries [ 18 , 19 ]. We found that NICE published faster guidance for solid organ therapies but not haematological malignancy therapies than SMC, although the reasons for this are unclear. NICE was significantly faster at publishing guidance than SMC for technologies that were assessed by both agencies during the study period, which was driven by faster times for solid organ cancer therapies, with no difference in timelines seen in therapies for haematological malignancy. This likely reflects the fact that technologies appraised by both agencies in the study period were likely to be completely novel therapies in high-impact areas or areas of unmet need, which may have led to an acceleration of timelines for both agencies, although this appears to have been more pronounced for NICE than SMC. The Cancer Drugs Fund (CDF) was set up in England in 2010 to improve access to novel cancer therapies. The CDF provided reimbursement for therapies that were pending appraisal by NICE or have been rejected by NICE due to a lack of cost-effectiveness or immature data for health economic models [ 20 ]. In 2016, the CDF was reformed due to multiple consecutive years of spending outside of its allotted budget and re-aligned with NICE [ 20 , 21 ]. These reforms included changes to the NICE appraisal process, such that the process began earlier, with initial submissions and reviews occurring prior to the drug receiving MA approval, thereby reducing delays to patient access after MA approval by the MHRA [ 20 ]. Previous work has indicated that 2014 reforms at the SMC for the approval of end-of-life and orphan indications had led to increased access to therapies for advanced cancer, although this only addressed approval rates and not timelines to approval [ 16 , 22 ]. The difference in processes between NICE and the CDF may provide some explanation as to why NICE continues to publish oncology HTA guidance faster than the SMC. Other factors that can account for discrepancies between the two agencies that have previously been published in the literature include the method of dealing with uncertainties about cost-effectiveness, comparator choice, clinical benefits [ 23 , 24 ], negotiation of patient access schemes and market entry agreements [ 18 ], consideration of indirect benefits of treatment and the innovative nature of treatments [ 25 , 26 ]. The increased recommendation rate observed in our studies compared to previous work likely reflects the ongoing consequences of reforms made in the 2010s at both organisations and further refinements in the approach of manufacturers and agencies to increase access to oncological therapies. Delays in publication of HTA guidance can occur for a variety of reasons. The type of technology appraised is relevant, as both therapies for oncology and orphan status are known to extend the appraisal process duration [ 27 , 28 ]. Initial draft guidance by NICE is negative in 60% of cases [ 29 ] and leads to significant delays in the issuance of final guidance [ 28 ]. The manufacturer's response to the draft guidance is also important in terms of the final outcome. In one study, 38% of preliminary negative decisions could receive recommendations in final guidance after the introduction or enhancement of a patient access scheme discount, which would be at the discretion of the manufacturer [ 28 ]. Drugs for advanced cancers are also known to require more committee meetings prior to the decision and a longer time from the first meeting until publication of final guidance than non-oncology therapies for both SMC and NICE [ 9 , 28 ]. Limitations This study has a number of limitations. We assessed only the time until publication of final guidance and did not take into account where draft guidance rejecting the drug was issued before the company resubmitted with final approval (which may come with substantial time delays). Additionally, the time to publication of results may be affected due to a lack of, or delay in, submission of data by the MA holder as well as delays in the SMC or NICE process, which our study does not distinguish between. We also assumed that the decision-making processes of NICE and SMC are independent; however, companies may change the contents of the submitted dossier, including their proposed pharmacoeconomic models, on the basis of feedback or interactions with the other agency. Furthermore, the initial models submitted by the manufacturer to the respective agencies may be different, which can lead to increased meetings and time until publication of the guidance or differences in the outcome of the HTA process. Finally, we only looked at decisions within the UK, although evidence suggests there is significant variation in HTA assessment and recommendations between G7 countries [ 30 ].
Conclusions Recommendation rates are similar for single health technology assessments of oncological therapies for both NICE and SMC (89.4% and 90.5%), with only a minority of therapies (12.6%) having discordance in recommendation outcomes between the two organisations. NICE published guidance significantly faster than the SMC for solid organ cancer therapies, but timelines were similar for haematological malignancies. This is most likely due to differences in process and methodology between the two agencies, in particular the role of the CDF in England.
Background and aims Pharmacoeconomic assessment of novel oncological therapies is an increasingly important factor in determining patient access to therapies. Organisations such as the National Institute for Health and Care Excellence (NICE) in England and the Scottish Medicines Consortium (SMC) in Scotland assess medications for their cost-effectiveness through health technology assessments (HTA) and provide guidance on whether the public health service should fund a therapy. We assessed six years of data to determine if there were any differences in timescales and decisions between NICE and SMC for new oncological therapies. Methods and results Time (days) from marketing authorisation (MA) to publication of final HTA guidance was calculated for single technology appraisals published by NICE and SMC between January 1, 2017, and December 31, 2022, for oncological therapies. We assessed 161 HTAs by NICE and 148 HTAs by SMC published in the study period. The median time from MA to publication of HTA guidance was 291 days (IQR 222-406) for SMC and 257 days (IQR 167-448) for NICE (p=0.054). For solid organ cancer therapies, NICE was significantly faster in publishing guidance, with a median of 231.5 days (IQR 148-392.25), compared to SMC, which took 273 days (IQR 202-378) (p=0.039). Overall recommendation of technologies was similar between the SMC and NICE (90.5% and 89.4%, respectively), with discordance in a minority of cases (12.6%). Conclusions Recommendation rates for single HTAs are similar between NICE and SMC for oncological therapies with discordance in a minority of cases. The time from MA to publication of HTA guidance was similar overall, but NICE was faster in publishing HTA guidance for solid organ cancer indications. Differences in methodology and process between the two organisations, in particular the presence of the Cancer Drugs Fund in England, may explain this difference in publication times.
CC BY
no
2024-01-15 23:43:45
Cureus.; 15(12):e50560
oa_package/85/53/PMC10787948.tar.gz
PMC10787951
38222194
Introduction Prostate cancer is the most common malignancy among men and is the second-leading cause of cancer death in men [ 1 ]. In contrast to other cancers that commonly spread to brain, it is unusual for prostate carcinoma to metastasize to the central nervous system (CNS), making it the subject of case reports due to the uncommon event of neurologic side effects. We report a case of a 72-year-old male who presented to the hospital with a chief complaint of diplopia in the setting of a recent onset of urinary incontinence and right-sided back pain and was subsequently diagnosed with prostate cancer, notably metastasizing to the right sphenoid bone, causing impingement of the oculomotor nerve. Unlike previously few reported cases of oculomotor nerve palsy due to prostate cancer with non-adenocarcinoma pathology, our case biopsy uncovered neuroendocrine and adenocarcinoma histology, which is a rare phenomenon, and the patient received palliative orbital radiotherapy. This case report highlights the significance of employing a multidisciplinary diagnostic approach and emphasizes the pivotal role of palliative radiotherapy in alleviating symptoms related to rare skeletal metastases. Additionally, it underscores the significance of utilizing advanced imaging techniques for the early detection of such rare instances. Increased awareness of these atypical manifestations can contribute to prompt intervention and enhance outcomes in similar cases.
Discussion Prostate cancer is the second most common malignancy among men, with a range of clinical presentations, often urinary symptoms or bone pain due to metastasis [ 2 ]. Diagnostic confirmation involves prostate biopsy displaying adenocarcinoma, small cell carcinoma, and/or neuroendocrine phenotypes with grading based on the Gleason scoring system [ 3 ]. Metastasis is usually to the bone (84%), and much less commonly to distant lymph nodes (10.6%), liver (10.2%), thorax (9.1%), and brain (3.1%) [ 4 ]. Our case is unique in that the patient’s initial presentation of prostate cancer was oculomotor nerve palsy with subsequent histologic analysis of the primary tumor showing both small cell neuroendocrine carcinoma along with adenocarcinoma with intermediate-to-high risk Gleason scores. Although intracranial metastasis from prostate adenocarcinoma is extremely rare, brain metastasis from other types of prostate tumors is much higher; however, in our case, FDG-PET shows that this is likely from adenocarcinoma and not from neuroendocrine prostate cancer (NEPC)/small cell. The utility of FDG-PET in identifying brain metastases has often been scrutinized in the literature, revealing limitations due to its low sensitivity. According to some reports, PET could only detect 61-68% of metastatic lesions compared to those identified by MRI. This highlights the superior diagnostic capability of MRI in the context of brain metastasis detection, especially in adenocarcinomas [ 5 ]. Prostate adenocarcinoma is characterized by morphological features resembling luminal prostate cells, is androgen-driven, and is typically associated with elevated serum prostate-specific antigen (PSA). On the other hand, neuroendocrine prostate cancer (NEPC) is an aggressive subtype that can emerge spontaneously or develop in advanced stages of prostate cancer, or often as a result of treatment resistance. Patients with pathologically confirmed NEPC commonly exhibit visceral metastases, low PSA levels, and frequent loss of the RB1 and TP53 genes [ 6 ]. Two previous case reports have reported oculomotor nerve palsies as the presenting symptom of prostate cancer, but neither describes adenocarcinoma histology or the use of palliative radiotherapy [ 7 , 8 ]. Among 27 previously diagnosed prostate cancer patients who received bone scintigraphy, only 1 (3.1%) had skull metastasis [ 9 ]. A recent retrospective study reported that patients who had skull metastases had significantly higher biopsy Gleason scores, higher clinical T-stage, and shorter overall survival [ 10 ]. While the patient in this case did not have brain metastasis, it has been noted that the tumor histology impacts the likelihood of brain metastasis with small cell and primary transitional cell carcinomas more likely to do so than adenocarcinoma [ 11 ]. Also, prostate cancer with brain metastasis often involves mutations of homologous recombination repair genes, including BRCA1, BRCA2, and many others [ 12 ]. Though cerebrovascular accident (CVA), trauma, myasthenia gravis, granulomatous lesions, and multiple sclerosis were differential diagnoses for his diplopia, the patient’s age, urinary symptoms, musculoskeletal pain, and initial diagnostic workup prompted further evaluation of prostate cancer. This patient’s imaging showed compressive effects on the oculomotor nerve at the right sphenoid bone. In addition, palliative radiation therapy has been successful in alleviating the immediate symptoms. Since 1940's androgen deprivation therapy (ADT) alone has been the standard of care for many years in men with metastatic prostate cancer. Due to the limited survival under this monotherapy, many new treatment options have been developed in recent years. Especially for hormone-sensitive prostate cancer, combination therapies of two or three agents of ADT, androgen receptor signaling inhibitors (ARSIs), and chemotherapy have proven effective, resulting in a substantial improvement in overall survival [ 13 ]. The latest findings from cohort 6 in the COSMIC-021 study have reignited enthusiasm in the field of immunotherapy for metastatic prostate cancer. The study revealed a notable 32% response rate and an impressive 80% disease control rate when utilizing the combination of chemotherapy and immunotherapy. These promising results underscore the potential of immunotherapeutic in addressing this complicated disease [ 14 ]. GnRH agonists such as leuprolide and anti-androgens such as bicalutamide can be used [ 15 ]. Chemotherapy regimens often include docetaxel, cabazitaxel, and corticosteroid [ 16 ]. Palliative radiotherapy offers a speedy, economical, and compelling approach to decreasing large numbers of focal symptoms, especially in advanced cancer. Numerous studies so far have proven significant pain control and improvement in quality of life in about 50-60% of patients after receiving palliative radiotherapy, like in our case [ 17 ]. Typically, adenocarcinoma is androgen-sensitive and treated by ADT, while NEPC is rare, not androgen- or hormone-sensitive, more aggressive, and treated by chemotherapy. As mentioned earlier in neuroendocrine tumors, PSA levels are low due to a lack of expression of androgen receptors; hence, serum PSA does not correlate with disease burden. In our case, elevated PSA gave a clear diagnostic cue towards prostate as a sight of primary; however, this correlation is more prominent in patients with adenocarcinoma and not NEPC. In the event that the patient with isolated NEPC with metastasis, further testing and workup will be required. The attribution of prostate cancer metastasis as the cause of this patient’s oculomotor nerve palsy occurred after extensive evaluation for other causes, and imaging of the pelvis was only completed after realizing the patient had cervical spine metastasis. It is also interesting to note that despite the presence of seemingly aggressive disease from NEPC, which had metastasized to the bone, our patient’s symptoms were discovered through metastasis from relatively less aggressive adenocarcinoma, which was widespread.
Conclusions Ptosis itself is a rare presenting feature of prostate cancer. Other case reports describe ptosis as a presentation, but they do not describe the neuroendocrine histology. In addition, routine stroke protocol MRI and CTA missed the lesion, while gadolinium-enhanced targeted MRI revealed lesions in both the spine and the orbit. This case emphasizes the need for contrast-enhanced, as well as focused imaging in patients presenting with diplopia with negative initial workup for stroke. Ptosis can be a sign of metastasis from other cancers, including breast cancer, head and neck cancer, thyroid cancer, lymphoid or neuroblastoma, and it is important to have a broad differential including metastatic disease in patients presenting with similar symptoms and negative workup who may otherwise be at risk of cancer. The case illustrates the importance of physical examination, diagnostic evaluation, and radiologic imaging in providing appropriate care to a patient. As a final note, this case report suggests that a more extensive study encompassing more patient samples could further enhance our understanding of the clinical nuances associated with ptosis as a presentation of prostate cancer. Such studies would aid in refining diagnostic and therapeutic strategies for similar presentations in the future.
We report a case of a 72-year-old male who presented to the hospital with a chief complaint of diplopia in the setting of a recent onset of urinary incontinence and right-sided back pain. He was subsequently diagnosed with prostate cancer, notably metastasizing to the right sphenoid bone, causing impingement of the oculomotor nerve. Our case is unique in that the patient’s initial presentation of prostate cancer was oculomotor nerve palsy with subsequent histologic analysis of the primary tumor showing both small cell neuroendocrine carcinoma along with adenocarcinoma. Also, the initial routine stroke protocol MRI and computed tomography angiography (CTA) missed the lesion, while gadolinium-enhanced targeted MRI revealed lesions in both the spine and the orbit. This case emphasizes the need for enhanced contrast as well as focused imaging in patients presenting with diplopia with a negative initial workup for stroke. Ptosis can be a sign of metastasis from other cancers and it is important to have a broad differential including metastatic disease in patients' presenting with similar symptoms and negative initial workup who may otherwise be at risk of cancer.
Case presentation A 72-year-old male with medical history of hypertension, hyperlipidemia, prediabetes, Lyme disease, coronary artery disease, hepatitis C, 50-pack-year smoking history, and diverticulitis was admitted for four days of worsening diplopia and right-sided ptosis. His physical examination revealed ptosis with asymmetric pupils (right 2.5 mm, left 3 mm), bilaterally reactive to light and accommodation. The right eye was depressed and abducted with a disconjugate gaze. On rightward gaze, diplopia improved but on leftward gaze, the right eye did not adduct, diplopia worsened, and nystagmus was noted. Cranial nerves I, II, and IV-XII were intact, and no other neurological deficits were observed. Initial lab workup was unremarkable in explaining his symptoms. Workup for stroke, including CT angiography of the head and neck (Figures 1A , 1B ) and stroke protocol MRI brain without contrast (Figures 1C , 1D ), were essentially within normal limits. Autoimmune workup revealed antinuclear antibody (ANA) titer 1:640 but negative for anti-SCL-70, anti-Smith, anti-dsDNA, anti-Ro/SSA and anti-La/SSB, c-ANCA, p-ANCA, rheumatoid factor, anti-sm/RNP, and acetylcholine receptor-blocking antibodies. The viral hepatitis panel was positive for hepatitis C antibodies but undetectable for hepatitis C virus (HCV) RNA. A CT scan of the chest with contrast showed bilateral subtle pulmonary nodules (the largest being 6 mm in the right upper lobe). Later, during his hospital stay, he complained of severe bilateral back pain, more in the mid-scapular region. He endorsed having had similar complaints intermittently for the last two weeks. MRI of the cervical spine (Figure 2 ) revealed multiple vertebral lesions and abnormal marrow signals concerning diffuse metastatic disease. On detailed review, he endorsed worsening urinary urgency, hesitancy, incomplete emptying, and nocturia despite taking tamsulosin for the last three months. He was supposed to follow up with the urologist for the urinary symptoms and check prostate-specific antigen (PSA) levels, but unfortunately, due to some reasons, it was delayed. His serum PSA levels during the hospitalization were found to be significantly elevated (226.3 ng/ml). Contrast MRI of the orbits revealed right sphenoid metastasis with extraosseous extension into the right cavernous sinus, causing compression of the third cranial nerve. CT abdomen and pelvis with contrast showed right prostate apex mass with lymph node involvement, liver metastasis, as well as extensive skeletal metastasis. He underwent ultrasound-guided prostatic biopsy, which showed small cell neuroendocrine carcinoma (3/12 cores) and adenocarcinoma (Gleason 7 8/12 cores, Gleason 9 1/12 cores). Immunohistochemistry was positive for TTF-1, synaptophysin, and chromogranin. 18 F-fluorodeoxyglucose-positron emission tomography (FDG-PET) showed increased FDG uptake in the apical and lateral areas, consistent with the area showing a small cell neuroendocrine tumor on prostate biopsy, while adenocarcinoma, which was seen to be extensively present throughout the prostate, did not demonstrate increased FDG uptake (Figure 3A ). Areas of orbital as well as epidural and vertebral metastasis did not demonstrate increased FDG uptake (Figures 3B , 3C ), while areas of liver metastasis did demonstrate increased FDG uptake (Figure 3D ). He was treated with dexamethasone and urgent palliative radiotherapy to the skull base (3000 cGy in 10 fractions) and C5-T1 (2000 cGy in five fractions) using a 3D conformal approach. His acute complaint of diplopia was eventually alleviated. He followed up with medical oncology outpatient, he was initiated on chemotherapy along with immunotherapy, which included carboplatin intravenous (I.V.) infusion on day 1, etoposide I.V. infusion on days one, two, and three, and atezolizumab I.V. infusion on day one, followed by maintenance therapy with atezolizumab I.V. infusion once every 21 days. He was also started on darolutamide (anti-androgen medication) 300 mg twice a day daily along with Lupron (hormone-modulating drug) every three months. Later on, lurbinectedin 3.2 mg (an alkylating agent) every three weeks was added to the chemotherapy regimen. In view of preventing skeletal-related events, he was started on bone anti-resorptive therapy called Xgeva and Percocet for pain. The patient's serum creatinine returned to baseline, and PSA decreased to 6.87 ng/mL. Re-staging CT after a couple of months showed a treatment response. He also received intermittent palliative radiation therapy, which considerably improved his pain, and he has been ambulating more steadily.
Mahvish Renzu and Ishaan Jay Bhatt are the first co-authors and equally contributed to the work.
CC BY
no
2024-01-15 23:43:45
Cureus.; 15(12):e50566
oa_package/e1/71/PMC10787951.tar.gz
PMC10787954
38222088
Introduction Mycobacterium tuberculosis complex (MTBC) causes tuberculosis. When a person coughs, sneezes, talks, or sings, droplet nuclei are produced, which spread from person to person through the air ( 1 , 2 ). Coughing for more than 2 weeks, fever, weight loss, and sputum production can occur in conjunction with hemoptysis, loss of appetite, night sweats, and fatigue, which are expressive clinical signs in patients positive for pulmonary tuberculosis ( 3 ). In 2021, approximately 1.6 million people died and 10.6 million people contracted tuberculosis (TB) worldwide. Low-and middle-income countries accounted for 80% of cases and deaths, with 23% of new cases in the World Health Organization (WHO) Africa region. TB is the second biggest killer among infectious diseases and the 13th leading cause of death worldwide ( 1 ). Globally, there were approximately 11 million people imprisoned in 2018. An increment of approximately 24% was observed between 2000 and 2018 globally. The imprisoned population in Africa has increased by 29% in recent years, and the tuberculosis burden in this region is the highest compared to other WHO regions ( 1 ). The prison system is a potential area for transmitting communicable diseases such as tuberculosis due to overcrowding, poor ventilation, inadequate lighting, illicit drug use, difficulty accessing health services, lack of or precarious basic sanitation housing infrastructure, and malnutrition ( 1 , 3–5 ). In developing countries, TB is more common in prisons than in the general population, and prisons in SSA are riskier due to the high number of incarcerated people per cell block, ventilation systems, nutrition-related issues, and high prevalence of human immunodeficiency virus (HIV) ( 6–9 ). TB prevalence among prisoners from 24 SSA countries ranges from 0.4 to 16.3% ( 10 ). Prison staff are at risk of contracting tuberculosis due to their interaction with their inmates, which leads to the spread of the disease to their families and communities. This suggests that tuberculosis in prison is a concern for society as a whole, not just for prisoners ( 7 ). Therefore, compared to other regions in the world, SSA is one of the regions with a high burden of TB; therefore, this systematic review and meta-analysis helps to update the prevalence of tuberculosis among prisoners, inform policymakers, and improve approaches to prisoners.
Materials and methods Reporting The results were reported using the Preferred Reporting Items for Systematic Reviews and Meta-Analyzes (PRISMA) statement ( 11 ). The article screening was based on the PRISMA 2009 statement, and the selection process has been shown using a PRISMA-P flow diagram. This review is registered in PROSPERO with registration number CRD42023428933. Search methods and strategies To identify potentially relevant articles, we performed exhaustive searches of electronic databases with no date limits on Google Scholar, Web of Science, PubMed/MEDLINE, Science Direct, PubMed/MEDLINE, and EMBASE. All searches were limited to articles written in English and human studies. We conducted a manual search for additional relevant studies using references from retrieved articles and related systematic reviews to identify original articles that may have been overlooked. The following keywords were used to generate search strings or terms: prevalence, magnitude, Tuberculosis, Pulmonary Tuberculosis, Mycobacterium infections, Prisoners, and sub-Saharan Africa. Advanced search databases were built with the above-mentioned terms in mind, using “Medical Subject Headings (MeSH) [((((“Prevalence”) OR “Burden” OR “Magnitude”) AND “tuberculosis” AND “prisoners”)) AND sub-Saharan Africa]. Inclusion and exclusion criteria All studies on the prevalence of tuberculosis among prisoners in sub-Saharan Africa were included. Furthermore, this systematic review and meta-analysis included all cross-sectional studies on prisoners published in English and conducted in sub-Saharan Africa. Review papers, case series, case reports, abstracts, and qualitative studies were also barred from consideration. Outcome measurement One major finding that emerged from this systematic review and meta-analysis is the estimation of the pooled prevalence of tuberculosis among prisoners in sub-Saharan Africa. A tuberculosis-positive patient has a Mycobacterium tuberculosis complex found in a clinical specimen, whether by smear, culture, or WHO-recommended rapid diagnosis (such as Xpert MTB/RIF). Data extraction and quality assessment Endnote citation manager software version X9 for Windows was used to import retrieved studies from the databases, and manual removal was performed for duplicated articles. All articles were screened by three independent reviewers for predefined inclusion and exclusion criteria (abstract and title), followed by a full-text review. If disagreements regarding the inclusion of studies could not be resolved, a fourth investigator was invited to reach an agreement. Excel spreadsheet software was used to extract the data from the included studies. The spreadsheet included the first author’s name, publication year, study design, country, sample size, diagnostic methods, and number of cases ( Table 1 ). Statistical analysis The analysis was carried out using the statistical software STATA Version 14.1 (StataCorp, College Station, Texas, United States), and heterogeneity was checked across studies by computing the I2 statistical test. If the I 2 values were 0, 25, 50, and 75%, we assumed no, low, medium, and high heterogeneity across studies. A meta-analysis using a fixed-effects model with 95% confidence intervals (CI) was performed to analyze the pooled prevalence of tuberculosis among prisoners (I 2 ,16.3% p = 0.188). A visual inspection of the funnel plot was performed to check for evidence of publication bias, followed by Begg’s rank and Egger’s tests, with a value of p of less than 0.05 used as a cut-off point. Leave-one-out sensitivity analysis was also performed to assess the impact of a small study. The analysis was carried out step-by-step, excluding the study, to assess the effect of each study on the pooled prevalence of tuberculosis. A forest plot was used to estimate pooled prevalence.
Results Selection of studies A search of the biomedical electronic databases yielded 352 published and unpublished studies. Although 244 duplicate articles were identified and removed, 108 were included in the screening. After 64 studies were removed based on title and abstract screening, 44 studies remained. Finally, 40 studies that met the eligibility criteria were included in the final analysis to estimate the pooled prevalence of tuberculosis among sub-Saharan African prisoners. The full selection process is illustrated in Figure 1 . Included studies characteristics Among the 40 included studies, there were 19 studies were from Ethiopia; 3 from South Africa; 3 from Nigeria, Malawi, and Zambia; 2 from the Democratic Republic of Congo (DRC), Uganda, and Ghana; and 1 from Cameroon, Tanzania, and Cote d’Ivoire. The included study sample size ranged from 84 (2) to 31,843 (3), with 80,608 prisoners. Observational and interventional studies published between 1997 and 2020 were included. All 40 articles had a cross-sectional design. All included studies were facility-based ( Table 1 illustrates the included studies’ baseline characteristics). The pooled prevalence of tuberculosis among prisoners in sub-Saharan Africa The pooled prevalence of tuberculosis among sub-Saharan African prisoners was 4.02% (95% CI: 2.68–5.36). The forest plot shows that statistical heterogeneity was low (I 2 = 16.3%; p 0.188). As a result, we used a fixed effects model to estimate the pooled prevalence of tuberculosis ( Figure 2 ). Sub-group analysis A sub-group analysis based on diagnostic methods and country setting was performed to identify potential sources of heterogeneity. It shows the highest detection of tuberculosis was by Gene Xpert, which was 4.97% (95% CI: 2.22–7.73); sputum smear microscopy was 3.53% (95% CI: 1.92–5.13) and culture was 2.88% (95% CI: 2.40–8.16) ( Figure 3 ). Thus, we observed country variation in the prevalence of tuberculosis in this study. The prevalence of TB was found to range between 7.10 (95% CI: 4.58–9.62) in Ethiopia and 1.37 (95% CI:-1.17–3.91) in Malawi ( Figure 4 ). Meta-regression Meta-regression was used to identify factors associated with the pooled prevalence of tuberculosis among prisoners while keeping continuous variables in mind. For the meta-regression, publication year and sample size were considered. Meta-regression analysis revealed no statistically significant relationship between the pooled prevalence of tuberculosis among prisoners and publication year or sample size ( Table 2 ). Publication bias To assess possible publication bias, a visually inspected funnel plot was used, which was statistically supported by Egger’s and Begg’s rank regression tests. The symmetrical distribution of the included studies in a large inverted funnel demonstrated the absence of a publication bias. With p -values of ( p = 0.26) and ( p = 0.15), respectively, the Egger and Begg rank tests revealed no publication bias among the included articles to estimate the pooled prevalence of tuberculosis among prisoners in sub-Saharan Africa ( Figure 5 ). Sensitivity analysis By excluding each study one at a time, a leave-out-one sensitivity analysis was used to determine the effect of a single study on the pooled prevalence of tuberculosis among prisoners in sub-Saharan Africa. According to the findings, no single study had a significant impact on the pooled estimate of tuberculosis among prisoners in sub-Saharan Africa ( Table 3 ). Trends of TB prevalence The trend analysis indicated that despite efforts to eradicate TB, the disease burden among prisoners in sub-Saharan Africa continued to rise from 1997 to 2020 ( Figure 6 ).
Discussion There is evidence that the number of people developing tuberculosis is increasing in many low-and middle-income countries, and between 2019 and 2021, the number of deaths from tuberculosis also increased ( 1 ). In prisons, infectious diseases such as tuberculosis may spread more easily because segregation criteria are based on criminal characteristics rather than on public health concerns ( 52 ). As a result, the goal of this systematic review and meta-analysis was to report the most recent estimated pooled prevalence of tuberculosis among prisoners in sub-Saharan Africa. The prevalence of tuberculosis among household contacts was 3.29% (95% CI: 2.35–4.23) ( 52 ). A recent systematic review has documented a 3- to 1,000-fold increase in the prevalence of TB in prisons compared to the general population ( 22 ). In SSA, it is estimated to be 6–30 times higher than that in the general population ( 53 ). According to the results of the current systematic review and meta-analysis, the pooled prevalence of tuberculosis among prisoners in sub-Saharan Africa was 4.02% (95% CI: 2.68–5.36). This finding is consistent with findings from Tajikistan 4.5% ( 54 ), South Africa 2.7% ( 51 ), and Ethiopia (4.0%) ( 55 ). This could be attributed to the similarity in tuberculosis diagnostic methods used in the incarcerated population. However, the pooled prevalence of tuberculosis in this systematic review and meta-analysis was lower than that in an Ethiopian systematic review and meta-analysis (8.33%) (2). Furthermore, it was lower than that in studies conducted in Brazil 27.8% ( 56 ), Malaysia 7.7% ( 57 ), Nepal 10% ( 55 ), Iran 7.9% ( 58 ), South Africa (8.8%) ( 11 ), and Zambia (6.4%) ( 26 ). The lower prevalence found in this study could be attributed to differences in geographical location and the number of rooms with prisoners with poor ventilation. The pooled prevalence of tuberculosis among prisoners in the current meta-analysis was higher than that in studies conducted in Brazil (1.89%) ( 59 ), Thailand (2.1%) ( 60 ), and Peru (2.5%) ( 61 ). The higher prevalence of tuberculosis in our study might be due to overcrowding and the difference in the incarcerated years of inmates. Sub-group analysis of the pooled prevalence of tuberculosis among prisoners in sub-Saharan Africa showed no statistically significant difference ( p = 0.188). Using diagnostic methods, tuberculosis was detected by Gene Xpert (4.97%), sputum smear microscopy (3.53%), and culture (2.88%). Xpert MTB/RIF’s suitability and feasibility as an MTB diagnostic method are attributed to its suitability and feasibility as a quick, reliable, controllable, simple, and cost-effective test ( 62 ). Gene Xpert uses DNA PCR technology to detect MTB and rifampicin resistance mutations simultaneously ( 63 ). The sub-group analysis of this review also showed that the prevalence of tuberculosis among prisoners was higher in Ethiopia (7.10%) compared to other countries in sub-Saharan Africa. The variation in the prevalence of pulmonary TB within countries in prisons could be due to differences in diagnostic techniques, screening methods, overcrowding, and sociocultural and socioeconomic factors among the study participants. An ongoing intervention for Tuberculosis (TB) in sub-Saharan Africa is the implementation of active case-finding and treatment programs within prisons. This involves screening all inmates for TB, providing treatment for those who test positive, and implementing infection control measures to prevent the spread of the disease within the prison environment. Additionally, TB preventive therapy is provided to high-risk inmates, such as those with HIV or other underlying health conditions, which helps to reduce the overall burden of TB within the prison population ( 64 ). Strengths and limitations of the study The strength of this review is that it follows the recommended PRISMA guidelines. We also rigorously searched the literature in different databases and identified eligible studies. Moreover, in the present review, the heterogeneity among studies was low. While interpreting the results of this systematic review and meta-analysis, we considered the limitations of this review. We were forced to compare our findings with those of primary studies in some parts of the discussion because of a lack of adequate systematic reviews and meta-analyzes. The other limitation of this review is that we only considered articles written in the English language, which may result in the exclusion of other articles. Last but not least, we found studies conducted in 13 SSA countries, which may not represent prisoners throughout the whole region.
Conclusion The pooled prevalence of tuberculosis among prisoners in sub-Saharan Africa was prominently high based on this systematic review and meta-analysis. Therefore, to reach the end of the global TB epidemic, improvement in the prison setting is important. Screening on entry to the prison, periodical TB symptom screening, TB prevention training and information dissemination among the health staff in the prison and the inmates, and immediate treatment of diseased prisoners are important these measure to be put in place. Finally, this will help with the early identification and diagnosis of tuberculosis, which will reduce multidrug-resistant tuberculosis occurrence.
Edited by: Belaineh Girma Belaineh, International Training and Education Center for Health (I-TECH), United States Reviewed by: Raymond Salanga Dankoli, World Health Organisation, Ukraine; Andargachew Kumsa Erena, Ministry of Health, Ethiopia Background Tuberculosis (TB) is a key community health problem in numerous settings, predominantly in sub-Saharan Africa (SSA). TB is the second most lethal infectious disease worldwide. Around 1.6 million people died from TB in 2021. TB prevention and control strategies are difficult to implement in prison, especially in sub-Saharan Africa, owing to overcrowding and poor ventilation. Thus, this systematic review and meta-analysis aimed to synthesize the estimated pooled prevalence of tuberculosis among prisoners in sub-Saharan Africa. Materials and methods Electronic biomedical databases such as Google Scholar, Web of Science, PubMed/Medline, EMBASE, and Science Direct were used to systematically explore candidate studies published until December 2022. Data extraction was performed using a Microsoft Excel spreadsheet. The estimated pooled prevalence of tuberculosis was determined using a fixed-effects model. Cochrane Q-test and I 2 statistics were used to check heterogeneity statistically across different studies. Begg’s rank and Egger’s tests were performed to assess evidence of possible publication bias. Results A total of 40 articles involving 59,300 prisoners were included in this systematic review and meta-analysis. The pooled prevalence of tuberculosis was 4.02% (95% CI: 2.68–5.36). We found the highest prevalence using Gene X pert as a diagnostic method, which was 4.97 (95% CI: 2.22–7.73). There is no evidence of publication bias. Conclusion The outcome of this review revealed a high prevalence of tuberculosis among prisoners in sub-Saharan Africa. To reach the “End Tuberculosis strategy” by 2030, early identification of cases through screening on entry and periodical active case finding is important. Moreover, prevention and prompt treatment after diagnosis must be implemented to limit transmission to the general population. Systematic review registration https://www.crd.york.ac.uk/prospero/#searchadvanced , identifier (CRD42023428933).
Data availability statement The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. Author contributions The study was conceptualized and developed by YS and TM, who also conducted data analysis and interpretation and wrote the first draft. GA and YS built the search strategy, extracted the data, and assessed the quality of the studies included. The writing was reviewed and edited by YS and TM. All authors contributed to the article and approved the submitted version.
We would like to thank the primary study authors who contributed to this systematic review and meta-analysis. Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
CC BY
no
2024-01-15 23:43:45
Front Public Health. 2023 Dec 28; 11:1235180
oa_package/12/39/PMC10787954.tar.gz
PMC10787955
38218830
Background Cervical cancer is the fourth most common female cancer globally after breast, colorectal, and lung cancers, and is the second most common female cancer in Low- and Middle-Income Countries (LMICs) after breast cancer [ 1 ]. In 2020, it was estimated that approximately 604,127 women were diagnosed with cervical cancer worldwide, with an estimated 341,831 deaths attributable to the disease [ 2 ]. The majority (90%) of cervical cancer-related deaths in 2020 occurred in LMICs among women aged 15–49 years, largely due to a lack of access to screening programs, lack of availability of the Human Papillomavirus (HPV) vaccine, poor health infrastructure, and delays in diagnosis [ 2 , 3 ]. Several risk factors are attributed to the development of cervical cancer, including HPV infection, low socioeconomic status, cigarette smoking, long-term use of oral contraceptives, marriage before the age of 18 years, early sexual intercourse, multiple sexual partners, and multiparity [ 4 , 5 ]. Nonetheless, HPV infection (primarily genotypes 16 and 18) is considered to be the central etiological risk factor for cervical cancer [ 6 ]. According to data sourced from 900 invasive cervical cancer cases originating from 22 countries, 93% of biopsies have HPV DNA [ 4 ]. Thus, cervical cancer can be primarily prevented by HPV vaccination and routine Pap smear screening [ 4 , 7 ]. While many countries have introduced national HPV vaccination programs over the last several decades, such programs remain rare in LMICs [ 7 ]. Furthermore, the implementation of Pap smear screening in LMICs is challenging due to poor attendance and lack of awareness of the importance of this type of screening in the prevention and early detection of cervical cancer [ 8 ]. Previous studies conducted in LMICs have indicated that women may encounter several barriers to cervical cancer screening, including poor knowledge regarding the importance of the HPV vaccine or Pap smear test, negative attitudes towards cervical cancer-related risk factors (particularly HPV infection), fear of the results, and concern regarding the gender of the doctor performing the screening test [ 9 – 11 ]. In Oman, the HPV vaccine is not yet included in the Expanded National Immunization Program and is therefore not systematically provided by government-funded health institutes under the national Ministry of Health [ 12 , 13 ]. In Oman, the incidence of cancer has increased over the past 20 years, a finding which may be attributed to population aging, rapid socioeconomic changes, and the increased prevalence of unhealthy lifestyle practices (e.g., tobacco use, physical inactivity, and unhealthy dietary habits), as well as advances in diagnostic and treatment modalities [ 14 ]. Among the population of 1.17 million women aged 15 years and older in Oman, there is a notable risk of developing cervical cancer; current estimates suggest that approximately 88 women are diagnosed with cervical cancer each year, resulting in 50 deaths [ 15 ]. Moreover, cervical cancer ranks as the fourth most common cancer among women of all ages in Oman and the third most common among women aged 15–44 years [ 14 , 15 ]. Unfortunately, there are currently no available data on the prevalence of HPV in the general population of Oman. However, in Western Asia, the region to which Oman belongs, it is estimated that about 2.5% of women in the general population are affected by cervical HPV infection at any given time [ 16 ]. Furthermore, 72.4% of invasive cervical cancers in this region are attributed to HPV types 16 or 18 [ 16 ]. Additionally, a recent study from 2020 indicated that the prevalence of HPV infection remains high in Oman at 17.8% [ 17 ]. In Oman, the absence of a well-structured national screening program for cervical cancer and limited availability of Pap smear testing at the primary care level are significant concerns. This lack of accessibility and awareness regarding the availability of the Pap smear test are believed to be the main reasons for the high incidence of cervical cancer in the Middle East region [ 9 , 18 ]. A previous study indicated that while most Omani women attending a tertiary teaching institute had heard of cervical cancer, they lacked specific knowledge regarding cervical cancer signs and symptoms, risk factors, and Pap smear testing [ 19 ]. Therefore, the present study aimed to assess knowledge, attitudes, and practices regarding cervical cancer and Pap smear screening among Omani women attending an outpatient clinic at a tertiary teaching hospital in Muscat, Oman. The study also aimed to establish correlations with various sociodemographic factors. These findings may be useful in informing future health promotion activities that aim to improve cervical cancer awareness in the general public and promote utilization of screening services in Oman.
Methods Study design and setting An observational, cross-sectional study was carried out from October 2019 to February 2020 at the outpatient clinic of the Department of Obstetrics and Gynecology at the Sultan Qaboos University Hospital (SQUH), Muscat, Oman. As a tertiary hospital, SQUH receives patients from all over the country. Study subjects and recruitment strategy A total of 380 Omani women aged 18–50 years old and attending the clinic for various reasons during the study period were recruited via a systematic random sampling strategy in which every second women registered with the clinic was selected for participation. Exclusion and inclusion criteria The inclusion criteria for the participants included all women of Omani nationality between 15 and 50 years of age attending the outpatient clinic for various reasons. However, those with learning difficulties, those who did not speak Arabic or English, those with emergency conditions or who were very sick, and those with no time to participate in the survey were excluded. Sample size The necessary sample size was calculated to be 374 women, based on an anticipated level of knowledge regarding cervical cancer and its screening (50%), with a 5% margin of error, 95% confidence level, 5% alpha error, and a design effect of 2. Survey instrument A validated, pre-tested, Arabic-language questionnaire was used for data collection purposes. The questionnaire had been previously used in similar studies performed in Oman [ 19 – 22 ]. This four-part questionnaire was self-administered and took approximately 15–20 minutes to complete. The first section covered the participants’ sociodemographic characteristics, including their age, education level, employment status, marital status, number of marriages, number of previous pregnancies, their husbands’ education level, and monthly family income. The second section assessed the presence of known risk factors for cervical cancer, such as smoking status, personal history of cervical cancer, and family history of cervical cancer. The third part of the questionnaire assessed the participants’ knowledge regarding cervical cancer, related risk factors, and appropriate screening. This section covered whether the participants had ever heard of cervical cancer and whether they believed that cervical cancer can be prevented, has a latent and asymptomatic period, can be detected in its early stages, is curable when detected early, is a genetic disease, is more likely if a family member has it, can be prevented by maintaining healthy sexual hygiene, if postmenopausal women and HPV-positive women are at risk of getting cervical cancer, whether cytological examination is the main screening method in the early stages of the disease, and whether the disease is caused by a specific HPV genotype. In addition, this section explored knowledge of cervical cancer-related risk factors, including HPV infection, early sexual activity, multiple sexual partners, multiparity, and smoking status. All questions in the second and third sections of the questionnaire were designed to elicit yes/no responses. The last part of the questionnaire evaluated the participants’ knowledge, attitudes, and practices related to cervical cancer screening and Pap smear testing. Participants were asked if they had ever heard of Pap smear screening, if they had previously undergone Pap smear testing, and whether they would be willing to undergo such testing. In turn, those who had never undertaken Pap smear testing were asked about their reasons for not doing so and their willingness to undertake such screening in the future. Awareness of the actual screening procedure was assessed, with participants being asked about the aim, usefulness, and importance of screening, the appropriate time for testing, the site of the test, whether one needed to be symptomatic to get the test, and when to stop the test. Most of the questions in this section also required yes/no responses; however, there were exceptions for certain questions, such as reasons for not taking action, screening aims, appropriate time for testing, site of testing, and when to stop the test. For these questions, respondents were provided with multiple-choice options to choose from. Scoring All knowledge-related items in the questionnaire were compiled, and a scoring system was created. Each correct response received a score of 1, resulting in a total score range of 0–30. Patients were then divided into two categories based on their total scores: not knowledgeable (scores of ≤15) and knowledgeable (scores of 16–30). Ethics Ethical approval for this study was obtained from the Medical Research and Ethics Committee of the College of Medicine and Health Sciences, Sultan Qaboos University (#SQU-EC/214/19, #MREC2013). All participants were briefed regarding the objectives of the study and were informed that their participation was voluntary in nature and that they had the right to withdraw at any time. Written informed consent was received from all of the women prior to their participation in the study. The participants’ anonymity and confidentiality were ensured at all times, with each participant assigned a unique identification number for the purposes of data analysis. Statistical analysis The data analysis was carried out using the Statistical Package for the Social Sciences (SPSS), version 23 (IBM Corp., Armonk, NY). Descriptive statistics were used to report the sample’s characteristics. For categorical variables, frequencies and percentages were reported, whereas means and standard deviations were used to present continuous variables. Crude and adjusted Odds Ratios (ORs) and corresponding 95% Confidence Intervals (CIs) were used to test correlations, and a p value of ≤0.05 was considered statistically significant.
Results Sociodemographic characteristics The mean age was 32.1 ± 7.6 years (range: 18–50 years), with a similar proportion of participants aged 18–30 years ( n = 182; 48%) and 31–50 years ( n = 198; 52%). More than half of the participants had an undergraduate-level education or higher ( n = 247; 65%). Most were unemployed ( n = 222, 58%) and had been married once ( n = 290; 76%), with few having been married more than once ( n = 7; 2%). Of the participants who had been married at least once, the majority of their husbands had secondary school diplomas ( n = 149; 51%) or undergraduate degrees ( n = 141; 49%). In terms of family income, 209 (55%) reported earning < 1000 Omani Riyals (OMR) per month (equivalent to <$2598 USD). The vast majority were non-smokers ( n = 377; 99%). Only two participants (1%) had a personal history of cervical cancer, while only one (< 1%) was aware of a family history of cervical cancer (Table 1 ). Knowledge of cervical cancer and pap smear testing In terms of cervical cancer-related knowledge, most participants ( n = 325; 86%) had previously heard of cervical cancer. Over half ( n = 224; 59%) believed that cervical cancer can be prevented and 251 (66%) thought it a genetic disease. The most frequently identified cervical cancer-related risk factors included multiple sexual partners ( n = 162; 43%), smoking ( n = 145; 38%), and HPV infection ( n = 91; 24%). Early sexual activity ( n = 41; 11%) and having three or more children ( n = 30; 8%) were the least frequently reported risk factors. With regards to knowledge, attitudes, and practices related to Pap smear testing, 209 women (55%) had previously heard of this screening method, although the majority ( n = 302; 80%) admitted that they did not undertake such testing on a regular basis. Only 78 women (21%) had themselves previously undergone Pap smear testing, although 283 (75%) reported a willingness to undergo such testing in the future. Various reasons were reported for not having previously undertaken Pap smear testing, including concern regarding being examined by a male doctor ( n = 287; 76%), feeling embarrassed ( n = 183; 48%), fear of the test results ( n = 172; 45%), fear of pain ( n = 139; 37%), being healthy ( n = 136; 36%), being busy ( n = 115; 30%), fear of the test itself ( n = 98; 26%), the unavailability of nearby health services ( n = 89; 23%), being discouraged by their partner ( n = 88; 23%), privacy concerns ( n = 87; 23%), fear of bleeding ( n = 69; 18%), being unaware of where to get the test ( n = 50; 13%), having no time ( n = 49; 13%), the expense of being tested ( n = 45; 12%), religious reasons ( n = 44; 12%), and being too old to be tested ( n = 34; 9%). Associations with sociodemographic characteristics Overall, the vast majority of participants ( n = 281; 74%) were not knowledgeable with regards to cervical cancer and Pap smear testing, receiving total scores of < 16 for all knowledge-related items in the questionnaire. Only 99 (26%) women were considered knowledgeable on these topics, with total scores of ≥16. Knowledge scores were significantly associated with several sociodemographic factors, including marital status (OR 2.84, 95% CI 1.42–5.67) and a previous awareness of cervical cancer (OR 4.82, 95% CI 1.68–13.83), with married women, those who had had one or two pregnancies, and those who had previously heard of cervical cancer being more knowledgeable compared to their respective counterparts. No significant associations were observed between knowledge scores and other sociodemographic characteristics of the participants, including age, educational and employment status, monthly income level, and number of previous pregnancies ( p > 0.05) (Table 2 ). Furthermore, significant associations were noted between Pap smear practices and several factors, including age (OR 2.13, 95% CI 1.18–3.84), marital status (OR 21.76,, 95% CI 2.91–162.63), and a previous awareness of cervical cancer (OR 2.44, 95% CI 0.90–6.54). Specifically, older women, those who were married, and those who had previously heard of cervical cancer more frequently reported having previously undertaken Pap smear testing compared to those who were younger, those unmarried, and those who had not heard of cervical cancer before (Table 3 ). Moreover, married women (OR 4.56, 95% CI 2.52–8.25) and those who had heard of cervical cancer before (OR 2.42, 95% CI 1.29–4.52) more frequently reported a willingness to undertake such testing in the future (Table 4 ).
Discussion Although cervical cancer is the one of the most common cancers among Omani women, there is as yet no established cervical cancer screening program in the country [ 12 – 15 ]. Moreover, Pap smear testing is often only provided to married women and is unavailable at the primary healthcare level, performed solely in secondary and tertiary care facilities for diagnostic purposes. Combined with the lack of a national HPV vaccination program, such concerns are critical because the onus for cervical cancer screening and Pap smear testing therefore rests on the individual patient. As such, public knowledge and awareness of the importance of such screening and cervical cancer risk factors is crucial for early identification and prevention purposes. Thus, the present study was conducted in order to assess knowledge, attitudes, and practices regarding cervical cancer and Pap smear screening and to establish correlations with various sociodemographic factors among a cohort of Omani women attending an outpatient clinic at a tertiary teaching hospital in Muscat, Oman. The findings indicated that cervical cancer knowledge was limited, with 86% having heard of it, but only 59% believing in its preventability. Regarding Pap smear testing, 55% were aware, but 80% did not undergo such testing regularly. Sociodemographic factors such as marital status and previous awareness of cervical cancer were associated with higher knowledge scores and increased likelihood of participating in Pap smear practices. In our study, although the vast majority of participants (86%) had previously heard of cervical cancer, only 55% had heard of Pap smear testing. A prior survey conducted in Oman similarly reported that 80% of participants had heard of cervical cancer [ 19 ]. This level of public awareness is high, possibly because such studies were conducted among patients, students, and employees at a tertiary care institute that routinely provides this service. According to Shrestha et al., only 18% of women who visited a tertiary care facility in Nepal knew of Pap smear testing, despite 66% having previously heard of cervical cancer [ 23 ]. In contrast, much higher rates of awareness (85 and 76%, respectively) were recorded among women accessing primary healthcare facilities in Qatar [ 24 ]. In Kuwait, 77% of married women visiting polyclinics had heard of Pap smear testing; similarly, 74% of Vietnamese American women surveyed in the USA had heard of the Pap smear test, although 90% were aware of cervical cancer [ 25 , 26 ]. Differences in knowledge across different countries or regions may be due to cultural or healthcare system factors. For example, although Qatar is also a Middle Eastern country with a similar sociocultural milieu to Oman, the researchers noted that Pap smear testing is available at most primary health centers due to the country’s well-woman clinic program [ 24 ]. In the current study, various sociodemographic factors were found to have a significant impact on knowledge of cervical cancer and Pap smear testing, including marital status and a previous awareness of cervical cancer. Similar findings have been reported in studies conducted in Qatar, Kuwait, and the USA [ 24 – 26 ]. In particular, married women were significantly more knowledgeable compared to their counterparts, as well as those who had previously heard of cervical cancer. Marital status likely influences cervical cancer-related knowledge in Oman due to family planning practices among married women, which would involve regular consultations with an obstetrician or gynecologist. These healthcare professionals often provide guidance on fertility, prenatal care, and childbirth preparation, possibly encompassing discussions about women’s general reproductive health, including cervical cancer. Indeed, this would tie in with the finding that knowledge scores were significantly associated with number of previous pregnancies. Studies from Kazakhstan have similarly shown that number of births is associated with increased awareness of HPV and cervical cancer screening practices [ 27 , 28 ]. Indeed, the impact of marital status on cervical cancer-related knowledge may be more pronounced in conservative, religious countries like Oman where premarital sexual relations are strictly prohibited. Alternatively, as previously described, Pap smear testing is often offered only to married women in Oman; this might also explain why this group were more likely to demonstrate knowledge of cervical cancer and Pap smear testing. However, previous studies from Oman have also reported associations with other factors, such as education level, employment status, and income [ 19 – 22 ]. These variations in findings could be due to differences in the research population and study design, as well as in the availability and accessibility of cervical cancer screening services in different healthcare settings and regions. In terms of knowledge regarding cervical cancer, our study reported results in line with previous studies performed in Qatar and Kuwait [ 21 , 22 ]. However, given that the majority of the participants (74%) had inadequate knowledge regarding cervical cancer and Pap smear testing, regardless of education level, this may indicate that the Omani public’s overall exposure to cervical cancer-related education is lacking. Unfortunately, only 24% of participants in the current study were aware of HPV infection as a risk factor for cervical cancer. This is concerning as previous research has identified a link between awareness of HPV as a major cause of cervical cancer and participation in Pap smear practices [ 27 , 28 ]. Only 21% of the women in our study revealed that they had previously undergone Pap smear testing themselves. Despite this, the majority (75%) reported that they would be willing to undergo such testing in the future, One of the limitations of this study includes the length of the questionnaire and the fact that the questionnaires were completed by participants while in the clinic’s waiting area, which can often be noisy and crowded. Moreover, some of the questions sought information on the participants’ past experiences and were related to potentially sensitive topics which could have resulted in recall or response bias. Secondly, the study was conducted in a single institution located in the capital city of Muscat; accordingly, the findings cannot be generalized to other institutions or regions of Oman, particularly more remote regions. Nonetheless, a strength of the study was that the participants involved in the study originated from all over the country as SQUH is one of the few tertiary-level institutions in Oman. Overall, the present study identified inadequate awareness on matters related to cervical cancer, its risk factors, and screening methods among a sample of Omani women attending a tertiary outpatient clinic. Accordingly, more efforts need to be made to increase general awareness of cervical cancer and its screening methods and risk factors through various channels, including school-level interventions and health promotion activities. In particular, healthcare professionals should be encouraged to deliver cervical cancer-related information to female patients of appropriate age groups as they would be able to immediately and accurately respond to any inquiries and correct any misconceptions. Commonly reported barriers to taking part in cervical cancer screening in Oman include fear of being examined by a male doctor, being embarrassed, fear of the results, fear of pain, being healthy, being busy, and fear of the procedure itself [ 19 – 22 ]. It is likely that these women would demonstrate more positive attitudes concerning Pap smear testing if they were to be reassured and given more detailed information concerning the test procedure itself. Much of this responsibility is likely to fall on the attending physician who should provide detailed information to female patients regarding cervical cancer screening, including its indications, ideal frequency, and possible complications. Moreover, healthcare professionals—whether at the primary, secondary, or tertiary care level—should seek to maintain appropriate channels of communication throughout the entire consultation and follow-up process in order to build up a trusting relationship with their patients, particularly given that such topics may be considered sensitive or embarrassing, especially in very conservative, religious societies.
Conclusions The burden imposed by cervical cancer on health systems remains considerable, especially in LMICs such as Oman which lack adequate HPV vaccination and cervical cancer screening coverage. As such, public awareness of these issues is paramount to ensure early identification and prevention. Overall, the present study identified suboptimal awareness on matters related to cervical cancer, its risk factors, and screening methods among a cohort of Omani women attending a tertiary outpatient clinic. Despite this, most participants were generally accepting of the idea of undergoing Pap smear testing in future. Thus, it is strongly recommended that the Omani Ministry of Health consider implementing a national screening program in order to lower the incidence of cervical cancer. Moreover, awareness and education campaigns are also urgently needed to effectively inform the public on the importance of and indications for cervical cancer screening. Such campaigns might be implemented utilizing a multimedia approach; in addition, collaboration with other government sectors, such the Ministry of Education, would be helpful in order to reach adolescents and younger women and to incorporate knowledge about cervical cancer warning signs and risk factors into national school curricula. To support the execution of such initiatives, further research is recommended, particularly studies exploring ways to reduce barriers to Pap smear screening and the impact of religious and cultural factors on cervical cancer awareness and screening practices. In addition, comprehensive research is needed to determine the prevalence of HPV infection in the general population of Oman, with a focus on identifying specific genotypes. This could inform decisions related to the inclusion of the HPV vaccine in national immunization strategies.
Objectives To assess knowledge, attitudes, and practices regarding cervical cancer and Pap smear screening among Omani women attending a tertiary clinic in Muscat, Oman, and to establish correlations with selected sociodemographic factors. Methods An observational, cross-sectional study was carried out among Omani women aged 18–50 years old attending the outpatient clinic of the Department of Obstetrics and Gynecology, Sultan Qaboos University Hospital, from October 2019 to February 2020. A validated Arabic-language questionnaire was utilized to collect data regarding the participants’ sociodemographic characteristics, their knowledge of cervical cancer and related risk factors, and their knowledge, attitudes, and practices related to cervical cancer screening and Pap smear testing. Results Of the 380 respondents, 86 and 55% had previously heard of cervical cancer and Pap smear testing, respectively; however, only 26% were knowledgeable concerning these topics. Knowledge scores were significantly associated with various sociodemographic factors, including marital status and a previous awareness of cervical cancer (odds ratio: > 1, p < 0.05). Only 21% had themselves previously undergone Pap smear testing; however, 75% reported being willing to undergo such screening in future. Conclusions Knowledge regarding cervical cancer-related risk factors and Pap smear screening was poor among a cohort of Omani women attending a tertiary clinic in Muscat, Oman. This may play a role in the increased frequency of cervical cancer cases observed in Oman over recent years. As such, a well-structured public education program is recommended to raise awareness of this issue. Keywords
Abbreviations Human Papillomavirus Low- and Middle-Income Countries Omani Riyals Papanicolaou Sultan Qaboos University Hospital Statistical Package for the Social Sciences Odds Ratio Confidence Interval United States of America Acknowledgements The authors thank all of the women who participated in this study. They are also grateful to the staff at the Department of Obstetrics and Gynecology at SQUH for their cooperation with the data collection procedures. Authors’ contributions T.M., K.M, and M.A conceived the presented research idea and went through literature review. T.M, and K.M under the supervision of M.A, M.K and R.K. designed the research methodology and the questionnaire format. T.M, and K.M were involved in the data collection and date entry. T.M, K.M, H.S and R.K analyzed and interpreted the results. T.M, R.K and H.S were a major contributor in writing the manuscript in consultation with M.A. and M.K. R.K, M.K and M.A were the research supervisors who guided T.M and K.M throughout the project. All authors read and approved the final manuscript. Funding No funding was received for this work. Availability of data and materials The raw datasets used and/or analyzed during this study are available from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate The study received ethical approval from the Medical Research and Ethics Committee of the College of Medicine and Health Sciences, Sultan Qaboos University, Oman (#SQU-EC/214/19, #MREC2013). Written informed consent was obtained from all participants prior to their inclusion in the study. The participants’ anonymity and confidentiality were ensured at all times, with each participant assigned a unique identification number for the purposes of data analysis. Consent for publication Not applicable as no details regarding individual patients have been reported in the manuscript. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:43:46
BMC Womens Health. 2024 Jan 13; 24:40
oa_package/c4/3a/PMC10787955.tar.gz
PMC10787956
0
Introduction Bruxism refers to the phenomenon of involuntary contraction of the masticatory muscles under non-physiological conditions, resulting in intermittent masticatory movements [ 1 ]. It is divided into awake bruxism and sleep bruxism. Awake bruxism occurs in a conscious state and is usually associated with emotions such as mental tension, anxiety, stress, anger, or depression. However, sleep bruxism occurs at night and is usually caused by sleep apnea or related to sleep disorders. Research have shown that bruxism is common in all age groups and has become an important factor for dental health. It can lead to rapid tooth wear, resulting in pulpitis, narrowing of the occlusal surface, temporomandibular joint disease, nervous system disease, and muscle pain [ 2 ]. The prevalence of sleep bruxism in adults ranges from 8 to 16%, whereas in children, it can be as high as 40% [ 3 , 4 ]. Unfortunately, the etiology of bruxism is complex and the pathogenesis is unclear [ 5 ]. Therefore, it is of great significance to detect sleep bruxism as early as possible in order to select the most appropriate treatment method. Polysomnography (PSG) is considered the gold standard for diagnosing bruxism, but it requires many sensors that increases the complexity causing discomfort to patients. However, electroencephalogram (EEG) provides information related to brain activities that helps to understand relationship between bruxism and brain function. In literature, various physiological signals such as electrocardiogram (ECG), Electromyography (EMG), and EEG have been used in the detection of sleep bruxism. Based on ECG, heart rate variability was used to assess sympathetic cardiac activity in patients with bruxism. Research indicates that patients with bruxism exhibit higher sympathetic cardiac activity compared to healthy controls [ 6 , 7 ]. Facial EMG can be used to record the potential activity of the patient’s masseter and temporal muscles at night, and the potential value and activity of the EMG can be used to determine whether bruxism occurs [ 8 , 9 ]. Research has shown that combining ECG and EMG to achieve classification of nocturnal bruxism has also achieved good results [ 10 ]. In addition, video capture can also be used to detect the occurrence of bruxism [ 11 ], or magnetic resonance imaging for bruxism examination [ 12 ]. In addition, EEG is a non-invasive signal, which can be easily obtained from the electrode. It can record neural activity in sleep through different frequency bands. This technology has been widely used as a standard to quantify potential neural activity in sleep research [ 13 , 14 ]. Researchers have found that most patients with bruxism have significant signs of increased electrical activity in the cerebral cortex during tooth grinding [ 1 ]. Sleep bruxism detection has also been conducted by analyzing EEG signals [ 15 – 18 ]. The research of Dakun Lai et al. [ 10 ]. shows that the power Spectral density (PSD) of EEG channels in patients with bruxism is significantly higher than that in normal people during rapid eye movement (REM) and awake sleep. Their research also shows that on the basis of the power spectrum characteristics of EEG signals, the fusion of EMG1 and ECG channel signals can achieve better results in the recognition of sleep bruxism patients. Bin Heyat et al. also demonstrate the effectiveness of EEG in detecting sleep bruxism [ 16 ]. In addition, the theta activity is also believed to be associated with the occurrence of bruxism [ 18 ]. Although the effectiveness of EEG signals has been proven in previous studies, there are problems obtaining EEG signals with too many electrodes causing difficulty in installation and makes patients discomfort. Therefore, identifying the neural correlates of bruxism via single-channel EEG may be of high clinical significance in managing the adverse consequences of sleep bruxism. The goal of this study is to fully extract the PSD features of EEG in patients with sleep bruxism, thereby attempting to obtain the most effective single channel EEG for identifying bruxism. We propose a new data processing algorithm based on the fusion of multiple EEG frequency band signal features. The algorithm first extracts the PSD values of different EEG frequency bands in REM sleep stage, and then extracts 28 features of these frequency bands in time domain, frequency domain and nonlinear. On the basis of fully extracting the features of EEG signals, the machine learning algorithm is used to identify patients with bruxism. After integrating these features, the classifier can obtain sufficient reference information, providing a reliable basis for the training and accurate classification using machine learning algorithms. The remaining sections of the paper are structured as follows: Section II covers data preparation, data processing, feature extraction, and statistical analysis. Subsequently, machine learning algorithm is applied to classify patients with bruxism. Section III presents the results from data analysis. In Section IV, the research results are thoroughly discussed and compared with existing methods and findings. Section V presents conclusions and prospects for future work.
Materials & methods To accurately describe the classification process of bruxism, we designed a data processing flow as shown in Fig. 1 . Data processing flow description: Data processing and feature extraction EEG Signal Reading: Read the EEG signals from the REM sleep stages of the subjects. Wavelet Decomposition: Utilize the DB5 wavelet to decompose the EEG signals into delta, theta, alpha, and beta frequency bands. Relative PSD Calculation: Calculate the relative PSD for each frequency band, capturing time-domain, frequency-domain, and non-linear features. Statistical analysis Normality Test: Perform a normality test on the extracted features to ensure their distribution is suitable for statistical analysis. Statistical Test Selection: Determine the appropriate statistical test (e.g., t-test, Mann-Whitney Rank-Sum test) based on the distribution. P-value Calculation and Significance Analysis: Carry out the selected statistical test, calculate the p-value, and determine the significance of the features. Classification and result statistics Classifier Selection: Choose an appropriate classification algorithm based on the features. Classification and Result Collection: Utilize the selected classifier to classify the data and collect the classification results. Optimization and Validation: Iterate over multiple rounds of experimentation and comparison to identify the optimal EEG channel and classify the data effectively. Through these three key processing steps, we aim to extract significant features from the EEG signals, perform statistical analysis to identify significant features, and utilize these features for effective classification of sleep bruxism. The end result would be the identification of the optimal EEG channel for classification and the resulting classification outcomes. Experimental protocol The data was acquired from cyclic alternating pattern (CAP) sleep database of PhysioNet [ 19 , 20 ]. It provides representative PSG records of 108 participants with various pathophysiological backgrounds, including 16 controls, 2 bruxism patients, 9 Insomnia, 5 Narcolepsy, 40 Nocturnal frontal lobe epilepsy, 10 Periodic leg movements, 22 REM behavior orders, and 4 Sleep disordered breathing. Each record includes three or more EEG signals, as well as EMG, airflow, respiratory effort, SaO 2 , and ECG signals, and each record has been carefully reviewed by expert neurologists for sleep stage and CAP annotations. The healthy controls who participated in the study exhibited no neurological disorders and were not taking any medications that could affect the central nervous system. All the bipolar EEG channels were sampled at 512 Hz and placed according to 10–20 international electrode placement system [ 21 ]. Since the EEG channels collected by each subject in the database are different, we can only select data from 4 health controls (participants: n3, n5, n10, and n11) and 2 bruxism patients (participants: brux1 and brux2) with the same EEG channels (F4C4, C4P4, Fp1F3, F3C3, and C4A1) in this study. According to the annotations made by neurologists, the length of each REM sleep segment was determined to be 30 s. In this study, the total number of REM sleep events is 1295 segments (duration is 38,850 s), including 276 segments of data from bruxism patients. Demographic information for the participants, along with REM sleep events, is presented in Table 1 . Data processing First, the “edfread” function is used to read each polysomnographic record, which helps to obtain the identification and specific data of each channel, providing a foundation for subsequent analysis. Then, the “ScoringReader” function is executed to obtain the identification code and duration of each sleep stage. This information is crucial for accurately determining an individual’s sleep state. Finally, EEG data from the REM sleep stage are extracted, which prepare for further in-depth research. After preparing the REM segment of the EEG signal, the next step is to calculate its PSD. First, the REM segment of the EEG signal is decomposed into different frequency bands using the DB5 wavelet, including (δ, 1–4 Hz), theta (θ, 4–8 Hz), alpha (α, 8–13 Hz), and beta (β, 13–30 Hz). Next, the PSD was calculated using Welch estimate with Hamming window size of 128 samples with 50% overlap and 256 discrete Fourier transform points for each frequency bands. Finally, the absolute PSD (APSD) were calculated. The relative PSD on each frequency band was obtained from the ratio of absolute power of each four frequency bands to the total power within the spectrum of 1–30 Hz [ 13 ]. The delta band is chosen beyond 1 Hz to minimize low frequency head movement and ocular artifacts below 1 Hz. The PSD analysis within the frequency band 1–30 Hz is chosen since EEG information relating to sleep relies within this spectrum. In order to conduct subsequent statistical analysis and obtain significant feature indicators for characterizing bruxism, we performed subsequent processing. Various features were extracted by time domain, frequency domain, and sample entropy (SampEn) for each REM sleep epochs from healthy controls and bruxism patients. For time domain features, Mean value, Standard Deviation (SD), and Root Mean Square (RMS) were calculated for each of the frequency bands [ 22 ]. For frequency domain features, the relative power spectral density (RPSD) including RPSD (δ), RPSD (θ), RPSD (α), RPSD (β), and ratios including (θ + α)/β, α/δ, α/θ, and α/β were computed as the additional features [ 23 ]. For nonlinear features, the Shannon entropy (SampEn) was calculated with default parameters (the maximum template length is 5 and the matching threshold is 0.2). Finally, a total of 28 features including 12 time domain features, 12 frequency domain features, and 4 non-linear analysis features were extracted. Statistical analysis The statistical analysis is performed to compare differences in EEG features from five EEG bipolar electrodes (channels) among healthy controls and bruxism patients. A Shapiro-Wilk test is used to examine the normality of data and a suitable parametric or non-parametric test is adopted based on the data distribution [ 24 ]. Since the data samples are independent, either two sample t-test as a parametric test for normalized data or Mann-Whitney Rank-Sum test as a non-parametric test for non-normalized data can be performed to obtain statistical comparison among two groups: healthy controls and bruxism patients. Classification and cross-validation The systematic empirical evaluation of various machine learning algorithms shows that Decision tree (Fine Tree) algorithm is the optimal algorithm for the classification of bruxism in CAP database. Therefore, unless stated otherwise, all the data discussed in this paper is the output of the Fine Tree classifier. Decision tree is a basic method of classification and regression. It achieves classification by dividing input features into different subsets layer by layer. The core idea of decision tree classifier is to determine the decision rules for classification by systematically dividing features, thus facilitating data classification. Due to its resemblance to the branches of a tree, this decision graph is called a decision tree. During the decision tree classification process, instances are segmented based on features and assigned to distinct categories. The main advantages of this method are model readability, easy to understand, fast classification, fast modeling, and prediction [ 25 ].
Result The normality test in the data have exhibited mixed behavior and the samples are independent between the two groups. Therefore, Mann-Whitney Rank-Sum test is used to test the significant difference. The significance level is set at the alpha criterion α = 0.05. The relative PSD in five EEG channels during REM sleep stage among healthy controls and bruxism patients are compared and their trends (mean ± SE) are illustrated in Fig. 2 . The statistical analysis comparing healthy controls and bruxism patients revealed a significantly lower relative delta power ( p < 0.05) in three channels (F4C4, Fp1F3, and F3C3), as well as a significant decrease in relative theta power across all five channels. In contrast, the relative alpha and beta power exhibited a significant increase ( p < 0.05) in all five channels for bruxism patients. The overall results depicted decrease in low frequency band power (delta and theta bands) and increase in high frequency band power (alpha and beta bands) for bruxism patients. In low frequency bands, the profound lower relative power was observed in theta band and more dominant in C4A1 and C4P4 channels whereas, the profound higher in relative power was observed in beta band for high frequency bands and more dominant in Fp1F3 and F4C4 channels. Among four frequency bands, relative beta power had most significant differences for sleep bruxism patients. The significance of 28 features from each channel was compared between healthy controls and individuals with bruxism. As an instance, Tables 2 , 3 , 4 and 5 compare all the features of participants under healthy controls and bruxsim for C4P4 channel. The Shapiro-Wilk test was used to observe the normality of data distribution and the result showed mixed behavior due to limited sample size. Therefore, Wilcoxon rank sum test was used for statistical analysis [ 22 ]. The test result at p ≤ 0.05 was considered significant. The results showed that SD (α), SD (β), RMS (α), RMS (β), SampEn (α), SampEn (β), RPSD (δ), RPSD (θ), RPSD (α), (θ + α)/β, α/δ, α/θ, α/β, and APSD (α) had significant differences between healthy controls and bruxism patients, demonstrating that these features are effective for classification. Most of the features used for the classification task were found to be significant. To visually illustrate the behavior of each feature across different channels, Fig. 3 displays all 28 features of participants from both healthy controls and bruxism patients for the C4P4 channel, as an example. Upon observing Fig. 3 , it can be inferred that, in comparison to other frequency bands, the delta frequency band exhibits significant amplitude values on all features, except for a smaller amplitude on the SampEn feature. There were significant differences in SD, RMS, SampEn and four ratio PSD characteristics between healthy controls and bruxism patients. Features related to the theta, alpha, and beta frequency bands exhibited distinct behavioral patterns, with some of them also demonstrating statistical significance. All the above mentioned 28 features (see Tables 2 , 3 , 4 and 5 for detail) can provide basis for correct identification of bruxism patients from different aspects. By utilizing these features, we evaluated various machine learning algorithms and ultimately concluded that Fine Tree is the optimal algorithm for bruxism classification in the CAP database. Fine Tree classifier with default parameters, in MATLAB (Mathworks Inc., MA, USA), was used for classification. Table 6 lists the sensitivity, specificity, accuracy, and positive predictive value (PPV) obtained from all the five channels in the parietal region using five-folds cross validation. It showed the C4P4 channel achieves the highest classification performance with sensitivity (95.59%), specificity (98.44%), accuracy (97.84%), and PPV (94.20%) in C4P4 channel.
Discussion The purpose of this study was to comprehensively investigate the utility of single EEG channel for classification of bruxism patients from healthy controls in REM sleep. Literature relies primarily on PSD changes in the EEG frequency bands, including delta, theta, alpha, and beta to estimate the characteristics of neural activities in the brain [ 7 , 10 , 15 ]. However, in this study, in order to obtain sufficient signal features, we combined time domain, frequency domain, and nonlinear features from the EEG signals. The artifacts in EEG signals are highly complex and can appear in any frequency bands. The effective identification and filtering of artifacts such as swallowing, yawning, jaw alignment, snoring, or apnea is challenging that requires further in-depth research. However, in our database, the recordings were annotated by expert neurologists with information on sleep stages (wakefulness, S1-S4 sleep stages, REM, and body movements), body position (left, right, prone, or supine), and duration (seconds). Among all these annotated EEG signals, we only extracted the REM segment of interest, and this data segment already excluded body movements and body positions. Further, to remove slow frequency motion artifacts, we used DB5 wavelet transform to decompose the single-channel EEG signals into multi-frequency bands and then obtained the EEG signals at selected frequency bands within 1–30 Hz. This further filter out slow frequency artifacts below 1 Hz and higher frequency artifacts above 30 Hz. Since EEG signals can directly reflect the neural activities of the human brain and more directly reflect the influence and characteristics of bruxism at the neural level, it is of great importance in the analysis of bruxism. In the present study, we used statistical methods to analyze EEG frequency bands to compare the PSD dynamics from EEG among bruxism patients and healthy controls during REM sleep. Since there are many EEG channels, it will be beneficial if the channels and frequency bands reflecting the occurrence of bruxism are selective that simplifies the design of EEG acquisition instrument having improved classification accuracy. Therefore, in this paper, we analyzed five different EEG channels and revealed the most effective frequency band of EEG channel in characterizing sleep bruxism. Tang Jin Cheng et al. [ 26 ] made important achievements in the research of brain-computer interface by extracting the mean, standard deviation, root mean square value and other time-domain features of EEG signals, revealing that these time-domain features are helpful to understand the data distribution characteristics of different EEG frequency bands. By conducting experiments on multiple channels of EEG signals, it was found that the standard deviation and variance features were the most significant [ 26 ]. Hayat [ 16 ] and Lai [ 10 ] analyzed the average, maximum, and minimum values of the average normalized power spectrum using two EEG channels, revealing the effectiveness of power spectrum in the detection of sleep bruxism. Among time-domain features, mean, SD, and RMS were typical approaches to measure the amplitude of EEG [ 27 – 29 ]. Therefore, we also attempted to extract the relevant time-domain features of EEG signals and evaluate their role in the identification of bruxism. In this study, we conducted statistical analysis on the time-domain features of PSD (mean, SD, and RMS) in each frequency bands of EEG and found that SD(δ), SD(θ), SD(α), SD(β), RMS(δ), RMS(θ), RMS(α), and RMS(β) had significant differences ( p < 0.001) between bruxism patients and healthy controls. The frequency domain features can represent the proportion of components in different frequency bands, thus representing the activity of cranial nerves [ 30 , 31 ]. Studies have also found that some ratios of different bands, which includes following derived indices (θ + α)/β and α/β [ 32 ], and (θ + α)/(α + β) and θ/β [ 23 ] are useful for EEG features analysis. In this study, the RPSD including RPSD (δ), RPSD (θ), RPSD (α), RPSD (β), and ratios including (θ + α)/β, α/δ, α/θ, and α/β were computed. We found that many frequency domain features have significant differences ( p < 0.001) that includes RPSD(θ), RPSD(α), RPSD(β), α/δ, α/θ, α/β, APSD(δ), APSD(θ), and APSD(α). The nonlinear feature (Shannon entropy, SampEn) is a measure of the complexity of a system, and it is very effective for the analysis of short-length time series. The SampEn of EEG signal can be used to reveal the potential regularity and periodicity of data, which has been proven to accurately distinguish patients diagnosed with depression from the control group, which can serve as a highly sensitive and clinically relevant marker [ 33 ]. Mahshid Dastgoshadeh et al. also showed that the SampEn feature of EEG signals is a good tool for detecting epilepsy [ 34 ]. Richman et al. have revealed that SampEn is more suitable for the study of biological time series signals [ 35 ]. In this study, we found that the behavior of SampEn (δ), and SampEn (β) had significant differences ( p < 0.001) and SampEn(α) also had significant differences ( p < 0.05) between bruxism patients and healthy controls. In order to maximize the extraction of EEG frequency band features, we extracted 28 time-domain, frequency-domain, and nonlinear features (SampEn). Then statistical methods were used to compare the ability of each feature in classifying bruxism. The statistical results show that most of the features are effective to identify bruxism patients from healthy controls, especially in the high frequency bands (theta, alpha, and beta) that vary significantly. This variation in the high frequency bands might be due to the activation of neural activities involved in clenching teeth [ 1 ]. Mastication is controlled by central nervous system of brain [ 5 , 36 ] and the cause of involuntary clenching during sleep bruxism is mostly unknown. The relative increase in high power around the frontal and central EEG channels in our results indicates high-frequency neural oscillations in the fronto-central brain regions, contributing to involuntary teeth grinding during sleep bruxism. Certainly, further analysis with larger sample size needs to be performed to support the claim with higher confidence level. Table 7 shows the comparison between the classification performance of this study and previous studies. In the literature, among the detection methods of bruxism patients, EMG, ECG and EEG have been used by researchers. The accuracy of identifying bruxism patient based on EMG can reach 82.8% [ 9 ]. However, after combining EEG, EMG and ECG signals, the bruxism recognition accuracy reached 97.21% [ 10 ]. This method needs to collect EEG, EMG and ECG signals at the same time that increases the difficulty in signal acquisition and signal processing, and is not suitable for using portable devices to identify bruxism patients. Therefore, researchers began to study the strategy of only using EEG to identify bruxism, but the recognition rate of bruxism was not ideal. When utilizing only time-domain features (maximum, minimum, and mean) of PSD, even with the fusion of features from two EEG channels, the accuracy of bruxism classification can only achieve 81.25% [ 16 ]. Due to the correlation between EEG channels, even using channel fusion does not effectively improve classification accuracy. However, if the features of a single channel EEG signal can be fully extracted, it may help to improve classification accuracy. Based on this assumption, we used the same database as in literature [ 10 , 16 , 17 ] for our study. Our experimental results showed that when using 28 time domain, frequency domain, and nonlinear features, we could achieve higher accuracy (97.84%) from single EEG channel (C4P4). Our results demonstrated bruxism detection using single channel EEG.
Conclusion While previous studies have put several schemes for bruxism recognition, the utilization of multi-channel data acquisition has rendered the detection system more complex and unsuitable for portable devices. Some studies still have insufficient feature extraction and low recognition rates. Our research findings will help explore the potential of designing a portable sleep bruxism detection system based on a single channel EEG. We propose a classification method for sleep bruxism using a single EEG channel combined with time domain, frequency domain, and nonlinear features. Experimental results showed that there are significant differences ( p < 0.05) in all the 28 features except for Mean(δ), Mean(θ), Mean(α), Mean(β), RPSD(δ), APSD(β), SampEn(θ) between bruxism patients and healthy controls. Investigation on classification had also confirmed that these features were useful for classification. Multi-channel EEG devices are inconvenient to place electrodes, while using a single channel EEG acquisition can effectively reduce the complexity of the detection device, facilitate the installation of the device, and reduce the discomfort caused by data acquisition. The results of our study also showed that the classification performance of different channels of the brain were different. Among the channels, the C4P4 channel had the best classification results (the accuracy can reach 97.84%). In the 10–20 international electrode placement system, C4P4 is located on the right side of the Parietal and Central of scalp. Our experimental results suggest that placing electrode in this area is conducive to the development of a single channel portable bruxism detection system. Admittedly, our sample size is limited to 2 bruxism patients and 4 healthy controls. A larger sample size and more uniform population distribution (age, sex, and body weight) will give people more confidence in extending the results to the prediction and monitoring of patients with bruxism. Although the CAP sleep database of PhysioNet only has fewer patients with bruxism, it has a long record time for each subject, and expert neurologists annotate the data, making it authoritative. The database has important academic research value and has been used by multiple groups for research on bruxism. In addition, EEG has been used to characterize sleep bruxism in this study, we have extracted more effective features, and achieved higher classification performance using single channel EEG. This finding not only further supports the previously reported findings, but also extends the ability of single channel EEG to recognize bruxism. Indeed, the classification of bruxism in clinical research is a continuously evolving process that requires the combination of various assessment methods to obtain more reliable results. Simply relying on database analysis is not sufficient for accurate classification of bruxism. Attempts have been made in the literature to use EMG, ECG, or EEG signals to characterize the features of bruxism occurrence [ 9 , 10 , 15 ]– [ 18 ]. Although these methods cannot fully replace the accurate judgments of doctors, they can serve as medical auxiliary equipment to provide monitoring and warnings to potential bruxism patients or assist dentists in diagnosis. These data and analysis methods still have certain reference value for the diagnosis of “bruxism”. Additionally, we also need to clarify that the proposed method is only suitable for sleep bruxism recognition and is not a general bruxism classification system. Therefore, it is very important to comprehensively consider various data and analysis methods in the research and diagnosis of bruxism to improve the accuracy and reliability of classification. In the future, additional physiological signals such as ECG, EMG, and SaO 2 will be analyzed across all five sleep stages. The study will also be expanded to assess the recognition ability of bruxism in different sleep stages.
Background In the classification of bruxism patients based on electroencephalogram (EEG), feature extraction is essential. The method of using multi-channel EEG fusing electrocardiogram (ECG) and Electromyography (EMG) signal features has been proved to have good performance in bruxism classification, but the classification performance based on single channel EEG signal is still understudied. We investigate the efficacy of single EEG channel in bruxism classification. Methods We have extracted time-domain, frequency-domain, and nonlinear features from single EEG channel to classify bruxism. Five common bipolar EEG recordings from 2 bruxism patients and 4 healthy controls during REM sleep were analyzed. The time domain (mean, standard deviation, root mean squared value), frequency domain (absolute, relative and ratios power spectral density (PSD)), and non-linear features (sample entropy) of different EEG frequency bands were analyzed from five EEG channels of each participant. Fine tree algorithm was trained and tested for classifying sleep bruxism with healthy controls using five-fold cross-validation. Results Our results demonstrate that the C4P4 EEG channel was most effective for classification of sleep bruxism that yielded 95.59% sensitivity, 98.44% specificity, 97.84% accuracy, and 94.20% positive predictive value (PPV). Conclusions Our results illustrate the feasibility of sleep bruxism classification using single EEG channel and provides an experimental foundation for the development of a future portable automatic sleep bruxism detection system. Keywords
Author contributions CW participated in data acquisition, post processing, algorithm development, data analysis, drafting, and writing of the manuscript. BG and CL made contributions in data analysis, feature extraction, and classification, while AKV and XX provided constructive suggestions for the revision of the manuscript. All authors read and approved the final manuscript. Funding This work was supported in part by the National Natural Science Foundation of China under Grant 82160347, Chaozhou Science and Technology Plan Project under Grant 202201GY01, and Scientific Research Fund of Hanshan Normal University under Grant XY202106, QD202321, and QD2021218. Special Project of Guangdong Province in Key Fields of Ordinary Colleges and Universities under Grant 2023ZDZX2038. Innovation Teams of Ordinary Universities in Guangdong Province under Grant 2023KCXTD022. Data availability The data that support the findings of this study are available from PhysioNet [ 19 , 20 ], which are publicly available. PhysioNet is a repository of freely available medical research data, managed by the MIT Laboratory. Availability of data and material All data generated or analyzed during this case are included in this published article. Declarations Ethics approval and consent to participate Not applicable. All data were publicly available in PhysioNet. Consent for publication Not applicable. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:43:46
BMC Oral Health. 2024 Jan 14; 24:81
oa_package/46/68/PMC10787956.tar.gz
PMC10787957
38218962
Background For decades, insecticide-treated nets (ITNs) have played an important role in preventing malaria cases and deaths in malaria-endemic countries. Between 2000 and 2015, the malaria case incidence declined by 27% owing mostly to mass distribution campaigns of ITNs in malaria-risk countries [ 1 , 2 ]. The malaria case incidence slightly increased due to disruption caused by the Coronavirus disease (COVID-19) pandemic in 2020. However, while ITNs are still being widely distributed in endemic countries every three years, recent malaria case incidence has not substantially reduced [ 3 ]. Various explanations may be posited to explain the lack of a downward trend in malaria case incidence, but one contributory factor could be related to the lack of physical durability provided by modern ITN products. Clearly, if nets are insufficiently robust and quickly accumulate holes that undermine their ability to provide physical protection, they can no longer perform their basic function. Field studies have repeatedly reported loss of ITN physical integrity within two years post-distribution, with many ITNs being so badly damaged, they are discarded [ 4 ]. Over the same period there has been a lack of performance standards in product specifications relating to the physical durability of nets, meaning manufacturers have not been given appropriate targets to address. This has created a disconnect between the engineering design of ITN products and their expected performance in the field. Meanwhile, the weighted average price for ITNs sharply decreased from USD 4.20 per ITN in 2011, down to USD 1.88 per ITN in 2019 due to greater cost transparency, volume, industry consultation and synchronised demand [ 5 ]. This raises the question as to whether the physical durability of ITNs has also been impacted by product design changes, in response to price pressures. With so much emphasis being placed on long-term insecticidal functionality, the critical role the net textile plays in providing physical protection has been given insufficient attention. If the textile forming the net is too weak to withstand forces it is exposed to during use, large holes can be expected to form rapidly, compromising the ITN’s ability to provide physical protection, regardless of the insecticide’s bioefficacy. The physical durability of an ITN therefore depends on its inherent strength and ability to resist damage, which is governed by textile properties, as well as how carefully it is used following distribution. The inherent Resistance to Damage (RD) of any ITN can be measured before it is distributed. It depends on the mechanical properties of the textile used to manufacture the net and can be determined by a suite of four laboratory textile tests, in which fabric properties linked to damage mechanisms observed in the field are measured under controlled conditions. Each of these properties is a measure of the ability of the net to withstand the mechanical forces it will be exposed to during normal household use. The RD methodology introduced by Wheldrake et al. [ 6 – 9 ] has also highlighted marked differences in the performance of existing WHO prequalified ITNs. The RD score characterises the inherent resistance to damage of an ITN, and is measured on a 0–100 scale, where a RD score of 100 indicates the highest performance. Knowing the RD score of an ITN product before it is distributed is, therefore, a useful indicator of its inherent robustness and ability to retain physical integrity during normal use. A link between the RD performance and survivorship has been confirmed by Kilian et al. who reported a correlation between RD scores and actual ITN performance in the field, where RD scores measured in the laboratory of > 50, resulted in substantially longer service life in the field [ 10 ]. Given the ongoing concerns about the physical durability of ITNs, the purpose of this study was to determine if there have been any changes in the inherent resistance to damage of ITNs from 2013 to 2020. Test results for WHOPES-recommended (2013) and WHO-prequalified (2020) branded ITNs sampled in 2013 and 2020, respectively, were compared based on the textile testing methodology of Wheldrake et al . [ 6 – 9 ]. The study involved many different branded ITNs, including WHO-PQ qualified nets available in 2020, and aimed to identify any trends or changes in performance over the seven-year period.
Methods ITN samples Existing data for nine WHOPES-recommended ITNs measured in 2013 by Wheldrake et al . [ 7 ] was compared with primary data for twenty-four WHO-prequalified ITNs sampled in 2020. This included ITN products that were available both in 2013 and 2020, as well as newer products that were not available in 2013. The purpose of including both was not just to compare performance changes in the same products, but also to determine the extent to which product development has led to performance improvements across all available ITNs. The ITNs were made by different suppliers: Vestergaard Frandsen; Tianjin Yorkool International Trading Co., Ltd; Bayer industry Co., Ltd; V.K.A Polymers Pvt Ltd; Sumitomo chemical; BASF Agro B.V. Arnhem; Disease Control Technologies; Tana netting FZ; Shobikaa Impex Private Ltd. and Moon Netting. Comparison was possible between several of the ITNs available in 2020 and 2013 because in most cases they were produced by the same supplier and were identically branded. Herein, the ITNs are anonymously labelled, and general product specifications are given in Table 1 and Fig. 1 , including their principal polymer compositions (as indicated on the product packaging), abbreviated as follows: high density polyethylene (HDPE); polyethylene terephthalate (PET) and polyethylene (PE). The area densities of the ITNs ranged from 27 g m −2 to 52 g m −2 . Ownership of the Net U product transferred to another company between 2013 and 2020 so this lineage was reflected in the coding, even though the ITN specification also changed. Laboratory testing of ITNs All ITN samples were tested in accordance with RD methodology, which involves four textile tests each of which reflects different mechanisms of damage that nets are exposed to in the field [ 6 ]. For each ITN, the sample preparation methods, test procedures and number of replicates recommended by Wheldrake et al . [ 7 , 9 ] were followed to obtain data for the bursting strength, snag strength, abrasion resistance and resistance to hole enlargement. For each ITN, three separate net samples were analysed with five measurements per sample. The sampling method also accounts for those products with different fabrics forming the roof and side panels. Resistance to Damage (RD) scores The RD methodology developed by Wheldrake et al . [ 8 ] enables a single performance metric to be obtained by aggregating the textile testing data for snag strength, abrasion resistance, bursting strength and hole enlargement resistance, together with human factors (to obtain aspirational performance values). The RD score characterises the inherent resistance to damage of the ITN on a RD = 0–100 scale, where RD = 100 represents the highest performance as shown in Eq. 1 [ 8 ]. where: RD = Resistance to damage. = Actual bursting strength (kPa). = Aspirational bursting strength (kPa) = Actual snag strength (N). = Aspirational snag strength (N). = Actual abrasion resistance strength (number of rubs). = Aspirational abrasion resistance (number of rubs). σ H = Hole enlargement resistance.
Results To enable comparison between 2013 and 2020, the mean data for each physical property associated with the RD methodology is reported, together with the aggregated RD scores for all ITNs. Bursting strength Although bursting strength alone is a poor predictor of physical integrity and resistance to hole formation [ 9 ], it is a useful measure when aggregated with other textile mechanical property data, and measurements form part of the RD score. Figure 2 reveals that in 2013 the mean bursting strength of ITNs ranged from 299 kPa to 626 kPa, and in 2020, from 281 kPa to 578 kPa. Except for Net N and Net L which performed significantly better (p < 0.05) (+ 37% and + 42%, respectively), the ITNs in 2020 performed similarly, or worse compared to 2013. Nets B, C, H and U had significantly lower bursting strengths (p < 0.05) in 2020 compared to 2013 (− 8%, − 7%, − 17%, − 22% respectively) whereas Nets Q, W and X exhibited similar bursting strength in 2020 and in 2013 (p > 0.05). Note that all ITNs passed the long-standing bursting strength requirement set by WHO of 250 kPa, with most achieving values > 400 kPa in 2020. This reflects the fact that each ITN product has its own bursting strength requirements and specifications, which can be substantially higher than 250 kPa. Snag strength Snagging is the most frequently encountered form of mechanical damage in ITNs retrieved from the field and is an inherent weakness of lightweight knitted fabrics [ 11 ]. Figure 3 illustrates marked differences in the mean snag strengths across ITNs (2013 and 2020) from as low as 25N to 56N. Nets L and N performed significantly better in 2020 (p < 0.05) than in 2013 (+ 56% and + 17% respectively), while the snag strength of Net B, H and U significantly decreased by approximately − 9, − 10 and − 22%, respectively in 2020 compared to 2013 (p < 0.05). In respect of Net U (2020), this is likely to be attributable to a decrease in the yarn linear density and the area density of the fabric compared to 2013. Abrasion resistance Figure 4 compares the abrasion resistance of ITNs in 2013 and 2020, specifically the rate at which nets failed as the number of rubs (abrasion cycles) applied to the net increased. Except for Net U, the abrasion resistance of the 2020 ITN samples was similar, or worse than the 2013 performance. Hole enlargement resistance A comparison of the hole enlargement resistance of ITNs in 2013 and 2020 is shown in Fig. 5 . A significant decrease in the performance of Net L, Net U and Net W was observed between 2013 and 2020 (p < 0.05) with an increase in the end hole size of + 63%, + 154% and + 47% respectively. However, Net N and Net Q showed significantly improved performance in 2020 (p < 0.05) with reductions in the end hole size of -8% -19% respectively. Large differences in hole enlargement resistance were evident between different ITN products, with the worst performing ITNs generally being made from polyethylene (PE) monofilament warp knitted tulle fabrics. Across all nets, end hole sizes ranged from only 9.4 mm for Net N (very resistant to hole enlargement), to 73 mm for Net J, which is so large it is likely to completely undermine the ITN’s ability to provide long-term physical protection. Resistance to damage (RD) scores for ITNs A comparison of RD scores for ITNs in 2013 and 2020 are reported in Fig. 6 . These scores highlight marked differences in the performance of different products. According to RD methodology, the performance of ITNs should be achieving RD > 50, and ideally approaching RD = 100. Comparing the results from 2013 and 2020, a reduction in the RD value was observed for Net B, Net C net Q, Net H, Net U, Net W and Net X of -12%, -7%, -6%, -33%, -35%, -14% and -34%, respectively. Only Net N and Net L showed improved RD values between 2013 and 2020 of + 5 and + 47%, respectively. In 2020, only Net B and Net T achieved RD scores > 50. Apart from Net N and Net L, the RD scores of ITNs in 2020 were lower than in 2013. Concerningly, in 2020, six out of twenty-four WHO-recommended ITNs (25%) produced very low RD scores of < 30, making them highly susceptible to mechanical damage in the field and compromising their ability to provide long-term physical durability in the field. The average RD score of all comparable ITNs brands (excluding Net U, which was subject to a known specification change in the intervening years) decreased from 40 in 2013 to 36 in 2020. Figure 7 compares the average RD scores of nets in 2013 and 2020 and indicates reductions from 57 to 44 (for HDPE) and from 31 to 24 (for PE). However, the average RD scores for PET nets did not follow a similar trend. Note that there are also differences in the type of yarn construction (mono and multifilament) as well as knitting patterns, which means caution is needed in making generalised conclusions about the performance of nets made from different polymer types. Certainly, it is evident that there has been no major improvement in average RD scores in the seven-year period for any of the nets grouped by polymer type.
Discussion Over the last decade, field study evaluations of ITNs have repeatedly reported unsatisfactory performance in terms of physical integrity and survivorship, due to nets becoming too torn or damaged to provide long-term physical protection for at least three years. Whilst progress has been made in developing new insecticidal formulations, relatively little attention has been paid to the importance of the net textile itself, even though it plays a vital role in providing physical barrier protection and substantial physical damage could result in the net being disposed of [ 2 ]. Apart from bursting strength, no pre-assessment of the inherent ability of an ITN to resist damage is normally required prior to its approval and distribution, even though holes and tears are generally regarded as inevitable. ITNs with inherently high resistance to damage can be expected to last longer in the field, promoting improved survivorship, and this basic hypothesis has been confirmed by Kilian et al. [ 10 ] who demonstrated that existing ITNs achieving RD scores > 50, improved survivorship in the field by approximately seven months. However, as evidenced in for example Fig. 5 (hole enlargement), some ITNs resist damage so poorly that their ability to provide long-term physical protection is highly questionable. The fact that existing ITN products are not ‘the same’ in terms of their resistance to damage may also partly explain why the physical durability of ITNs generally appears to be so variable. The disconnect between the performance standards ITNs need to meet in the lab, and their required performance in the field, is in stark contrast to the insecticide, which is subject to thorough laboratory testing, as well as semi- and full field testing to ensure bioefficacy before use. Elsewhere, textiles used to make products for protecting people from potential harm, such as personal protective equipment (PPE) are subject to much stricter performance standards in the laboratory than ITNs, to ensure they are fit for purpose. ITN textiles need to be more thoroughly tested in the laboratory, based on appropriate performance standards, measuring properties that are relevant to the required performance in the field. In this regard, it is positive that WHO PQ-Vector Control team has made constructive steps to consider adoption of additional textile performance standards to drive improvements in the physical durability of ITNs. The pressing need for better ITN performance standards related to physical durability is highlighted by comparing the inherent resistance to damage data for ITNs sampled in 2013 and 2020, which reveals that the performance of many ITN products has not improved in seven years (Figs. 2 , 3 , 4 , 5 and 6 ). In 2020, only two of the twenty-four ITNs examined achieved an RD score > 50, while 25% of the ITN products available achieved a very low RD score of < 30 (Fig. 6 ). The ITNs with RD scores > 50 (Net B and T) were made from fabrics having relatively high area densities, mesh counts and linear density. By contrast in 2013, six of the sixteen WHOPES-recommended ITNs available, achieved RD values of > 50 [ 7 ]. The generally lower RD values observed in 2020 compared to 2013 is also highlighted when comparing ITNs in terms of their polymer composition (Fig. 7 ), although comparisons between nets made from HDPE, PE and PET are not straightforward because the knitting pattern is not consistent between products. Low RD values are likely to increase the vulnerability of ITNs to physical damage in the field, such that they will be prone to accumulating holes more quickly, limiting their long-term service life. Given the inherent weakness of current ITNs, particular focus is needed on improving abrasion resistance (Fig. 4 ), snag strength (Fig. 3 ) and hole enlargement resistance (Fig. 5 ), to markedly increase overall resistance to damage. This is likely to require major upgrades to the specifications of textile fabrics used to manufacture ITNs, including but not limited to polymer grades, filament linear density, knitting pattern, as well as basis weight (g/m 2 ). Of course, this will have cost implications, but these should be balanced against resulting increases in the reliability and value of ITNs in providing longer service life. At the same time, user perceptions and experience are key to ensuring on-going product acceptance and so the design of more durable ITNs should also consider factors such as the aesthetics and thermophysical comfort of products. The availability of robust, more damage resistant ITNs is not only a technical issue, but one of market dynamics. The design of more physically durable, longer-lasting ITN products is feasible, but there is little incentive to pursue improvements or innovate given that prices and margins are so low. As evidenced in this study, even though existing WHO Prequalified ITNs are not ‘the same’ in terms of their inherent resistance to damage, there is no recognition of this in procurement policies. Progressive price reductions in ITNs over many years combined with specifications that do not reward quality in terms of product performance standards, are disincentivising innovation and risk stagnating product innovation. Price pressures also incentivise cost-cutting, which is likely to negatively impact product performance. Evidence of changes to product specifications since 2013 (Table 1 ) leading to a fall in ITN performance are apparent in the present data. For example, the drop in bursting strength (Fig. 2 ), abrasion resistance (Fig. 4 ) and snag strength of Net U from 2013 to 2020 is likely to be attributable to the reduced fabric basis weight (-12%), together with decrease in both the mesh count and filament linear density of approximately, -36% and -20%, respectively. Raising the bar by upgrading textile performance standards for ITNs, combined with better pricing policies could act as a positive incentive for innovation, and drive major improvements in physical durability in the field. Upgraded textile performance standards would also ensure that better performance is built in to ITN products prior to use, instead of physical durability issues being highlighted when it is too late, or after lengthy field studies. Practically, this means implementing additional textile testing requirements and quantitative performance standards for snag strength, abrasion resistance and hole enlargement resistance, alongside bursting strength. It is the implementation of the individual textile testing methods that is key. Aggregating the resulting data to determine RD would be an optional step. Such an approach is also likely to lead to a more cost and time-efficient process for ITN development and evaluation, ensuring ITNs become longer-lasting more quickly, and truly fit for purpose.
Conclusion The inherent resistance to damage (RD) of ITNs has not markedly improved in the seven-year period from 2013 to 2020, and the performance of some ITN products has declined. This is likely to be a contributory factor in the high rates of ITN attrition and poor survivorship that have been repeatedly reported in field studies over the last decade. Of the WHO pre-qualified ITNs tested in 2020, only two achieved an RD score > 50, and six scored < 30, indicating a high degree of inherent weakness amongst currently available products. There are also large differences in the performance of existing ITNs, such that they cannot all be considered ‘the same’. More rigorous textile testing of ITN products is required, to provide new performance standards for snag strength, hole enlargement resistance and abrasion resistance to encourage development of more physically durable, longer-lasting ITN products, capable of protecting users more effectively. The resulting metrics would also assist procurers and countries to make informed choices.
Background For at least a decade, concerns have been raised about the physical durability of insecticide-treated nets (ITNs) and their ability to remain in good condition for at least three years. To discover if the resistance to damage (RD) of ITNs has improved or not, the RD scores of ITNs sampled in 2013 and 2020 were compared. Methods The RD scores and disaggregated textile performance data for nine ITNs recommended by the WHO pesticide evaluation scheme (WHOPES) measured in 2013 were compared with WHO-prequalified ITNs sampled in 2020. This included assessment of newer ITNs not available in 2013, to determine the extent to which product development has led to performance improvements across all available ITNs in the intervening years. Results The resistance to damage of ITNs has not generally improved from 2013 to 2020, and in some cases performance is worse. The average RD score of comparable ITNs brands decreased from 40 in 2013 to 36 in 2020. Of the nets available in 2020, only two of the twenty-four ITN products tested achieved an RD score of > 50, while six ITNs had very low RD scores of < 30, highlighting a serious inherent, and literal weakness in many WHO-prequalified ITNs. Conclusions The long-term physical durability of ITN products cannot be expected to improve while their resistance to damage remains so low, and major upgrades to the performance standards of textile materials used to make ITNs, as well as incentives to develop stronger ones are urgently required. Keywords
Acknowledgements We gratefully acknowledge the generous support of the Bill & Melinda Gates Foundation in the development of this work. We also wish to thank Dr. Angus Spiers and Dr. Edward Thomsen of Innovation to Impact (i2i) for their helpful, insightful comments and suggestions, which greatly assisted preparation of the manuscript. Thanks, are also due to Manoj Rathod, Rachel Douglas-Phillips, Dr. Fadi Junaid, Nicole Mitchell and Jack Street for their assistance with laboratory textile testing. Author contributions AW, EG and SJR developed the methodology and contributed to the data analysis and preparation of the manuscript. EG, HA, EZ and VC contributed to the data analysis and preparation of the manuscript. All authors read and approved the final manuscript. Funding This work was supported, in whole or in part, by the Bill & Melinda Gates Foundation [INV-003160]. Under the grant conditions of the Foundation, a Creative Commons Attribution 4.0 Generic License has already been assigned to the Author Accepted Manuscript version that might arise from this submission. Availability of data and materials Datasets are available on reasonable request to NIRI, UK. Declarations Competing interests The authors declare that they have no competing interests.
CC BY
no
2024-01-15 23:43:46
Malar J. 2024 Jan 13; 23:19
oa_package/bd/48/PMC10787957.tar.gz
PMC10787958
38218911
Background The incidence of thyroid mass has rapidly increased in the past few decades [ 1 , 2 ]. Papillary thyroid carcinoma (PTC), the most common type of thyroid malignancy, is prone to lymph node metastasis (LNM). As it is acknowledged that PTC has quite a low mortality rate and LNM has a less significant factor influencing survival rate [ 3 ], several recent studies found that LNM still negatively affected long-term recurrence [ 4 – 6 ]. The most common site of LNM in PTC is the central compartment, followed by the lateral compartment. According to the latest American Thyroid Association (ATA) guidelines and National Comprehensive Cancer Network (NCCN) guidelines, lateral neck dissection (LND) is only recommended for patients who were diagnosed with lateral lymph node metastasis (LLNM) before or during the operation. Prophylactic LND is not suggested. However, the sensitivity of evaluations, including preoperative imaging examination and intraoperative frozen pathology, is limited [ 7 , 8 ], and occult LNM may lead to relapse and secondary operation. Previous studies reported that the incidence of occult LLNM in PTC patients could reach 18.6-64% [ 9 – 11 ]. Thus, constructing a precise predictive model for LLNM seems essential in improving the efficacy of operations. It is proved that age is one of the most important prognostic factors of overall survival and disease-specific survival in PTC [ 12 – 14 ]. Several studies have also reported that the risk of LLNM decreased as age increased simultaneously [ 15 – 17 ]. However, the exact tendency of the risk ratio of LLNM as age changes is still unclear. Therefore, the cut point of patient age to stratify the risk of LLNM has not been well established yet. In addition, studies focusing on investigating distinct predictors of LLNM according to patient age were limited, leading to a lack of evidence to support more specific surgical strategies for PTC patients of different ages. As the risk of LLNM changes with age, a novel prediction model based on age stratification might be more accurate and benefit clinical practice. Our study aimed to identify the differences in clinicopathological characteristics, especially age, between patients with or without LLNM, to determine the optimal breakpoint of age for a more precise prediction model of LLNM, and to reveal differences in risk factors between patients of distinct age stages.
Materials and methods Patients This institutional ethics committee-approved retrospective study was conducted by searching the Department of Head and Neck Surgery, Cancer Hospital of Chinese Academy of Medical Sciences, and Peking Union Medical College databases from February 2016 to January 2020. Patients were included if they met all the following inclusion criteria: (i) primary thyroid lesion was confirmed PTC by pathology; (ii) patients had undergone thyroid lobectomy or total thyroidectomy and LND by the same experienced clinician; (iii) patients had no history of neck surgery, or radioactive treatment; (iv) patients had no history of other systematic malignant tumors. Clinical data were collected on demographics (age, gender, etc.) and tumor characteristics (tumor size, tumor location, multifocality, etc.). Informed consent was obtained from all participants. We explained intraoperative and postoperative risks to all patients in detail before surgery to ensure they fully understood the disease and surgical methods. Surgery We performed total thyroidectomy on patients with bilateral PTC. For unilateral PTC patients, total thyroidectomy or lobectomy plus isthmusectomy were performed depending on patients’ wishes. When unilateral PTC patients met one of the following conditions, total thyroidectomy was recommended: tumor size > 4 cm, multifocality in one lobe, contralateral benign nodules, or distant metastasis (according to guidelines of the Chinese Thyroid Association). All patients received central lymph node dissection. LND of level II-IV or II-V was performed on patients pathologically confirmed to have LLNM by fine-needle aspiration or intraoperative frozen biopsy in the lateral compartments. Additionally, when LNM in the central compartment was found before or during surgery, we performed LND of level III and IV based on the original thyroid collar incision. After surgeries, all specimens had pathologically examinations for LNM diagnosing. Statistical analysis All statistical analyses were performed using SPSS version 26.0 (IBM Inc, Armonk, NY, USA) and R version 4.2.1( www.r-project.org ). Measurement data were expressed in mean ± standard deviation, and independent samples t-test was used for comparative analysis; categorical variables were expressed as numbers [percent (%)] and were compared by the chi-square test. Variables with p value less than 0.05 in univariate analysis were included in multivariate logistic regression analysis to identify independent risk factors. P < 0.05 was considered statistically significant. The impact of age on LLNM was evaluated by logistic regression analysis. A locally weighted scatterplot smoothing (LOWESS) curve was used to fit and visualize the tendency of the OR of LLNM according to age. The structural break point of the fitting curves was considered as the optimal cutoff point and was determined by ‘changepoint’ package using R.
Results Patient characteristics A total of 499 patients were enrolled in this study. Among them, 396 (88.2%) were confirmed to have LLNM, and 103(22.9%) were absent from LLNM. A total of 163 patients (32.7%) undergone prophylactic lateral neck dissection of level III and IV as LNM in the central compartment was found, and 36.8% of these patients were confirmed to have occult LLNM. The details of characteristics were presented in Table 1 , which showed statistical differences between the two groups with regard to age( p = 0.003), gender( p = 0.005), tumor size( p = 0.009), tumor location( p = 0.011), multifocality( p < 0.001), bilaterality( p = 0.002), extra-thyroidal extension( p < 0.001), extra-nodal extension( p < 0.001), and central lymph node metastasis (CLNM) ( p < 0.001). No statistically significant difference between the two groups was observed in thyroiditis. Predictive value of age for LLNM A logistic regression analysis was performed to identify further the relationship between age and LLNM (Table 2 ). The results indicated that age was significantly related to LLNM ( p = 0.043). As shown in Fig. 1 , the LOWESS curve fitting the trend of odds ratio as above demonstrated that the risk of LLNM decreased as age grew. Correspondingly, structural breakpoints of the fitting curve identified by R using the ‘changepoint’ package confirmed that optimal age cut points were 30 and 45. Patients were divided into three groups based on cut points of 30 and 45 years old to verify the predictive value of age. The likelihood of LLNM was significantly different between the three groups, and older patients tended to have a lower risk of LLNM. After multivariate analysis, the risk of LLNM in the patient groups divided using the identified cut points remained distinctly different ( p = 0.020) (Table 3 ). P for interaction was also calculated between age group and other risk factors of LLNM, apart from extra-thyroidal extension, which showed an interaction p-value of 0.011 with age, there was no statistically significant interaction between age group and other risk factors. Distinct risk factors of LLNM according to age As age played a significant role in LLNM, we further investigated and compared independent predictors of patients of different ages. For patients younger than 30 years old, the univariate analyses revealed that sex( p = 0.039), tumor size( p = 0.046), tumor location( p = 0.047), multifocality( p = 0.017), extra-thyroidal extension( p = 0.010), extra-nodal extension( p = 0.011), and CLNM( p = 0.003) were related to LLNM. Further multivariate analyses showed that sex( p = 0.033), tumor size( p = 0.027), tumor location( p = 0.020), and CLNM( p = 0.019) were independent predictors for LLNM. Surprisingly, apart from CLNM, which had an 11.011(95%CI: 1.475–82.214) times the OR of LLNM compared to those without CLNM, tumors located in the upper lobe of glands had a 25.780(95% CI: 1.651-402.522) times the OR of LLNM (Table 4 ). The c-index of the prediction model was 0.811 (95%CI: 0.730–0.892). As for patients between 30 and 45 years old, tumor size( p = 0.049), multifocality( p < 0.001), bilaterality( p < 0.001), extra-nodal extension( p = 0.015), and CLNM( p < 0.001) were found to be related to LLNM. After including all factors above in multivariate analyses, CLNM( p = 0.007) was identified as the only predictor of LLNM and showed a 2.990(95% CI: 1.344–6.648) times the OR (Table 5 ). CLNM had a c-index of 0.626 (95%CI: 0.547–0.705) in predicting LLNM. For patients over 45 years old, tumor location( p = 0.026), extra-thyroidal extension( p < 0.001), extra-nodal extension( p = 0.001), and CLNM( p < 0.001) were significant in univariate analyses. Further multivariate analyses confirmed that tumor location( p = 0.013), extra-thyroidal extension( p < 0.001), extra-nodal extension( p = 0.042), and CLNM( p < 0.001) were all independent risk factors to LLNM. And extra-thyroidal extension showed the highest OR of LLNM of 13.005(95%CI: 3.202–11.371) times (Table 6 ). The c-index of the prediction model was 0.852 (95%CI: 0.781–0.923).
Discussion Previous studies have reported many risk factors related to LLNM in PTC, including age, sex, tumor size, tumor location, extra-thyroidal invasion, multifocality, and CLNM [ 18 , 19 ]. As for age, younger patients tend to be more likely to have LLNM than older ones [ 15 – 17 ]. This is similar to our findings. In our study, the figure that demonstrated the trend of the OR of LLNM decreased rapidly as age grew, especially after the specific age threshold. This suggested that a separate prediction system of LLNM should be constructed for patients of different ages to achieve better sensitivity and specificity. It was generally agreed that younger patients were more likely to have LNM. Still, the exact relationship between the risk of LNM and a decrease in age was shown in one large population research based on the SEER database, which aimed to investigate the impact of age on the risk of LNM; patients were divided into five subgroups by age. The result showed that younger patients had an increased predisposition for LNM [ 20 ]. However, the comparison groups were divided according to the cut points of age defined by authors and failed to demonstrate the difference in risk of LNM between groups. We provided a similar result that younger patients were more likely to have LLNM. Apart from that, our division of patients based on age was more accurate and reliable, since the cut points of age were determined based on analyzing the risk of LLNM in each patient subgroup stratified by age in 5-year intervals. The structural breakpoint of the OR fitting curve was identified using the R package, and it found that 30 and 45 were the optimal cut points for the possibility of LLNM. Many studies have found that CLNM was related to LLNM, and one Meta-analysis showed that the risk of LLNM of CLNM was 7.64 times more than those patients without CLNM [ 21 ]. Some researchers suggested that when the number of CLNM > 3, LND should be more actively performed [ 22 , 23 ]. Additionally, our previous study showed that the risk of occult metastasis in the lateral compartment was much higher for patients with CLNM [ 24 ]. Tumor located in the upper pole was another risk factor for LLNM due to an abundant blood supply and direct lymphatic vessels between the upper lobe of the thyroid and lateral neck [ 25 ]. Extra-thyroidal extension was an independent predictor for LLNM revealed by several studies as an extension may indicate the tumor’s aggressiveness and lead to metastasis. For patients with occult LLNM, insufficient treatment may lead to residual tumor or relapse, and secondary operation carries a higher risk of surgical complication and leads a psychological and economic burden to patients. Therefore, our study could be interpreted as an implication for a change in surgical management. In the condition of CLNM being observed before or during surgery, detection of the lateral compartment should be considered. For patients younger than 30 years old, male gender and large tumor size, especially tumor located in the upper lobe, were risk factors that required more attention and appropriate expansion of the extent of surgery when needed. The sufficient surgical extent was more critical for patients older than 45 years old with a poorer prognosis. When finding tumors with extra-thyroidal extension during operation, including gross and minimal, a careful evaluation of potential LLNM was highly recommended. Despite these findings, we acknowledge that there were still some limitations in this study. First, there is an inherent bias in all single-center and retrospective studies. Second, our sample size was relatively small, especially the group of patients without LLNM, which could reduce the statistical power of this research. Some merits of this study should also be noted. First, the preoperative evaluation of suspicious lymph nodes was evaluated through high-resolution US, which the same experienced experts performed. This avoided bias of diagnosis due to different experienced physicians to the greatest extent. Second, our dissection of levels III and IV was based on the original thyroid incision and did not involve the accessory nerve, and this would not increase the risk of skin numbness and shoulder dysfunction. Third, a group of pathologists described and documented the characteristics of PTC that may be related to LNM.
Conclusion This study indicated that the optimal age cut points to stratify the risk of LLNM were 30 and 45 years old. CLNM was a prominent risk factor for further LNM in all patients, and LND should be more actively performed when CLNM was confirmed. We also revealed other distinct risk factors of LLNM in patients with different age stages. Especially for younger patients with tumors in the upper lobe and older patients with extra-thyroidal extension tumors, more aggressive detection of the lateral neck and more intensive surgical treatment should be considered.
Introduction Studies have revealed that age is associated with the risk of lateral lymph node metastasis (LLNM) in papillary thyroid cancer (PTC). This study aimed to identify the optimal cut point of age for a more precise prediction model of LLNM and to reveal differences in risk factors between patients of distinct age stages. Methods A total of 499 patients who had undergone thyroidectomy and lateral neck dissection (LND) for PTC were enrolled. The locally weighted scatterplot smoothing (LOWESS) curve and the ‘changepoint’ package were used to identify the optimal age cut point using R. Multivariate logistic regression analysis was performed to identify independent risk factors of LLNM in each group divided by age. Results Younger patients were more likely to have LLNM, and the optimal cut points of age to stratify the risk of LLNM were 30 and 45 years old. Central lymph node metastasis (CLNM) was a prominent risk factor for further LNM in all patients. Apart from CLNM, sex( p = 0.033), tumor size( p = 0.027), and tumor location( p = 0.020) were independent predictors for patients younger than 30 years old; tumor location( p = 0.013), extra-thyroidal extension( p < 0.001), and extra-nodal extension( p = 0.042) were independent risk factors for patients older than 45 years old. Conclusions Our study could be interpreted as an implication for a change in surgical management. LND should be more actively performed when CLNM is confirmed; for younger patients with tumors in the upper lobe and older patients with extra-thyroidal extension tumors, more aggressive detection of the lateral neck might be considered. Keywords
Author contributions H.Z. designed the study and wrote the manuscript. L.Z designed the study and analyzed the data. Z.H., S.W., and P.S. collected data. Y.D. prepared the manuscript. L.N. performed all ultrasound examinations. Z.L. designed the study and edited the manuscript. All authors reviewed the manuscript. Funding This work was supported by a grant provided by the Capital Clinical Features Application Research (Z171100001017211). This work was supported by a grant provided by National Natural Science Foundation of China (82373439). Data availability The data that support the findings of this study are available on request from corresponding authors. Declarations Competing interests The authors declare no competing interests. Ethical approval and consent to participate This study was approved by the ethics committee of the Cancer Hospital of Chinese Academy of Medical Sciences (reference number: NCC2016ST-23). We explained intraoperative and postoperative risks to all patients in detail before surgery to ensure they fully understood the disease and surgical methods. Informed consent was obtained from all participants. Consent for publication Not Applicable (NA).
CC BY
no
2024-01-15 23:43:46
BMC Surg. 2024 Jan 13; 24:24
oa_package/13/9e/PMC10787958.tar.gz
PMC10787959
38218874
Introduction The temporomandibular joint (TMJ) is classified as a bicondylar diarthrosis. The joint is composed of the mandibular condyle and the glenoid fossa [ 1 – 4 ]. The TMJ is vital in guiding jaw movement and managing daily activities such as swallowing, chewing, and speaking [ 5 – 7 ]. Temporomandibular disorders (TMD) are a range of both musculoskeletal and degenerative conditions [ 8 – 12 ]. The leading causes of TMD are an altered position of the intra-articular disc or abnormal muscle hyperactivity. The main symptoms include joint clicks, limitation of movement, and facial muscle tension, known as orofacial pain [ 13 , 14 ]. It is estimated that 25% of the population shows signs of TMD, while only a tiny percentage shows the need for treatment. Furthermore, the prevalence of symptoms is much higher in women than men [ 15 ]. The average age of patients with symptoms varies between 20 and 50 years. More than 70% of TMD patients show joint disc malposition as the cause. This condition is called an internal disorder. The progression of the disease is poorly understood; however, most affected individuals show a degenerative condition known as osteoarthritis or osteoarthrosis, depending on whether there is an inflammatory state. A clinical study on TMD patients with symptoms at the opening and closing of the mouth, on palpation, showed osteoarthritis phenomena at the joint level. Several studies have demonstrated by the magnetic resonance that the articular discs of asymptomatic patients in their normal anatomical position during movements have morphological change processes at the condyle and eminence level due to adaptive procedures [ 16 , 17 ]. While in symptomatic patients, there is a critical alteration also at the bone level. The observed changes are abrasion and deterioration of the articular cartilage and bone remodeling. Treatment options vary according to the joint's internal imbalance severity and osteodegenerative phenomena. There are non-invasive or minimally invasive treatments for patients in the early stages of internal disorders [ 18 – 21 ]. While in advanced or complex cases, patients require joint replacements or invasive therapies [ 22 ]. The first phase of the pathology passes through the remodeling of the articular disc and bone heads. This process is one of adaptation and distributes excessive stress loads better. This is an adaptive process and therefore remains physiological. When the adaptive and remodeling capacity of the disc is exceeded, osteoarthritis phenomena occur. Significant alterations include alteration of bone components such as flattening of the joint eminence, decreased glenoid fossa, and flattening of the articular disc [ 23 , 24 ]. Degenerative arthritis can result from a lack of adaptation or excessive or prolonged stress. All the alterations of the TMJ start from an imbalance and an alteration of the articular disc. The various pathological transitions from internal disorders to osteoarthritis are not understood; however, there is a correlation between the two conditions. The progression and onset of TMD are poorly understood; however, Wilkes has divided the passage into 5 phases. In stage 1, there is a painless click at the beginning of the opening and the end of the closing. There is a displacement of the disc forward and an inability of the disc to return to its original position. The bone bases appear unaltered. In stage 2, click symptoms and orofacial pain are present. The magnetic resonance shows a slight disc deformation and displacement forward; however, the disc in the maximum opening returns to the physiological position [ 25 , 26 ]. Stage 3 is associated with frequent orofacial pain; jaw blocking during movement is more common. There is a thickening of the articular disc. At the beginning of phase 3, the disc is recaptured but not entirely and in response, the disc deforms under the thrust of the condyle forward. During stage 4, symptoms are more persistent and include chronic pain and limitation in movement. The disc appears very thickened and is not recaptured during the maximum opening. There are also osteoarthritis phenomena. Stage 5 is the most advanced. Patients have persistent chronic pain, crackles, and movement restrictions. Non-invasive treatment of TMDs is the therapy of choice [ 27 ]. There are several non-invasive therapies available for the treatment of TMD. One of these therapies is drug treatment. TMDs affect a substantial portion of the global population, with prevalence rates ranging from 5 to 12%. The diverse nature of TMDs, which may involve myofascial pain, arthralgia, disc displacement, and osteoarthritis, contributes to the complexity of their clinical presentation. Patients often experience symptoms such as pain during mastication, restricted jaw movement, joint noises, and referred pain to the head and neck region. Moreover, TMDs can lead to additional comorbidities, including headaches, sleep disturbances, and psychological distress, underscoring the need for effective therapeutic approaches. Pharmacological interventions in the management of TMDs target the underlying pathophysiological mechanisms, aiming to alleviate pain, improve joint function, and enhance overall patient well-being. Non-steroidal anti-inflammatory drugs (NSAIDs) remain a cornerstone of pharmacotherapy, offering analgesic and anti-inflammatory effects through inhibition of cyclooxygenase enzymes. Muscle relaxants, such as benzodiazepines and cyclobenzaprine, act by reducing muscle hyperactivity and relieving muscle-related symptoms. Tricyclic antidepressants (TCAs) and selective serotonin-norepinephrine reuptake inhibitors (SNRIs) have demonstrated efficacy in managing TMD-related pain, likely through their modulation of central pain pathways and neurotransmitter balance. Botulinum toxin injections, a relatively novel approach, target muscle hyperactivity and have shown promise in reducing pain associated with TMDs. Pharmacological interventions represent a vital component of the multifaceted management approach for TMDs. These interventions aim to alleviate pain, restore joint function, and enhance patients' quality of life. As our understanding of the pathophysiology of TMDs continues to evolve, it is imperative to critically evaluate the efficacy, safety, and long-term outcomes of various pharmacological agents. This review aims to consider the different drugs used in the treatment of TMD [ 28 ]. In addition, a meta-analysis was conducted regarding the different pharmacological treatments of temporomandibular pain. The purpose of this systematic literature review with meta-analysis is to evaluate the main pharmacological treatments of pain caused by TMD. The purpose of this systematic literature review with meta-analysis is to evaluate the main pharmacological treatments of pain caused by TMD. In fact, the purpose of the meta-analysis is to evaluate which is the best pharmacological treatment of pain caused by temporomandibular disorders.
Materials and methods Eligibility criteria The following population (including animal species), Exposure, Comparator, and Outcomes (PECO) were used to determine the eligibility of all documents: P) Participants are patients with TMD E) Exposure consisted of patients with TMD treated with different types of drugs, nonsteroidal anti-inflammatory drugs, corticosteroids, antidepressant drugs, centrally acting muscle relaxants, anticonvulsants, benzodiazepines and to whom pain was assessed by VAS scale. C) Comparisons are patients with TMD and treated with placebo. O) Outcome is to evaluate the effectiveness of different types of drugs on TMD pain. As a secondary outcome is to evaluate the effectiveness on TMD treated with the different drugs compared with placebo. Only papers providing data at the end of the intervention were included. Exclusion criteria were: 1) history of Temporomandibular joint (TMJ) trauma; 2) patients suffering from any inflammatory disorders or rheumatic diseases (e.g., rheumatoid arthritis, psoriatic arthritis); 3) patients with fibromyalgia; 4) patients with headache/migraine; 5) patients with a congenital abnormality or neoplastic conditions in TMJ region; 6) cross-over study design; 7) studies written in a language different from English; 8) full-text unavailability (i.e., posters and conference abstracts); 9) studies involving animal: 10) review article; 11) case report. Inclusion criteria are: patients with TMD; RCTs; observational studies; clinical trial. Search strategy The study made use of major scholarly databases (PUBMED, WEB of SCIENCE, LILACS). The time period taken into account for the electronic search was from January 3, 2000, to January 2, 2023. Following the method outlined in Table 1 , papers were systematically searched for in the databases of PubMed, Web of Science, and Lilacs. In addition, a manual search of earlier systematic reviews on the same subject was also done. The connector "AND" was used to unite the words "drug" and "temporomandibular disorders". MESH was utilized to assist with the web search (Medical Subjects Headings). This systematic review was conducted according to Preferred Reporting Items for Systematic Reviews (PRISMA) guidelines and the Cochrane Handbook for Systematic Reviews of Interventions. The systematic review protocol has been registered on the International Prospective Register of Systematic Reviews (PROSPERO) with the following number CRD 42022316112 [ 29 ]. Data extraction Two reviewers (M.G.; R.F.) independently extracted data from the included studies using a customized data extraction on a Microsoft Excel sheet. In case of disagreement, a consensus was reached through a third reviewer. The following data were extracted: 1) First author; 2) Type of study; 3) Sample; 4) Type of drugs; 5) Data used for meta-analysis; 6) Results. Quality assessment The risk of bias in papers was assessed by two reviewers using Version 2 of the Cochrane risk-of-bias tool for randomized trials (RoB 2). Any disagreement was discussed until a consensus was reached with a third reviewer [ 30 ]. Statistical analysis The following factors were analyzed: VAS scale pre and after therapy. The mean values and standard deviation (SD) for each variable in each cohort were extracted. A meta-analysis was carried out when it was feasible. A 95% confidence interval (CI) for the mean difference was determined. Plots depicting forests were made to show the findings. Because of the high heterogeneity of the studies, the random effects model was used. Heterogeneity and the I 2 statistic describes the percentage total variation across studies that is due to heterogeneity rather than change. The suggested interpretation of I 2 is as follows: 0%–40% may represent low heterogeneity, 30%–60% may represent moderate heterogeneity, 50%–90% may represent substantial heterogeneity and 75%–100% considerable heterogeneity. The pooled analyses were performed using the software SPSS version 27 (IBM Corp. (2020). IBM SPSS Statistics for Windows (Version 27.0) [Computer software]. IBM Corp). In this meta-analysis, used Pre-Calculated Effect Sizes models was used in order to assess the variances between the VAS scale before and after therapy with different drug classes. he reported summary statistics were calculated as random-effects models based on the assumption of heterogeneity between studies. Pooling was done according to the DerSimonian and Laird method, using inverse variance. The p - value was set at 0.05 The Risk ratio between the two groups was measured. The Ta et Dionne study was divided into Ta et Dionne in which it evaluated the effects of naproxen and Ta et Dionne 1 in which the effects of celecoxib were evaluated.
Results Study characteristics At the end of the research, 1698 studies were identified. During the first screening phase, 190 articles were eliminated because they were not in English, precisely 130 from PubMed, 15 from Web of Science, and 45 from Lilacs. In addition, 668 items were excluded because they were duplicates. Ultimately, clinical trials and randomized trials were selected through special filtering. Therefore, 671 articles were excluded. In the final stage, 178 articles were selected, and abstracts were read. In the final stage of selection, 170 articles were excluded. One hundred fifty-eight did not meet the PECO, and 20 articles because they were off-topic (they treated oral-facial pain or did not meet the inclusion criteria) (Fig. 1 ). The selected studies come from various parts of the world and are heterogeneous regarding drug use. They consider a heterogeneous population of age. In the studies reviewed, adults were considered apart from Arabshahi's study, which evaluated children. In this systematic review, 531 patients were assessed; the population is heterogeneous regarding the type of TMD. Main findings Ta and Dionne studied the pharmacological efficacy of NSAIDs. They performed a double-blind, randomized, placebo-controlled study in which they administered one group of celecoxib (100 mg twice daily), the second group of naproxen (500 mg twice daily) and the placebo control group for six weeks. This study showed that naproxen effectively reduces orofacial pain symptoms due to TMD compared to celecoxib. The study Singer [ 31 ] evaluated the efficacy of different drugs on gold facial myogenic pain thanks to a double-blind, randomized and controlled study. Thirty-nine subjects were recruited, including 35 women and four men, with orofacial pain due to TMD for at least three months and marked distress on palpation of the chewing muscles. Patients were treated for pain with one of the drugs to be tested: placebo, diazepam, ibuprofen, or a combination of diazepam and ibuprofen. Muscle pain, maximal opening and limitation of the movement were assessed at times 0 at two weeks and four weeks, respectively. Pain, as measured by a visual analogue scale, was significantly decreased in the diazepam and diazepam plus ibuprofen groups but not in the ibuprofen or placebo groups. Analysis of variance showed a significant pharmacological effect for diazepam but not ibuprofen, indicating that pain relief was attributable to diazepam. No significant changes were observed in muscle tenderness, interincisal opening or plasma beta-endorphin level [ 31 , 32 ]. The study by List [ 32 ] evaluated the effectiveness of injecting an intra-articular dose of morphine. Fifty-three patients with joint pain were recruited. This randomized, double-blind study evaluated patients before treatment and a follow-up one week after treatment. The pain intensity was recorded using the VAS scale at the mouth's maximum opening and resting position. Intra articulaar injection was made into one TMJ containing either 1.0 mg morphine-HCl, 0.1 mg morphine-HCl, or saline (placebo). The pain was recorded in a diary three days before surgery and five days after. Pain assessed by VAS was significantly reduced 1–10 h after injection. The VAS score was lower in the 0.1 mg morphine group than in the 1.0 mg morphine group ( P < 0.043) and in the placebo group ( P < 0.021) [ 33 ]. Arabshahi's study evaluated the effects of corticosteroid injection into the temporomandibular joint in children with idiopathic arthritis and TMJ inflammation. Twenty-three children aged 4 to 16 years were recruited and received intra-articular corticosteroid injections (triamcinolone acetonide [ n = 16] or triamcinolone hexacetonide [ n = 7]). Pain and maximum incisal opening before and after treatment were evaluated as benchmarks. From the follow-up results of 13 patients with joint pain, 10 had complete resolution of symptoms with a significance of P < 0.05. All patients showed a reduction in mouth opening. After injection, the maximum aperture improved by 0.5 mm in 10 patients with a significance of p = 0.0017. Post-injection oedema occurred in only two patients [ 34 ]. In the Alstergren research, 22 patients (29 joints) with specific or nonspecific temporomandibular joint (TMJ) arthritis received a single intra-articular glucocorticoid (GC) injection. At follow-up appointments, 2–3 or 4–6 weeks following therapy, the impact on subjective symptoms, clinical signs in the craniomandibular system, and joint aspirate concentration of neuropeptide Y-like immunoreactivity (NPY-LI) were assessed. The medication improved the symptoms and clinical indicators in patients with the particular inflammatory joint condition, and 2–3 weeks later, the TMJ level of NPY-LI decreased. After 2–3 weeks, there was also a variable clinical improvement and NPY-LI level decrease in the patients with unspecific inflammatory joint illness, but not statistically significantly [ 35 ]. Several studies show the analgesic effect of tricyclic antidepressants in chronic pain. The study of Rizzatti-Barbosa showed that amitriptyline in the dose of 25 mg daily decreased symptoms in patients with chronic orofacial pain. While increasing the amount to 50–75 mg/day showed no substantial analgesic effects. The quantities of tricyclic antidepressants used to treat pain are much lower than those used to treat depression [ 36 ]. Another class of antidepressants used are SSRIs. They were introduced in the 1980s and have become the most widely used drugs for treating depression. This study of Kimos was to assess gabapentin's analgesic effects on myalgia. Fifty participants were randomly assigned to one of two research groups in this 12-week randomized controlled clinical trial: 25 received gabapentin, and 25 received a placebo. Palpation Index, VAS-measured pain, and a VAS-measured impact of myalgia on daily functioning were the outcome measures used (VAS-function). Thirty-six subjects completed the trial. Clinically and statistically, gabapentin was more effective than a placebo in reducing patient-reported pain, masticatory muscle hyperalgesia, and the impact of myalgia on daily functioning (Gabapentin = 51.04%; placebo = 24.30%; P = 0.037; gabapentin = 67.03%; placebo = 14.37%; P = 0.001; gabapentin = 57.70%; placebo = 16.92%; P = 0.022) [ 37 ]. The study of Gilron evaluates the administration of pregabalin in the wider variety of neuropathic pain etiologies in this multicenter experiment. Pregabalin was administered to 256 participants in this enriched enrolment randomized withdrawal trial in a single-blind, flexible-dose for four weeks when stable concomitant analgesics were permitted. A total of 157 patients were randomized and treated, double-blind, to receive either pregabalin ( n = 80) or a placebo ( n = 77) for five weeks, and 135 of them (65%) reported a pain improvement of at least 30%. Of the single-blind responders who were randomly assigned, 81% received a placebo, and 86% received pregabalin. At the double-blind endpoint, the pregabalin group's mean (SD) pain scores were 2.9 (1.9), and the placebo groups were 3.5 (1.7) ( P = 0.002). These small but significant pregabalin-placebo differences were seen in each patient category with a diabetic peripheral neuropathy or postherpetic neuralgia diagnosis ( P = 0.03) and those with other diagnoses ( P = 0.02). Sleep disruption, Hospital Anxiety and Depression Scale Anxiety and Depression subscales, and other secondary measures showed significant variations. 28 out of 80 (35.0%) pregabalin users and 28 out of 77 (36.4%) placebo users discontinued the double-blind phase due to a noticeable increase in pain. In the single-blind stage, adverse events were consistent with the pregabalin tolerability profile. They resulted in the discontinuation of 9 patients, five in the placebo group and two in the pregabalin group [ 38 ] (Table 2 ). Quality assessment and risk of bias Using RoB 2, the risk of bias was estimated and reported in Fig. 2 . Regarding the randomization process, 88% of the studies ensured a low risk of bias. However, 16% of the studies excluded a performance bias, but 84% reported all outcome data, 89% of the included studies adequately excluded bias in the selection of reported outcomes, and 50% excluded bias in self-reported outcomes. Overall, 6 of the eight studies were shown to have a low risk of incurring bias . Meta-analysis The meta-analysis was conducted by SPSS software. Continuous Outcomes with Pre-Calculated Effect Sizes was performed because are continuous variables, using the p value at 0.05. In this meta-analysis, in order to statistically analyse the data, only studies with a control group and taking into account the VAS before and after therapy were taken into account, therefore the studies taken into account for statistics are 4. In order to conduct this meta-analysis, we evaluated and considered the change in VAS 6 weeks later in order to assess which is the best pharmacological treatment for TMD pain. The overall effect, reported in the forest plot (Fig. 3 ). The Forrest plot found no significant variation in pain symptomatology assessed by the VAS scale among the different therapies analyzed (Overall:33.47; C.I. 5.79–61.79). In Ta et Dionne we analyzed naproxen and celecoxib. In the study by Rizzatti Barbosa et al. we evaluated the effects of amytriptyline on the VAS scale. In the study by Gilron et Al. the effects of pregabalin were evaluated. The Meta-Analysis with Continuous Outcomes with Pre-Calculated Effect Sizes resulted in the rejection that there is intergroup variability (p.0.02). Effect Size Estimates for Individual Studies was reported in Fig. 3 .
Discussion In light of the results of the meta-analysis and literature review, this article aims to evaluate which pharmacological treatment is most effective for the treatment of TMD pain. Therefore, we evaluated the main drugs used and compared their change in pain at 6-week follow-up. This meta-analysis showed no signigicative differences. Therefore, the clinician's choice on the type of drug treatment should be based on a number of variables including the patient's overall assessment, the presence of comorbidities. The following part will evaluate all the advantages and disadvantages of each drug, and then we can arrive at choosing a drug that can create fewer interactions and fewer side effects for the TMD patient. NSAIDs inhibit cyclooxygenases and thus prevent the formation of prostaglandins. They were the most prescribed drugs for orofacial pain. NSAIDs treat patients with mild to acute TMJ inflammation, especially in disc dislocation cases, without reduction or acute trauma. These drugs must be taken for a minimum of 2 weeks. Several NSAIDs are used extensively in the dental field, such as ibuprofen, naproxen, diflunisal and ketorolac. The superiority of any NSAID over the others has not been demonstrated. The significant side effect is on the gastrointestinal level. NSAIDs can cause ulcers and bleeding in the gastric tract. Studies in the USA found that about 16,000 people die from gastrointestinal side effects [ 39 – 47 ]. Therefore, administering an NSAID to a patient with active gastrointestinal problems is not recommended. Naproxen and Ibruprofen appear to be the safest at the cardiovascular level. Ibuprofen also has fewer gastrointestinal effects. Another pharmacological possibility for patients at risk of gastric bleeding is using COX-2 inhibitors; however, patients should not have cardiovascular or cerebrovascular risk factors. NSAIDs can also interact with other types of drugs. For example, the clearance of Lithium decreased following the intake of NSAIDs. Therefore, there may be an increase in the serum concentration of Lithium and an increase in toxicity [ 48 ]. Combining NSAIDs with angiotensin conversion inhibitors or loop diuretics can cause acute kidney injury. NSAIDs reduce renal blood flow and the excretion of these drugs. Therefore, if NSAIDs are taken for more than five days, the effects of antihypertensive medications such as diuretics may be enhanced [ 49 ]. NSAIDs have an antiplatelet effect by inhibiting the synthesis of thromboxane. Therefore, they should be used cautiously in patients on anticoagulant therapy (e.g., warfarin) [ 50 ]. Widely used drugs are opioids, and their action against moderate pain is very effective. In the treatment of TMD, their use is mild to severe pain. The most widely used opioid drugs are codeine, oxycodone, and hydromorphone for severe pain. If the oral route is not preferred, one can opt for the fentanyl patch. The study did not show a variation between injectable and oral opioids [ 51 ]. Corticosteroid drugs have a chemical similarity to cortisol which is produced by the adrenal glands. These drugs are used to treat moderate to severe TMD. They block phospholipase A2, decreasing the production of prostaglandins and leukotrienes. Corticosteroids can be injected directly into the joint capsule or orally. Usually, intra-articular cortisone solutions are diluted with a local anaesthetic. Other drugs used in the treatment of TMD are centrally-acting muscle relaxants. They are used and administered in patients with chronic pain. These drugs cause drowsiness and are therefore taken before going to bed. The most common muscle relaxants are carisoprodol, cyclobenzaprine, metaxalone and methocarbamol [ 52 ]. Antidepressive drugs are widely used for the management of TMD pain. Tricyclic antidepressant drugs and selective serotonin reuptake inhibitors appear to be the most important. Several studies evaluate the efficacy of tricyclic antidepressants in the management and control of chronic pain. Furthermore, patients with chronic pain also suffer from depression and sleep disturbances. The exact mechanism of action is not fully known; however, it is probably given by inhibiting the serotonin reuptake and noradrenaline at the synaptic level of the central nervous system. By blocking the reuptake at the back of the horn, there is an increase in the availability of these neurotransmitters, which block the transmission of pain. The drugs used are amitriptyline, nortriptyline and desipramine [ 53 ]. Adverse events of SSRI drugs are sweating, dizziness, blurred vision, and constipation. With an increase in the bioavailability of endogenous catecholamines, the administration of epinephrine can cause an overdose reaction. In addition, monoamine oxidase inhibitors, given together with tricyclic antidepressants, can lead to serotonin syndrome with fever, ataxia and severe hypertension. These drugs cause gastrointestinal problems, headaches, sexual dysfunction, dry mouth and sweating. These drugs should be used with caution. These patients need combined management with their treating physician. Anticonvulsant medications are often used for neuropathic pain. Their analgesic mechanism remains unclear. These drugs inhibit excessive neuronal activation. The sites of action are the voltage-gated ion channels. Gabapentin and pregabalin are used for orofacial pain. They have a chemical structure like GABA, the primary inhibitory neurotransmitter. However, none of these drugs acts on the GABA receptor [ 54 ]. Benzodiazepines are drugs used to treat sleep disorders and muscle disorders. These drugs are associated with tolerance and dependence; therefore, their long-term use is not recommended. These act on GABA receptors which mediate inhibitory transmission in the central nervous system. These drugs act on the chloride receptor, opening it and thus promoting neuronal hyperpolarization. They are mainly used as anxiolytics, but they also have their use as muscle relaxants. Their use for the treatment of epilepsy is also well established. Several clinical trials have evaluated the efficacy of these drugs in treating TMD compared to a placebo. These drugs have been discouraged due to adverse reactions such as sleepiness. Furthermore, these drugs show tolerance and dependence whereby the sudden discontinuation leads to symptoms including anxiety, agitation, restlessness, insomnia and convulsions. They should not be used in patients with myasthenia gravis and glaucoma. These drugs are degraded by the CYP 4503A4 and therefore show numerous interactions with other medications. Also, some agonist drugs of this cytochrome, such as grapefruit juice, can reduce the metabolism of benzodiazepines and thus increase their bioavailability. They are delicate drugs that experienced medical personnel must administer [ 55 ]. The main side effects of opioids are sedation, dizziness, nausea, vomiting, constipation, physical addiction, tolerance and respiratory depression. All these symptoms are accentuated in geriatric patients. The use of opioids with other central nervous system depressants, such as benzodiazepines, may have additive effects and cause increased sedation. Opioids are not recommended due to their tolerance and dependence. Therefore, the prescription of this category of drugs should be limited. Furthermore, no clinical evidence exists that long-term opioid therapy is more effective than other treatments [ 56 , 57 ]. The structure of cyclobenzaprine is like tricyclic antidepressants. This drug is contraindicated in patients with hyperthyroidism, congestive heart failure, arrhythmias, and recent heart attacks. Low doses of cyclobenzaprine have positive effects on pain. Therefore, no more than 10 mg before bedtime, cyclobenzaprine is prescribed in low doses. The treatment plan should be 30 days, followed by a 2-week off period during which the patient's symptoms are evaluated. However, chronic therapies with cyclobenzaprine should be managed with the treating physician. Therefore having evaluated all pharmacological mechanisms and side effects we were able to assess that as an effect on pain all drugs are effective. Moreover, given the side effects, the clinician should well categorize the pain and start through NSAIDs which are the drugs of first choice and with less side effects and then vary the therapy in case of failure. Limitations of this meta-analysis The limitation of this meta-analysis is that we analyzed the different types of drugs together as we wanted to analyze which treatment was the most effective. However, we did not take into consideration the duration of treatment of each study. The limitation of this study lies in having evaluated different drug therapies for the treatment of TMDs.
Conclusions The study includes different types of pharmacological treatment for TMD and therefore we cannot state that there is a first choice drug for the treatment of pain. We can state that NSAIDs are the most widely used drugs. However, we can conclude from the review and the meta-analysis that NSAIDs are undoubtedly very effective drugs in the treatment of acute pain and are undoubtedly the safest drug class. Opiods are the substitute drugs for NSAIDs in the case of patients with previous gastrointestinal bleeding or in the case of acute moderate/severe TMJ pain Corticosteroids are always used for the treatment of acute moderate/severe pain, however, the first choice is an intra-articular injection. Myorelaxants are the drugs of choice either for acute contractions and/or contractures or are used to treat chronic pain. Another drug class used are antidepressants; they are used for chronic pain and in patients refractory to bite therapy. Anticonvulsants are drugs to treat neuropathic pain and thus chronic TMJ pain. Benzodiazepines are still drugs used in the treatment of chronic myofascial pain, however, they are drugs that need careful use and their usefulness also lies in alleviating sleep disturbances. In fact, the clinician should help himself with criteria such as DC/TMD for diagnosis and diagnostic classification. Therefore, in conclusion, the clinician's skill lies in identifying the type of dysfunction and knowing how to choose drugs also on the basis of the patient's other comorbidities. Therefore, the gnathological framing of the type of dysfunction is fundamental and helps the clinician to choose the appropriate drug therapy. In addition, pharmacological treatment must be supported by functional therapy, physiotherapy and behavioural therapy.
Background Temporomandibular disorders (TMD) are manifested by soreness in the jaw joint area and jaw muscles, clicks or creaks when opening or closing the mouth. All these symptoms can be disabling and occur during chewing and when the patient yawns or speaks. Several classes of drugs are used to treat symptoms. This review aims to assess which drug suits the different signs. Methods Pubmed, Web of Science and Lilacs were systematically searched until 01/02/2023. Clinical trials were selected that dealt with drugs used in temporomandibular dysfunction Results Out of 830 papers, eight studies were included. The Meta-Analysis with Continuous Outcomes with Pre-Calculated Effect Sizes resulted in the rejection that there is intergroup variability (p.0.74). Conclusions Treatment of orofacial pain is still a significant challenge for dentistry. We can conclude that there is no drug of first choice in the treatment of temporomandibular pain. However, the clinician must distinguish the type of pain and the aetioloic cause of the pain so that the patient can be treated and managed pharmacologically. Keywords
Acknowledgements Not applicable. Authors’ contributions Conceptualization, GM, RF, VR; methodology, GM, RF, and SC; software, GM, RF, and MDB; validation, GM, RF, and VR; formal analysis GM, RF, and GC; investigation, AB; data curation, GM, RF; writing—original draft preparation, GM, RF, and MC; writing—review and editing, GM, RF; visualization, MC, AB, GC; supervision, MC, AB, GC, and GM; All au-thors have read and agreed to the published version of the manuscript. Funding This research received no external funding. Availability of data and materials Data generated and analysed during study will be available from Rocco Franco upon reasonable request. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:43:46
BMC Oral Health. 2024 Jan 13; 24:78
oa_package/3f/7b/PMC10787959.tar.gz
PMC10787960
38218837
Background Since laparoscopic pancreatoduodenectomy (LPD) was first reported in 1993 [ 1 ], LPD has been increasingly adopted worldwide for the treatment of benign and malignant tumors surrounding the duodenum, ampulla, lower common bile duct and head of the pancreas [ 2 ]. However, bleeding, abdominal infection, and even potentially life-threatening pancreatic fistula remain as existing challenges [ 3 ]. Several studies have demonstrated that the postoperative pancreatic fistula (POPF) rate remains high, with a related mortality rate of 3–8%; thus, POPF is known as the Achilles’ heel of the Whipple procedure [ 4 – 6 ]. A variety of approaches to reduce the incidence of POPF due to transanastomotic stenting, fibrin glue use, pancreaticogastrostomy and pancreaticojejunostomy (PJ) have been developed by pancreatic surgeons [ 7 – 9 ]. However, none of these approaches can completely avoid POPF. Recently, it has been reported that Blumgart anastomosis, a well-accepted procedure among pancreatic surgeons, reduced the incidence of POPF by enhancing the adhesion between the pancreatic parenchyma and intestine [ 10 , 11 ]. However, precise needle handling and prevention of suture tangling during LPD are still needed. We herein introduce our modified Blumgart approach to secure these structures with adjustable adhesions using a homemade crochet needle in LPD.
Methods Study population We performed the modified Blumgart PJ technique using a homemade crochet needle for LPD in February 2019 at the Department of Hepatobiliary and Pancreatic Surgery, The Second Affiliated Hospital, Zhejiang University School of Medicine. The inclusion criteria were resectable benign and malignant tumors surrounding the duodenum, ampulla, lower common bile duct and head of the pancreas. The exclusion criteria were malignant tumors with distant metastases and invading the superior mesenteric vessels on preoperative radiologic evaluation. We applied the enhanced recovery after surgery (ERAS) pathway in the LPD from December 2017, whereas the standard perioperative care protocol was used before. We retrospectively collected clinicopathologic variables (sex, age, body mass index (BMI), pancreatic texture, pancreatic duct diameter, and histopathological findings), operative details (total operating time, time needed for PJ, and intraoperative bleeding volume), and postoperative hospitalization data (amylase level on postoperative days 3 and 5, incidence of postoperative complications, incidence of clinically relevant postoperative pancreatic fistula (CR-POPF including grades B and C pancreatic fistula), hospitalization length, and 90-day mortality rate). All laparoscopic procedures were performed by the same surgeons following the same criteria and using the same anastomosis technique. This study was approved by the Ethics Review Committee of the Second Affiliated Hospital of Zhejiang University School of Medicine. Surgical procedures Patient position and trocar distribution A supine position was adopted for the patient, with his or her legs spread and head elevated above the feet (at a 30-degree incline). The resection was performed with five trocars (Fig. 1 ): two 12-mm trocars (right and left upper quadrants), two 5-mm trocars (the right and left flank patterns) and one 10-mm trocar (umbilical). Self-made crochet needle Our homemade crochet needle consists of 4 − 0 prolene (Ethicon Inc, Somerville, NJ, USA), an injection needle with a diameter of 0.9 mm and a length of 80 mm (Zhejiang Kindly Medical Equipment Co., Ltd., China) and a frame constructed with 3 M Tegaderm transparent film (3 M company, USA). First, the two ends of the prolene thread were passed through the injection needle and fixed at the tail of the injection needle using a 3 M Tegaderm transparent film dressing. Finally, a closed circle with a circumference of approximately 3 cm was formed on the tip of the injection needle (Fig. 2 ). During our operation, the posterior wall of the two pancreaticojejunostomy U-shapes was fixed outside the body using a homemade crochet needle to adjust the tension at any time and reduce interference from the threads under laparoscopy, especially in obese patients. PJ procedure Preparation of the pancreatic stump and jejunal loop Typically, dissociation of 1–2 cm of the pancreatic stump borders was needed to perform PJ later (Fig. 3 A). The main pancreatic duct was verified by either performing careful visual inspection for thick ducts or using a slender tube for narrow ducts. The closer end of the jejunum loop was closed. The jejunal limb was moved to the right of the middle colic vessels in a retrocolic fashion, while the blind end was placed close to the pancreatic remnant. Pancreaticojejunal anastomosis A large 4 − 0 prolene suturing needle was used to vertically enter the pancreas stump 1 cm from the ventral side, extending out from the dorsal side of the edge of the pancreas (Fig. 3 B). After the needle has been inserted horizontally along the long axis of the jejunum, it is advanced 1 cm within the seromuscular layers of the jejunum. Next, the needle protrudes 1 cm from the pancreatic dorsal side to the ventral side. A U-shaped suture was created, and the needle was cut. Then, both ends of the thread were lifted out of the body with our homemade crochet needle, and the two threads were fixed with a vascular clamp to facilitate adjusting the tension between the pancreas and jejunum (Fig. 3 C). Furthermore, a similar second U-shaped suture encompassing the main pancreatic duct that extended between the pancreatic parenchyma and the jejunal seromuscular layer was created (Fig. 3 D), and both ends of the thread were also lifted out of the body with our homemade crochet needle (Fig. 4 ). Finally, a third U-shaped suture was created with both ends of the thread and fixed with a hemolock (Fig. 3 G). This maneuver has an advantage over the classical technique because at this point, the posterior faces of both the jejunum and pancreas are not yet sutured so the duct-to-mucosa (DTM) anastomosis can be made. The posterior semicircle sutures of the DTM were placed at the 4, 6, and 8 o’clock positions using 5 − 0 PDS II (Ethicon Inc, Somerville, NJ, USA) (Fig. 3 E). With traction on the externalized transpancreatic stitches, the jejunal loop can be reinserted into the pancreatic posterior face, while the stitches on the posterior face of the TMD served as anchors for the loop. It is usually necessary to place a stent into the pancreatic duct. Later, stitches were placed on the anterior face at the 10, 12, and 2 o’clock positions of the DTM in the same manner (Fig. 3 F). Once the DTM anastomosis was completed, the three U-shaped sutures were knotted sequentially. Then, a single layer of continuous sutures was made between the pancreatic stump and the anterior seromuscular layer of the jejunum using the 3/0 barbed suture Stratafix (Ethicon Inc, Somerville, NJ, USA) (Fig. 3 H and I). After PJ was completed, biliary and gastric reconstructions were sequentially performed. Two external drainage tubes were routinely placed around the hepaticojejunostomy and PJ. Postoperative management The drain output was recorded each day after the operation. The amylase level in the drainage fluid was measured on postoperative days 3 and 5 and at any time when POPF was suspected. One abdominal CT scan was routinely conducted on postoperative day 5. To prevent infection after surgery, broad-spectrum antibiotics and anti-anaerobic drugs were used for 72 h postoperatively. All patients received octreotide after the operation to decrease the volume of pancreatic external secretion. POPF was diagnosed and graded according to the definition from the International Study Group on Pancreatic Fistula (ISGPF) (2016 version). If the amylase content of any measurable drainage on or after postoperative day 3 was greater than 3 times the upper normal serum value or the drains were either left in place for > 3 weeks or repositioned endoscopically or percutaneously, a grade B POPF was considered. On the other hand, grade C POPFs were those that needed reoperation or led to single/multiple organ failure and/or mortality. The postoperative complications included delayed gastric emptying, abdominal hemorrhage, gastrointestinal anastomotic hemorrhage, bile leakage, infection, chylous fistula and mortality. On postoperative days 5, if the amylase content of any measurable drainage is lower than 3 times the upper limit of normal serum amylase and follow-up abdominal CT shows no sign of abdominal fluid in the operative region, the drainage tube can be removed. Statistical analysis Student’s t test was performed to compare continuous variables represented as mean ± standard deviation (SD). Categorical data are described as numbers (percentages) and were compared using the chi-square test or Fisher’s exact test. Furthermore, the differences between non-POPF group and POPF group were evaluated by t tests in the case of normally distributed variables or by chi-square test in the case of categorical data. Statistical significance was determined by a P value of 0.05. All analyses were conducted with SPSS 22.0 software (SPSS, Inc., Chicago, IL, USA).
Results Demographic and clinicopathological features of patients There were 96 patients who underwent LPD with the new technique, including 54 men and 42 women with a mean age of 63.38 ± 10.41 years (Table 1 ). The average BMI of the patients was 22.52 kg/m 2 . The leading indication for LPD was pancreatic ductal adenocarcinoma (n = 48, 50%), followed by duodenal papillary carcinoma (n = 21, 21.9%), distal cholangiocarcinoma (n = 12, 12.5%), ampullary carcinoma (n = 7, 7.3%), intraductal papillary mucinous neoplasm (n = 4, 4.2%), solid pseudopapillary tumor (n = 2, 2.1%), pancreatic neuroendocrine tumor (n = 1, 1.0%), and duodenal adenoma (n = 1, 1.0%). The average total operative time, intraoperative blood loss and postoperative hospital stay were 445.30 min, 198.43 mL and 13.68 days, respectively. The mean operation time for PJ was 66.28 min. Grade B POPF occurred in 13 patients (13.5%), while 1 grade C POPF was observed. One patient with postoperative abdominal hemorrhage was cured after reoperation to achieve homeostasis. Six patients (6.3%) suffered from chylous fistula, 4 patients (4.2%) suffered from delayed gastric emptying, 3 patients (3.1%) suffered from pneumonia, 2 patients (2.1%) suffered from bile leakage, 2 patients (2.1%) suffered from abdominal infection, and 1 patient (1.0%) suffered from gastrointestinal anastomotic hemorrhage, which were all treated with conservative therapy. There were no operative or in-hospital deaths. Comparisons between the non-POPF and POPF subtypes When the patients in these subgroups were compared (Table 2 ), the incidence of soft pancreas was higher in the POPF group than in the non-POPF group (P = 0.016). Furthermore, the incidence of pancreatic ductal adenocarcinoma or pancreatitis was lower in the POPF group than in the non-POPF group (P = 0.013). However, no significant differences were observed in age (P = 0.094), BMI (P = 0.575), operative time (P = 0.419), intraoperative bleeding (P = 0.610) or pancreatic duct diameter (P = 0.270). The fistula risk score of non-POPF group was significant lower than that of POPF group (3.18 ± 1.69 vs. 5.50 ± 1.34, P < 0.001).
Discussion In recent decades, LPD has been adopted in many medical centers for the radical treatment of both benign and malignant pancreatic and periampullary disease. To date, the mortality and morbidity of LPD have significantly declined, and the data show that patients who underwent LPD in high-volume centers achieved a better prognosis than those treated in low-volume centers [ 12 – 15 ]. However, the overall postoperative morbidity rate remains high, and the most fatal complication is POPF. The POPF rate in recent literature ranges from 10 to 29% [ 16 , 17 ], and this complication can also prolong patient hospitalization, increase mortality and increase costs. Some studies have revealed various risk factors for POPF, such as the PJ technique, pancreatic texture, and duct diameter. Moreover, many efforts have been made to decrease POPF incidence. In fact, many improvements to the PJ technique have been developed to minimize the rate of POPF. End-to-end and end-to-side approaches are the main approaches to PJ after pancreatoduodenectomy (PD) in most medical centers. Peng et al. reported a study that consisted of 150 patients who underwent PJ, with a 0% rate of POPF [ 18 ]. Although this method achieved the lowest POPF rate, it has not been repeated in subsequent foreign studies. Maggiori et al. reported a POPF rate of 36% using Peng’s technique [ 19 ]. Furthermore, duct-to-mucosa anastomosis has been improved in various ways. Karavias et al. reported a POPF rate of 7.9% after using their method called true duct-to-mucosa anastomosis. Although it was not mentioned in the report, the PJ time seemed prolonged because mucosal eversion was performed [ 20 ]. Interestingly, triple-layer duct-to-mucosa PJ was introduced by Su et al. with a POPF rate of 4%. The three layers included the pancreatic duct to jejunal mucosa, the pancreatic capsular parenchyma to the jejunal seromuscular and the pancreatic capsular parenchyma to the jejunal serosa [ 21 ]. However, despite the numerous PJ techniques, no prospective randomized controlled trials (RCTs) have been carried out to determine the best approach. Recently, Blumgart anastomosis, a new DTM anastomosis procedure well accepted among pancreatic surgeons, was reported to reduce the incidence rate of POPF [ 10 , 11 ]. However, this pancreatic anastomosis procedure has disadvantages that mainly limit the extent of LPD. First, multiple small sutures must be left untied, causing confusion in the field of vision. Second, it is difficult to create the posterior face of the DTM anastomosis when the posterior face’s capsular stitches have been previously tied [ 22 – 24 ]. Therefore, we applied a modified Blumgart method using a homemade crochet needle to facilitate PJ in LPD. With our proposed modification, the pancreatic duct and jejunal orifice are aligned correctly during DTM after the application of external traction through the homemade crochet needle. The space between the posterior wall of pancreatic remnant and jejunal loop can be exposed by adjusting the tension of the external threads, which can facilitate DTM. Moreover, there are few small sutures left untied, which makes the surgical field clearer. Finally, the homemade crochet needle does not leave scars after puncture and can be used for puncture at many places on the abdominal wall. In the present study, the CR-POPF rate (grade B-C) was only 14.6%, and 1 grade C POPF was observed using our PJ procedure; these rates are lower than those in most reported studies. Only one patient with postoperative abdominal hemorrhage was cured after reoperation, thereby achieving homeostasis. There were no operative or in-hospital deaths. Thus, this new PJ procedure is feasible for achieving a safe LPD. This study has several potential limitations due to its retrospective design. It is necessary to perform a prospective, randomized study that includes more patients and centers in order to validate the rate of pancreatic fistula following this type of anastomosis.
Conclusions We demonstrate the feasibility and safety of a modified Blumgart method using a homemade crochet needle to facilitate PJ in LPD. However, randomized controlled trials are needed to further verify the feasibility of the present PJ technique in LPD.
Background Among the safest procedures for anastomosis in pancreaticoduodenectomy, Blumgart pancreaticojejunostomy is associated with low rates of postoperative pancreatic fistula (POPF) and postoperative complications. However, this technique is difficult to perform during laparoscopic pancreaticoduodenectomy (LPD). This study presents a modified Blumgart method using a homemade crochet needle to facilitate laparoscopic pancreaticojejunostomy and evaluates its safety and reliability. Methods From February 2019 to October 2022, 96 LPD surgeries with the new technique were performed by the same surgeons in the Second Affiliated Hospital of Zhejiang University School of Medicine. The operative details (operative time, pancreaticojejunostomy time, POPF rate, postoperative complication rate, mortality rate) were analyzed along with clinical and pathological indicators (pancreatic duct diameter, pancreatic texture, and histopathological findings). Results There were 54 men and 42 women with a mean age of 63.38 ± 10.41 years. The intraoperative bleeding volume, operative time and postoperative length of hospital stay were 198.43 ± 132.97 mL, 445.30 ± 87.05 min and 13.68 ± 4.02 days, respectively. The operation time of pancreaticojejunostomy was 66.28 ± 10.17 min. Clinically relevant POPFs (grades B and C) occurred in 14.6% of patients. Only one patient had postoperative abdominal hemorrhage and was cured after reoperation. There were no operative or in-hospital deaths. With our proposed modification, the pancreatic duct and jejunal orifice are aligned correctly during duct-to-mucosa (DTM) after the application of external traction through the homemade crochet needle. The space between the posterior wall of pancreatic remnant and jejunal loop can be exposed by adjusting the tension of the external threads, which can facilitate DTM. Conclusions A modified Blumgart method using a homemade crochet needle could be technically feasible and safe during LPD. A randomized control trial is needed to confirm these findings. Keywords
Acknowledgements Not applicable. Author contributions BZ contributed to the conceptualization and design of the study and drafting of the manuscript. ZG and YT contributed to the collection and analysis of the data. SY helped supervise and interpret the data. BZ and SY conceived and supervised the whole project. All authors read and approved the final manuscript. Funding This work was supported by grants from the Zhejiang Province Medical and Health Science and Technology Program (No. 2023KY109) and the Zhejiang Province Basic Public Welfare Research Project (No. 2018C37115). Data Availability The datasets analyzed during the current study are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate Our study was approved by the Ethics Review Committee of the Second Affiliated Hospital of Zhejiang University School of Medicine. Written informed consent was obtained from all subjects in our study. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Abbreviations Laparoscopic pancreaticoduodenectomy Postoperative pancreatic fistula Pancreaticojejunostomy Body mass index Clinically relevant postoperative pancreatic fistula Duct-to-mucosa Standard deviation
CC BY
no
2024-01-15 23:43:46
BMC Surg. 2024 Jan 13; 24:22
oa_package/58/d0/PMC10787960.tar.gz
PMC10787961
38218815
BMC Urology (2023) 23:209 10.1186/s12894-023-01378-4 The author name “Yin Guonan” was incorrectly written as “Yin Gounan” and the same has been updated. The original article has been corrected.
CC BY
no
2024-01-15 23:43:46
BMC Urol. 2024 Jan 13; 24:15
oa_package/75/8a/PMC10787961.tar.gz
PMC10787962
38218864
Introduction Dirofilariasis is a parasitic zoonosis caused by nematodes of the genus Dirofilaria . There are over 20 dirofiliaris species, which mostly infect dogs, cats, and wild carnivores and are being transmitted by multiple mosquito species [ 1 ]. In these mosquito vectors, microfilariae mature into infectious larvae after which transmission takes place when the mosquito takes a blood meal. Canidae are the main reservoir but incidentally the nematodes can be transmitted to humans. In human beings, complete sexual maturation of the microfilaria cannot occur due to the host defense, preventing the expression of larvae in the blood stream. Therefore, no further transmission to other hosts takes place [ 2 ]. Most cases have been described of the species Dirofilaria repens and Dirofilaria immitis regarding human infections [ 3 ]. D. immitis causes deep organ infections, mostly in the pulmonary arteries and right ventricle of the heart, and D. repens mainly causes subcutaneous and ocular infection [ 3 ]. Most human infections by D. repens occur in the Mediterranean region and other subtropic and tropic places in Europe, Asia, and Africa [ 2 , 3 ]. Sri Lanka is the most afflicted country in Asia, where 30–60% of the dog population is infected with D. repens in some parts of the country [ 4 ]. The most reported symptoms of subcutaneous dirofilariasis are migrating subcutaneous lesions, intermittent painful erythema, and itching [ 5 ]. Subcutaneous dirofilariasis should be treated with complete surgical extirpation of the lesion. Antihelminthic drugs are usually only required when immunodeficiency is present, since a normal functioning immune system is able to prevent reproduction and transmission [ 2 , 5 ]. On ultrasound, subcutaneous dirofilariasis presents as a hypoechoic nodular lesion with an internal serpiginous structure with internal parallel echogenic walls and an anechoic center. Sometimes active motion of microfilariae is visible, similar to the “filarial dance sign” seen in lymphatic filariasis caused by filarial nematodes of the species Wuchereria bancrofti , Brugia malayi , and Brugia timori [ 6 ]. Human subcutaneous dirofilariasis by D. repens is scarcely reported in nonendemic areas such as Belgium, and in almost all cases the diagnosis is made after excisional biopsy, often with initial misdiagnosis, significant treatment delay, and medication error. To the best of our knowledge, this case is the first well-reported human subcutaneous dirofilariasis preoperatively diagnosed on ultrasound in Western Europe.
Discussion Human subcutaneous dirofilariasis is a zoonosis caused by infection due to nematodes of the genus Dirofilaria , with the majority caused by D. repens . They are transmitted by mosquito vectors and Canidae are the main reservoir. Incidentally, transmission to humans are dead ends in Dirofilaria infestation due to the fact that the worms do not attain maturity and are unable to express microfilaria in the blood stream. Therefore, serological tests are usually not useful and systemic therapy is not indicated. Eosinophilia is an inconsistent finding depending on immune response [ 7 ]. Subcutaneous dirofilariasis should be considered if a patient presents with a migrating subcutaneous nodule, also in nonendemic areas. Ultrasonography is the first-line imaging technique with high specificity showing a cystic nodule with an internal serpiginous structure consisting of parallel echogenic lines. Active twirling movement of the serpiginous structure can be seen, as described in lymphatic filariasis as the “filarial dance sign.” Magetic resonance imaging (MRI) can rarely be valuable if extension to the muscles or joints is expected [ 7 , 8 ]. The subcutaneous lesion should be treated with total surgical extirpation of the lesion. Anthelmintic treatment (e.g., albendazole) is not recommended in most cases but can be useful for immunocompromised patients or migratory lesions, especially in the face, because these drugs promote fixation, after which the lesion can be surgically removed [ 2 , 5 ].
Conclusion Despite the characteristic imaging features of subcutaneous dirofilariasis on ultrasound, these lesions are usually removed surgically with significant delay and without preoperative ultrasonographic investigation, especially in nonendemic areas. Early and correct diagnosis prevents significant patient distress and prevents medication error (e.g., inappropriate antibiotic use). Hopefully, this case report will create more knowledge of the characteristic ultrasound appearance of subcutaneous dirofilariasis and more awareness of the disease in general, leading to better patient care with early and correct diagnosis and treatment.
Background Subcutaneous dirofilariasis is a parasitic zoonosis commonly described in Canidae but rarely seen in humans. Most physicians are unfamiliar with this disease, especially in nonendemic areas, which can lead to medication error and diagnostic and treatment delay. To the best of our knowledge, no previous case of subcutaneous dirofilariasis preoperatively diagnosed on ultrasound has been described in Western Europe. Case presentation A 25-year-old Belgian male patient presented with a subcutaneous nodule in the epigastric region. Ultrasound investigation showed a typical cystic lesion with an internal serpiginous structure with echogenic lines, and there was active twirling movement of this serpentine structure during investigation, pathognomonic for subcutaneous dirofilariasis. Surgical extirpation was performed, and the diagnosis was histopathologically confirmed. Conclusion Subcutaneous dirofilariasis has a characteristic appearance on ultrasound but is not well known in nonendemic areas, often leading to diagnostic delay and initial incorrect treatment. More knowledge of this disease and of its characteristic ultrasound appearance will hopefully lead to better patient care. Keywords
Case presentation A 25-year-old native male patient from Flanders (Belgium) with no relevant medical history presented to the general practitioner (GP) with an infrasternal, clinically palpable nodule in the epigastric region with localized irritation and itching. A cyst or benign lymphadenopathy was suspected and initially no treatment was started but at the patient’s request, an ultrasound examination was conducted to exclude other pathology. Approximately 6 months before symptom onset, the patient had been in France and central Italy. The leukocyte count and eosinophil count was within normal limits. Symptoms started about 2 weeks before ultrasonography. On ultrasound, there was a cystic structure with a diameter of circa 8 mm with an internal tubular serpiginous structure with parallel echogenic lines (Fig. 1 A). During the investigation, there was active twirling movement of the serpentine structure, similar to the “filarial dance sign” seen in lymphatic filariasis. The lesion showed no vascularization during Doppler investigation, and there was no penetration through the fascia with an intact appearance of the underlying musculature (Fig. 1 B). The preoperative diagnosis of a subcutaneous dirofilariasis was made and surgical extirpation was performed (Fig. 2 ). The excisional specimen was send to the Institute of Tropical Medicine, where histopathological assessment confirmed a subcutaneous dirofilariasis caused by D. repens .
Acknowledgements We thank our patient for consenting for publication and providing detailed information regarding the case. Author contributions CV performed the ultrasound investigation of the patient and SK collected the data and wrote the manuscript. CV supervised the manuscript. Both authors read and approved the final manuscript. Funding No funding has been received for this article. Availability of data and materials All data analyzed during this study are included in this published article. Declarations Ethics approval and consent to participate Individual approval and consent to participate was obtained. Consent for publication Written informed consent was obtained from the patient for publication of this case report and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal. Competing interests The authors declare no conflicts of interests.
CC BY
no
2024-01-15 23:43:46
J Med Case Rep. 2024 Jan 14; 18:16
oa_package/af/7c/PMC10787962.tar.gz
PMC10787963
38218862
Background Acute appendicitis is a common abdominal emergency that requires prompt diagnosis and treatment. For over a century, open appendectomy was the only standard treatment for appendicitis. However, recent studies have challenged the necessity of surgery in uncomplicated cases of appendicitis, and nonoperative management (NOM) with antibiotics alone has emerged as a promising alternative [ 1 – 4 ]. Although appendectomy has long been considered the gold standard operative management (OM) for acute appendicitis, there is growing interest in NOM with antibiotics in both adults and children [ 5 ]. While nonoperative management may offer certain advantages over appendectomy, such as decreased morbidity and shorter recovery time, there are concerns regarding the efficacy and safety of this approach. For instance, nonoperative management may be associated with a higher rate of recurrent appendicitis and an increase in the duration of hospital stay [ 6 ]. Thus, it is important to evaluate the efficacy and safety of nonoperative management compared to appendectomy in uncomplicated cases of appendicitis. Despite years of experience performing surgery to treat uncomplicated appendicitis, there is still a shortage of data that can be used to compare NOM and OM, making the choice between the two more challenging. This systematic review and meta-analysis of RCTs purpose was to compare NOM and OM in terms of efficacy, costs, length of hospital stay, quality of life and complications in a population of adults.
Material and methods A systematic literature review was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and as outlined in a predefined protocol (PROSPERO 2023: CRD42023413780) [ 7 ]. Literature search strategy The PubMed, Scopus, and Cochrane Library databases and ClinicalTrials.gov, Google Scholar were screened without time restrictions up to November 23rd, 2023 using the Mesh major topic “appendicitis” and “surgery” and Mesh terms “appendectomy” and “conservative treatment”. The search query is available in the Additional file 1 . Articles without free full text availability were searched through the University of Milan digital library in order to realize a complete research. The bibliographies of potentially relevant studies that were identified were manually searched for additional studies. Additionally, all studies that cited the primary studies were screened for inclusion on Google Scholar. We did not apply language or publication status restrictions. Eligibility criteria The study selection criteria encompassed randomized controlled trials (RCTs) that investigated the comparison between antibiotic treatment and appendectomy in adult participants, presenting with uncomplicated acute appendicitis diagnosed either clinically or radiologically. Exclusion criteria consisted of non-randomized studies and studies that included patients with complicated appendicitis or children. Study selection Two investigators (FB, GB) performed the literature search independently with the aid of Rayyan systematic review software [ 8 ]. Cases of disagreement were resolved by a third investigator (LC). In cases where multiple reports were found for the same study, data from all reports were utilized as necessary, while ensuring that there was no duplication of study participants. Data extraction Data extraction was performed independently by two authors (F.B. and G.B.), with any discrepancies resolved through consultation with a third senior author (L.C.). Data were gathered and recorded in a digital database, including information on the baseline characteristics of the studies, including characteristics of patients as follows: exam blood test, Alvarado score [ 9 ], LOS, recurrence at 1 year, and efficacy of the treatment performed. Outcome measures Primary outcome measures Complication-free treatment success: the success of the initial treatment (nonoperative management or operative management) was evaluated based on an uncomplicated course, with no occurrence of postoperative complications (complications or recurrences for NOM; postoperative complications for surgical intervention) Treatment efficacy based on 1-year follow-up: the efficacy of nonoperative management (NOM) was defined as achieving a definitive improvement without the need for surgery within a median follow-up of 1 year. Lack of efficacy in the NOM group included two scenarios: the persistence of acute appendicitis during hospitalization (referred to as index admission NOM failure, characterized by non-resolving appendicitis with persistent or worsening symptoms during the primary hospital stay) and recurrence of acute appendicitis. For OM, efficacy is defined as the resolution of symptoms following surgical treatment. Postoperative complications: the analysis involved evaluating the number and rates of various postoperative complications. Secondary outcome measures The study analyzed the number and rates of patients treated with a laparoscopic approach in both groups. Total costs: This encompassed the overall medical and surgical costs associated with the primary hospital stay. Length of primary hospital stay: This refers to the number of days of inpatient admission during the initial hospitalization. Quality of life following antibiotic therapy (AT) and surgical therapy (ST) was assessed. Assessment of risk of bias To assess any potential bias in the studies included in the analysis, the researchers (F.B. and G.B.) utilized the risk of bias tool developed by the Cochrane Collaboration [ 10 ]. The studies were evaluated based on criteria such as selection bias, performance bias, detection bias, and attrition bias. A total risk of bias score was then determined based on these domains, with the levels categorized as low risk of bias, high risk of bias, or unclear risk of bias. Statistical analysis Data from the individual eligible studies were entered into a spreadsheet for further analysis. Review Manager (RevMan) (Version 5.4.1. Copenhagen: The Nordic Cochrane Center, the Cochrane Collaboration, 2011). Risk Ratio (RR) was calculated for discrete variables with 95% confidence intervals (c.i.) calculated using a Mantel–Haenszel random-effects model. Mean Difference (MD) were calculated for continuous variables with 95% c.i. using an inverse-variance random-effects model. Statistical significance was taken at P < 0.05 using two-tailed testing. Heterogeneity among the trials was determined by means of the Cochrane Q value and quantified using the I 2 inconsistency test [ 10 ]. Trial sequential analysis Cumulative meta-analyses of trials face a susceptibility to stochastic errors due to inadequate data and repetitive testing as the data accumulates [ 11 , 12 ]. Trial sequential analysis (TSA) was employed, for primary outcome measures, to evaluate the statistical robustness of the data in a cumulative meta-analysis. TSA served as a means to gauge whether the existing evidence was sufficiently conclusive. The adjusted required information size (RIS) was computed using a significance level (alpha) of 0.05 (two sided) and a power (1—beta) of 0.20 (corresponding to 80% power). This calculation involved a control group proportion derived from the outcomes of our meta-analysis for binary outcomes. The decision to seek additional evidence from additional trials can be determined by assessing whether the cumulative Z-curve crosses trial sequential monitoring boundaries (TSMB) or the futility zone. Trial sequential analysis version 0.9 beta ( http://www.ctu.dk/tsa ) was used for all these analyses [ 13 ].
Results Figure 1 displays the PRISMA flowchart. Eight RCT fulfilled the inclusion criteria and were included in the meta-analysis (publication dates 1995–2022). In total, 3213 patients were allocated to NOM ( n = 1615) or OM ( n = 1598). General characteristics of patients as reported in the studies are shown in Table 1 . Study characteristics There was a significant amount of heterogeneity observed among the studies included in the analysis, particularly in terms of the diagnostic criteria used to define uncomplicated appendicitis. Additionally, there was substantial heterogeneity found in the type of antibiotics administered, the duration of administration, and the various outcomes that were evaluated. Risk of bias Figure 2 shows the RoB (Risk of Bias) analysis, indicating the assessment of bias in the included studies. In terms of study quality assessment, the included RCTs exhibited varying levels of risk across different domains. Out of the 8 RCTs analyzed, 6 studies reported a low risk of selection bias as they adequately described random sequence generation and allocation concealment [ 14 , 17 – 19 , 21 , 22 ]. However, the risk of selection bias remained unclear in two studies, where insufficient information was provided [ 4 , 15 ]. Concerning attrition bias, two studies were deemed to have a high risk due to inconsistencies in the reported numbers in tables and text [ 14 , 15 ]. Additionally, two studies were identified as having a high risk of selective reporting due to the lack of predefined endpoints [ 15 , 19 ]. The meta-analysis portrays a robust picture with most of the included studies exhibiting a low risk of bias across crucial domains. This underscores the reliability of our results, affirming the study's overall credibility. A graphical representation of the risk of bias assessment is provided in Additional file 1 of the manuscript. In terms of potential publication bias, no significant indications were observed graphically, as evidenced by the funnel plots. For further details and visual representations, the funnel plots are available as Additional file 1 accompanying this paper. The risk of language and geographic bias in this study is deemed low, as the nature of the research conducted, and the comprehensive analysis undertaken help mitigate any potential skew toward specific languages or regions. Complication-free treatment success (Fig. 3 ) All studies included in the analysis provided data to evaluate the effectiveness of the treatments [ 4 , 14 , 15 , 17 – 19 , 21 , 22 ]. The results showed that antibiotic treatment had a significantly lower treatment efficacy rate (70.45%, 1066 of 1513) compared to appendectomy (84.49%, 1248 of 1477). The risk ratio (RR) was 0.80 (95% confidence interval 0.71 to 0.90, p < 0.00001), indicating a statistically significant difference between the two treatment approaches. Furthermore, a substantial level of heterogeneity was observed in the meta-analysis, with an I -squared value of 87%, suggesting significant variation among the included studies. Trial sequential analysis of 8 trials comparing NOM vs. OM for overall treatment efficacy. The cumulative Z-curve crossed the conventional boundary for benefit and required information size but did not cross the trial sequential monitoring boundary for benefit, suggesting that the current evidence is statistically significant but does not support a superiority of OM and further trials will not change this conclusion. A diversity adjusted required information size of 2805 patients was calculated (Fig. 4 ). Treatment efficacy at 1-year follow-up (Fig. 5 ) Seven studies included in the analysis provided data to evaluate the effectiveness of the treatments at 1-year follow-up [ 4 , 14 , 15 , 17 – 19 , 22 ]. The results showed that antibiotic treatment had a significantly lower treatment efficacy rate (64.51%, 540 of 837) compared to appendectomy (96.8%, 788 of 814). The risk ratio (RR) was 0.69 (95% confidence interval 0.61 to 0.77, p < 0.00001), indicating a statistically significant difference between the two treatment approaches. Furthermore, a substantial level of heterogeneity was observed in the meta-analysis, with an I -squared value of 81%, suggesting significant variation among the included studies. Trial sequential analysis of 7 trials comparing NOM vs. OM for treatment efficacy at 1-year follow-up. The cumulative Z-curve crossed the conventional boundary for benefit, the trial sequential monitoring boundary for benefit and required information size, suggesting that the current evidence is conclusive and further trials will not change this conclusion. A diversity adjusted required information size of 611 patients was calculated (Fig. 6 ). Length of primary hospital stay (Fig. 7 ) All studies reported LOS at index hospital admission [ 4 , 14 , 15 , 17 – 19 , 21 , 22 ]. The analysis showed that there was no statistically significant difference between antibiotic treatment and appendectomy in terms of their effect on the duration of hospital stay. The mean difference was − 0.58 days (95% confidence interval − 1.59 to 0.43, p = 0.26), indicating that the difference observed was not statistically significant. However, there was a high level of heterogeneity among the included studies, with an I -squared value of 99%, suggesting an important variability in the results across studies. Costs (Fig. 8 ) The pooled analysis of primary costs included 3 studies [ 15 , 17 , 20 ]. Overall, NOM resulted in significantly lower costs when compared to OM (sample size: 599; MD − 214.6; 95% CI − 218.51 − 210.69; P < 0.00001, I 2 = 0%). There was a low level of heterogeneity among the included studies, with an I -squared value of 0%, suggesting a negligible variability in the results across studies. Postoperative complications (Fig. 9 ) Eight studies reported post-treatment complications [ 4 , 14 , 15 , 17 – 19 , 21 , 22 ]. There was no statistically significant difference in the rate of post-treatment complications between participants treated with antibiotics (8.98%; 145 out of 1613) and those who underwent appendectomy (10.88%; 173 out of 1590). The risk ratio (RR) was 0.66 (95% confidence interval 0.41 to 1.04, p = 0.07), indicating that the difference observed was not statistically significant. However, there was a considerable level of heterogeneity among the included studies, with an I -squared value of 69%, suggesting some variability in the results across studies. Trial sequential analysis of 8 trials comparing NOM vs. OM for postoperative complications. The cumulative Z -curve did not cross both the conventional boundary and the trial sequential monitoring boundary but crossed the required information size, suggesting that there are no significant differences in terms of complications and further trials difficulty will change this conclusion. A diversity adjusted required information size of 787 patients was calculated (Fig. 10 ). Quality of life Three studies provided data regarding quality of life [ 17 , 20 , 21 ]. However, a pooled analysis could not be done because of numerous scales utilized to evaluate the outcome. In the Talan et al. trial, NOM patients had higher physical SF-12v2 scores than OM patients at the 2-week and 1-month follow-up intervals (median 54 vs. 44). On the contrary, individuals who had OM both at 2 weeks (median 58 vs 55) and at 1 month follow-up (median 56 vs 55) had higher scores for the mental SF-12v2. The study “A Randomized Trial Comparing Antibiotics with Appendectomy for Appendicitis” (CODA trial), in a single time point of 30 days following randomization, reported QoL using the EQ-5DTM (EuroQoL Group, Rotterdam, The Netherlands), demonstrating no difference between antibiotic therapy and appendectomy (mean 0.92; SD 0.13 vs. mean 0.91; SD 0.13). O’Leary et al. assessed quality of life (QoL) using the same scale but at four different points in time (one week, one month, three months, and twelve months after randomization). However, data were reported with participants divided into three groups (appendectomy, antibiotic treatment, and failed antibiotic treatment with subsequent appendectomy). When compared to the group that underwent successful antibiotic therapy, the appendectomy group's mean QoL at 12 months was substantially higher (mean 0.976; CI 0.962 to 0.990 vs. mean 0.888; CI 0.856 to 0.920).
Discussion This study, including 3213 patients and 8 RCTs [ 2 , 4 , 14 , 15 , 17 – 24 ], is, to our knowledge, the largest meta-analysis of randomized controlled trials conducted thus far encompassing an adult population. The results demonstrate that antibiotic therapy as a first-line treatment has a failure rate of 29.5% during the initial hospitalization, 35.4% at 1-year follow-up, a non-statistically significant difference in terms of length of stay (LOS), a comparable rate of complications and significantly lower costs compared to surgical treatment. Several meta-analyses over the previous years have highlighted that surgical treatment is associated with an increased rate of complications, such as the study by Podda et al. [ 25 ], published in 2019. On the contrary, two recent studies [ 6 , 26 ] did not observe a lower rate of complications in the conservatively treated group. Our study aligns with these latter findings. This is likely attributed to the higher number of laparoscopic appendectomies performed more recently. As compared to open technique, laparoscopic appendectomy has been shown to significantly reduce wound infection rates [ 27 ]. In our analysis the rate of laparoscopic appendectomies performed was 68.44%, as reported by 6 RCTs. Furthermore, recent trials included in our study predominantly analyzed laparoscopic appendectomies, with a percentage of 100% for a trial [ 14 ], 96% [ 21 ], and 90% [ 17 ], respectively. In contrast, previous studies, particularly the Antibiotic Therapy vs Appendectomy for Treatment of Uncomplicated Acute Appendicitis (APPAC) trial and the study conducted by Styrud et al., primarily consisted of open procedures. Another important factor that could influence these results is the presence of appendicoliths. In the CODA trial, participants who were randomized to antibiotic medication and had an appendicolith experienced problems with a rate of 14% compared to 2% in those who did not [ 21 ]. This latter trial and the study by Vons et al. [ 22 ] included patients with appendicoliths diagnosed by CT scan. The other trials had a heterogeneous diagnostic protocol, so several patients with appendicoliths may have remained unrecognized. In conclusion, we can affirm that NOM is safe, as it has a comparable rate of complications to laparoscopic appendectomy. However, there was heterogeneity in diagnostic assessment, antibiotic regimens and treatment duration among the various studies, which could impact the results. The higher number of laparoscopic appendectomies may have also influenced the outcome regarding LOS. It is well-established in the literature that LOS is shorter when the procedure is performed laparoscopically, leading to an equivalence in LOS with conservative treatment [ 27 ]. It was not possible to perform a subgroup analysis due to lack of the necessary data. However, it would be important, in the future, to have RCTs that perform totally laparoscopic appendectomies, as Ceresoli et al. did, or that perform a subgroup analysis to explore the differences between laparoscopic and laparotomy appendectomies for this outcome. However, this result is certainly influenced by significant heterogeneity in the implementation and management of conservative therapy. Indeed, there is inconsistency among the studies regarding the type of antibiotic used, the duration of intravenous administration, and subsequent oral administration. For example, in the CODA trial [ 21 ], an initial bolus, administered on the first day, was followed by oral therapy from the second to the tenth day. On the other hand, O’Leary et al. [ 17 ] continued that the antibiotic therapy until a clear clinical improvement of the patient was achieved. Regarding the results concerning the complication-free treatment success during the initial hospitalization, they significantly favor surgical intervention. The conservative treatment has an efficacy rate of 71.84% in the index hospitalization. Undoubtedly, a clear advantage of appendectomy is the ability to remove the pathogenic cause with a negligible risk of stump appendicitis [ 28 ]. Conversely, this is not possible with conservative treatment, which carries a significant risk of lifetime recurrence, estimated between 6.7% and 8.6% [ 29 ]. The treatment effectiveness assessed at one-year follow-up demonstrates a greater effectiveness of surgery compared to conservative treatment; this latter has an efficacy of 67.3% at one year compared to 97.4% for appendectomy. It is important, therefore, to determine whether a conservative treatment with a lower efficacy measured at one-year follow-up and with comparable rates of complications, can be considered acceptable and feasible as a first-line treatment. It is true that approximately one-third of patients experience a recurrence within the first year. However, according to the 5-year follow-up results of the APPAC trial, patients can be successfully treated again with antibiotic therapy, and if surgery is required, it does not appear to be associated with increased complications or technical difficulty. In fact, when Salminen et al. [ 30 ] published the 5-year follow-up findings of the APPAC randomized clinical trial in 2018, they addressed the issue of the paucity of research on the long-term clinical efficacy of antibiotics, which had previously been seen as one of the most significant barriers to the widespread adoption of NOM for uncomplicated appendicitis. Only 2.3% of patients undergoing surgery for recurrent appendicitis were found to have complicated forms of the disease and the overall complication rate was significantly lower in the antibiotic group than in the appendectomy group (6.5% vs. 24.4%, P = 0.001) among patients who were initially treated with antibiotics for uncomplicated appendicitis. Recently, Pàtkovà et al. have published a cohort study regarding the long-term outcomes of NOM [ 31 ]. This study drew patients from two RCTs included in this meta-analysis: Eriksson et al.'s study [ 4 ] published in 1995 and Styrud et al.'s study [ 19 ] published in 2006. The article concludes that over the course of two decades, more than half of the patients treated through NOM did not experience recurrences, and there is no evidence of long-term risks associated with NOM, except for the recurrence itself. The long-term follow-up confirmed the feasibility of NOM as a surgical alternative. It would be very important to have new RCTs that analyze the results of the comparison between NOM and OM in the long term, in order to draw more robust conclusions on the topic. Therefore, given these circumstances, an informed patient choice is crucial, in our opinion. In a study published by Hanson et al. [ 32 ] in 2018, 9.4% of the surveyed population responded that they would opt for nonoperative management (NOM) in the case of appendicitis. This number increased to 14.5% when asked about choosing for their children. The study focused on discussing the failure rates of NOM, and indeed, the authors themselves speculate that different numbers would have been obtained if the success rates were presented to patients. A more recent study, published in 2021 by Bom et al., presents very different results. Approximately half of the participants in the average population sample expressed a preference for antibiotics as a treatment for uncomplicated appendicitis, even if it entailed a higher risk of recurrence, in order to avoid surgery initially. Additional rigorous qualitative research will be necessary to investigate the factors behind the strikingly different outcomes observed in these two studies and to gain a deeper understanding of patient preferences in various situations. We are faced with two therapies that are equivalent in terms of safety, with one being less expensive, less effective, and non-invasive, while the other is more expensive, more effective, and invasive. Beyond the decision of which therapy should be considered first-line, the outcome that could matter the most is the patient's quality of life. Regarding this latter outcome, the diversity of presented results highlights the need for more literature. To establish more reliable analyses, it is crucial to use homogeneous scales across various trials. It is interesting to notice, despite the limitations outlined above, that in the three studies examined in one case, there is no difference in QoL between NOM and OM, and in the remaining two, the surgery appears to be associated with higher QoL. Interpreting these results for clinical application requires consideration of several limitations. The significant heterogeneity limits confidence, variations in intervention expertise and the broad timespan of included RCTs may introduce confounding factors. Our study encompassed RCTs spanning a significant time frame from 1995 to 2022. Over this period, there were significant advancements in surgical techniques, diagnostic imaging, and antibiotic selection, resulting in noticeable variations in treatment protocols across the included studies. These variations were evident in the use of different antibiotics and the progression of surgical techniques from predominantly open appendectomies to primarily laparoscopic procedures throughout the chronology of the included RCTs. These variations have resulted in a high degree of heterogeneity among the studies, which constitutes a significant limitation, potentially biasing the analysis.
Conclusions This meta-analysis and trial sequential analysis provide evidence that NOM with antibiotics is safe and, in the majority of cases, successful. NOM is equivalent to surgery in terms of complications and LOS while also incurring lower costs. While NOM's efficacy is lower than surgery, it does not seem to increase long-term complications. In relation to the three primary outcomes examined in our study, the evidence gleaned from current literature can be regarded as conclusive. It is highly unlikely that new RCTs focusing on these outcomes will substantially alter the existing body of evidence available to date. Thus, offering NOM and discussing its risks and benefits with the patient is reasonable based on this data. Further scientific efforts should be directed toward the attempt to provide surgeons with tools that allow the early identification of those patients who might respond adequately to NOM.
Background The aim of this study is to provide a meta-analysis of randomized controlled trials (RCT) comparing conservative and surgical treatment in a population of adults with uncomplicated acute appendicitis. Methods A systematic literature review was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines . A comprehensive search was conducted in MEDLINE, Embase, and CENTRAL. We have exclusively incorporated randomized controlled trials (RCTs). Studies involving participants with complicated appendicitis or children were excluded. The variables considered are as follows: treatment complications, complication-free treatment success at index admission and at 1 year follow-up, length of hospital stay (LOS), quality of life (QoL) and costs. Results Eight RCTs involving 3213 participants (1615 antibiotics/1598 appendectomy) were included. There was no significant difference between the two treatments in terms of complication rates (RR = 0.66; 95% CI 0.61—1.04, P = 0.07, I 2 = 69%). Antibiotics had a reduced treatment efficacy compared with appendectomy (RR = 0.80; 95% CI 0.71 to 0.90, p < 0.00001, I 2 = 87%) and at 1 year was successful in 540 out of 837 (64.6%, RR = 0.69, 95% confidence interval 0.61 to 0.77, p < 0.00001, I 2 = 81%) participants. There was no difference in LOS (mean difference − 0.58 days 95% confidence interval − 1.59 to 0.43, p = 0.26, I 2 = 99%). The trial sequential analysis has revealed that, concerning the three primary outcomes, it is improbable that forthcoming RCTs will significantly alter the existing body of evidence. Conclusions As further large-scale trials have been conducted, antibiotic therapy proved to be safe, less expensive, but also less effective than surgical treatment. In order to ensure well-informed decisions, further research is needed to explore patient preferences and quality of life outcomes. Supplementary Information The online version contains supplementary material available at 10.1186/s13017-023-00531-6. Keywords
Supplementary Information
Acknowledgements The authors acknowledge support from the University of Milan through the APC initiative. Author contributions FB, GB and JV wrote the main manuscript text and LC, FD, PF and CF prepared Figures 1, 2 and 3. PD and LA reviewed the manuscript. All authors reviewed the manuscript. Funding None to declare. Availability of data and materials Data-sharing requests will be considered by the management group upon written request to the corresponding author. If agreed, deidentified participant data will be available, subject to a data-sharing agreement. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests None to declare.
CC BY
no
2024-01-15 23:43:46
World J Emerg Surg. 2024 Jan 13; 19:2
oa_package/69/a6/PMC10787963.tar.gz
PMC10787964
38218797
Background After a chest trauma, patients often have a wide range of chest injuries. Therefore, injuries of the myocardium and valvular apparatus, if not clinically manifested immediately, can remain unrecognized for a long period of time. Traumatic tricuspid regurgitation is usually well tolerated in the acute phase, which is why surgical treatment of the tricuspid valve is performed much later than the onset of the injury. Transthoracic echocardiography (TTE) plays an important role in the diagnosis of valvular apparatus injuries, thus enabling early adequate treatment. A very small number of pericardial ruptures caused by chest wall trauma has been described in clinical practice, with various presentations. We present the case of a patient who was injured in a traffic accident, and manifested signs and symptoms of tricuspid valve and pericardial rupture with partial cardiac herniation 10 years later.
Discussion and conclusions Although chest injuries in a car accident are common, injuries of valvular structures are very rare (less than 1%) and usually present late [ 1 – 3 ]. The right ventricle is located just behind the sternum and therefore prone to injury, caused by the pressure forces to the front or back of the chest. The mechanism of tricuspid valve injury is usually due to a deceleration force transmitted to the chest and heart, especially if the force acted during late diastole, thus leading to a rapid increase in right ventricular intracavitary pressure, which may lead to the rupture of the papillary muscle or tendinous chords [ 1 , 4 ]. The mechanism of delayed rupture of the tricuspid valve is usually due to contusion of the papillary muscle, followed by haemorrhage, inflammation, and necrosis that can lead over time to rupture of the valvular apparatus [ 5 , 6 ]. Rupture of the papillary muscle usually presents acutely and is therefore treated very quickly surgically [ 7 ]. In contrast, rupture of tendinous chords has a much milder clinical course and often remains unrecognized after the injury [ 8 ]. Therefore, wide time periods are described in the literature during which the rupture of tricuspid valve was detected and corrected [ 7 – 10 ]. Our patient belongs to the group of late ruptures, with significant tricuspid regurgitation and signs of right ventricular failure. We may assume that the rupture of tricuspid chords occurred earlier (before the patient reported symptoms), so that tricuspid regurgitation (and volume overload) lasted longer and caused both - degeneration of the anterior leaflet and RV dilatation and dysfunction, leading to RV-related heart failure. When these symptoms became prominent, together with symptoms caused by the compression of the edges of the ruptured pericardium on the coronary arteries, the patient presented to the emergency department. As the anterior leaflet of tricuspid valve suffered significant degeneration due to loss of support and huge motions in large blood stream, it became shortened and thickened. Myocardial injuries, in addition to rupture of the valvular apparatus, may include myocardial contusion, rupture of a free wall or septum, and pericardial effusion. The highest percentage of traumatic injuries to the valvular apparatus was observed on the aortic and mitral valves, due to higher pressures in the left heart [ 11 ]. Echocardiography has a significant role, especially in patients with minimal clinical symptoms. This technique also serves to adequately describe anatomical disorders that occur after an injury, which is of great importance to the cardiac surgeon, in order to select an adequate surgical technique. A limitation of TTE is the fact that these patients usually have significant chest injuries, including haemothorax and pneumothorax, which makes their echocardiographic windows less adequate for interpretation compared to patients without chest injury. Prolonged haemodynamic instability of the patient prompts the physician to repeat the TTE examination or consider a TEE [ 7 ]. Both TTE and TEE may not be ideal in some cases [ 9 ]. In our patient, pericardial rupture with LV protrusion was not seen on echocardiography (probably because of the elastic forces of the LV wall), which indicates the necessity of other diagnostic procedures, such as chest computed tomography (CT) scan and cardiac magnetic resonance (CMR) imaging in symptomatic post-trauma patients, which is advised by other authors, as well [ 12 ]. The rupture of the pericardium in blunt chest trauma is also very rare [ 13 ]. Deceleration forces are usually responsible for the occurrence of pericardial defect, since the base of the heart is more fixed to the pulmonary vasculature and aorta, while the apex is more mobile, causing the rupture mostly on the lateral side of pericardium [ 13 ]. Pericardial rupture is seen in less than 0.5% of patients presenting after blunt trauma, and cardiac herniation through a pericardial defect is a potential complication of this injury [ 14 ]. In some occasions, herniation of the heart can be asymptomatic and go unrecognized [ 15 ]. On the other hand, major cardiac herniation can cause torsion of the great vessels, included inferior vena cava and strangulation of the herniated heart, causing cardiogenic shock and sudden death [ 14 , 16 , 17 ]. Also, if pneumopericardium occurs, air within a limited potential space can result in cardiac tamponade and hemodynamic instability [ 17 ]. Pericardial rupture is difficult to diagnose by echocardiographic techniques because of tiny structure of pericardium. Some indirect signs such as pneumopericardium or hemopericardium might be of help but could not prove definite diagnosis [ 16 ]. Chest CT scan enables timely recognition of pericardial rupture. The defect in the pericardium outlined by air may be directly visible on CT. If there is accompanying cardiac herniation, constriction by the pleuro-pericardial defect can be visible like a collar or waist [ 15 , 17 ]. Also, cardiac tamponade can be seen, as compression of the heart chambers by the air in the pericardial space which results in a small heart size [ 18 ]. CMR imaging plays an important role in the assessment of pericardial injuries and cardiac herniation. The best way to visualize the pericardium is by using T1 weighted imaging during systole [ 19 , 20 ]. This visualization method could make very good distinction between the pericardial and myocardial tissue. Besides that, CMR imaging is superior to CT because it generates motion pictures and can estimate regional wall motion abnormalities. These cine MR images could identify motions of the heart which is dislocated from the pericardial sac through the pericardial tear, indicating possible dynamic obstruction of the ventricles, as well as major blood vessels. However, even the CMR imaging has limitations. The parietal pericardium may be incompletely visualized, especially over left sided chambers, where pericardial rupture happens very often, because of scarcity of surrounding fat [ 21 ]. The cardiac herniation visualised by the CMR imaging is often intermittent and limited by the changes in the decubital position of the patient [ 21 ]. The question of when to operate the patient with traumatic tricuspid regurgitation that occurred during a chest injury remains open. The best results were achieved with early use of surgical techniques in patients with severe tricuspid valve regurgitation [ 22 ]. In contrast, in patients who were presented to a cardiac surgeon late, atrophy of the papillary muscle and chords and significantly increased amplitude of tricuspid valve leaflet movement were noted. Therefore, it is considered that surgical treatment of these patients, before the development of right ventricular failure, prevents further complications and maintains a stable sinus rhythm. Reparative techniques for treating traumatic tricuspid valve injuries today involve the use of synthetic materials to replace the ruptured chord or papillary muscle [ 22 – 24 ]. Traumatic injuries of the tricuspid valve and pericardium are often unrecognized in a timely manner, leading to late complications and right heart failure. Transthoracic and transoesophageal echocardiography play a crucial role in the recognition and proper treatment of these entities, though cardiac CMR may be needed in some cases. Early surgical treatment of unstable patients with severe tricuspid regurgitation prevents further complications and maintains a stable sinus rhythm.
Discussion and conclusions Although chest injuries in a car accident are common, injuries of valvular structures are very rare (less than 1%) and usually present late [ 1 – 3 ]. The right ventricle is located just behind the sternum and therefore prone to injury, caused by the pressure forces to the front or back of the chest. The mechanism of tricuspid valve injury is usually due to a deceleration force transmitted to the chest and heart, especially if the force acted during late diastole, thus leading to a rapid increase in right ventricular intracavitary pressure, which may lead to the rupture of the papillary muscle or tendinous chords [ 1 , 4 ]. The mechanism of delayed rupture of the tricuspid valve is usually due to contusion of the papillary muscle, followed by haemorrhage, inflammation, and necrosis that can lead over time to rupture of the valvular apparatus [ 5 , 6 ]. Rupture of the papillary muscle usually presents acutely and is therefore treated very quickly surgically [ 7 ]. In contrast, rupture of tendinous chords has a much milder clinical course and often remains unrecognized after the injury [ 8 ]. Therefore, wide time periods are described in the literature during which the rupture of tricuspid valve was detected and corrected [ 7 – 10 ]. Our patient belongs to the group of late ruptures, with significant tricuspid regurgitation and signs of right ventricular failure. We may assume that the rupture of tricuspid chords occurred earlier (before the patient reported symptoms), so that tricuspid regurgitation (and volume overload) lasted longer and caused both - degeneration of the anterior leaflet and RV dilatation and dysfunction, leading to RV-related heart failure. When these symptoms became prominent, together with symptoms caused by the compression of the edges of the ruptured pericardium on the coronary arteries, the patient presented to the emergency department. As the anterior leaflet of tricuspid valve suffered significant degeneration due to loss of support and huge motions in large blood stream, it became shortened and thickened. Myocardial injuries, in addition to rupture of the valvular apparatus, may include myocardial contusion, rupture of a free wall or septum, and pericardial effusion. The highest percentage of traumatic injuries to the valvular apparatus was observed on the aortic and mitral valves, due to higher pressures in the left heart [ 11 ]. Echocardiography has a significant role, especially in patients with minimal clinical symptoms. This technique also serves to adequately describe anatomical disorders that occur after an injury, which is of great importance to the cardiac surgeon, in order to select an adequate surgical technique. A limitation of TTE is the fact that these patients usually have significant chest injuries, including haemothorax and pneumothorax, which makes their echocardiographic windows less adequate for interpretation compared to patients without chest injury. Prolonged haemodynamic instability of the patient prompts the physician to repeat the TTE examination or consider a TEE [ 7 ]. Both TTE and TEE may not be ideal in some cases [ 9 ]. In our patient, pericardial rupture with LV protrusion was not seen on echocardiography (probably because of the elastic forces of the LV wall), which indicates the necessity of other diagnostic procedures, such as chest computed tomography (CT) scan and cardiac magnetic resonance (CMR) imaging in symptomatic post-trauma patients, which is advised by other authors, as well [ 12 ]. The rupture of the pericardium in blunt chest trauma is also very rare [ 13 ]. Deceleration forces are usually responsible for the occurrence of pericardial defect, since the base of the heart is more fixed to the pulmonary vasculature and aorta, while the apex is more mobile, causing the rupture mostly on the lateral side of pericardium [ 13 ]. Pericardial rupture is seen in less than 0.5% of patients presenting after blunt trauma, and cardiac herniation through a pericardial defect is a potential complication of this injury [ 14 ]. In some occasions, herniation of the heart can be asymptomatic and go unrecognized [ 15 ]. On the other hand, major cardiac herniation can cause torsion of the great vessels, included inferior vena cava and strangulation of the herniated heart, causing cardiogenic shock and sudden death [ 14 , 16 , 17 ]. Also, if pneumopericardium occurs, air within a limited potential space can result in cardiac tamponade and hemodynamic instability [ 17 ]. Pericardial rupture is difficult to diagnose by echocardiographic techniques because of tiny structure of pericardium. Some indirect signs such as pneumopericardium or hemopericardium might be of help but could not prove definite diagnosis [ 16 ]. Chest CT scan enables timely recognition of pericardial rupture. The defect in the pericardium outlined by air may be directly visible on CT. If there is accompanying cardiac herniation, constriction by the pleuro-pericardial defect can be visible like a collar or waist [ 15 , 17 ]. Also, cardiac tamponade can be seen, as compression of the heart chambers by the air in the pericardial space which results in a small heart size [ 18 ]. CMR imaging plays an important role in the assessment of pericardial injuries and cardiac herniation. The best way to visualize the pericardium is by using T1 weighted imaging during systole [ 19 , 20 ]. This visualization method could make very good distinction between the pericardial and myocardial tissue. Besides that, CMR imaging is superior to CT because it generates motion pictures and can estimate regional wall motion abnormalities. These cine MR images could identify motions of the heart which is dislocated from the pericardial sac through the pericardial tear, indicating possible dynamic obstruction of the ventricles, as well as major blood vessels. However, even the CMR imaging has limitations. The parietal pericardium may be incompletely visualized, especially over left sided chambers, where pericardial rupture happens very often, because of scarcity of surrounding fat [ 21 ]. The cardiac herniation visualised by the CMR imaging is often intermittent and limited by the changes in the decubital position of the patient [ 21 ]. The question of when to operate the patient with traumatic tricuspid regurgitation that occurred during a chest injury remains open. The best results were achieved with early use of surgical techniques in patients with severe tricuspid valve regurgitation [ 22 ]. In contrast, in patients who were presented to a cardiac surgeon late, atrophy of the papillary muscle and chords and significantly increased amplitude of tricuspid valve leaflet movement were noted. Therefore, it is considered that surgical treatment of these patients, before the development of right ventricular failure, prevents further complications and maintains a stable sinus rhythm. Reparative techniques for treating traumatic tricuspid valve injuries today involve the use of synthetic materials to replace the ruptured chord or papillary muscle [ 22 – 24 ]. Traumatic injuries of the tricuspid valve and pericardium are often unrecognized in a timely manner, leading to late complications and right heart failure. Transthoracic and transoesophageal echocardiography play a crucial role in the recognition and proper treatment of these entities, though cardiac CMR may be needed in some cases. Early surgical treatment of unstable patients with severe tricuspid regurgitation prevents further complications and maintains a stable sinus rhythm.
Background Although chest trauma happens very often, accompanying tricuspid valve injuries occur rarely and may be manifested by scarce symptoms and signs. Pericardial rupture with cardiac herniation is even a bigger rarity. Transthoracic echocardiography plays a key role in the diagnosis of valve injuries but is of limited value in cardiac herniation. Case presentation We present the case of 58-year-old man who experienced severe chest trauma in a car accident. Symptoms of right heart failure occurred 10 years after the injury, due to the loss of tricuspid leaflet support caused by the rupture of tendinous chords with significant tricuspid regurgitation. Intraoperatively, old posttraumatic pericardial rupture into left pleura was also found, with partial cardiac herniation and pressure of the edge of pericardium on all left-sided coronary arteries simultaneously. The patient was successfully operated and is free of symptoms 4 years later. Conclusions This case emphasizes the importance of timely diagnosis and underlines a mechanism that leads to delayed rupture of the tricuspid valve apparatus. Repeated echocardiography in all patients who experienced chest trauma could be of great importance. Also, given the limited value of echocardiography in posttraumatic pericardial rupture and cardiac herniation, cardiac computed tomography should be performed. Supplementary Information The online version contains supplementary material available at 10.1186/s12872-024-03716-2. Keywords
Case presentation A 58-year-old patient was admitted to the hospital due to chest pain, dyspnea and fatigue on physical exertion. The symptoms started two weeks before admission. His previous history was unremarkable, except for a car accident 10 years ago, with chest trauma and fracture of two ribs. On physical examination, the patient was cyanotic, with signs of right ventricular failure and pansystolic murmur at the lower left sternal border; the blood pressure was 110/80 mmHg, heart rate was 110/min. The electrocardiogram (ECG) showed a right bundle branch block (Fig. 1 ). TTE revealed flail of the anterior leaflet of tricuspid valve due to chordal rupture, leaving the anterior half of the leaflet completely unsupported. Also, there was a moderate tethering of the septal and posterior leaflet. (Fig. 2 , Supplementary material 1 and 2 ). Massive tricuspid regurgitation, registered by Colour Doppler (Fig. 3 , Supplementary material 3 ), was considered to be post-traumatic, based on the exclusion of other causes. Ebstein anomaly was excluded by echocardiography (the septal displacement index of the insertion of tricuspid to mitral leaflet was 0.76 cm/m2 (i.e. < 0.8 cm/m2). Other causes were ruled out by normal values of laboratory blood tests, including inflammatory markers and blood cultures (infective endocarditis, carcinoid), as well as the absence of any gastrointestinal symptoms, invasive cardiac intervention or radiation therapy. Detailed TTE and transoesophageal echocardiography (TEE) examination also revealed enlarged right ventricle (RV diastolic diameter was 5,1 cm from the PLAX view, compared to LV diameter of 4,8 cm), moderate reduction of right ventricular function, and TV annular dilatation (4,4 cm) with reduced motion amplitude. Left ventricular shape and dimensions were within normal limits and no wall motion abnormalities were observed, apart from paradoxical movements of the septum, caused by right ventricular overload. There was no significant pericardial effusion. (Fig. 4 , Supplementary material 4 ). After 10 days of intensive heart failure therapy, the patient was transferred to the cardiac surgery department for reconstruction of the tricuspid valve. Coronary angiography, performed prior to cardiac surgery, revealed the presence of an unusual finding of multiple dynamic stenotic lesions at the same level of all left-sided coronary vessels, predominantly on the first, second and third obtuse marginal branches of the circumflex coronary artery (Fig. 5 , Supplementary material 5 ). After medial sternotomy, an old pericardial rupture with thick edge was observed (Fig. 6 a). Pericardial opening was measuring 7 × 6 cm in size and the heart was herniated through the pericardial defect. The edge of the ruptured pericardium compressed the coronary arteries along the line, which was in accordance with the previous coronary angiographic findings. The heart was returned to normal position and the pericardial rupture was sutured. Intraoperatively, it was confirmed that the anterior leaflet of the tricuspid valve completely lost its support, due to the rupture of the main common chord at the level of papillary muscle (Fig. 6 b). Based on the measures calculated by transesophageal echocardiography, the CorMatrix patch was constructed and a tube was formed for the reconstruction of the tricuspid valve. The leaflets were then excised and the CorMatrix patch was sutured in three places to the bases of the papillary muscles and proximally to the tricuspid annulus. The specific tricuspid surgery with Cor Matrix patch was performed because the leaflet tissue was insufficient in size to cover the valve area and to perform neochordal implantation. On control transthoracic echocardiography, a tube connected to the annulus and the base of the papillary muscles was confirmed (Fig. 7 , Supplementary material 6 ). Mild residual tricuspid regurgitation persisted, with right ventricular systolic pressure 38 mmHg. After recovery, the patient was discharged home, without symptoms and signs of right heart failure. Electronic supplementary material Below is the link to the electronic supplementary material.
Acknowledgements Not applicable. Author contributions N.R. and M.P. participated in manuscript writing, literature search, data collection, critical revision. M.R.R. participated in data analysis, critical revision, literature search. I.B., O.P., N.L., E.K., A.D. and M.R. participated in literature search, data collection, and data analysis. D.M., a principal investigator of this study participated in manuscript writing, literature search, critical revision and gave final approval. All authors reviewed the manuscript. Funding There was no funding for this study. Data availability Please contact the corresponding author regarding data availability. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Informed consent was obtained from all subjects for publication of identifying information/images in an online open-access publication. Competing interests The authors declare no competing interests. Abbreviations Cardiac magnetic resonance computed tomography Electrocardiogram left ventricle parasternal long axis view right ventricle Transoesophageal echocardiographic examination Transthoracic echocardiography tricuspid valve
CC BY
no
2024-01-15 23:43:46
BMC Cardiovasc Disord. 2024 Jan 13; 24:44
oa_package/76/0c/PMC10787964.tar.gz
PMC10787965
38218849
Background Migration is now a common reality for tens of millions of people around the world. Among people who have left their country of origin to settle in another country, some portion will engage in problematic drug use [ 1 ], whether initiating in their country of origin or in the host country. Data on drug use in migrant populations are scarce and inconclusive. For example, a review identified eight studies comparing drug use in first-generation migrants to drug use in the general population of which two found higher levels in migrants, four found lower levels in migrants and two found different results depending on the type of drugs [ 2 ]. However, these studies were highly heterogeneous in their methodologies and the type of populations and drugs they included. Although, generally, migrant groups seem to have lower rates of substance use than host populations, several risk factors make them particularly vulnerable to engaging in problematic drug use [ 3 ]. Persons who migrated often face challenges in the country they settled in, including limited job opportunities and acculturation difficulties driven, in part, through poverty (as they tend to occupy lower socio-economic positions in society), language barriers, mental health problems, and the consequences of trauma [ 4 – 6 ]. The latter can stem from pre-migration traumatic experiences related to political conflict, war, or economic deprivation that prompted their migration, as well as the potentially traumatic migration journey itself [ 7 – 9 ]. Although exact numbers on the burden of drug use among migrants are inconclusive, recent global events have likely impacted the number of migrants using drugs in the European Union (EU). As a result of the war with Russia, millions of people from Ukraine, one of the countries with the highest levels of injecting drug use globally, have been entering the EU [ 10 ]. In addition, globally, migrants’ and refugees’ self-reported use of alcohol and drugs increased by 20% during the COVID-19 pandemic [ 11 ]. In March 2023, the European Monitoring Centre for Drugs and Drug Addiction (EMCDDA) recognized migration as a “megatrend”—a long-term driving force that will likely have significant future influence—in its work addressing the increasingly complex and dynamic drug markets in Europe through 2030 [ 12 ]. However, the EMCDDA European drug report 2023 only made a limited mention of drug use among migrants [ 13 ]. In summary, migrants with problematic drug use are a neglected group and should be provided with adequate support.
Each year, thousands of migrants enter the EU. Data on drug use in migrant populations are scarce and inconclusive. However, several risk factors make them particularly vulnerable to engaging in problematic drug use. In this perspective, we summarize the limited information that is available on migrants who use drugs and make a case as to why it is essential to improve access to health and social services, including harm reduction services, for this population. With this aim, we call for the co-creation of integrated services that better address the needs of migrants who use drugs in Europe. Keywords
Limited access to services for migrants who use drugs Providing adequate health and social services to migrants who use drugs (MWUD) is not only a matter of human dignity and sound public health, but also of their rights under European and international law [ 14 ]. However, MWUD face significant barriers when trying to access health and social services, including in European health systems. These barriers are often rooted in a range of factors, including cultural differences such as language and (mis)understanding of the health system [ 15 ], legal status [ 15 , 16 ], and discrimination [ 16 , 17 ]. By denying migrants access to health and social services, irrespective of drug use, these fundamental rights are violated and perpetuate inequity, stigma, and discrimination in addition to contributing to poorer health outcomes. Public health benefits of improved healthcare for migrant who use drugs Improving access to health and social services, including harm reduction services, for migrants who use drugs is crucial for promoting public health. MWUD may be at a higher risk of infectious diseases, such as HIV and viral hepatitis, due to unsafe drug use practices [ 18 – 20 ], poor living conditions and, in some cases, higher prevalence of these diseases in their country of origin. For example, in 2022, the prevalence of HIV in the EU increased, which has been largely attributed to refugees living with HIV that arrived from Ukraine [ 21 ]. By providing MWUD with access to harm reduction services, such as needle and syringe programs, opioid substitution therapy, and HIV and viral hepatitis B and C testing and treatment, the spread of infectious diseases will likely be reduced. Further, investment in harm reduction services is cost-effective in the promotion of public health [ 17 , 22 , 23 ]. Recommendations from civil society experts on improving health and social services MWUD may experience social marginalization or isolation due to their drug use and/or their migrant status [ 24 ]. This can result in stigma, discrimination, and exclusion from social, economic, and political opportunities. Further, unaddressed health and social needs may lead to the development or exacerbation of serious mental health conditions. Inclusive practices for treating vulnerable and marginalized groups can help improve the social and mental health of MWUD [ 25 ]. For example, including interpreters or cultural mediators in healthcare services improves the quality of care for patients [ 26 , 27 ]. Civil society and health experts working with migrants who use drugs in the European Union recently published recommendations in four areas as part of an EU-funded project “Services for vulnerable migrants who use drugs in the EU (SEMID-EU)” [ 18 ] (Table 1 ). Including the voices of migrants who use drugs The civil society and health experts agreed that addiction services available in EU countries are often not sensitive to the specific needs of migrants [ 18 ]. Further, a literature search conducted in 2022 revealed no published studies on the self-reported needs of MWUD [ 2 ]. One study from Norway, later published in 2023, interviewed MWUD about their drug use and help seeking barriers; however, this study only included six participants [ 28 ]. Despite the recommendations of many international organizations to involve people who use drugs in research and program and policy development, implementation of these efforts continues to stall [ 29 ]. Evidence shows that migrant involvement has a positive impact on research, service adaptations, policy dialogues, and the social and personal circumstances of migrants when they are involved [ 30 ]. To this end, the SEMID-EU project team conducted a community-based participatory research to identify and explore the specific needs of MWUD in the EU aiming to improve availability of and access to services for this population. A report including the findings from 98 interviews with MWUD with 45 different nationalities living in Amsterdam, Athens, Berlin, or Paris provided a nuanced overview of the interrelatedness between problems these populations are facing related to their migration background, drug use, and additional factors such as homelessness. In addition, it described barriers and good practices for accessing healthcare and harm reduction services, including the differences between settings and different migrant populations [ 31 ]. Call to co-create integrated services that better address the needs of MWUD Improving availability of and access to health and social services for all MWUD is essential for upholding human rights, promoting public health, and facilitating social integration and is especially urgent with the current number of migrants in the EU from Ukraine. Considering that the societal vulnerability of many migrants who use drugs leads to issues on multiple life domains, such as (mental) health problems, housing, and financial issues, an integrated, holistic approach is needed that offers support across these domains. To achieve these goals, policymakers must recognize the importance of providing MWUD with access to adequate services and work to eliminate the barriers that prevent them from accessing the care they need, while engaging migrants who use drugs at every step of the process.
Abbreviations European Union Migrations who use drugs Services for vulnerable migrants who use drugs in the EU Acknowledgements Authors acknowledge support to ISGlobal from the Spanish Ministry of Science, Innovation and Universities through the “Centro de Excelencia Severo Ochoa 2019–2023” Programme (CEX2018-000806-S) and from the Government of Catalonia through the CERCA Programme. Author contributions LvS and TMW wrote the first draft of this perspective and all co-authors provided feedback. Funding This paper was written as part of the project SErvices for vulnerable MIgrants who use Drugs in the EU (SEMID-EU) and funded by the European Union’s Justice Programme—Drugs Policy Initiatives (Grant number 101045837). Availability of data and materials Not applicable. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests JVL reports research grants to his institution from AbbVie, Gilead Sciences and MSD, and speaker fees from AbbVie, Gilead Sciences, Intercept, Janssen, MSD, Novo Nordisk, and ViiV, and an advisory board fee from AbbVie and Novavax, all unrelated to this work. CAP participated in a podcast financed by Gilead Sciences, unrelated to this work.
CC BY
no
2024-01-15 23:43:46
Harm Reduct J. 2024 Jan 13; 21:9
oa_package/af/99/PMC10787965.tar.gz
PMC10787966
38218809
Introduction Since the advent of echocardiography, researchers have focused on cardiac masses. These masses can be divided into non-neoplastic masses (such as thrombi, vegetations, calcifications, or other rare conditions), pseudotumors (lesions not originating from a neoplastic transformation of a specific cell type), benign tumors, or malignant tumors. Non-neoplastic masses account for 75% of all cases [ 1 , 2 ]. Although cardiac tumors are rare, primary cardiac tumors have a prevalence of 0.001–0.03%, while metastatic cardiac tumors occur 10–1,000 times more frequently (2.3–18.3%) [ 3 – 5 ]. Primary cardiac tumors are classified based on histological characteristics into benign or malignant. A previous study indicated that the distribution of cardiac tumors was 34% in the left atrium, 26% in the right atrium, 6% in the left ventricle, 7% in the right ventricle, and 27% in other locations [ 6 ]. Cardiac masses located in any chamber adjacent to large blood vessels or pericardium may require treatments, such as surgical removal or chemoradiotherapy, depending on the histopathological type, extent of invasion, and patient risk stratification [ 7 ]. Early detection and accurate differentiation of cardiac masses can lead to prolonged survival and improved quality of life for affected patients. Several imaging modalities are used to assess cardiac masses; these modalities include transthoracic echocardiography (TTE), transesophageal echocardiography (TEE), cardiac magnetic resonance (CMR), positron emission tomography (PET), computed tomography (CT) [ 8 ], CT-PET [ 9 ], etc. [ 10 – 12 ]. However, no guidelines or consensus have been established on the best diagnostic approach due to the diversity of cardiac masses. A recent comprehensive review suggests that TTE is typically the first choice for cardiac mass examination, and CMR provides high-resolution imaging for further evaluation if a mass is suspected. PET can be useful for staging malignancies and guiding biopsy location [ 13 ]. TTE is a valuable tool for determining the presence, size, shape, echogenicity, mobility, attachment point, and hemodynamic effects of cardiac masses and has a sensitivity of 93%. However, TTE may not be sufficient in some cases where image quality is suboptimal or the echoes are complex. Accurate differentiation between benign and malignant tumors using TTE can be challenging, with an accuracy of less than 70% [ 3 , 14 , 15 ]. To address these limitations, contrast echocardiography (CE) has emerged as a promising technology in recent years. Although published guidelines suggest that CE can improve image quality and aid in differentiating between benign and malignant lesions, most studies on CE diagnosis of cardiac masses are case reports or retrospective/small-sample-sized prospective cohorts [ 15 – 17 ]. The present study aimed to evaluate the diagnostic accuracy of CE in patients with suspected cardiac masses and address the insufficient evidence for differential diagnosis using CE.
Materials and methods This prospective study was conducted in four tertiary hospitals in China including First Affiliated Hospital of Guangxi Medical University, Hunan Provincial People’s Hospital, First Affiliated Hospital of University of South China and Xiangyang No. 1 People’s Hospital, Hubei University of Medicine. Study participants Adult patients who underwent TTE between April 2018 and July 2022 and were suspected to have cardiac masses were included in this consecutive cohort study. Exclusion criteria were allergies to albumin, blood products, or ultrasound enhancing agents. Patients with severe heart failure (New York Heart Association Class IV), severe arrhythmia, respiratory failure, severe liver or kidney dysfunction, or mental illness or epilepsy were also excluded. Echocardiographic image acquisition Each patient underwent echocardiographic examinations in the left lateral position by using a Philips iE33 ultrasound system (Philips Medical Systems, Bothell, WA, USA) and a TTE probe (S5–1, 1–5 MHz) by an echocardiographist with over 10 years of TTE experience at each center. All images and measurements were obtained in accordance with the echocardiography guideline [ 18 ]. Following the TTE examination, all patients underwent CE according to the most recent published guidelines [ 19 ]. CE protocol The study protocol was designed in accordance with the most recent guideline for CE [ 20 ]. Commercially available ultrasound enhancing agents (SonoVue; Bracco, Plan-Les-Ouates, Switzerland) were utilized during CE. The left ventricular opacification (LVO) mode was activated with a low mechanical index of 0.2 and 30-Hz frame rates. Subsequently, 0.8 mL of prepared ultrasound enhancing agents were rapidly injected via peripheral vein, followed by a slow (10–20 s) 3–5 mL saline flush as necessary to achieve optimal delineation of the left ventricular cavity and cardiac masses. Morphological and hemodynamic features of cardiac lesions were observed and digitally saved in this mode. The myocardial CE (MCE) mode was activated with a very low mechanical index of 0.08 and 30-Hz frame rates. After the left ventricle and myocardium were filled, the ultrasound enhancing agents were continuously infused with a dedicated Vueject R syringe pump (Bracco, Milano, Italy) at a rate of 1 mL/min. Intermittent flash technique (high mechanical index of 1.0) was employed to destroy the microbubbles. High mechanical index ultrasound impulse was transmitted between 5 and 10 frames to destroy the microbubbles. Perfusion was verified after contrast replenishment following the impulse to prevent false-positive readings caused by saturation artifact. The imaging results of the masses and adjacent normal myocardium before and after the flash were saved. Echocardiographic image analysis The study used qualitative and quantitative analyses. For patients with an echocardiographic suspicion of cardiac masses, qualitative analysis included observing echogenicity (uniform/non-uniform), boundary (well-demarcated/not well-demarcated), base morphology (narrow with peduncle/narrow with notch/broad), mass perfusion (no perfusion/mild perfusion/intense perfusion) [ 21 ], motility (absent/present), and pericardial effusion (absent/present) [ 22 , 23 ]. Two physicians with 10 years of experience in echocardiography jointly made a diagnosis based on the above qualitative indicators. Quantitative analysis was conducted using QLAB software (version 13.0; Philips Medical Systems, Andover, MA, USA). The area of the masses was measured when the long maximum diameter was visible, and the peak intensity of the masses and adjacent myocardium were measured as A1 and A2, respectively, with a ratio of A1 to A2 to differentiate between malignant and benign tumors (Figs. 1 , 2 and 3 ). Follow-up and validation All patients were prospectively followed up until March 1, 2022 to determine all-cause mortality by reviewing their medical records, conducting telephone interviews, and performing outpatient examinations every 6 months. Three types of cardiac masses were identified. (I) Pseudomass is defined as a variant or prominent normal structure, including Eustachian valve or Chiari network, Crista terminalis, and Coumadin ridge. The diagnosis was confirmed using CMR, and no morphological changes were observed in follow-up imaging [ 24 ]. (II) Thrombus is defined as a distinct mass of echoes visible throughout systole and diastole. The diagnosis was confirmed based on one of the following two criteria: (i) a significant reduction in size or complete resolution after anticoagulation therapy, with confirmation of thrombus upon follow-up TEE or CT; or (ii) pathological confirmation [ 25 ]. (III) All tumors were confirmed by surgery or biopsy and classified as benign or malignant based on the 2015 World Health Organization classification of tumors of the heart and pericardium [ 26 ]. Statistical analysis Continuous parameters were presented as mean ± standard deviation, while non-normally distributed parameters were shown as median (interquartile range, IQR). Independent sample t-test was used to evaluate differences in continuous parameters among groups, while Mann–Whitney U test was used for non-normally distributed parameters. Pearson’s Chi-squared test or Fisher’s exact test was used to compare categorical parameters among groups. Receiver operating characteristic (ROC) analysis was used to determine the differentiating capacity of variables for cardiac masses. Youden’s J statistic was used to identify the optimal cut-off value. The area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. The level of statistical significance was set at P < 0.05.
Results Population characteristics Between April 2018 and July 2022, a total of 49,354 TTEs were performed at six departments, of which 153 (0.31%) examinations were conducted on patients with suspected cardiac masses. Eight patients with allergic constitution refused CE (Fig. 4 ). A total of 145 patients with a median age of 59.4 years (IQR: 51.2–63.9 years) and including 90 (62.0%) men were enrolled. Table 1 summarizes the baseline demographic and clinical characteristics of all patients. Of the 145 patients, two did not have any cardiac masses, four had a cardiac pseudomass, 43 had a cardiac thrombus, 66 had a benign tumor, and 30 had a malignant tumor. These findings indicated that the history of previous cardiovascular disease and malignancy varied significantly among the four groups. Three cases of cardiac pseudomass were attributed to the hypertrophy of the interatrial septum, while one case was due to the hypertrophy of the papillary muscle. Anticoagulation therapy was administered to all patients diagnosed with a cardiac thrombus, and none underwent pathological analysis. Solitary thrombi were observed in all cases. Among the 43 patients with thrombi, 72.1% (31/43) experienced dissolution, and 27.9% (12/43) had a significant reduction in thrombus volume. Benign tumors were confirmed through surgery (60/66) and biopsy (6/66). Malignant tumors were confirmed through surgery (9/36) and biopsy (27/36). Following the administration of contrast enhancement (CE), two investigators diagnosed cardiac masses based on qualitative characteristics such as echogenicity, boundary, base, mass fusion, motility, and pericardial fusion. The results revealed that out of 145 cardiac masses, 140 were consistent with the final diagnosis, yielding a diagnostic accuracy of 96.6%. Of the 140 consistent diagnoses, 2 cases had no mass, 4 were pseudomasses, 43 were thrombi, 63 were benign tumors, and 28 were malignant tumors. However, instances of misdiagnosis were recorded. Three benign tumors erroneously identified as malignant tumors, and two malignant tumors were misclassified as benign tumors (Fig. 4 ). Comparison and differentiation of cardiac tumors from thrombi The tumor group exhibited a significantly larger area, higher rate of non-uniform echogenicity, wider base, higher perfusion intensity, and higher A1/A2 compared with the thrombus group ( P < 0.05, Table 2 ). When the cut-off value for A1/A2 was set to 0.499, the AUC for A1/A2 was 0.977 (95% CI: 0.947–1.000). The sensitivity, specificity, accuracy, PPV, and NPV were 0.979, 0.884, 0.957, 0.959, and 0.957, respectively (Fig. 5 ; Table 3 ). Comparison and differentiation of malignant tumors from benign tumors Compared with the benign group, the tumor group exhibited a larger area, a higher rate of non-uniform echogenicity, an indistinct boundary, a wider base, presence of motility, and higher A1/A2 ( P < 0.05, Table 4 ). The AUC for A1/A2 was 0.950 (95% CI: 0.894–1.000) when the cutoff value was set to 1.58. The sensitivity, specificity, accuracy, PPV, and NPV were 0.933, 0.939, 0.938, 0.875, and 0.969, respectively (Fig. 6 ; Table 3 ).
Discussion This study reports that the diagnostic performance of CE is notable in patients with suspected cardiac masses. CE exhibited high sensitivity and specificity in distinguishing cardiac tumors from non-neoplastic cardiac masses. It outperformed conventional TTE and showed comparable accuracy with pathological analysis in discriminating between malignant and benign tumors. Cardiac masses pose a significant threat to patients, and improving their diagnostic efficiency is an essential objective for radiologists and cardiologists. The management of the diagnostic approach is also important for clinicians [ 4 , 14 ]. Currently, TTE, TEE, cardiac CT and CMR are commonly used in diagnostic procedures. In the approach to cardiac masses, some echocardiographic parameters could provide good diagnostic accuracy if integrated in weighted and not weighted scores [ 9 , 27 ]. CMR is the subject of intense research and exhibits excellent accuracy in differentiating cardiac thrombi from tumors and distinguishing between benign and malignant neoplasms in various retrospective and prospective studies. Although prospective studies have highlighted useful imaging characteristics such as tumor size, invasiveness, irregular border, and late heterogeneous gadolinium enhancement, they have been limited to qualitative or semi-quantitative analysis. Therefore, a diagnostic imaging technique with quantitative parameters should be developed to ease the burden on CMR and reduce the workload of pathologists [ 10 , 12 , 23 ]. TTE is still the primary imaging modality used to evaluate cardiac masses [ 15 ]. However, conventional TTE has limitations in accurately assessing the characteristics of cardiac tumors, particularly in differentiating benign from malignant tumors; it also heavily relies on the experience of the radiologist and cardiologists [ 28 ]. To overcome this limitation, CE has become an essential part of echocardiography because it can improve the accuracy of left ventricular ejection fraction measurement and provide quantitative analysis of cardiac masses. Kirkpatrick et al. [ 29 ] demonstrated the diagnostic utility of A1 and A2 values by using CE in cardiac masses in 2004; subsequent studies provided evidence for the differential diagnostic value of A1/A2. For example, Xia et al. [ 30 ] found a significant difference in A1/A2 between malignant and benign tumors, while Mao et al. [ 31 ] revealed that A1/A2 > 1 had a high diagnostic accuracy in differentiating benign masses from malignant metastatic tumors in a cohort study. Differentiation between cardiac tumors and thrombi In this study, it was discovered that CE demonstrated remarkable accuracy in diagnosing intracardiac thrombi. When diagnosing a thrombus, setting A1/A2 with a cut-off value of 0.499 exhibited a specificity of 97.9% and a sensitivity of 88.4%. The A1/A2 value for the majority of thrombi was close to zero. However, in three cases, the A1/A2 values were significantly higher (1.39, 1.62, and 1.47) possibly due to fresh thrombi. In previous research, the loose texture of fresh thrombi and the ability of ultrasound-enhancing agents to enter from the periphery during CE resulted in a higher A1/A2. By contrast, old thrombi have a dense texture, and microbubbles of the ultrasound-enhancing agents cannot penetrate; as such, the A1/A2 values are near zero. Differentiating between fresh and old thrombi is crucial because fresh thrombi are more easily removed and less attached to the left ventricular wall; this structure makes them more brittle due to their collagen-poor organization. As a result, careful evaluation of the risk of fresh thrombus shedding is necessary. Uenishi et al. [ 32 ] demonstrated another perfusion phenomenon, where ultrasound-enhancing agents did not penetrate the interior of the thrombus (81.8%, 27/33) or remained only at the periphery (12.1%, 4/33). They also found that the agents typically perfused the periphery (44.7%, 21/47) or even the entire cardiac tumor (48.9%, 23/47). However, additional samples are required to confirm the perfusion patterns observed in the present study and the findings of Uenishi et al. in the future. Differentiation between cardiac malignant tumors from benign tumors This study demonstrates that CE could effectively distinguish between most benign and malignant tumors through qualitative and quantitative diagnostic methods. CE can enhance image quality and assess blood supply within tumors. Malignant tumors typically have abundant blood supply, and benign tumors have a sparse blood supply [ 33 ]. In previous studies, an A1/A2 cut-off value of 1.0 was utilized to differentiate malignant from benign tumors [ 29 , 34 , 35 ]. However, some benign tumors may have an A1/A2 value that is close to or slightly higher than 1, such as 1.32 in hemangioma, 1.08 in rhabdomyoma, 0.84 in fibroma, and 0.92 in hemangioma, and 1.06–1.15 in myxoma. Some malignant tumors contain necrotic tissues, which can result in an A1/A2 value of less than 1, as seen in 3.6% of cases in the study of Mao et al. The present study suggests that a cut-off value of 1.58 is better than 1 for differentiating malignant from benign tumors by using A1/A2. For less experienced radiologists, using A1/A2 with a cut-off value of 1.58 would result in good diagnostic accuracy. The size of the tumor area is beneficial in distinguishing between malignant and benign tumors, consistent with prior research. This study possesses several strengths, including a novel diagnostic approach for distinguishing cardiac masses, a prospective study design, and a relatively large sample size. The use of a simple, rapid, and highly reproducible quantitative parameter (A1/A2) can greatly assist in clinical diagnosis, particularly for radiologists without extensive experience in using TTE to diagnose cardiac masses [ 36 ]. Other modalities In addition to TTE, transesophageal (TEE) echocardiography, cardiac CT, CMR, and 18Ffluorodeoxyglucose (18 F FDG)-PET have a complementary and mutually reinforcing role in assessing cardiac masses. TEE can serve as a valuable tool in diagnosing cardiac masses. Previous studies demonstrated that ultrasound-enhancing agents can enhance the diagnostic accuracy of cardiac thrombi during TEE for patients with atrial fibrillation [ 6 ]. Xia et al. also reported that combining TEE with CE can detect suspected cardiac masses and had an accuracy of 97.8–100%, particularly in distinguishing between benign and malignant lesions [ 30 ]. With the availability of various tissue characterization imaging sequences, CMR has distinctive advantages in noninvasively diagnosing cardiac masses. In a recent study involving 213 pediatric cardiac masses, CMR demonstrated the following diagnostic accuracies for cardiac tumors: 94% for fibromas, 71% for rhabdomyomas, and 50% for myxomas [ 37 ]. Cardiac CT may serve as an alternative to CMR, particularly in cases where other imaging techniques are non-diagnostic or contraindicated [ 38 ]. Cardiac CT is particularly useful for evaluating calcified masses compared with other imaging modalities. Previous studies demonstrated that the diagnostic accuracy of cardiac CT in predicting the malignant nature of cardiac masses could be more than 90% [ 9 ]. However, the use of cardiac CT has some limitations, including radiation exposure, a low risk of contrast-induced nephropathy, and a restricted soft tissue and temporal resolution in comparison with magnetic resonance imaging. Studies have suggested that cardiac CT can distinguish between cardiac tumors and thrombi [ 39 ]; however, further research with a large sample size is required to confirm this finding. 18 F-FDG PET/CT is confirmed as an extremely powerful tool to provide substantial information regarding the nature of cardiac masses. A recent study reported that the accuracy of 18 F-FDG PET/CT in predicting the benign or malignant nature of cardiac masses exceeds 91%. In particular, the study emphasized the value of PET in cases with inconclusive diagnoses following cardiac CT, specifically among patients exhibiting three or four abnormal CT findings. In these instances, the presence of all PET parameters below the specified cut-off values indicates a benign mass, while the identification of at least one abnormal PET characteristic reliably indicates malignancy [ 9 ]. Limitations This study has several limitations that need to be addressed. First, the participating hospitals were tertiary, which may have introduced selection bias because secondary hospitals typically treat thrombi with a well-demarcated boundary and low echocardiographic suspicion. Second, more cases of pseudomass should be included in future analyses because their low representation in the current study resulted in limited conclusions. Third, the recruitment period was short, and long-term follow-up may be necessary to determine whether A1/A2 can predict the prognosis for patients with cardiac tumors. Finally, the analysis did not explore the diagnostic performance of CE by less experienced radiologists, which may be an underlying confounder in this study.
Conclusions CE can be a promising tool in accurately differentiating cardiac masses by combining qualitative and quantitative analyses. However, additional studies with larger sample sizes are needed to validate these findings due to the limited sample size and the potential for underlying confounders.
Background Cardiac masses can encompass a variety of conditions, such as tumors, thrombi, vegetations, calcific lesions, and other rare diseases. Treatment and management of these types of cardiac masses differ considerably. Thus, accurately distinguishing among thrombi, benign tumors, and malignant tumors in the heart is of great importance. Contrast echocardiography (CE) has emerged as a promising technology. Although published guidelines suggest that CE can enhance image quality and assist in differentiating between benign and malignant lesions, most studies on CE diagnosis of cardiac masses are limited to case reports or retrospective/small-sample-sized prospective cohorts. This study aims to evaluate the diagnostic accuracy of CE in patients with suspected cardiac masses and address the insufficient evidence for differential diagnosis using CE. Methods Between April 2018 and July 2022, a prospective multicenter study was conducted, which included 145 consecutive patients suspected to have cardiac masses based on transthoracic echocardiography. All patients underwent CE examinations. The echocardiographic diagnosis relied on qualitative factors such as echogenicity, boundary, morphology of the base, mass perfusion, pericardial effusion, and motility as well as quantitative factors such as the area of the masses and the peak intensity ratio of the masses to adjacent myocardium (A1/A2). Results The final confirmed diagnoses were as follows: 2 patients had no cardiac mass, 4 patients had pseudomass, 43 patients had thrombus, 66 patients had benign tumors, and 30 patients had malignant tumors. The receiver operating characteristic (ROC) analysis indicated that an optimal A1/A2 cutoff value of 0.499 distinguished a cardiac tumor from a thrombus, with AUC, sensitivity, specificity, PPV, and NPV of 0.977, 97.9%, 90.7%, 95.9%, and 95.1%, respectively. The optimal A1/A2 cutoff value of 1.583 distinguished a cardiac tumor from a thrombus, with AUC, sensitivity, specificity, PPV, and NPV of 0.950, 93.3%, 93.9%, 87.5%, and 96.9%, respectively. Conclusions Combined with qualitative and quantitative analyses, CE has the potential to accurately differentiate among different types of cardiac masses. Keywords
Acknowledgements The authors wish to thank Manying Xie, radiologist, for the ultrasound image analysis, Sufen Zhou and Ling Gan for the excellent technical assistance, and Caihong Chang and Hongzhi Guo for their cooperation in organizing cardiac mass data. Author contributions ZJQ and HZ designed this project, QTW wrote the first draft of this article, and completed Figs. 1 , 2 and 3 . BW revised this article and completed Figs. 4 , 5 and 6 . XFZ, XZ, and SC, as multicenter units, provided data on cardiac tumors according to operating standards. YJB, LJ and QQY jointly completed Tables 1, 2, 3 and 4 and assisted in data analysis.All authors reviewed the manuscript. Funding This work was supported by the Key R&D foundation of Hubei Province and Science (2022BCE004), the Clinical Medical Technology Innovation Guidance Project of Hunan Provincial Department of Science and Technology (2021SK50921), and the Technology Foundation of Xiangyang (2021YL27 and 2022YL26A). Data availability The data used to support the findings of this study are available from the corresponding author upon request. Declarations Ethics statement and consent to participate This study was approved by the Ethics Committee of Xiangyang No. 1 People’s hospital (2018KYLX61). All methods were performed in accordance with the relevant guidelines and regulations compatible with the Declaration of Helsinki. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Abbreviations Area Under Curve Contrast echocardiography Cardiac magnetic resonance Left ventricular opacification myocardial contrast echocardiography Positive predictive value Negative predictive value Receiver operating characteristic curve transesophageal echocardiography Transthoracic echocardiography
CC BY
no
2024-01-15 23:43:46
BMC Cardiovasc Disord. 2024 Jan 13; 24:43
oa_package/ab/a6/PMC10787966.tar.gz
PMC10787967
38218800
Introduction Critical obstetric hemorrhage is a significant cause of maternal mortality, currently accounting for 27% of maternal deaths worldwide and is the leading cause of maternal mortality in Japan, accounting for 22% of maternal deaths [ 1 , 2 ]. Because disseminated intravascular coagulation (DIC) can easily complicate even moderate amounts of blood loss, especially in cases of obstetric hemorrhage with underlying disease [ 3 ], a decision must be made to perform a pregnancy-related hysterectomy at the appropriate time. In particular, the proportion of pregnant women with an underlying condition of placenta accreta spectrum (PAS) has continued to increase with the recent increase in cesarean section deliveries, and the associated number of PAS cesarean hysterectomies has also increased. These findings are evident in data from two recently published large multinational cohort studies [ 4 ], in which PAS cesarean hysterectomies performed as the final step in a management protocol for massive hemorrhage associated with PAS disorders are associated with considerable maternal morbidity and mortality reportedly high. We need to be aware of this condition and the difficulties in its diagnosis and management and be prepared for the surgical procedure and its management. Despite the increasing trend, the occasions when PAS cesarean hysterectomy must be performed are extremely rare. Therefore, many surgical procedures and general management algorithms have been proposed by quality centers with a multidisciplinary approach, but unfortunately, not all of them are based on pathologically confirmed cases of PAS [ 5 ]. Despite the high morbidity and mortality associated with hysterectomy, many studies unanimously suggest that it should be performed under multidisciplinary management. Recently, a multidisciplinary management algorithm has gained attention, proposing the involvement of gynecologic oncologists in the surgical management of PAS cesarean hysterectomies. The rationale behind this proposal is that the changes occurring in the female reproductive system during pregnancy add complexity to PAS cesarean hysterectomies [ 6 ]. The purpose of this study, which is a single-center, retrospective study, is to evaluate the diagnostic, surgical, and management of cases in which obstetricians and gynecologic oncologists performed multidisciplinary management and PAS cesarean hysterectomy, to determine the impact on maternal morbidity, and to help establish this troublesome treatment modality in the future.
Materials and methods Cases and surgical procedures A case series study of single, non-normal pregnancies pathologically confirmed as PAS in hysterectomized uteri between 2013 and 2022 at the University of Fukui. The primary endpoint was to assess intraoperative and postoperative complications associated with hysterectomy. The definition of complications was determined based on Clavien-Dindo classification [ 7 ]. Secondary endpoints were to examine preoperative diagnostic ability and risk factors for PAS. In addition, we examined and evaluate the utility of management interventions implemented in the sequence of events leading to obstetric crisis hemorrhage and PAS cesarean hysterectomy. We examined whether these endpoints differed between patients who underwent hysterectomy with a Shock Index (S.I.) > 1.5 at hysterectomy (Group I; S.I.>1.5) and those who underwent total hysterectomy before reaching this state (Group II; S.I. ≤ 1.5) [ 8 , 9 ]. Our surgical procedure for PAS cesarean hysterectomy, while comparable to non-obstetric hysterectomy in general steps, exhibits distinctive characteristics that are tailored to the complexity of PAS. The operation is meticulously planned and executed by our multidisciplinary team, which is led by experienced gynecologic oncologists. The following outlines the full operative steps. Preoperative planning and team assembly A comprehensive preoperative plan is formulated, with a multidisciplinary team at the helm to ensure all necessary expertise is available. This team includes, but is not limited to, gynecologic oncologists, anesthesiologists, neonatologists, and urologists. Vascular access and monitoring Central venous access is established to facilitate rapid fluid administration and central venous pressure monitoring. An arterial line is placed for continuous blood pressure monitoring and regular blood gas analysis. The patient’s hemodynamic status is closely monitored throughout the procedure. Adjustments to medication regimens are made in real-time, based on ongoing assessments of blood loss, uterine tone, and the patient’s overall condition. Anesthetic management Anesthesia is initiated with either spinal or epidural anesthesia. In cases where a total hysterectomy is anticipated, the patient is transitioned to general anesthesia to ensure patient immobility and optimal pain control. Patient positioning The patient is positioned in the lithotomy position to provide the surgical team with adequate access to the operative field. Hemostatic measures preparation Intrauterine balloon catheters (such as the Atom Uterine Compression Balloon) are prepared for potential rapid deployment to control uterine bleeding. Uterine compression sutures with blunt needles are on standby for immediate use if necessary. Iliac artery balloon occlusion is prepared in collaboration with the radiology department. Toe SpO2 monitors are also attached to monitor peripheral perfusion. When employed, the balloon is expanded for 15 min with 5-min intervals, without heparinization to mitigate the risk of bleeding. Blood products and transfusion management Transfusion preparations include the availability of autologous blood (4–8 units) and allogeneic blood with 10 units of red blood cells and 10 units of fresh frozen plasma, ensuring readiness for potential massive blood loss. Insertion of bilateral ureteral stents In all planned surgeries, bilateral ureteral stents are inserted preoperatively. This step is crucial for the identification and preservation of the ureters during the surgery. Uterine incision and fetal extraction The uterus is typically incised transversely, unless otherwise indicated by the placental position. The method of fetal extraction is determined by fetal position and placental location, with care taken to avoid placental disruption. Administration of ecbolics and other medications Immediately after the delivery of the fetus, ecbolics are administered to control bleeding. Intravenous oxytocin is administered as the first-line agent. A 10 units of bolus dose is given initially, followed by a continuous infusion to sustain uterine contractions. The dose is titrated based on the response of the uterus and the clinical judgment of the attending anesthesiologist and obstetrician. The use of additional ecbolics, such as methylergonovine or carboprost, is considered if the response to oxytocin is inadequate or if there is a contraindication to its use. The selection of these agents is tailored to the individual’s clinical status, including blood pressure and any pre-existing medical conditions. Simultaneously, the uterine incision wound is quickly sutured simply to hemostat and promote uterine contractions. Use of energy devices Energy devices such as Bipolar scissors (Ellman-Japan, Osaka, Japan) and HARMONIC FOCUS® (ETHICON, Bridgewater, NL, USA) are employed for cutting and coagulation, which minimizes blood loss and enhances precision in tissue dissection. The upper uterine ligament is fully ligated and severed. The ovaries are typically preserved. Bladder and uterine cavity preparation To facilitate a complete hysterectomy, the lateral cavity of the bladder is meticulously expanded, and the lateral cavity of the uterus is secured before dissection begins. This preparation is critical for safely accessing the surgical planes. Cystocele dissection and release of the uterus Following the separation of the bladder, the sacrouterine ligaments are transected, and the Douglas fossa peritoneum is incised to release the uterus posteriorly and laterally. The “holding-up uterus” method Our unique ‘holding-up uterus’ technique is then employed. The surgeon places one hand in the cysto-uterine fossa and the other in the Douglas fossa to grasp and lift the entire uterus. This method not only provides superior visualization but also creates necessary distance between the ureter and the cervix, facilitating the safe liberation of the ureter. Identification and division of uterine artery With the uterus lifted, the uterine artery can be clearly identified and safely divided. This step is critical to controlling the blood supply to the uterus and ensuring hemostasis. Completion of hysterectomy The cervix is amputated at the predetermined site, identified by inserting a finger into the posterior fornix and lifting the cervix. The uterus is then completely removed. Hemostasis verification A thorough examination of the surgical field is conducted to confirm complete hemostasis. If any bleeding points are identified, they are addressed immediately with additional sutures, electrocautery, or application of hemostatic agents as required. The bilateral uterine arteries and the site of the cervix, particularly, are inspected meticulously given their potential as primary sources of hemorrhage. Application of adhesion prevention agents Once hemostasis is confirmed, adhesion prevention agents are applied. These may include hyaluronic acid-based gels or oxidized regenerated cellulose, which are strategically placed over areas prone to adhesion formation, such as the raw surfaces created by dissection. This step is crucial for reducing the risk of postoperative adhesions, which can lead to chronic pain and ileus. Each step is conducted with utmost precision, keeping in mind the altered anatomy due to PAS. Our technique is notable for the emphasis on preoperative stenting of the ureters, the use of advanced energy devices, and the strategic ‘holding-up uterus’ maneuver, all of which are pivotal for the success of these complex surgical procedures. (Fig. 1 and Additional file 1 ) Statistical analysis Continuous variables were described using mean ± standard deviation. To evaluate normally distributed data, the student t-test was utilized, and Mann-Whitney’s U test was used for between-group comparisons to evaluate data that were not normally distributed. PAS and diagnostic efficiency in ultrasonography and MRI examinations were performed with the binomial distribution test, and significance between the two was performed with Fisher’s exact definite test. The IBM Statistical Package for the Social Sciences (SPSS, version 22, SPSS Inc., Chicago, IL, USA) software was used in all analyses. P-values < 0.05 were judged as significant.
Results Twelve patients who were managed conservatively for obstetric crisis hemorrhage at the University of Fukui Hospital from 2013 to 2023, but who ultimately underwent PAS caesarean hysterectomy. Pathology included four cases of simple adherent placenta (FIGO Grade 1), seven cases of invasive placenta (FIGO Grade 2) and one case of placental penetration (FIGO Grade 3) (Sup. Tables 1 , 2 ) [ 10 ]. The group that underwent total hysterectomy with S.I. > 1.5 (Group I) had 6 cases with a total blood loss of 5490 mL (± 1821 mL), and the group that underwent total hysterectomy by S.I. ≤ 1.5 (Group II) had a total blood loss of 1959 mL (± 909 mL). With regard to intraoperative and postoperative complications, there were no serious complications above Grade 3 or deaths in Group I, although significantly more complications occurred in Group I. The only intraoperative complication was a partial bladder injury: one in Group I and the other in Group II. Both were completely healed by surgical repair sutures at the same time (Table 1 ). None of the three cases in which PAS was suspected preoperatively and combined with uterine artery embolization (UAE) had any serious complications. In addition, a total hysterectomy with UAE performed immediately after cesarean section and additional prophylactic intravascular balloon catheter placed within 7 days after surgery was safe. Three out of three patients who underwent emergency common iliac artery balloon occlusion (CIABO) and then primary total hysterectomy with S.I. > 1.5 developed postoperative Grade 2 thrombosis. Three patients in group I had preoperative suspicion of placenta accreta and underwent scheduled surgery, and three patients had no suspicion of placenta accreta and underwent emergency total hysterectomy. Three patients in group II also had preoperative suspicion of placenta accreta, while three had no suspicion of placenta accreta. Preoperative intervention for obstetric crisis hemorrhage due to PAS was more common in group I, and intervascular radiology was comparable in both groups (Table 2 ). At our institution, initial screening was performed with ultrasound, and MRI was the second imaging modality of choice when PAS was suspected. In the present study, there were 6 suspected cases on ultrasound and 4 suspected cases on MRI. The present study did not predict the diagnosis of PAS or cases with S.I. > 1.5 on risk factor assessment or preoperative imaging evaluation. The diagnostic efficiency of ultrasound and MRI examinations performed at our institution with PAS was 0.5 (50%), with a binomial distribution binomial distribution test, the point estimate for preoperative diagnosis of PAS by ultrasound was 0.5 (50%), with a 95% confidence interval of 0.218 (21.8%) to 0.782 (78.2 (%). For MRI, the rate was 0.444 (44.4%), with 95% confidence intervals ranging from 0.121 (12.1%) to 0.767 (76.7%). There was no difference in diagnostic efficiency between the two, with a P value = 0.3306 in Fisher’s exact definite test. The history of the patients in the present study was characterized by the fact that one patient had undergone three cesarean sections and two patients had undergone one cesarean section. Six patients had total placenta previa. Other surgical procedures that caused surgical damage to the uterine wall, such as surgical hysteroscopy, and suction curettage were observed in three cases. Notable history includes one patient with a history of UAE, two patients with a uterine cavity length of less than 5 cm at non-pregnancy, and two patients with systemic lupus erythematosus [ 11 ]. In vitro fertilization (using cryopreserved embryos) was performed in 7 cases. The only significant difference was that 2 assisted reproductive technology (ART) pregnancies with a pre-pregnancy uterine cavity length of less than 5 cm were observed in Group I (Table 3 ). There was no significant difference between the two groups in pre-operative management, but there was more pre-operative management in Group I (Table 2 ).
Discussion Our “Holding-up uterus” method during PAS cesarean hysterectomy, even in critical situations with S.I. >1.5, facilitates ureteral emancipation, identification of the uterine artery, and facilitates the Pelosi method [ 12 ] in which the bladder is finally detached after cutting the vaginal wall, avoiding adhesions on the anterior bladder from the previous cesarean-section. In cases of severe adhesions, the posterior vaginal canal may be opened first, facilitating the treatment of the basal ligament and dissection of the bladder and anterior vaginal wall, which may be an extremely useful method for complete hysterectomy. PAS cesarean hysterectomy has become the gold standard as the final step in the management protocol for massive hemorrhage associated with PAS disorders. However, this primary radical surgical treatment is associated with a high incidence of maternal surgical-related adverse events, particularly massive hemorrhage, and damage to surrounding organs (40–50%) and maternal death (approximately 7%) [ 13 , 14 ]. Pregnancy-related hysterectomy for PAS is considerably more technically challenging than hysterectomy for uterine atony because of the higher risk of adjacent organ injury [ 15 ]. Urinary tract injuries have been reported in 29% of surgeries, with lacerations of the bladder reported in 76%, ureteral injuries in 17%, and urogenital fistulas in 5% [ 16 ]. The procedure for pregnancy-related hysterectomy is identical to that for non-pregnancy hysterectomy [ 17 ]. However, during the operation, one must be aware of the changes that occur in the female reproductive organs during pregnancy [ 17 ]. As the uterus enlarges, it becomes more difficult to manipulate it and to visualize the entire pelvis. In addition, the ureters may become tortuous and dilated, resulting in significant hydroureteria. Tissue fragility and edema increase. Most importantly, uterine blood flow increases 10- to 30-fold in late pregnancy, and pregnant women with underlying diseases such as PAS are more prone to complications of DIC, even with moderate blood loss [ 3 ]. The reason is that it has been reported to shorten operative time and decrease blood loss. We use energy devices during hysterectomy. The reason is that it has been reported to reduce operative time and blood loss [ 18 ]. And supra-hysterectomy is often performed because of the short operating time required under conditions of critical bleeding, the increased risk of ureteral injury during emergency surgery, and, in the case of placenta previa, the fully dilated cervix, which makes identification of the transition from the cervix to the vagina difficult. However, FIGO recommends performing a supra-hysterectomy because of the potential risk of malignancy in the residual cervix and the consequent need for periodic cervical cytology, and because the residual cervix is a cause of postoperative bleeding (placenta previa PAS) [ 19 ]. It has also recently been shown that identification and clamping of the bilateral uterine arteries and removal of the uterus as low as possible at the inferior margin of the placenta, avoiding the ureter, reduces maternal bleeding morbidity the most [ 20 ]. For this reason, we try to perform total hysterectomies using the “Holding-up uterus” technique even in emergency situations. In this context, our “Holding-up uterus” technique during PAS cesarean hysterectomy facilitates ureteral emancipation, identification of the uterine artery, and dissection of the adherent bladder. Currently, even in critical situations of Group I (S.I. > 1.5), the operation can be performed in a short time. One case of bladder injury was observed in each Group I and II, but it could be easily repaired. Another proposed radical surgery is delayed hysterectomy. In this procedure, the uterus is closed after delivery, leaving the placenta in uterus, the mother’s abdomen is closed, and then a total hysterectomy is performed 3–12 weeks later. The rationale for this procedure is that uterine perfusion is reduced after delivery, even if the placenta is left in uterus, and the subsequent surgery is less risky for the woman due to uterine retraction and decreased vascularity [ 21 ]. However, abdominal closure for the purpose of delayed total hysterectomy may be followed by massive bleeding due to partial abruption of the placenta. Prophylactic intravascular balloon catheters have been proposed to reduce intraoperative bleeding during hysterectomy. It improves maternal morbidity and allows the surgeon to operate in a “cleaner” and more visible operative field. However, the incidence of potential complications is high, including the risk of vessel rupture, thromboembolism development, and impaired blood supply to the lower extremity [ 22 ]. In addition, PAS is associated with extensive abnormal neovascularization, and occlusion of some pelvic vessels may increase blood loss from collateral blood vessels [ 6 , 22 , 23 ]. Furthermore, two RCTs comparing balloon catheter placement in the iliac artery versus no intervention at all found no difference in the number of packed red blood cells transfused to patients, and a recent RTC comparing bilateral internal iliac artery ligation versus control found no difference regarding intraoperative blood loss [ 6 , 22 , 23 ]. In our study, in the critical situation of group I (S.I. >1.5), total hysterectomy with intravascular balloon insertion was more likely to cause postoperative venous thrombosis. The most important risk factor for the development of PAS has been shown to be the number of previous cesarean Sect. [ 6 ]. In the present study, one patient had had three cesarean sections and two had had one cesarean section. In vitro fertilization (using cryopreserved embryos), a risk factor that has received much attention recently, was used in seven cases. In terms of preoperative assessment factors, 67% of pregnancies in both groups were ART pregnancies. The only significant difference was that two ART pregnancies with hypoplastic uteri were observed in Group I. Although the definition of hypoplastic uterus is not known, both pregnancies were ART pregnancies with small pre-eclamptic uteri and a cervix to uterus size ratio of 1:1. The uterine cavity length was about 5 cm. There was no difference in diagnostic efficiency between PAS and diagnostic efficiency between ultrasound and MRI examinations performed at our hospital. This result did not differ significantly from the prenatal detection rate of PAS by ultrasound in two large population-based studies in the U.K. and U.S [ 24 , 25 ]. Nearly half of PAS is diagnosed only at birth, and even among pregnant women who underwent prenatal MRI testing, over a quarter were detected at birth [ 17 ]. The inclusion of imaging and clinical factors has been reported to improve the prenatal diagnosis of PAS [ 17 ] and should be actively investigated in the future. “Holding-up uterus” method has shown several significant strengths in the management of PAS cesarean hysterectomy. Firstly, it has proven to be beneficial in providing improved operative visualization. By holding up the uterus, the technique allows for a clearer view of the ureters and uterine arteries, which is crucial in the context of PAS where normal anatomical landmarks may be distorted. This is of particular importance as it facilitates the Pelosi method and other critical steps such as ureteral emancipation. Moreover, the method has been associated with reduced operative time and blood loss, which are critical outcomes in PAS cesarean hysterectomy. This reduction is not only beneficial for the patient’s immediate surgical outcome but also has long-term implications on their recovery process. The facilitation of complex surgical steps such as the dissection of the adherent bladder and identification of the uterine artery is another advantage that cannot be overstated. This simplification is vital, especially in the backdrop of the high incidence of maternal surgical-related adverse events. The ability of the technique to be effectively utilized even in emergency situations where the shock index is high is a testament to its robustness. This is underscored by our findings that even in group I (S.I. > 1.5), the operation can be executed in a short time frame, indicating that the method is adaptable to critical situations. Additionally, the practice of performing total hysterectomies aligns with recommendations to reduce potential risks associated with residual cervical tissue. However, the limitations of our technique are as important to consider as its strengths. Organ injury remains a substantial risk in PAS cesarean hysterectomy, and while our method aids in minimizing this risk, it does not eliminate it. The surgeon’s experience and the technique’s learning curve are additional factors that could affect the outcomes of the surgery. Our findings also suggest that the generalizability of the “Holding-up uterus” method might be limited by variations in surgical practices across different institutions. Furthermore, the comparative data on the “Holding-up uterus” method versus other techniques is not extensive. This lack of robust comparative data could be seen as a limitation as it does not allow for a conclusive argument for the superiority of our method. While we have observed benefits in our own practice, additional comparative studies are required to validate these findings further. Lastly, despite the method being designed to minimize complications, the potential for unforeseen surgical difficulties and postoperative complications remains. This highlights the necessity of vigilance and preparedness for managing such events should they occur. While our “Holding-up uterus” method demonstrates considerable promise, particularly in facilitating ureteral emancipation, identification of the uterine artery, and dissection of the adherent bladder, there is a need for a careful evaluation of the risks and benefits. Future studies should aim to provide a more comprehensive comparison with other techniques, evaluate the learning curve associated with the method, and explore strategies to mitigate the inherent risks of PAS cesarean hysterectomy.
Conclusions The primary surgical treatment for PAS is pregnancy-related hysterectomy, which is technically challenging due to the increased risk of adjacent organ injury. The “Holding-up uterus” technique during PAS cesarean hysterectomy facilitates ureteral emancipation, identification of the uterine artery, and dissection of the adherent bladder, making the total hysterectomy easier to perform even in critical situations. Prophylactic intravascular balloon catheters have been proposed to reduce intraoperative bleeding during hysterectomy, but their use may lead to potential complications. The most important risk factor for the development of PAS is the number of previous cesarean sections. Future research should focus on collecting high-quality data from well-designed prospective studies on a multidisciplinary team approach to diagnosis (prenatal imaging) and management strategies.
Background Placenta accreta spectrum (PAS) cesarean hysterectomy is performed under conditions of shock and can result in serious complications. This study aimed to evaluate the usefulness of the “Holding-up uterus” surgical technique with a shock index (S.I.) > 1.5. Methods Twelve patients who underwent PAS cesarean hysterectomy were included in the study. Results Group I had S.I. > 1.5, and group II had S.I. ≤ 1.5. Group I had more complications, but none were above Grade 3 or fatal. Preoperative scheduled uterine artery embolization did not result in serious complications, but three patients who had emergency common iliac artery balloon occlusion (CIABO) and a primary total hysterectomy with S.I. > 1.5 had postoperative Grade 2 thrombosis. Two patients underwent manual ablation of the placenta under CIABO to preserve the uterus, both with S.I. > 1.5. Conclusions The study found that the “Holding-up uterus” technique was safe, even in critical situations with S.I. > 1.5. CIABO had no intervention effect. The study also identified assisted reproductive technology pregnancies with a uterine cavity length of less than 5 cm before conception as a critical factor. Supplementary Information The online version contains supplementary material available at 10.1186/s12893-024-02311-8. What does this study add to the clinical work •The holding-up uterus technique following periuterine cavity expanded enables safe placenta accreta spectrum (PAS) cesarean hysterectomy. •PAS hysterectomy with holding-up uterus is effective even in situation of critical bleeding or shock. Supplementary Information The online version contains supplementary material available at 10.1186/s12893-024-02311-8. Keywords
Electronic supplementary material Below is the link to the electronic supplementary material.
Author contributions Conceptualization, J.T. and Y.Y.; data acquisition, A.S. and J.T.; data analysis, interpretation, and statistical analysis, J.T., D.I. and Y.Y.; writing (original draft preparation), and review and editing, M.O., H.K., D.I., H.T., N.T., T.K. All authors have read and agree to the published version of the manuscript. Funding Not applicable. Data availability The data presented in this study are available in Tables 1 , 2 and 3 and supplemental Tables 1 , 2 . Declarations Ethics approval and consent to participate The study was conducted in accordance with the Declaration of Helsinki, and approved by the Institutional Review Board of the University of Fukui Hospital (IRB Number: 20170100). Informed consent was obtained from all subjects involved in the study. Consent for publication Not applicable. Competing interests The authors declare no competing interests.
CC BY
no
2024-01-15 23:43:46
BMC Surg. 2024 Jan 13; 24:23
oa_package/42/ad/PMC10787967.tar.gz
PMC10787968
38218901
Introduction Stroke is a leading cause of long-term disability in the United States, affecting more than 800,000 people per year [ 1 ]. Unilateral paralysis (hemiparesis) affects up to 80% of stroke survivors, leaving many to struggle with activities of daily living (ADLs) including the ability to manipulate objects such as doors, utensils, and clothing due to decreased upper-extremity muscle coordination and weakness [ 2 ]. Restoration of hand and arm function to improve independence and overall quality of life is a top priority for stroke survivors and caregivers [ 3 ]. Intensive physical rehabilitation is the current gold standard for improving motor function after stroke. Unfortunately, 75% of stroke survivors, caregivers, and health care providers report that current upper extremity training practice is insufficient [ 4 ]. The development of user-centric neurotechnologies to restore motor function in stroke survivors could address these unmet clinical needs through a range of different mechanisms, such as improving motivation, enhancing neuroplasticity in damaged sensorimotor networks, and enabling at-home therapy. Assistive technologies (AT) hold potential to restore hand function and independence to individuals with paralysis [ 5 ]. ATs, including exoskeletons and functional electrical stimulation (FES), can assist with opening the hand and also evoke grips strong enough to hold and manipulate objects [ 6 ]. Additionally, these systems have been used therapeutically during rehabilitation to strengthen damaged neural connections to restore function [ 7 ]. A wide variety of mechanisms to control ATs have been investigated including voice [ 8 ], switch [ 9 ], position sensors [ 10 ], electroencephalography (EEG) [ 11 ], electrocorticography (ECoG) [ 12 ], intracortical microelectrode arrays (MEA) [ 13 ], and electromyography (EMG) [ 14 ]. Unfortunately, no single system has simultaneously delivered an intuitive, user-friendly system with a high degree-of-freedom (DoF) control for practical use in real-world settings [ 4 ]. Recent advances in portable, high-density EMG-based (HDEMG) systems have the potential to overcome several of these barriers and deliver an intuitive and entirely non-invasive AT control solution [ 15 , 16 ]. While various EMG-based ATs exist, including the commercially available MyoPro Orthosis [ 15 ], most of these systems use a small number of electrodes and rely on threshold-based triggering [ 14 ]. Consequently, these systems have limited DoF control which constrains their practical use. Conversely, HDEMG systems consisting of dozens of electrodes and leveraging machine learning approaches to infer complex movement intention can provide high DoF control, significantly expanding functional use cases as well as increasing the proportion of the stroke population that could benefit from these technologies [ 16 – 19 ]. Currently, HDEMG systems are primarily research systems and are not optimized for usability, including being difficult to set up, requiring manual placement of electrodes, and being non-portable and bulky, which can hinder the successful translation of technologies [ 4 ]. To address these limitations, we developed the NeuroLife ® EMG System to decode complex forearm motor intention in chronic stroke survivors while simultaneously addressing end user needs. The EMG system was designed to be used as a control device for various end effectors, such as FES systems and exoskeletons. Additionally, the system was specifically designed to meet user needs in domains previously identified as high-value for stroke survivors: donning/doffing simplicity, device setup and initialization, portability, robustness, comfortability, size and weight, and intuitive usage [ 4 ]. The sleeve is a wearable garment consisting of up to 150 embedded electrodes that measure muscle activity in the forearm to decode the user’s motor intention. A single zipper on one edge of the sleeve allows for a simplified and streamlined donning and doffing by the user and/or a caregiver. The sleeve design facilitates an intuitive setup process as embedded electrodes that span the entire forearm are consistently placed, eliminating the need for manual electrode placement on specific muscles. The lightweight stretchable fabric, similar to a compression sleeve, was chosen to enhance comfort for long-term use. The sleeve connects to backend Intan hardware housed in a lightweight, 8 × 10′′ signal acquisition module appropriate for tabletop upper-extremity rehabilitation. Overall, these design features help address critical usability factors for ATs [ 4 ]. In this work, we demonstrate that our EMG system can extract task-specific myoelectric activity at high temporal and spatial resolution to resolve individual movements. Based on EMG data collected from seven individuals with upper extremity hemiparesis due to stroke, trained neural network machine learning models can accurately decode muscle activity in the forearm to infer movement intention, even in the absence of overt motion. We demonstrate the viability of this technique for online decoding, as two subjects used the system for closed-loop control of a virtual hand. This online demonstration is a promising step towards using HDEMG sleeves for high DoF control of ATs based on motor intention. Finally, we present usability data collected from study participants that highlight the user-centric design of the sleeve. These data will be used to inform future developments to deliver an effective EMG-based neural interface that meets end user needs.
Methods Subjects Seven individuals (3 female, 4 male; 60 ± 5 years) with a history of stroke participated in a study that recorded EMG using the NeuroLife EMG System while attempting various hand and wrist movements. Additionally, data were collected from seven able-bodied individuals (4 female, 3 male; 27 ± 1 years) to serve as a general comparison of EMG data and to benchmark decoding algorithms. Able-bodied subjects were employees of Battelle Memorial Institute, but none were authors of this work. Data were collected as part of an ongoing clinical study being conducted at Battelle Memorial Institute that was approved by the Battelle Memorial Institute Institutional Review Board. All participants provided written informed consent before participation, in accordance with the Declaration of Helsinki. Demographics of study subjects with stroke are provided in Table 1 (data on able-bodied participants can be found in Additional file 1 : Table 1 ). Eligibility criteria were set to recruit adult chronic stroke survivors with hemiparesis affecting the arm and hand that were able to follow 3-step commands and communicate verbally. Specific inclusion and exclusion criteria are listed in the Additional file 1 : Methods. During the first session prior to EMG data collection, standardized clinical assessments were performed by a licensed occupational therapist in all subjects with stroke. These included the upper extremity section of the Fugl-Meyer (UE-FM) to assess upper extremity motor impairment, the Box and Blocks test to assess manual dexterity, and the Modified Ashworth test to assess spasticity of the finger, wrist, and elbow flexors. Based on predetermined exclusion criteria, an eighth subject was removed from data analysis due to hemispatial neglect affecting their ability to consistently follow movement cues. Experimental setup Subjects sat facing a computer monitor with their arms placed on a table, and the sleeve on the paretic arm for participants with stroke (Fig. 1 ). The sleeve was placed on the right arm for able-bodied subjects, regardless of handedness. The sleeve comprises a stretchable fabric with an embedded array of electrodes (Additional file 1 : Fig. S1). Depending on the forearm size of the participant, a small, medium, or large sized sleeve was used containing 128 electrodes (64 channel pairs), 142 electrodes (71 channel pairs), or 150 electrodes (75 channel pairs), respectively. Each electrode is 12 mm diameter, spaced 25 mm apart, and wrap the forearm from elbow to wrist. With a flexible and lightweight nylon-Lycra hybrid material, the sleeve wears like a compression sleeve and weighs 180, 195, and 220 g for the small, medium, and large sleeves, respectively. A zipper on the ulnar edge of the sleeve allows for easy donning and doffing. Prior to donning, an electrode solution spray (Signaspray, Parker Laboratories, Fairfield, NJ) was applied to the subject’s forearm to improve signal quality. Bipolar EMG signals were sampled at 3 kHz with a gain of 192 V/V using an Intan Electrophysiology Amplifiers (Intan RHD2000, Intan Technologies, Los Angeles, CA) [ 20 ]. An embedded electrode in the sleeve near the elbow was used as a reference for all bipolar amplifiers. The sleeve was connected to a custom-built, 8 × 10′′ footprint, EMG signal acquisition module, which then connected to a laptop computer (Fig. 1 and Additional file 1 : Figure S1a). The subjects were instructed to attempt a series of hand, wrist, and forearm movements. A series of images of the desired hand movement was presented on a computer monitor, and the subjects were instructed to attempt each movement shown to the best of their ability. Subjects were instructed to attempt the movement at 25–50% of their subjective maximal effort to minimize muscle fatigue and co-contractions throughout the session. The following movements were collected during the session: Hand Close (Power Grip), Hand Open, Index Extension, Thumb Flexion, Thumb Extension, Thumb Abduction, Forearm Supination, Forearm Pronation, Wrist Flexion, Wrist Extension, Two Point Pinch, and Key Pinch. These movements were identified by a licensed occupational therapist as highly relevant functional movements for dexterous hand use, and these movements have been used in similar studies [ 21 ]. Recording blocks consisted of a single movement repeated 10 times (referred to as “single blocks”), or multiple movements repeated within a single recording block (referred to as “mixed blocks”). Every block began with an 8 s rest period, followed by alternating movement and rest periods. During mixed blocks, a collection of movements (e.g., Hand Close, Hand Open, Forearm Supination) were randomly presented to the subject with interleaved rest periods. Before beginning the block, subjects were shown the movement(s) in the upcoming block. For subjects with stroke, the time for each movement was randomly selected from a uniform distribution between 4 and 6 s, and rest time was randomly selected between 4 and 6 s. For able-bodied subjects, the movement and rest times were both set randomly between 2 and 3 s. The cue and rest times were shortened in able-bodied subjects due to faster movement times and the expectation of simpler decoding compared to the subjects with stroke. In the last recording session, a usability questionnaire assessing user needs (adapted from [ 4 ]) was given to stroke subjects to evaluate the usability of the current sleeve design (responses from subjects are presented in Additional file 1 : Table 5). We collected data from each stroke subject across 3–4 sessions lasting < 2 h each. Data from all sessions were used to train the classifiers, with the last half of the data from the final session held out for testing. The sleeve was not doffed before collection of the test dataset in the final session. The total amount of training data per movement for subjects with stroke are shown in Additional file 1 : Figure S3. For able-bodied experiments, data were collected in a single session with a total of 10 repetitions for each movement. The first 5 repetitions were used for training, and the last 5 repetitions were used for testing, without doffing the sleeve between. This structure was designed to simulate an envisioned use case in which a decoding algorithm would be calibrated for a rehabilitation session using both previous session data and data from a short same-day calibration protocol. To assess each subject’s ability to perform the movements without any assistance, each movement was scored by a licensed occupational therapist based on a scoring scheme adapted from the Action Research Arm Test (ARAT) [ 18 ]. The “observed movement score” was ranked using the following categories: 0 = no movement; 1 = incomplete range of motion; 2 = complete range of motion but impaired; 3 = normal. Pre-processing, windowing, and feature extraction The EMG data were bandpass filtered (20–400 Hz, 10th order Butterworth filter), and a 60 Hz notch filter was applied similar to previous studies [ 23 ]. The root mean square (RMS) was extracted using consecutive 100 ms data windows with no overlap (Fig. 2 B, C ). For decoding of movement intent during a given time window, the current window and three preceding windows were used, totaling 400 ms of RMS data used for each prediction. Next, the training data were normalized (mean = 0, variance = 1) and the testing data were normalized using the mean and variance from the training data. Classification of movement intention in stroke participants was performed in two different ways: (1) using the 2.5 s center window during a cue or rest period, or (2) on the continuous timeseries data. For the 2.5 s center window method, the middle 2.5 s of each cue and rest period during a block was extracted (Fig. 3 A). This resulted in a total of 22 predictions of 100 ms binned RMS data per cue (2.5 s with the first three 100 ms bins removed for containing out-of-window data at the beginning of the cue). This method was applied to both the training and testing datasets to reduce noise from motion artifact and transient muscle activation by removing the transition periods, similar to previous studies [ 24 ]. This dataset was used to evaluate different machine learning models for decoding the user’s movement intent. In able-bodied subjects, classification was performed as described above but with a 1.5 s center window during a cue or rest period, resulting in a total of 12 predictions. These data are presented in Fig. 3 for participants with stroke and Additional file 1 : Figure S5 for able-bodied subjects. For decoding of continuous timeseries data, we performed a dynamic cue shifting technique to account for the variability in the subject’s ability to respond to the onset and offset of cues. Latency between cue onset and the onset of EMG activity is a persistent problem within decoding that can lead to significant deficits in algorithm performance and is exacerbated in data recorded from subjects with neurological impairments such as stroke. Traditionally, these onset and offset variabilities are handled by shifting cues a predetermined amount of time based on reaction times [ 25 ] or by assigning each cue manually [ 21 ]. However, these methods still fail to capture the full distribution of onset and offset variability. Here, we use an automated approach to dynamically shift cue labels to match the EMG activity. The average EMG signal was aligned with the intended cue times, and residuals were calculated between the EMG signal and the signal mean for each cue segment. The transition point between segments was then iteratively optimized to minimize the sum of squared residuals (Additional file 1 : Figure S4). Cue timings were shifted up to a maximum time of 2 s beyond the intended cue time. Classification Classification was performed using all recording blocks (single and mixed). Importantly, the testing consisted of the final 4 recording blocks of data collected for that subject. In other words, none of the training set occurred later in time than the testing set to prevent data leakage of time dependent signal fluctuations that could significantly influence decoding performance. Three classifiers were compared: a logistic regression (LR) model [ 26 ], a support vector machine (SVM) [ 27 ], and a neural network (NN). For the LR and SVM models, data were additionally preprocessed using principal component analysis for dimensionality reduction, keeping components that accounted for > 95% of the variance. LR and SVM models were trained using the scikit-learn toolbox [ 28 ] in Python 3.8. To optimize hyperparameters for both LR and SVM, a grid search on the training data with fivefold cross validation was applied to tailor a specific model for each subject. Hyperparameter C was varied from 1e-4 to 1e4 for LR, and hyperparameters C and Gamma were varied from 1e-4 to 1e4 for SVM. The best performing model hyperparameter combinations for each were selected for evaluation. The NN was developed in Python 3.8 using the FastAI package [ 29 ]. FastAI defaults were used for training except where noted. The model architecture takes an input of a flattened N channels × 4 array from the N channels of the sleeve and 4, 100 ms windows of mean RMS signal. The input layer connects to two fully-connected dense layers, with size 1000 and 500 respectively, with batch normalization and the ReLU activation function between layers. The final layer had 13 classes corresponding to the 12 cued movements and rest. Finally, a Softmax activation function was applied to the model outputs to provide prediction probabilities for each of the movements. The predicted movement for a given time point was the movement with the greatest prediction probability. The training procedure used label smoothing cross entropy loss (p = 0.9) and the Adam optimizer. During training, dropout was applied to each layer with 20% probability to prevent overfitting. The learning rate was optimized using the FastAI learning rate finder tool [ 29 ]. Each model was trained for 400 epochs with early stopping criterion, using the one cycle training policy from FastAI. To simulate massed practice rehabilitation exercises, participants repeated movements with interleaved rest. We evaluated the decoding algorithms with two complementary metrics relevant to this use case and commonly used for similar applications. Accuracy was defined as the percentage of 100 ms time bins predicted by the classifier to be the same as ground truth similar to our group’s previous decoding study [ 30 ]. Accuracy is a standard classification metric and provides a high temporal resolution metric of performance. Chance level accuracy was determined based on the percentage of labels equal to the majority class (Rest; 50.41%), which represents the accuracy of a naïve classifier. Since the majority class prediction yields the highest chance level accuracy of any random strategy in a 13-class problem (e.g. stratified, uniform, or majority), we chose to use this method to represent the naïve decoder option for all reported chance levels. When decoding a subset of the full 12 movement set (Fig. 4 ), the rest cues directly before each target movement were sampled to maintain rest at 50% of the sampled dataset to avoid biasing the results. We also present success rate as a decoding performance metric, similar to previous studies [ 25 ]. A movement is considered successful if there is at least 1 s continuous period within a cue that is correctly decoded as the intended movement. The success rate is then calculated as the percentage of cues which are considered successful. This metric approximates an observer rating each cue as a binary success or failure and is more aligned with how a user would perceive performance. Real-time demonstration In separate sessions, we tested the performance of the decoder online in two stroke subjects (Subjects 13,762 and 30,458) to demonstrate the ability to predict a user’s motor intention in real-time. An occupational therapist identified a bottle pouring task as an appropriate massed practice therapy task shared for both of these subjects based on their personal abilities. The movements Hand Open, Hand Close, and Forearm Supination were further chosen by the therapist as movements for which the NeuroLife EMG system may be programmed to control FES or an exoskeleton to assist the subjects. We used a decoder trained to classify these movements and rest for this real-time evaluation of a simulated use case. The same NN architecture described above was used during this real-time demonstration. Data collected from this online decoding session is referred to as the real-time demonstration dataset. The NN decoder was built using two blocks of training data that was collected during the online decoding session. Each block contained 5 repeats of three movements (Hand Close, Hand Open, Forearm Supination) with interleaved rests. EMG data was filtered using identical filters as the offline method described above. RMS was extracted using consecutive 100 ms data windows with no overlap, and the current window and three preceding windows were used, totaling 400 ms of RMS data for each prediction. Training data were normalized (mean = 0, variance = 1) and the online data were normalized using the mean and variance from the training data. Cue labels were shifted by 300 ms to account for reaction time of the participant. To avoid unintentional flipping between states in the online system, the NN class output probabilities needed to exceed a threshold of 0.6 to change the decoder class prediction from the previous prediction. Additionally, a stable decoder output was required to change decoder state, therefore two consecutive samples of the same prediction class were required to update the final class prediction. An experimenter then prompted each subject with randomized cues with the online decoder running. A virtual hand on the computer screen reflected the real-time movement detection. An experimenter manually labeled cues provided to the participant and the decoding accuracy was calculated. Videos of these blocks for each subject can be found in Additional file 1 : Media 1 and 2. Following the online session, NN models were re-trained on the same real-time demonstration dataset using the same methods described above for a comparison of online and offline performance. These data are presented in Fig. 6 . Statistical analysis All comparisons were planned in the experimental design a priori . Normality of distributions were tested using Lilliefors tests. Significant differences were determined using paired t-tests (Fig. 3 C) and unpaired t-tests (Figs. 3 D, 4 A) and where appropriate. Significant differences for multiple comparisons were determined using one-way ANOVAs followed by Tukey HSD tests (Fig. 3 C, D ). Alpha of 0.05 was used for single comparisons. To correct for multiple comparisons, a Bonferroni-corrected alpha of 0.0167 was used for Fig. 3 D and an alpha of 0.025 was used for Fig. 5 A. The p-value for the correlations were determined using Wald Test with t-distribution of the test statistic (Additional file 1 : Figure S10). Statistical tests for each comparison are noted in the text. Statistical analysis was performed in Python 3.8 using SciPy and Statsmodels. In all figures, * indicates p < 0.05, ** indicates p < 0.01, and *** indicates p < 0.001. Error bars indicate mean ± SEM in all figures. Usability assessment Usability is a critical factor in the long-term adoption of an AT. Inconveniences of setup and comfort, as well as frustrations with reliability can often lead to eventual device abandonment. Therefore, in our final EMG recording session with each participant, we collected initial usability data of the NeuroLife Sleeve for use in chronic stroke survivors to help guide future development efforts. The questions posed to subjects here were adapted to investigate overarching themes mentioned by stroke survivors, caregivers, and HCPs for the use of an assistive technology [ 4 ]. Subjects answered each question on a 1 to 5 scale, and questions were targeted at the following categories: simple to apply, comfort for long-term use, freedom of movement during use, functionality / lightweightness and portability, potential for clinical and home use, and overall aesthetic design of the device (Additional file 1 : Table 5). Subjects were instructed to consider a use case in which the NeuroLife EMG System (sleeve and signal acquisition module) is used to control a FES or exoskeleton system when responding. For usability metrics with more than one question (e.g. simple to apply), the mean value was scored for that assessment.
Results Movement intention can be inferred from forearm EMG activity of subjects with stroke using the NeuroLife EMG system Removing the transition periods and focusing on periods of consistent activity yielded a standardized dataset to compare performance of various models (Fig. 3 A). Heatmaps of EMG activity across the sleeve are shown for one subject with stroke (Fig. 3 B). These heatmaps highlight the visual differences between forearm EMG activity across the various movements. In contrast to the heatmaps of able-bodied subjects (Additional file 1 : Figure S2), EMG activity is less localized in the heatmaps of subjects with stroke. This trend is consistent across stroke severity, with more severely impaired subjects having less localized forearm EMG activity (Additional file 1 : Figure S9). These results are consistent with previous reports of lack of independent muscle control following stroke [ 31 ]. To validate our decoding pipeline, we tested decoding performance in able-bodied subjects across all 12 movements with the expectation of highly accurate decoding using three different approaches. Overall, the NN obtained 96.8 ± 0.5% accuracy and outperformed LR and SVM models, which had 91.5 ± 0.8% and 90.8 ± 1.2% accuracy, respectively (Additional file 1 : Figure S5; One-Way ANOVA: F[3, 10] = 4015, p = 9.02 × 10 –16 ; paired t-test NN vs. LR, p = 5.8 × 10 –5 ; NN vs. SVM, p = 1.6 × 10 –3 ). These decoding results are consistent in the dataset comprised of subjects with stroke attempting all 12 movements, where the NN obtained 77.1 ± 5.6% accuracy, and outperforms the LR (69.0 ± 5.4%) and SVM models (66.6 ± 6.9%) (Fig. 3 C; One-Way ANOVA: F[4, 12] = 64.02, p = 5.41 × 10 –8 ; paired t-test NN vs. LR, p = 9.1 × 10 –4 ; NN vs. SVM, p = 9.3 × 10 –3 ). In summary, the NN outperforms the LR and SVM when decoding forearm EMG activity to infer movement intention. All subsequent analyses were performed in subjects with stroke using the NN for decoding. Next, we investigated the relationship between the subject’s ability to perform a movement unassisted and our ability to accurately decode that movement. Generally, decoding performance improved as the observed movement score increased (Fig. 3 D; One-way ANOVA: F[3, 80] = 13.38, p = 3.7 × 10 –7 ). A comparison of decoding accuracy based on movement score was computed using a Tukey HSD test (Additional file 1 : Table 3). For movements with visible motion (score ≥ 1), the overall decoding accuracy was 85.7 ± 3.2%, whereas for movements where the subject had no visible motion (score = 0) the accuracy dropped significantly to 27.3 ± 3.2% (Chance: 4.0%) (Movement Ability: Movement score = 0 vs. Movement score = 1–3: unpaired t-test, p = 3.9 × 10 –9 ). We also investigated the relationship between decoding accuracy and the assessed clinical metrics (Additional file 1 : Figure S10). We observed moderate and significant correlations (Wald Test) between decoding performance and the UEFM Hand subset, and both the MAS wrist and fingers scores. In summary, these data suggest that EMG decoding performance decreases as impairment increases across a variety of clinical metrics assessing various aspects of dysfunction. Next, we investigated decoding performance of individual movements in subjects with stroke. The confusion matrix with individual movements for one subject is shown in Fig. 3 E. The best performing movements across subjects were Wrist Flexion and Index Extension with an average accuracy of 68.7 ± 2.2%. On average across subjects, the worst performing movements were Forearm Supination and Thumb Abduction, with an average accuracy of 39.4 ± 9.9% (Additional file 1 : Figure S6). The success rate per movement type for one subject is presented in the right column of the confusion matrix (Fig. 3 E). The overall grand average success rate across all movements and rest achieved 75.9 ± 4.2%. The top movements quantified by successes/attempts were Index Extension: 23/30 Wrist Flexion: 20/30, and the bottom movements were Forearm Supination: 14/43 and Thumb Abduction: 17/43. Decoding movement subsets to achieve high performance in subjects with severe stroke impairments As the decoding performance of our algorithms was dependent on the presence of visible movement in our subjects, we next investigated the association of hand impairment severity based on the Upper Extremity Fugl-Meyer Hand Subscore (UEFM-HS) with observed movement scores and decoding performance (Fig. 4 A). Both the observed movement score and decoding performance in subjects with severe hand impairment (UEFM-HS < 3) were significantly different than in individuals with moderate or mild hand impairment (UEFM-HS ≥ 3) (Average movement score: unpaired t-test UEFM-HS < 3 vs. UEFM-HS ≥ 3, p = 0.02; Decoding accuracy: unpaired t-test UEFM-HS < 3 vs. UEFM-HS ≥ 3, p = 0.006). To better understand if a smaller subset of movements could be decoded in the presence of severe impairment, we assessed if sufficient signal was present to decode general muscle activity during cued movement periods compared to rest. Practically, this decoding scheme would enable an individual with severe hand impairment to control an AT with a single movement. We separated the problem into two classes (Rest vs. Move), where the “Move” class consists of the 12 different movements combined into one class (Fig. 4 B). The NN decoder was able to achieve high performance in individuals with severe hand impairment with 86.7 ± 2.6% accuracy and 85.2 ± 3.6% success rate (Successes/Attempts; Rest: 164/185, Move: 151/185). These results indicate that the surface EMG collected from individuals with severe hand impairment is sufficient for binary scenarios. Encouraged by the binary decoder performance, we extended our analysis to include key functional movements for restoring grasp function, namely Rest, Hand Close, and Hand Open (Fig. 4 C). In this 3-class scenario, the rest periods before each movement were downselected from the full 12 movement dataset to keep chance accuracy decoding at 50%. With these key movements in individuals with severe hand impairment, the decoding performance achieved 85.4 ± 6.4% accuracy and 88.0 ± 7.7% success rate (Successes/Attempts; Rest: 45/46, Hand Close: 22/23, Hand Open: 14/23). While decoding the movements to enable Hand Close and Hand Open is ideal for intuitive control of an AT, alternatively decoded movements with the greatest performance can be mapped to the most impactful functional movements. Thus, we tested decoding only the top performing movements for each subject (Fig. 4 D). When comparing Rest and the top two movements for each individual, decoding performance achieved 91.0 ± 3.9% with a grand average success rate of 90.6 ± 4.2% (see Additional file 1 : Table 4 for full details). This performance was comparable to the decoding performance of individuals with UEFM-HS ≥ 3 on 12 movements (87.6 ± 3.4%) and provides a reasonable alternative for subjects with more severe impairments. Decoding continuous forearm EMG data in real-time scenarios in chronic stroke survivors To demonstrate the utility of the NeuroLife EMG System to interpret muscle activity from the forearm to act as a control signal for assistive devices, we tested our decoding algorithms in a continuous dataset. Following a stroke, the ability to contract and relax muscle groups is slowed and highly variable [ 32 ], which consequently makes automated labeling of cues using a static time shift (e.g., 800 ms) for training machine learning models imprecise. To account for this cue onset and offset variability, we first performed a dynamic cue shifting technique to automatically shift cue labels to match EMG activity (Additional file 1 : Figure S4A). An average of 843 ± 95 ms of cue data per cue change or a grand average of 16.1 ± 1.0% of the full cue data stream across all subjects was shifted using this technique (Additional file 1 : Figure S4B). To verify this method, we compared decoding performance with and without cue shifting. Dynamic cue shifting significantly improved decoding performance achieving 74.7 ± 5.0% overall with no cue shift achieving 62.5 ± 6.7% (Fig. 5 A; Cue Shift: paired t-test Dynamic vs. None, p = 0.020). However, we found no significant difference in decoding accuracy between dynamic cue shifting and a static 800 ms cue shift (70.5 ± 5.4% decoding accuracy) representing an estimate of the average dynamic shift (Fig. 5 A; Cue Shift: paired t-test Dynamic vs. Static 800 ms, p = 0.22). One subject (13,762) had an increase in decoding performance from a static shift, while the rest of the subjects experienced a decrease or no change in performance, suggesting that the dynamic cue shift was the most robust technique for our analyses. Using the dynamic cue shifting technique, we investigated decoding performance of individual movements in the continuous dataset. The confusion matrix with individual movements for a single subject is shown in Fig. 5 B. The best performing movements across subjects were Wrist Flexion and Wrist Extension, with an average accuracy of 61.2 ± 5.0%. The worst performing movements across subjects were Forearm Supination and Thumb Abduction, with an average accuracy of 29.5 ± 9.0%. A continuous time series plot of all movement probabilities is shown in Fig. 5 C Shaded regions indicate the cued movement with the probability of the movement type decoded based on motor intention. To assess whether the NN decoder could be used in real-time situations, inference testing was conducted using a Surface Book 2 with NVIDIA GeForce GTX 1060 GPU. The NN decoder was trained using cued movement data collected at the beginning of the session, totaling 1526 sample bins (2.54 min) for subject 13,762 and 3000 sample bins (5.0 min) for subject 30,458. The trained NN decoder was exported and loaded in using the Open Neural Network Exchange (ONNX) Runtime [ 33 ] for inference testing. NN forward model prediction times on average took less than 1 ms (307 ± 49 μs). Taking the entire preprocessing pipeline into consideration in addition to the NN forward prediction, the total inference time was 23.1 ± 4.4 ms. Since the resulting inference time is under 100 ms (time bin for RMS feature calculation), the NN model was deemed suitable for real-time inference. We next tested the decoder online to verify closed-loop control of a virtual hand on two stroke subjects (13,762 and 30,458) using the NN model. The confusion matrix with individual movements tested during the online testing for Subject 30,458 are shown in Fig. 6 A. The best performing movement was Forearm Supination, with an overall accuracy of 97.1%. The continuous time series plot of movement probabilities for Subject 30,458’s online decoding session is shown in Fig. 6 C. Finally, videos from the online sessions are shown in Additional file 1 : Media 1 & 2, with the user following along with movements cued from an experimenter, and the decoded motor intention controlling a virtual hand on the computer monitor. These videos demonstrate the online decoding accuracy and responsiveness of the system and highlight the utility of the NeuroLife EMG System for eventual closed-loop control of upper-extremity devices. The NeuroLife Sleeve meets usability needs of chronic stroke survivors Summary data from the usability questionnaire suggest that the NeuroLife Sleeve meets many user needs (Fig. 7 ). Subjects answered questions on a scale of 1 to 5 with higher values indicating stronger agreement. In general, subjects were optimistic that they could don and doff the NeuroLife Sleeve with the help of a caretaker in their home (3.60 ± 0.28). Concerns were generally centered around the pre-application of the conductive spray and relative positioning of the system, which we are actively addressing in our next design iteration. During sessions, subjects had the sleeve donned for > 1.5 h, and all participants reported general satisfaction with the overall comfort of the device (4.57 ± 0.20). The sleeve was designed with a lightweight stretchable fabric, and participants were generally satisfied with the ability to move their arm while the sleeve was donned (4.07 ± 0.32). Subjects were highly confident (4.07 ± 0.22) that they could wear the sleeve during functional light activities around their home, suggesting that the sleeve is non-restrictive, lightweight, portable, and promising for home use. A commonly overlooked barrier to widespread adoption of assistive technologies is user acceptance of the overall look and feel of the device [ 4 ]. All subjects were extremely satisfied with the overall design of the sleeve (4.36 ± 0.24). In general, they were all very excited for the opportunity to use the sleeve with the “general favorability” metric receiving the highest score of 4.79 ± 0.15. In summary, the usability results from the current study provide promising early data that the NeuroLife Sleeve can meet end user needs with directions on where to improve for future iterations.
Discussion In this study, we demonstrate decoding of motor intention using the NeuroLife EMG System in people with upper-extremity hemiparesis due to chronic stroke. Based on high-density surface EMG data collected from the forearm, 12 functional hand, wrist and forearm movements were classified with high accuracy. Overall decoding accuracy was associated with the subject’s ability to perform the movement (quantified here as observed movement score), with greater functional movement corresponding with higher decoding accuracy. Even in movements with little to no movement capacity (movement score ≤ 1), the system was able to accurately differentiate movement intent, albeit with some decrease in performance. Furthermore, we report that decoding performance was associated with a variety of different aspects of impairment such as overall motor impairment and spasticity. We also demonstrate online decoding of 3 task-relevant movements and rest for closed-loop control of a virtual hand, highlighting the decoding accuracy, speed and responsiveness of the system. Usability data demonstrated that the sleeve is comfortable and lightweight, allowing stroke survivors to wear the sleeve for extended periods of time without restricting their movement. In summary, this work demonstrates the NeuroLife EMG System’s utility as a wearable, user-friendly device to infer movement intention in stroke survivors with severe motor impairments. Previous studies have demonstrated decoding of motor intention using surface EMG in the upper extremity in chronic stroke survivors [ 21 , 34 – 36 ]. In these studies, a range of machine learning techniques, impairment levels of the participants with stroke, and types of movements were investigated. The classification accuracy we measured was comparable to previous work with similar movement sets, although differences in study methodology restrict direct comparison. We found that a NN model outperformed the LR and SVM models in both able-bodied and stroke subjects across all movements. However, decoding accuracy decreased in stroke subjects with severe motor impairments. Specifically, we find our hardware and NN decoding techniques provide high performance in able-body (96.8% accuracy) and stroke (85.7% accuracy) participants if visible movement was observed. Included in the 85.7% accuracy are movements where participants had incomplete or impaired range of motion, indicating that we could consistently decode the subject’s intent to move despite their inability to properly complete the movement. These movements would be strong candidates for improvement with an EMG-controlled assistive device. Our complete 12-movement survey is helpful for understanding what movements may be decodable for each subject and may be appropriate for facilitating ATs in individuals with moderate or mild hand impairments. However, those with severe hand impairments are unlikely to be able to accurately control that many movements. Instead, it may be desirable to use only a subset of movements customized to the individual, that they can accurately control. A dynamic cue shifting technique may present a more robust and automated solution to account for differences among subjects. Improvement in decoding performance from using dynamic cue shifting is likely due to: (1) improved accuracy of the timing of cue onset and offsets in the training data which gives a better representation of each movement and thus better decoding performance, and (2) more accurate testing alignment and better testing parameters. These results suggest that cue labeling can substantially affect overall decoding performance in online decoders, and intelligent cue labeling can improve overall performance. Though we only briefly assessed our system’s online decoding capabilities, our initial results suggest that online EMG decoding of motor intention is possible, though more subjects and functional movements are needed to increase robustness. Recent studies have shown encouraging results using a limited set of manually placed electrodes, which may account for some performance differences compared to our results [ 21 , 35 ]. Moreover, localizing electrodes to muscle activity critical to grasp production can be an effective strategy to minimize system complexity. The optimization of electrode placement and reduction of hardware complexity is a planned future direction for the NeuroLife Sleeve. Additional studies have used similar numbers of channels as we have presented [ 21 , 37 , 38 ]. However, the system presented here streamlines system setup with a single zipper closure aligned by easily recognizable anatomical landmarks, the elbow and radial styloid process. Usability around donning/doffing is a key concern for adoption of AT, and systems with extensive setup procedures risk poor acceptance in clinics, rehabilitation settings, and the home. Prior studies have also shown that time domain features, such as RMS, combined with NN approaches can outperform more classical statistical or machine learning approaches [ 21 ]. Our results agree with these findings, further supporting that high density EMG recordings have sufficient complexity to leverage the recent developments in deep learning. We extend the findings of previous studies by presenting an easy-to-don and doff wearable device that removes the need for manual placement of electrodes. This reduces the necessary setup time and ensures consistent placement of recording electrodes across sessions. Additionally, we present data to support the real time performance of the decoding paradigm. This study provides evidence that the device can decode motor intention with high performance across a variety of subjects, and we demonstrate decoding speed that is fast enough to reliably perform real-time inference alongside data collection. Notably, processing in this context did not involve removing transition periods during training or online testing, indicating the system’s robustness to motion artifact. Finally, we present a viable, automated cue shifting method that removes the necessity for manual relabeling and improves system performance. Usability is an important factor for clinical technologies to assist with stroke rehabilitation by supporting motivation for consistent and active training. While existing AT solutions show promising results, these systems tend to focus on the technology and often fall short in the user-centric designs. Most clinical ATs involve manual placement of patch electrodes and long calibration procedures which limits the amount of practice that can be achieved within a given rehabilitation session. Furthermore, many systems are bulky and lack portability, which can limit patient adoption for use outside of rehabilitation training and into the home [ 39 ]. The system evaluated here uses a lightweight wearable and reusable sleeve connected with a ruggedized cable to an 8 × 10′′ signal acquisition module. Further work is required to ensure this system’s reliability a variety of home and non-laboratory contexts, but here we demonstrate that the NeuroLife EMG System can address many usability concerns of current technologies while providing robust decoding of motor intention. In combination with soft exoskeletons or FES, the sleeve can drive intention-based training coupled with functional movements in a user-centric form factor. Based on user feedback from the current study, the sleeve design meets various end user needs. The design allows for use on either arm, and the stretchable, lightweight fabric design was reported by participants to be comfortable without limiting natural arm movements. Aesthetically, subjects were pleased with the sleeve design and advocated that they would use the system at home for rehabilitation and activities of daily living given the opportunity. Participants mostly agreed that the sleeve was straightforward to don and doff during the study with the help of the researchers and believed that they could apply the sleeve with the help of a caretaker. However, participants identified the simplicity to apply the sleeve as an area that is currently lacking, and participants were not confident in being able to apply the sleeve independently without assistance. This is an identified area for future development and will be the focus of next design iterations to enable at-home use. Despite this current usability limitation, participants indicated that not only would they feel comfortable performing rehabilitation therapy at home but are excited for the possibility of using the sleeve as a therapy tool indicated by the highest score for general favorability. This study expands the scope of previous EMG decoding studies by presenting the performance of a novel algorithm across a wider range of subjects, UEFM score of 7 to able-bodied, in offline and online contexts while highlighting the importance of usability. The data collection was designed to simulate a realistic use case in which EMG-controlled ATs are used to assist in tabletop task-oriented upper-extremity rehabilitation. This study indicates the practicality and usability of AT control using this EMG system and highlights the shortcomings of decoding in severely impaired subjects and low observed movement scores. These findings will inform future work for the field of EMG decoding and may inspire new approaches for EMG-controlled ATs in the space of rehabilitation suitable for severely impaired stroke survivors. The present study provides an initial demonstration of the NeuroLife EMG System to decode motor intention in chronic stroke survivors while simultaneously meeting needs, but some limitations merit consideration. We did not age match the able-bodied subjects to the stroke subjects, which may have affected comparisons between the two groups of subjects. Data was not collected from the non-paretic arm in the stroke subjects, although we do provide data from able-bodied subjects to demonstrate high-accuracy decoding to validate our approach. While the reported results indicate that the Neurolife EMG System can be used to decode motor intention in a package that meets end user needs, there is still room for improvement in various areas, including refinement of decoding algorithms, the sleeve design and related hardware, and eventual applications. Future work refining decoding algorithms will focus on overall improvements to decoding performance by leveraging many of the advancements made in recent years in the field of deep learning [ 40 ]. We will investigate the use of more sophisticated neural network models, including recurrent neural networks (RNNs), transformers optimized for time series modeling which could improve overall decoding accuracy, specifically for participants with limited movement capability [ 41 , 42 ]. Advanced neural network models may also aid in our ability to identify altered states of muscle activity, including spasticity and fatigue [ 43 , 44 ]. To better address inter-session variability, we will apply various machine learning techniques including unsupervised learning, data augmentation, and domain adaptation [ 45 – 48 ] to fully leverage multiple datasets to reduce setup and calibration times for new users. This study used about 30 min of intrasession data for training of 13-class decoders, but with further development of these techniques, similar performance may be achieved with a much shorter decoder recalibration sequence after an initial training session. We demonstrated high performance using only 3 min of training data in a 4-class online decoding scenario, which may be acceptable for some use cases. Our current model architecture does not consider the spatial information available in the sleeve. A future direction for feature extraction and decoder architecture is to include features that capture this relational data between electrodes to create decoders that are less sensitive to positional changes, such as convolutional, transformer, or graph neural networks. Improvements to data quality itself can be accomplished with visual reinforcement to subjects. An online decoding system that displays the decoded intention may be more beneficial to subject engagement over the image cues used in the current study. While we provide the initial proof-of-concept demonstration of the NeuroLife EMG System here, the data collected during the study was not representative of how the system will be ultimately deployed as an assistive device. For example, in the current study, subjects kept their elbow stationary on the table during movements and did not interact with objects, both of which can significantly influence forearm EMG activity and thus decoding performance. Future studies will focus on capturing training data in more complex situations, such as during reach and grasp tasks and object manipulations, to develop decoders robust to movement such as the spatial decoders described above. We also assumed a class distribution based on the target use-case of occupational therapy in which Rest periods occur in between movements (~ 50% of the time). However, this method oversamples the Rest class, which can mask poor performance of other movements which are more functionally relevant thus limiting comparisons to other decoding studies. For this study, we determined the chance level using a naïve decoder (i.e., always predicting the majority class). We acknowledge that this baseline is dependent on the class distribution, thus we also present the success rate metric which is designed to approximate a therapist judging binary success for each movement cue (and ignoring rest periods). Future work will examine different decoders and metrics, including those decoders presented here in Fig. 3 C. Similarly, the decoding performance presented here was in the absence of assistive device control. Commonly used assistive devices, including FES and exoskeletons, may interfere with EMG activity when active and thus can significantly affect decoding performance [ 34 , 49 ]. Our group is working to integrate FES functionality within the same EMG recording electrodes to eliminate the need for additional hardware, such as an exoskeleton or additional patch electrodes. Future work from our group will focus on developing algorithms that can decode EMG during FES activity. Furthermore, integration with assistive technologies will change the sleeve form factor as well as the backend hardware. The usability assessment in this study focused primarily on the sleeve component of the EMG system. Future work will include optimization of the backend hardware for space efficiency and portability with studies evaluation of the complete system usability. With a technology that incorporates EMG and FES into a single consolidated sleeve, the system has the potential to help support motor recovery and assist in ADLs [ 14 , 25 ].
Conclusion The focus of this study was to validate the NeuroLife EMG System by decoding hand, wrist, and forearm movements and collect usability data from subjects with stroke. We demonstrate accurate EMG decoding of 12 different movement classes with a neural network in both able-bodied and stroke subjects. Decoding accuracy in stroke subjects was associated with the movement ability of each subject. The decoding results were consistent with similar myoelectric intention-based studies. We demonstrate online decoding and closed-loop control of a virtual hand with high accuracy, speed, and responsiveness. Finally, we present data on the common usability factors of assistive devices including the simplicity, comfortability, portability, and weight of the sleeve. Overall, all subjects reported good to outstanding ratings for each of the usability categories, indicating that the NeuroLife EMG System can provide accurate decoding of upper extremity motor intention while meeting the usability needs of end users.
Objective Seventy-five percent of stroke survivors, caregivers, and health care professionals (HCP) believe current therapy practices are insufficient, specifically calling out the upper extremity as an area where innovation is needed to develop highly usable prosthetics/orthotics for the stroke population. A promising method for controlling upper extremity technologies is to infer movement intention non-invasively from surface electromyography (EMG). However, existing technologies are often limited to research settings and struggle to meet user needs. Approach To address these limitations, we have developed the NeuroLife ® EMG System, an investigational device which consists of a wearable forearm sleeve with 150 embedded electrodes and associated hardware and software to record and decode surface EMG. Here, we demonstrate accurate decoding of 12 functional hand, wrist, and forearm movements in chronic stroke survivors, including multiple types of grasps from participants with varying levels of impairment. We also collected usability data to assess how the system meets user needs to inform future design considerations. Main results Our decoding algorithm trained on historical- and within-session data produced an overall accuracy of 77.1 ± 5.6% across 12 movements and rest in stroke participants. For individuals with severe hand impairment, we demonstrate the ability to decode a subset of two fundamental movements and rest at 85.4 ± 6.4% accuracy. In online scenarios, two stroke survivors achieved 91.34 ± 1.53% across three movements and rest, highlighting the potential as a control mechanism for assistive technologies. Feedback from stroke survivors who tested the system indicates that the sleeve’s design meets various user needs, including being comfortable, portable, and lightweight. The sleeve is in a form factor such that it can be used at home without an expert technician and can be worn for multiple hours without discomfort. Significance The NeuroLife EMG System represents a platform technology to record and decode high-resolution EMG for the real-time control of assistive devices in a form factor designed to meet user needs. The NeuroLife EMG System is currently limited by U.S. federal law to investigational use. Supplementary Information The online version contains supplementary material available at 10.1186/s12984-023-01301-w.
Supplementary Information
Acknowledgements The authors would like to thank our development and management teams at Battelle Memorial Institute including Jesse Keckler, Nick Annetta, Sam Colachis, and Josh Branch for their engineering contributions, Charli Hooper for her assistance with data collection, and Andrew Sweeney for his contribution to the manuscript graphics. Financial support for this study came from Battelle Memorial Institute. Author contributions EM conducted the experiments, performed analysis, prepared the figures, and wrote the manuscript. DG, performed analysis, prepared figures, and wrote the manuscript. MD, LW conducted the experiments and reviewed the manuscript. NT performed analysis and prepared figures. BS prepared figures and reviewed the manuscript. IB, DF reviewed the manuscript. All authors reviewed the manuscript. Funding Funding for the study was provided by Battelle Memorial Institute. Availability of data and materials The data that support the findings of this study are available upon reasonable request from the authors. Declarations Ethics approval and consent to participate Data were collected as part of an ongoing clinical study (IRB0773, IRB 0779) being conducted at Battelle Memorial Institute that was approved by the Battelle Memorial Institute Institutional Review Board. All participants provided written informed consent before participation, in accordance with the Declaration of Helsinki. Consent for publication Consent was obtained from participants for publication of these data. Competing interests All authors declare no conflict of interest.
CC BY
no
2024-01-15 23:43:46
J Neuroeng Rehabil. 2024 Jan 13; 21:7
oa_package/f2/cb/PMC10787968.tar.gz