Deciphering Abnormal Platelet Subpopulations in Inflammatory Diseases through Machine Learning and Single-Cell Transcriptomics

doi:10.21203/rs.3.rs-4014672/v1

Download PDF

Research Article

Deciphering Abnormal Platelet Subpopulations in Inflammatory Diseases through Machine Learning and Single-Cell Transcriptomics

https://doi.org/10.21203/rs.3.rs-4014672/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

The study focuses on understanding the transcriptional heterogeneity of activated platelets and its impact on diseases like sepsis, COVID-19, and systemic lupus erythematosus (SLE). Recognizing the limited knowledge in this area, our research aims to dissect the complex transcriptional profiles of activated platelets to aid in developing targeted therapies for abnormal and pathogenic platelet subtypes. We analyzed single-cell transcriptional profiles from 47,977 platelets derived from 413 samples of patients with these diseases, utilizing Deep Neural Network (DNN) and eXtreme Gradient Boosting (XGB) to distinguish transcriptomic signatures predictive of fatal or survival outcomes. Our approach included source data annotations and platelet markers, along with SingleR and Seurat for comprehensive profiling. Additionally, we employed Uniform Manifold Approximation and Projection (UMAP) for effective dimensionality reduction and visualization, aiding in the identification of various platelet subtypes and their relation to disease severity and patient outcomes. Our results highlighted distinct platelet subpopulations that correlate with disease severity, revealing that changes in platelet transcription patterns can intensify endotheliopathy, increasing the risk of coagulation in fatal cases. Moreover, these changes also seem to impact lymphocyte function, indicating a more extensive role for platelets in inflammatory and immune responses. This study sheds light on the crucial role of platelet heterogeneity in serious health conditions, paving the way for innovative therapeutic approaches targeting platelet activation, which could potentially improve patient outcomes in diseases characterized by altered platelet function.

COVID-19

sepsis

platelets

single-cell RNA-seq

machine learning

Platelets, also known as thrombocytes, are cell fragments that lack a nucleus. Recent research has shown that platelets play a critical role in the immune system [1, 2], acting as first responders during infections and regulating the activation and post-infection deceleration of host immunity. They modulate the function of immune cells through physical adhesion or by releasing cytokines and chemokines, influencing the activation [3, 4], proliferation [5], differentiation [6], pathogen clearance [7], and cytokine response of other immune cells [8, 9]. Abnormal platelet activity has been implicated in numerous diseases involving immune system dysregulation [10–13].

In sepsis, platelets are activated and play key roles in immune response and coagulation. While the overall platelet count drops, the activated fraction of platelets increases in sepsis due to thrombin-mediated activation [14]. The immature platelet fraction (IPF) is also higher in sepsis patients compared to healthy individuals, which could be part of the ongoing inflammatory response to sepsis [15, 16].

In severe COVID-19, similar to sepsis, there is a trend of dropping platelet counts, particularly in patients with severe disease [17]. A meta-analysis of 7,613 COVID-19 patients revealed that patients with severe disease had a lower platelet count than those with non-severe disease. This decrease in platelet count could suggest a worsening thrombotic state, with lower nadir platelet counts associated with increased mortality [18]. At the same time, there is evidence of increased platelet activation in COVID-19. Platelets are hyperactivated in COVID-19, but the mechanisms promoting this activation are not fully understood [19]. Damaged lung tissues and pulmonary endothelial cells may activate platelets in the lungs, resulting in aggregation and formation of microthrombi [20]. Furthermore, there is growing evidence of platelet hyperactivation contributing to the severity of COVID-19. Increased levels of platelet and platelet-leukocyte aggregates have been linked to COVID-19 severity.

Single-cell RNA sequencing (scRNA-seq) has been used to study the transcriptome of isolated peripheral blood platelets from patients with various diseases, including periodontitis and diabetes [21], ovarian cancer [22], and COVID-19 [23]. Utilizing single-cell analysis, we aimed to identify and characterize subpopulations of abnormal platelets, investigate the molecular details of their pathogenic changes, and develop testable hypotheses regarding their contribution to patient outcomes, including fatal or survival.

Integrated Single-cell Transcriptome Atlas of Peripheral Blood Mononuclear Cells (PBMCs) from COVID-19, Sepsis, and Systemic Lupus Erythematosus Patients

We gathered single-cell RNA-seq datasets of peripheral blood mononuclear cells (PBMCs) from COVID-19 [24–32], sepsis [12, 33], and systemic lupus erythematosus (SLE) [34] patients, for a total of 413 samples, in order to study and compare the immunological dysregulation caused by these diseases. In the COVID-19 studies, individuals with severe influenza, lung infections but testing negative for COVID-19 were identified as the "hospitalized patients with similar symptoms" (SSH) group and we included them in our analysis. We categorized the samples into six groups based on the severity of the disease: healthy control (HC), convalescence (CV), mild (ML), moderate (MD), severe (SV), fatal (FT) and created a separate group for systemic lupus erythematosus (SLE). For the outcome analyses we grouped patients who survived from the CV, ML, MD and SV groups into a survivor (S) group, with the patients with unannotated outcomes forming another group labeled as the unknown group (Fig. 1A).

From the 413 samples, a total of 47,977 individual platelets were extracted. These platelets were classified based on disease severity, there were 3,205 cells from HC, 3,695 cells from CV, 7,359 cells from ML, 4,330 cells from MD, 19,805 cells from SV, 9,414 cells from FT, and 169 cells from SLE. By the outcome group, there were 3,205 from HC, 9,414 from FT, 25,750 from S, and 9,608 from unknown outcomes. Overall, there are 38,673 platelet datasets from COVID-19, 2,508 from SSH, 3,422 from sepsis, and 169 from SLE.

Removal of Doublets in scRNA-seq Data and Identifying the Cell Types

While platelets can form aggregates with monocytes within peripheral blood mononuclear cells (PBMCs), particularly in states of inflammation, the platelets used in our study were isolated as single cells. To reduce the events when two or more cells are incorrectly classified as a single entity, we excluded cells exhibiting more than 6,000 unique genes expressed [35]. Furthermore, we ensured the accuracy of platelet identification by using both source data annotation and the platelet cell marker PPBP. To enhance our analysis, we employed bioinformatics tools such as Seurat [36] and SingleR [37]. Seurat, a powerful tool for single-cell RNA sequencing data analysis, facilitated cell type classification by projecting reference metadata onto a query object without altering the query expression data. After identifying the anchor points, we used Seurat's TransferData() function, which provided us with a matrix of predicted IDs and prediction scores. This process allowed us to categorize our query cells based on the reference data. SingleR was another integral part of our toolkit. This tool functions by comparing the expression profile of each single cell in the query dataset with each of the reference expression profiles. By calculating the similarity between the query cell and each reference cell, SingleR identifies the most likely cell type for each single cell in the query dataset. This comparison allows for precise and granular cell type annotation, which is essential for detailed single-cell analyses like ours.

Integrating the Datasets

Expression data for platelets was extracted from all the datasets identified in the previous step. Using HGNChelper [38], we harmonized the nomenclature of the genes from different datasets and identified the 3000 most variable genes from each. For the integrated analysis, the top 3000 highly variable genes were retained. Harmony [39] was executed with PCA embeddings (30 PCs) as input and the default parameters to eliminate batch effects among twelve datasets (nine COVID-19 datasets, two Sepsis datasets, and one SLE dataset). Seurat was then applied to the Harmony embeddings to determine the clusters. In the round of clustering, the resolution for Louvain clustering was set to 0.4.

Differential Expression Analysis

The MAST method [40] from the Seurat package (implemented in the FindAllMarkers function) was used with the default parameters to perform the differential gene expression analysis and identified the DEGs. We used patients’ categories, as described before, to compare DEGs between disease severity states.

Machine Learning models

eXtreme Gradient Boosting (XGB) [41], is an advanced decision-tree-based algorithm that enhances the traditional decision tree model by building a series of successive trees where each new tree corrects the errors made by the previous ones, effectively reducing predictive errors through a technique known as gradient boosting. XGB's architecture allows for the incorporation of new input data, in this case, gene expression data from platelets classified according to survival or fatal outcomes, to refine the model iteratively. XGB's inherent feature selection capability, attributed to its bagging property, enables us to use all gene expression values to determine the most informative features automatically. To optimize the model's parameters, a grid search was employed to explore a wide array of hyperparameter combinations, such as learning rate, number of trees (estimators), and tree complexity (maximum depth), settling on a learning rate of 0.1 and a maximum depth of 9 for our model. The model's predictive power was evaluated using a confusion matrix. The XGBoost model was developed within the R environment, utilizing the XGBoost package.

Parallelly, the Deep Neural Network (DNN) model was developed using the Keras and TensorFlow Deep Learning frameworks. The model's hyperparameters, including the number of epochs, batch size, choice of optimizer, weight initializer, quantity of hidden layers, and number of nodes per hidden layer, were meticulously tuned through a comprehensive Grid Search approach. This method tested numerous hyperparameter combinations to determine those that yield the highest accuracy. The significance of each feature within the DNN model was assessed based on its impact on model accuracy when its values were permuted; a more substantial decline in accuracy indicates a higher feature importance. This DNN model implementation was performed in R, leveraging the capabilities of the 'caret' package.

Comparison of Module Scores

We measured the extent to which individual cells expressed certain predefined expression gene sets using cell module scores. All calculations and comparisons of module scores were performed using the AddModuleScore function from the Seurat package with the default parameters. We compared the expression of modules extracted from the GO.db database [42]. Additionally, we used literature-based modules for oxidative phosphorylation (OXPHOS), glycolysis [43], and MHC class II. We disregarded genes without detectable expression in our data.

Pathway Enrichment Analysis

We analyzed Gene Ontology biological process terms (GO-BP) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways using R package clusterProfiler [44] to identify the genes up- and down-regulated in disease samples as compared to HC. The "common up" and "common down" gene sets at the disease level were identified by intersecting the upregulated or downregulated DEGs for COVID-19, sepsis, and SLE compared to HC. We performed Medical Subject Headings (MeSH) enrichment using the R package DOSE [45] for each cluster's upregulated genes. We also calculated scores for the Molecular Signatures Database (MSigDB) signature gene set collection [46] using the R package GSVA [47].

Calculation of Ligand-Receptor Interaction Scores Between Platelets and Other Cell Types

Cell-cell ligand-receptor interactions were evaluated using the scoring system proposed by Kumar et. al. [48]. The score uses ligand-receptor interactions from iTALK ligand-receptor database [49]. The ligand-receptor interaction score between cell type A and cell type B is calculated as a product of average receptors’ expression across all cells of cell type A and the average ligands’ expression across all cells of cell type B.

$$Interaction score(receptor, ligand, cell type A, cell type B) = \frac{1}{{n}_{cell type A}}{\sum }_{i \in cell type A}^{}{e}_{i, receptor}\times \frac{1}{{n}_{cell type B}}{\sum }_{j \in cell type B}^{}{e}_{j, ligand}$$

e_i,j = expression gene j in cell i

n_celltypeA = number of cells in the cell type A

Trajectory Inference of Transition in Platelets Subclusters

We presented the platelets trajectory analysis using monocle3 [50]. Trajectories were calculated and the cells displayed based on monocle3 pseudotime approach rooted in the previously identified transitional platelets. Integrated gene expression matrices from each dataset were first exported from Seurat into monocle3 to construct a new_cell_data_set. The exported matrices were aligned using the single-cell batch correction method package batchelor [51]. Then, they were subjected to the standard PCA to preprocess the data with the number of dimensions set at 100. The dimension reduction and clustering of cells were all set to monocle3 default settings. Finally, we used reversed graph embedding to learn the principal graph from the reduced dimension space using the learn_graph function.

PBMC Composition Changes with Patient Disease Severity and Outcome.

First, we analyzed the relative populations (fractions) of immune cell subsets in peripheral blood mononuclear cells (PBMCs) in relation to disease severity, outcome, and disease types. The population proportions of platelets, precursor cells, and erythroblasts in PBMCs increased with the disease and outcome severity (Fig. 1B, Supplementary Fig. 1A, B), while the T cell fraction decreased (Fig. 1B). B cell fractions were significantly increased in SLE patients, which is a known phenotype in SLE patients (Fig. 1C); monocyte and neutrophil fractions were significantly increased in convalescence patients (Fig. 1E, F); however, fractions of the dendritic cells decreased as outcome severity increased (Fig. 1G). We evaluated ratios of different cell types as a criterion of separating FT and S patients and found that the ratio of platelets to T cells, (Pla-T ratio) had the highest area under the curve (AUC) at 0.754, with the 0.063 ratio between FT and S patients (Fig. 1H).

XGBoost and Deep neural network modeling identifies biomarkers of survival and fatal platelets.

In our study, we employed two machine learning approaches—a Deep Neural Network (DNN) and an eXtreme Gradient Boosting (XGB) algorithm—to analyze and interpret with the objective of identifying key biomarkers. These biomarkers are indicative of patient survival and the presence of fatal platelets, which are critical in the prognosis and treatment of life-threatening conditions. When comparing the two models, the XGB model has a slightly higher accuracy, while the DNN model shows marginally better performance in F1 score, precision, and recall. This could suggest that while XGB is slightly better at correctly identifying both positive and negative classes, DNN is better at identifying positive cases when the outcome is positive and also better at not missing positive cases (Fig. 2A). We only obtained only the features that are within the top 5% of importance or gain from their respective models. We also looked into genes have the absolute log2fc > 1. The overlap of 21 biomarkers suggests that there is a consensus between the three models on these biomarkers' significance in relation to the outcome of interest (e.g., survival or fatal platelets) (Fig. 2B). In order to discern the specific features that could potentially serve as biomarkers indicative of either survival or fatal outcomes, we meticulously labeled each gene within a volcano plot, as presented in Fig. 2C of our study. The biomarkers identified in survival platelets include AIF1, FOS, CD74, JUN, JUNB, HLA-DRA, MNDA, RPL39, RPS21, RPS18, EEF1A1, RPS28, RPL34, S100A8, S100A11, S100A12. The biomarkers identified in fatal platelets include HBA2, HBB, HPSE SLC25A37 TMCC2. We conducted a gene ontology enrichment analysis on the identified genes. For genes associated with survival, the enrichment analysis revealed a strong connection to processes involved in cytoplasmic translation and the active immune response, as detailed in Fig. 2D. In contrast, genes related to fatal outcomes were predominantly associated with pathways involved in coagulation and reactive oxygen species (ROS) metabolic pathways, as depicted in Fig. 2E.

Platelets Amplify Endotheliopathy and Disseminated Intravascular Coagulation in Fatal Patients

Platelets have been shown to play a role in the development and progression of endotheliopathy and disseminated intravascular coagulation (DIC) [52, 53]. Platelets can bind to and activate endothelial cells, releasing pro-inflammatory and pro-coagulant molecules that can contribute to the development of DIC. Additionally, platelets can also contribute to thrombus formation and further damage to the endothelium [54].

We looked into the expression of the Integrin Subunit Alpha 2b (ITGA2B) gene, which is involved in platelet-endothelial cell interactions by binding to the Integrin Subunit Alpha V (ITGAV) [55]. The patients who ultimately passed away had the highest levels of ITGA2B expression, followed by the patients who were in severe disease state (Fig. 3A). On the pathway level, the reorganization of actin cytoskeleton (GO:0007010), also followed the same severity trend (Fig. 3B). DIC is a complex condition characterized by abnormal clotting and bleeding due to the activation of the coagulation cascade and the depletion of clotting factors and platelets [56]. Several other GO terms associated with DIC, such as blood coagulation (GO:0007596), inflammatory response (GO:0006954), extracellular matrix disassembly (GO:0022617), and platelet activation (GO:0030168) expression had highest expression in FT/SV patients. All modules had lower expression in CV and SLE patients (Fig. 3C). Fig. 3. Differential expression of platelets affecting endotheliopathy across disease severity states

Platelet Subpopulations Associated with Disease Severity

The standard analysis of scRNA-seq focuses on identifying clusters that overlap with the standard cell types. Here we carried the clustering further to understand the details of platelet population changes in diseases. By trial and error, we identified similarity thresholds that resulted in stable clusters with the best separation and the minimal signature pathway overlap, resulting in thirteen clusters, designated C0 through C12 (Fig. 4A, B). We computed the composition of each cluster in terms of the contributions from different outcome groups. The C11 cluster was strongly associated with the FT group, with 78% of all C11 cells found in this group. Clusters C3, C5, and C9 had the highest proportion of HC (17%), while clusters C6 and C11 had the lowest (1%). Cluster C3 had the highest proportion of mild group at 25%. In clusters C6, C7, C10, and C12, the severe group comprised over fifty percent of total platelets (Supplementary Table 1). Over 70% of platelets in clusters C6, C8, and C10 come from the samples from the survivors (Supplementary Table 2). Based on these results, we designate clusters C4, C9, and C11 as “fatal”, C8 as “convalescent”, and C6 and C10 as “survival” (Fig. 4C, D).

Characteristics of Platelet Subpopulations

We used DEG and Gene Set Enrichment Analysis (GSEA) to identify molecular factors and pathways that could help us differentiate between the clusters (subpopulations) platelet identified in the previous step. The sets of genes enriched in the fatal platelet clusters C4 and C11 were significantly different from that in other platelet clusters (Fig. 5A) and in some cases show opposite trends from that in the convalescent and survival clusters. The fatal cluster C4, whose signature module comprises HPSE, WFDC1, and PF4 genes, and the convalescence cluster C8, whose signature module consists of RPLP0, RPS6, and RPS23, have several pathways that trend in the opposite direction. Fatal cluster C11, whose signature module includes TPT1, POLR2L, and CSRP1, had the highest energy consumption pathway scores, including oxidative phosphorylation (OXPHOS) and glycolysis (Supplementary Fig. 2A, B), but the lowest inflammatory response score (Supplementary Fig. 2C). The other fatal cluster C9, whose signature modules include HBB, HBA2, and HBA1, showed the weakest interferon response, including alpha and gamma interferons (Supplementary Fig. 2D, E). Coagulation, epithelial-mesenchymal transition (EMT), and the apical junction are all at their highest in C4 but at their lowest in C8 (Fig. 5B-D). At the same time, MYC targets v1 and v2, which contain nuclear-encoded genes involved in mitochondrial biogenesis [59], are the lowest in C4 but highest in C8 (Supplementary Fig. 2F, G). Based on these findings, we can conclude that platelets from the fatal cluster C4 are highly active in angiogenesis, coagulation, and endotheliopathy, while having the lowest RNA processing, cell division, and mitochondrial biogenesis, while those from the convalescence cluster C8 had the opposite trend in all these pathways.

Cluster C6, associated with survival, was characterized by elevated expression of genes CRBN, CD58, and SRSF3, and exhibited the most significant activity in the Notch signaling pathway alongside minimal IL6/JAK/STAT3 signaling pathway activity, as shown in Supplementary Figs. 2H and I. Conversely, survival cluster C10, distinguished by a gene module including CD74, HLA-DRA, and CD79A related to antigen presentation, demonstrated higher activity in allograft rejection and MYC targets v2 pathways (Supplementary Figs. 2J and G). Both clusters C6 and C10 showed reduced activity in angiogenesis, coagulation, epithelial-mesenchymal transition (EMT), and apical junction pathways, which were notably upregulated in the fatal cluster C4, as detailed in Fig. 5B-D.

Then, we examined the GO terms up-regulated in the fatal clusters C4, C9, and C11. The key pathways up-regulated in fatal cluster C4 include wound healing, platelet activation, hemostasis, and coagulation, which is consistent with the known observations that thrombotic problems are a major cause of morbidity and mortality in COVID-19 patients [60]. Oxygen transport, hydrogen peroxide metabolism, gas transport, and erythrocyte development are enriched terms for C9, reflecting the hypoxic environment of C9 platelets (Fig. 5F). ATP metabolic activities such as oxidative phosphorylation, cellular respiration, and aerobic respiration are among the enriched terms in C11; this is consistent with the C11 GSEA results, which suggested that C11 platelets are inactive (Supplementary Fig. 2L). Platelets in quiescence are known for requiring ATP for their basic function. According to a study, glycolysis produces up to 65% of the required ATP in inactivated platelets, with mitochondria providing the remainder [61]. C4 is the most active cluster in FT groups. Therefore, we examined C4-enriched genes associated with diseases, such as arteriosclerosis, bacterial endocarditis, thrombosis, and frontotemporal lobar degeneration, a neurodegenerative disorder (Fig. 5G). Among the mechanisms shared by survivor clusters C6, C8, and C10 are platelet translation regulation, mRNA processing, and ribosomal RNA biogenesis (Fig. 5H, Supplementary Fig. 2M, N). Additionally, C10 was enriched for the GO terms homeostasis lymphocyte activation cells, and leukocyte antigen cell-cell adhesion (Supplementary Fig. 2N).

Pseudotime Trajectory Analysis Identifies Platelet Signature Dynamics in Survival or Fatal Disease Outcomes

The fatal cluster C4 appears to be critical for the negative outcome groups, so we evaluated possible events leading to its emergence. Pseudotime analysis, although unable to recover real time dynamics of cell populations, provides hints as to their order along the developmental trajectories [60]. The possible precursor of C4 is the C0 cluster, which is also close to the C1 and C6 clusters in the trajectory map (Supplementary Fig. 3A). C0 cluster has two possible developmental routes, fatal C4 or survival C1and C6 (Fig. 6A, B), indicating that the platelets in the C0 could be targeted by early intervention to inhibit their development into the C6 phenotype.

Then, we examined DEGs between C4 and C0 and C0 and C1. AKR7A2, CALD1, CALR, CD36, CSRP1, CYBA, ENSA, FCGR2A, HBG2, HCST, HMGN1, HPSE, MMRN1, NDUFA4, PF4, RPLP1, RPS9, SAMD14, and TPT1 genes are the consistently up-regulated, with a log2Fold change greater than 0.4 and an adjusted p value less than 0.05, from C0 to C4, and C1 to C0 (Fig. 6C, D). For the further analysis we defined a fatal platelet module consisting of these genes. Among the fatal module enriched GO terms are regulation of body fluid levels, coagulation, phagocytosis, and tumor necrosis factor (TNF) production (Supplementary Fig. 3B). The DEGs consistently down-regulated between C4 vs. C0 and C0 vs. C1 were ADIPOR1, CDKN1A, MAP3K7CL, MMD, NEAT1, NUTF2, PTMA, RAB31, SLC40A1, TMEM140, TSC22D3, they were used to define the platelet survival module. However, no GO, KEGG, or Reactome pathways were enriched in the survival module. The scores for the fatal and survival modules were then calculated for all disease clusters and disease severity levels. As expected, cluster C2 had the lowest fatal score, followed by C1 and then survivor cluster C6. C11 had the greatest fatal score, followed by C4 and C9 (Fig. 6E). C2 had the greatest survival module score, followed by C1 and survivor cluster C6, while C11 and C4 had the lowest scores for the fatal modules (Fig. 6F).

The fatal and survival module scores could also serve as indicators of the disease outcome. The patient group with the highest fatal scores was the fatal group, followed by the severe group, while the mild group received the lowest score on the fatal module, followed by the moderate group. The HC group was positioned in the center of the fatal module (Fig. 6G). The groups with the lowest survival module ratings were fatal and severe groups. However, the highest scores were in the SLE and mild groups (Fig. 6H). When platelets encounter an immunological disorder response, their expression shifts away from HC, as demonstrated by the findings. In addition, we investigated the composition of platelet subclusters and discovered that the fraction of fatal cluster C4 was the best indicator among clusters for distinguishing between S and FT patients (Fig. 6I), with an AUC of 0.749. When C4 platelets exceeded 3.36 percent of the PBMC platelet composition, patients were at risk of death. In this case, detecting the presence of platelets C4 in a timely manner could help in the patient's survival.

These findings show the complexity of COVID-19 and sepsis in relation to gene and pathway signatures in platelets. This is especially true for patients with severe active disease, as well as those recovering from it and surviving it. This research could lead to a better understanding of platelet processes in COVID-19 and sepsis patients, which could help guide therapeutic options for both patient subgroups.

Unique and Shared Gene Expression Changes in Platelets from COVID-19, SSH, Sepsis, and SLE Samples

Comparing COVID-19, SSH, sepsis, and SLE platelets to HC platelets, we identified differentially expressed genes (DEGs) and related pathways (whose up- and down-regulations were analyzed using the GO and KEGG databases). There were 45 DEGs that were up-regulated in all four diseases studied here, including IFI27L2, IFITM2, IFITM3, and S100 family genes (S100A8 and S100A9) and genes for ATP synthase, such as ATP5E and C9orf16. BEX3, LCN2, RHEB, and TMEM219 are genes associated with apoptosis (Supplementary Table 4) (Supplementary Fig. 4A). 271 DEGs were downregulated in COVID-19, 515 DEGs were downregulated in SSH, 901 DEGs were downregulated in Sepsis, and 747 DEGs were downregulated in SLE. 129 genes were downregulated in all four diseases (Supplementary Fig. 4B). The 129 genes included 68 ribosome-related genes. CD52 is a therapeutic target and predictive biomarker for sepsis (14). CD3E, CD48, CD7, LCK, LEF1, PTPRC, and TCF7 are required for the activation of T cells (Supplementary Table 4).

There were 294 pathways up-regulated in COVID-19, 346 in SSH, 394 in sepsis, and 139 in SLE as compared to HC (Fig. 7A). There were 47 pathways that were consistently up-regulated in the four considered diseases. There were 120 down-regulated pathways in COVID-19, 381 in SSH, 710 in sepsis, and 657 in SLE (Fig. 7B). Among them, 81 pathways were consistently down-regulated by all four considered diseases.

To better interpret the functional meaning of the consistently up/down-regulated pathways among the four diseases, we used Revigo [58] to summarize the 47 up-regulated pathways and 81-down-regulated pathways from COVID-19 vs. HC, SSH vs. HC, Sepsis vs. HC, and SLE vs. HC. Pathways up-regulated in platelets consistently among the four considered diseases include GO: neutrophil mediated immunity, KEGG: Parkinson disease, and GO: ATP metabolic process (Fig. 7D). All translation processes are down-regulated, including the GO: nuclear-transcribed mRNA catabolic process, nonsense-mediated decay, GO: regulation of translational initiation, GO: protein localization in the endoplasmic reticulum, KEGG: Ribosome, and GO: response to interleukin-4 (Fig. 7D). Interleukin-4 has several biological activities, including the stimulation of activated B cells and T cell proliferation, mediated by pathways such as T cell activation regulation and response [59].

We then focused on the pathways that were up-regulated in both COVID-19 and sepsis relative to HC. KEGG: Parkinson disease, KEGG: Huntington disease, KEGG: Prion disease, KEGG: Alzheimer disease, KEGG: Pathways of neurodegeneration - multiple diseases, and KEGG: Amyotrophic lateral sclerosis was consistently up-regulated in platelets from COVID-19 and sepsis (Supplementary Fig. 4C). COVID-19 has been recently associated with neurological diseases [60], even though the molecular mechanism of this association is not clear. As with sepsis, COVID-19 may act as a significant inflammatory insult, increasing the brain's susceptibility to neurodegenerative illnesses, cognitive decline, and the likelihood of acquiring dementia later in life [61]. However, up-regulation of the pathways seen in neurons in neurogenerative diseases in platelets may suggest an effect of common regulatory mechanisms rather than a direct effect of abnormal platelets on the brain.

The KEGG pathways related to Endocytosis and Fc gamma R-mediated phagocytosis were found to be up-regulated, indicating an active engagement of platelets in the internalization of virions. This internalization is facilitated by Toll-like receptors binding to virion-released lysosomal ligands such as single-stranded RNA, double-stranded RNA, and CpG DNA, which trigger platelet activation and the release of granules, leading to the exposure of P-selectin and the formation of platelet-leukocyte aggregates. Additionally, pathways linked to protein targeting, translation, and ribosomes were identified, alongside a noted down-regulation in the context of sepsis for KEGG: Coronavirus disease - COVID-19. This down-regulation is primarily attributed to the diminished expression of ribosomal proteins, a pattern observed in both COVID-19 and sepsis conditions, as elaborated in Supplementary Fig. 4C.

Then, we investigated the pathways that are enriched in COVID-19 and sepsis fatal patients (FT) in comparison to survivors (S) (Fig. 7C). GO: neutrophil mediated immunity and GO: ATP metabolic process were up-regulated in FT patients and were likewise up-regulated in disease vs. HC pathways. GO: response to endoplasmic reticulum stress, GO: response to hypoxia, and GO: intrinsic apoptotic signaling pathway in response to oxidative stress are up-regulated stress pathways. While the KEGG: Bacterial invasion of epithelial cells was also among the up-regulated pathways, we concluded that only the part of this pathway involved in cytoskeleton rearrangement is upregulated (see the discussion section). GO: antigen processing and presentation of peptide antigen via MHC Class I were also up-regulated in fatal patients of COVID-19 and sepsis. Platelet MHC Class I mediates CD8 + T-cell suppression during sepsis, according to a previous study [62]. Pathways such as GO: protein localization to the endoplasmic reticulum, GO: regulation of translational start, and GO: regulation of RNA stability were down-regulated. In addition, lymphocyte activation pathways, such as GO: regulation of T cell proliferation, GO: interferon-gamma-mediated signaling pathway, and GO: regulation of interleukin-12 production, were down-regulated in FT patients. Interferon-gamma is mostly secreted by activated lymphocytes, including CD4 T helper type 1 cells and CD8 cytotoxic T cells [63], whereas interleukin-12 is known as T cell-stimulating factor [64] (Fig. 7C). These data indicate that platelets from the disease cohorts exhibited less immunological activation, fewer translational activities, and more neurodegenerative tendencies, such as the KEGG enrichment for Parkinson's disease (Fig. 7D). Compared to survivor platelets, the aforementioned tendencies became more pronounced in FT platelets, which began to exhibit the ability to invade epithelial cells (Fig. 7C).

Pathway Enrichment Related to Disease Severity in Platelets

Given the dynamic gene expression changes in platelets in multiple diseases, we evaluate what gene expression modules were significantly changed depending on disease severity. Two modules that diminish with disease severity are MHC class II genes (Fig. 8A) and translation initiation (Fig. 8B). MHC Class II scores are much higher in convalescent patients than in healthy controls, indicating that MHC Class II can be utilized as an indicator of patient recovery. High case fatality rates of COVID-19 reported in some countries have been linked to inadequate MHC class II presentation and, consequently, a weak adaptive immune response against these viral envelope proteins, according to studies [65]. The lowest score for translation initiation modules (GO:0006413) was found in the platelets of patients who did not survive the diseases, indicating a halt in protein translation due to fatal illness. An interesting observation is the presence of pathways implicated in neurodegeneration (GO:0070843) in severely sick COVID-19 patients (Supplementary Figs. 5A-5H). Except for axonal transport modules, platelets from severe and fatal COVID-19 patients show all the trends observed in neurodegeneration diseases' major biological processes. In both sepsis and COVID-19 patients, neurodegeneration-related pathways became more severe as the disorders advanced. The scores for blood coagulation (GO:0007596), platelet activation (GO:0030168) (Fig. 3D), oxidative phosphorylation (OXPHOS) (Fig. 8C), and glycolysis (Fig. 8D) (35) modules confirmed the hypothesis that platelet coagulation and energy consumption are functionally linked to the severity of sepsis disease and the progression of COVID-19 disease [66, 67]. Moreover, platelets in convalescent patients had higher glycolysis scores, which corresponded to module scores in response to oxygen radicals (Supplementary Fig. 5D), indicating that platelets in convalescent patients were also hypoxic. Hypoxia induces oxidative damage to neural cells and causes widespread neurodegeneration [68].

Interferon response modules, such as response to type I IFN (GO:0034340) (Fig. 8E), IFN- β (GO:0035456) (Fig. 8F), and IFN-γ (GO:0003341) (Fig. 8G), had the highest scores in moderate patients and the lowest scores in HC. The interferon (IFN) protein family is crucial for the immune response against viruses and other infections. IFNs have been demonstrated to play a significant role in preventing SARS-CoV-2 infection in the context of COVID-19, but they have also been linked to severe symptoms [69]. The findings may explain the contradictory reports of COVID-19 patients with impaired and robust type I IFN responses. Although robust type I IFN responses have been reported in patients with severe COVID-19 [70], studies have demonstrated that type I IFN responses are severely impaired in the peripheral blood of patients with severe or critical COVID-19, as indicated by low levels of type I IFNs and interferon-stimulated genes [71]. Our data indicate that severe and fatal patients had IFN levels much lower than moderate patients, but greater than mild patients and healthy controls. In contrast to individuals with other disorders, COVID-19 patients have continuously elevated IFN levels. The above data conclude that the higher expression of MHC Class II and translational initiation expression in platelets mean better outcomes in patients. Coagulation and higher ATP synthesis from platelets, means worse outcomes for patients. As for interferon response, with both protective and deleterious effects being reported, we confirmed the theory that severe COVID-19 is associated with decreased IFN signaling [72, 73].

The Platelets’ Crosstalk with Monocytes and Lymphocytes

An important role of platelets is communication with other cell types during formation of the thrombi, both regular ones forming during the hemostasis or potential abnormal ones, forming during sepsis and COVID-19. Using the ligand and receptor database from iTalk [49], we evaluated these interactions by computing the product of average ligand and receptor expressions in the corresponding cell types from peripheral blood mononuclear cells (PBMC) (see Materials and Methods). Platelet-monocyte interaction was evaluated and found to have the highest score in fatal patients relative to other outcomes (Fig. 9A). Consistently with our previous sepsis study [12], platelet-monocyte interaction scores were significantly elevated in FT sepsis patients. This phenomenon was also noticed in COVID-19-severe patients. Compared to control participants and mildly infected individuals, ICU-admitted COVID-19 patients had higher platelet-monocyte aggregate levels [67]. Thus, we postulate that the aggregation of platelets and monocytes is linked with the severity and mortality of sepsis and COVID-19.

Chemokine receptor 2 (CCR2), CCR5, and their selective ligands, chemokine ligand 2 (CCL2), and CCL3, have been found to promote the trafficking of leukocytes to sites of inflammation and regulate their activation [73]. Also, the CCL2-CCR2 and CCL3-CCR5 ligand-receptor systems in differentiating T cells have been identified [74]. CCL2-CCR2 and CCL3-CCR5 interactions between platelets and T cells revealed that SLE had the highest expression of the CCL3-CCR5 system between platelets and T cells, a characteristic of SLE [75]. Compared to other outcome groups, the CCL3-CCR5 system’s expression was the lowest among patients who did not survive the diseases (Fig. 9B). The results are associated with T cell differentiation module scores (GO:0030217) (Fig. 9F). The CCL2-CCR2 and CCL3-CCR5 ligand-receptor systems are more prevalent in surviving patients than in HC (Fig. 9C).

CD40 is recognized to play a crucial role in B lymphocyte proliferation and differentiation [76]. And CD40 ligand CD40LG expression is low or undetectable on the surface of resting platelets but is highly expressed upon platelet activation [77]. Also, platelets are noted to directly influence adaptive immune responses via the secretion of CD40 and CD40L molecules [2]. Platelets and B cell interaction analysis revealed that healthy controls had the highest CD40 induction relative to other groups, whereas patients who did not survive had the lowest CD40 induction from platelets – B cell interaction (Fig. 9E). The module score for B cell proliferation (GO:0042100) (Fig. 9G) also supported this conclusion. SLE also had the highest expression of CD40LG and CD40 interaction between platelets and B cells compared to other diseases (Fig. 9D). These data implicate that in SLE, platelets may induce B cell activity, consistent with previous studies reporting the putative effects of activated platelets in SLE pathogenesis [13].

For this study, we compiled single-cell transcriptome datasets from multiple cohorts of patients with inflammatory diseases with infectious and non-infectious roots (COVID-19, sepsis, SSH, and SLE). Extensive evidence points to a key role for platelets in all these diseases, severe COVID-19 [10, 78], fatal sepsis [12] and SLE [13] respectively. In our analysis we identified both shared and unique signatures in platelet transcriptomics in patients with these diseases. We analyzed alterations in immune cell subsets in peripheral blood mononuclear cells (PBMCs) and found many features related to disease severity, outcome, and disease type. The ratio of platelets to T cells (Pla-T ratio) was shown to be a useful criterion for separating patients with fatal and survival outcomes (Fig. 1H). Our study utilized DNN and XGB machine learning models to pinpoint biomarkers correlating with patient survival and the occurrence of fatal platelets, finding consensus on 21 key biomarkers, with XGB slightly outperforming in accuracy and DNN in F1 score, precision, and recall. Gene ontology enrichment revealed that survival-linked genes are involved in cytoplasmic translation and immune response, while fatal-linked genes are associated with hypoxia and coagulation processes (Fig. 2).

Analysis of the upregulated pathways shows that pathways characteristic of the development and progression of endotheliopathy and disseminated intravascular coagulation (DIC) are upregulated in platelets. The patients who passed away had the highest expression of the gene involved in platelet-endothelial cell interactions and highest expression in the modules related to DIC, including actin cytoskeleton reorganization (Fig. 3). Platelet actin cytoskeleton reorganization plays a role in this process, as the cytoskeleton helps to determine the shape and function of platelets and is essential for platelet activation and aggregation and changes observed here contribute to the pathological variants of this process.

We identified subgroups of platelets overrepresented in the convalescent, surviving, and fatal patients. Specifically, the three types of platelets were highly active in FT patients - the coagulation cluster C4, hypoxic cluster C9, and quiescence cluster C11. This suggests that a combination of anticoagulants, anti-hypoxic therapies, arteriosclerosis treatments, and selective serotonin reuptake inhibitors (SSRI) should be considered for the removal of fatal platelets. We examined the pseudotime trajectory of platelet clusters and identified C0 as the cluster preceding fatal cluster C4. C0 can be differentiated by following the path leading to the emergence of either fatal (C4) or survival (C1 leading to C6) cluster. We propose that targeting C0 and modifying its developmental path may prove to be an effective strategy to treat immune instability in sepsis or COVID-19. Expression of MYL9, FCER1G, and PARVB are the signatures of cluster C0 (Fig. 4E). MYL9 was investigated in relation to platelet dysfunction [79]. Comparing the fatal cluster C4, its ancestor C0, and its alternative, survival cluster C1 to each other, we identified two gene modules associated with fatal outcome and survival and we labeled them accordingly (Fig. 6C, D). Some genes in the fatal module, such as heparanase (HPSE) had already been studied as a target of therapy aimed at selectively neutralizing platelets to mitigate disease severity. HPSE expression and activity was shown to increase in platelets during clinical sepsis at both transcriptomic and proteomic level [80]. The same phenomenon was also observed in COVID-19 patients’ plasma [81]. Thus, in fatal and survival module genes identified in this study, there are potential clinical therapeutic targets, diagnostic and prognostic biomarkers in platelets that are worth investigating for the future studies.

The study also compared gene expression in four diseases (COVID-19, SSH, Sepsis, and SLE) to healthy controls (HC) and found consistent changes in gene expression, including stress pathways (Fig. 7D) and decreased levels of MHC class II gene expression in fatal cases (Fig. 8A). The presence of pathways also seen in neurodegeneration processes in the brain are also upregulated in severe COVID-19 patients and in all immune response disorders. Similarly, the ATP metabolic process is up-regulated but ribosome biogenesis and lymphocyte activation are commonly down-regulated (Fig. 7C).

COVID-19 [82, 83], sepsis [61], and SLE [84] are all strongly associated with neurodegenerative diseases, but platelets have not been evaluated for their potential involvement. They could get into the central nervous system by breaking the blood-brain barrier for a number of reasons, such as infection, neuroinflammation, neurodegeneration, and traumatic brain injury, which could allow platelet entry to the CNS. Platelet changes in activation and aggregation have also been documented in Lewy body diseases, amyotrophic lateral sclerosis and multiple sclerosis [85, 86]. However, we don’t know if abnormal platelets are directly involved in triggering neurodegeneration, or if they respond to the overall immune environment in a way similar to neurons undergoing degeneration. In any case, this study illustrates the potential clinical significance of platelets as peripheral diagnostic biomarkers and, potentially, therapeutic targets for neurodegenerative diseases.

Platelet-monocyte aggregation has been studied as a hallmark of severe COVID-19 [78, 82, 83] and sepsis [12, 87, 88]. The findings agree with our cell-cell interaction study (Fig. 9A), which again provides the molecular mechanism of this effect. In addition, we discovered that platelets could influence lymphocyte activation, proliferation, and differentiation through interactions with other immune cells, implying that platelets can modulate lymphocyte function and contribute to inflammatory and immune responses.

Overall, the study highlights the role of platelets in the development and progression of disease and the importance of monitoring changes in immune cell subsets and gene expression for predicting disease outcome and severity. We were able to study the etiology of dysfunctional platelets in sepsis, COVID-19, and SLE patients due to the high number of clinical samples included in our integrated analysis. Our research demonstrates the significance of platelets' dysfunction in immunological imbalance disorders and supports the utility of platelet-directed therapies to treat multiple immune disorders.

Data and Code Availability Statement

The published article includes all datasets generated or analyzed during this study. The Methods section provides details about the steps, functions, and parameters that were used for the analysis. Custom scripts foranalyzing data are available at https://github.com/xqiu625/PlateletSubpop-ML-ScTranscriptomics. Source data are provided inthis paper.

If you want to re-analyze the data in this paper, you can get any extra information you need from the lead contact.

Availability and Implementation

In this manuscript, we have not developed or introduced new software tools. Instead, our focus has been on the utilization and analysis of existing data, software, tools, and packages, all of which are readily available in the public domain. Each resource employed in our study is open-source, ensuring that our methods are transparent, reproducible, and accessible to the broader scientific community. For ease of reference and to facilitate further research, comprehensive citations and descriptions of these resources are provided throughout the manuscript. This approach aligns with the principles of open science, encouraging collaborative advancements and knowledge sharing within the field of bioinformatics and beyond.

Author Contributions

XQ and AG: conceptualization MN and AG: supervision. XQ: formal analysis. XQ: writing and editing. MN, AG and LJ: revision. AG: writing and editing and funding acquisition. All authors read and approved the final manuscript.

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.

Acknowledgments

This research was supported by the UCR School of Medicine (to A.G.). We thank the research computing at University of California, Riverside's High Performance Computer Cluster for providing supercomputing resources for this project.

Koupenova, M., et al. (2018) Circulating Platelets as Mediators of Immunity, Inflammation, and Thrombosis. Circ Res 122, 337–351
Maouia, A., et al. (2020) The Immune Nature of Platelets Revisited. Transfus Med Rev 34, 209–220
Jin, R., et al. (2013) Soluble CD40 ligand stimulates CD40-dependent activation of the β2 integrin Mac-1 and protein kinase C zeda (PKCζ) in neutrophils: implications for neutrophil-platelet interactions and neutrophil oxidative burst. PLoS One 8, e64631
Gudbrandsdottir, S., et al. (2013) Activated platelets enhance IL-10 secretion and reduce TNF-α secretion by monocytes. J Immunol 191, 4059–4067
Ali, R.A., et al. (2015) Platelets: essential components of the immune system. Curr Trends Immunol 16, 65–78
Chatterjee, M., et al. (2015) Platelet-derived CXCL12 regulates monocyte function, survival, differentiation into macrophages and foam cells through differential involvement of CXCR4-CXCR7. Cell Death Dis 6, e1989
Portier, I. and Campbell, R.A. (2021) Role of Platelets in Detection and Regulation of Infection. Arterioscler Thromb Vasc Biol 41, 70–78
Xiang, B., et al. (2013) Platelets protect from septic shock by inhibiting macrophage-dependent inflammation via the cyclooxygenase 1 signalling pathway. Nat Commun 4, 2657
Duffau, P., et al. (2010) Platelet CD154 potentiates interferon-alpha secretion by plasmacytoid dendritic cells in systemic lupus erythematosus. Sci Transl Med 2, 47ra63
Gorog, D.A., et al. (2022) Current and novel biomarkers of thrombotic risk in COVID-19: a Consensus Statement from the International COVID-19 Thrombosis Biomarkers Colloquium. Nat Rev Cardiol 19, 475–495
Assinger, A., et al. (2019) Platelets in Sepsis: An Update on Experimental Models and Clinical Data. Front Immunol 10, 1687
Qiu, X., et al. (2021) Dynamic changes in human single-cell transcriptional signatures during fatal sepsis. J Leukoc Biol 110, 1253–1268
Linge, P., et al. (2018) The non-haemostatic role of platelets in systemic lupus erythematosus. Nat Rev Rheumatol 14, 195–213
Shannon, O. (2021) The role of platelets in sepsis. Res Pract Thromb Haemost 5, 27–37
Tauseef, A., et al. (2021) Role of immature platelet fraction (IPF) in sepsis patients: A systematic review. J Family Med Prim Care 10, 2148–2152
Enz Hubert, R.M., et al. (2015) Association of the immature platelet fraction with sepsis diagnosis and severity. Sci Rep 5, 8019
Bergersen, K.V., et al. (2023) Health disparities in COVID-19: immune and vascular changes are linked to disease severity and persist in a high-risk population in Riverside County, California. BMC Public Health 23, 1584
Wool, G.D. and Miller, J.L. (2021) The Impact of COVID-19 Disease on Platelets and Coagulation. Pathobiology 88, 15–27
Puhm, F., et al. (2022) Platelet activation by SARS-CoV-2 implicates the release of active tissue factor by infected cells. Blood Adv 6, 3593–3605
Xu, P., et al. (2020) Mechanism of thrombocytopenia in COVID-19 patients. Ann Hematol 99, 1205–1208
Lee, H., et al. (2023) Single-cell analysis of platelets from patients with periodontitis and diabetes. Res Pract Thromb Haemost 7, 100099
Liu, C.J., et al. (2023) Platelet RNA signature independently predicts ovarian cancer prognosis by deep learning neural network model. Protein Cell 14, 618–622
Garma, L.D., et al. (2022) Integrated analysis of transcriptomic data reveals the platelet response in COVID-19 disease. Sci Rep 12, 6851
Wilk, A.J., et al. (2020) A single-cell atlas of the peripheral immune response in patients with severe COVID-19. Nat Med 26, 1070–1076
Lee, J.S., et al. (2020) Immunophenotyping of COVID-19 and influenza highlights the role of type I interferons in development of severe COVID-19. Sci Immunol 5
Schulte-Schrepping, J., et al. (2020) Severe COVID-19 Is Marked by a Dysregulated Myeloid Cell Compartment. Cell 182, 1419–1440.e1423
Arunachalam, P.S., et al. (2020) Systems biological assessment of immunity to mild versus severe COVID-19 infection in humans. Science 369, 1210–1220
Ren, X., et al. (2021) COVID-19 immune features revealed by a large-scale single-cell transcriptome atlas. Cell 184, 1895–1913.e1819
Liu, C., et al. (2021) Time-resolved systems immunology reveals a late juncture linked to fatal COVID-19. Cell 184, 1836–1857.e1822
Stephenson, E., et al. (2021) Single-cell multi-omics analysis of the immune response in COVID-19. Nat Med 27, 904–916
Bernardes, J.P., et al. (2020) Longitudinal Multi-omics Analyses Identify Responses of Megakaryocytes, Erythroid Cells, and Plasmablasts as Hallmarks of Severe COVID-19. Immunity 53, 1296–1314.e1299
Combes, A.J., et al. (2021) Global absence and targeting of protective immune states in severe COVID-19. Nature 591, 124–130
Jiang, Y., et al. (2020) Single cell RNA sequencing identifies an early monocyte gene signature in acute respiratory distress syndrome. JCI Insight 5
Mistry, P., et al. (2019) Transcriptomic, epigenetic, and functional analyses implicate neutrophil diversity in the pathogenesis of systemic lupus erythematosus. Proc Natl Acad Sci U S A 116, 25222–25228
Haque, A., et al. (2017) A practical guide to single-cell RNA-sequencing for biomedical research and clinical applications. Genome Med 9, 75
Hao, Y., et al. (2021) Integrated analysis of multimodal single-cell data. Cell 184, 3573–3587.e3529
Aran, D., et al. (2019) Reference-based analysis of lung single-cell sequencing reveals a transitional profibrotic macrophage. Nat Immunol 20, 163–172
Oh S, A.J., Al-Dulaimi R et al. (2020) HGNChelper: identification and correction of invalid gene symbols for human and mouse. F1000Research 2020
Korsunsky, I., et al. (2019) Fast, sensitive and accurate integration of single-cell data with Harmony. Nat Methods 16, 1289–1296
Finak, G., et al. (2015) MAST: a flexible statistical framework for assessing transcriptional changes and characterizing heterogeneity in single-cell RNA sequencing data. Genome Biol 16, 278
Chen, T.a.G., C. (2016) Xgboost: A scalable tree boosting system. Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining pp. 785–794
Carlson, M. (2019) GO.db: A set of annotation maps describing the entire Gene Ontology. R package version 3.8.2.
Frederick, M., et al. (2020) High expression of oxidative phosphorylation genes predicts improved survival in squamous cell carcinomas of the head and neck and lung. Sci Rep 10, 6380
Wu, T., et al. (2021) clusterProfiler 4.0: A universal enrichment tool for interpreting omics data. Innovation (Camb) 2, 100141
Yu, G., et al. (2015) DOSE: an R/Bioconductor package for disease ontology semantic and enrichment analysis. Bioinformatics 31, 608–609
Liberzon, A., et al. (2015) The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst 1, 417–425
Hanzelmann, S., et al. (2013) GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics 14, 7
Kumar, M.P., et al. (2018) Analysis of Single-Cell RNA-Seq Identifies Cell-Cell Communication Associated with Tumor Characteristics. Cell Rep 25, 1458–1468.e1454
Wang, Y., et al. (2019) iTALK: an R Package to characterize and illustrate intercellular communication. bioRxiv, 507871
Qiu, X., et al. (2017) Reversed graph embedding resolves complex single-cell trajectories. Nature methods 14, 979
Haghverdi, L., et al. (2018) Batch effects in single-cell RNA-sequencing data are corrected by matching mutual nearest neighbors. Nat Biotechnol 36, 421–427
Horton, L.E., et al. (2020) Endotheliopathy and Platelet Dysfunction as Hallmarks of Fatal Lassa Fever. Emerg Infect Dis 26, 2625–2637
Laursen, M.A., et al. (2018) Platelet function in disseminated intravascular coagulation: A systematic review. Platelets 29, 238–248
Popescu, N.I., et al. (2022) Disseminated intravascular coagulation and its immune mechanisms. Blood 139, 1973–1986
Coenen, D.M., et al. (2017) Platelet interaction with activated endothelium: mechanistic insights from microfluidics. Blood 130, 2819–2828
Boral, B.M., et al. (2016) Disseminated Intravascular Coagulation. Am J Clin Pathol 146, 670–680
Nakazato, T., et al. (2009) Gendoo: functional profiling of gene and disease features using MeSH vocabulary. Nucleic Acids Res 37, W166-169
Supek, F., et al. (2011) REVIGO summarizes and visualizes long lists of gene ontology terms. PLoS One 6, e21800
Silva-Filho, J.L., et al. (2014) IL-4: an important cytokine in determining the fate of T cells. Biophys Rev 6, 111–118
Xu, E., et al. (2022) Long-term neurologic outcomes of COVID-19. Nat Med 28, 2406–2415
Widmann, C.N. and Heneka, M.T. (2014) Long-term cerebral consequences of sepsis. Lancet Neurol 13, 630–636
Guo, L., et al. (2021) Platelet MHC class I mediates CD8 + T-cell suppression during sepsis. Blood 138, 401–416
Castro, F., et al. (2018) Interferon-Gamma at the Crossroads of Tumor Immune Surveillance or Evasion. Front Immunol 9, 847
Germann, T., et al. (1993) Interleukin-12/T cell stimulating factor, a cytokine with multiple effects on T helper type 1 (Th1) but not on Th2 cells. Eur J Immunol 23, 1762–1770
Liang, C., et al. (2021) Population-Predicted MHC Class II Epitope Presentation of SARS-CoV-2 Structural Proteins Correlates to the Case Fatality Rates of COVID-19 in Different Countries. Int J Mol Sci 22
Vardon-Bounes, F., et al. (2019) Platelets Are Critical Key Players in Sepsis. Int J Mol Sci 20
Litvinov, R.I., et al. (2021) Altered platelet and coagulation function in moderate-to-severe COVID-19. Sci Rep 11, 16290
Taylor, C.T., et al. (2016) Hypoxia-dependent regulation of inflammatory pathways in immune cells. J Clin Invest 126, 3716–3724
Galbraith, M.D., et al. (2022) Specialized interferon action in COVID-19. Proc Natl Acad Sci U S A 119
Zhou, Z., et al. (2020) Heightened Innate Immune Responses in the Respiratory Tract of COVID-19 Patients. Cell Host Microbe 27, 883–890.e882
de Lusignan, S., et al. (2020) Integrating molecular point-of-care testing for influenza into primary care: a mixed-methods feasibility study. Br J Gen Pract 70, e555-e562
Hadjadj, J., et al. (2020) Impaired type I interferon activity and inflammatory responses in severe COVID-19 patients. Science 369, 718–724
Galani, I.E., et al. (2021) Untuned antiviral immunity in COVID-19 revealed by temporal type I/III interferon patterns and flu comparison. Nat Immunol 22, 32–40
Fantuzzi, L., et al. (2019) Dual CCR5/CCR2 targeting: opportunities for the cure of complex disorders. Cell Mol Life Sci 76, 4869–4886
Turner, J.E., et al. (2012) Protective role for CCR5 in murine lupus nephritis. Am J Physiol Renal Physiol 302, F1503-1515
Spriggs, M.K., et al. (1992) Recombinant human CD40 ligand stimulates B cell proliferation and immunoglobulin E secretion. J Exp Med 176, 1543–1550
Rahman, M., et al. (2012) Metalloproteinases regulate CD40L shedding from platelets and pulmonary recruitment of neutrophils in abdominal sepsis. Inflamm Res 61, 571–579
Gu, S.X., et al. (2021) Thrombocytopathy and endotheliopathy: crucial contributors to COVID-19 thromboinflammation. Nat Rev Cardiol 18, 194–209
Songdej, N. and Rao, A.K. (2017) Hematopoietic transcription factor mutations: important players in inherited platelet defects. Blood 129, 2873–2881
Eustes, A.S., et al. (2021) Heparanase expression and activity are increased in platelets during clinical sepsis. J Thromb Haemost 19, 1319–1330
Buijsers, B., et al. (2020) Increased Plasma Heparanase Activity in COVID-19 Patients. Front Immunol 11, 575047
Goshua, G., et al. (2020) Endotheliopathy in COVID-19-associated coagulopathy: evidence from a single-centre, cross-sectional study. Lancet Haematol 7, e575-e582
Barrett, T.J., et al. (2021) Platelets amplify endotheliopathy in COVID-19. Sci Adv 7, eabh2434
Liu, F.C., et al. (2015) Inverse Association of Parkinson Disease With Systemic Lupus Erythematosus: A Nationwide Population-based Study. Medicine (Baltimore) 94, e2097
Leiter, O. and Walker, T.L. (2020) Platelets in Neurodegenerative Conditions-Friend or Foe? Front Immunol 11, 747
Zhang, W., et al. (2013) Contribution of blood platelets to vascular pathology in Alzheimer's disease. J Blood Med 4, 141–147
Chang, J.C. (2019) Sepsis and septic shock: endothelial molecular pathogenesis associated with vascular microthrombotic disease. Thromb J 17, 10
Rondina, M.T., et al. (2015) Platelet-monocyte aggregate formation and mortality risk in older patients with severe sepsis and septic shock. J Gerontol A Biol Sci Med Sci 70, 225–231

No competing interests reported.

SupplementaryMaterials.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Deciphering Abnormal Platelet Subpopulations in Inflammatory Diseases through Machine Learning and Single-Cell Transcriptomics

Status:

Version 1

Abstract

Figures

Introduction

Materials and Methods

Removal of Doublets in scRNA-seq Data and Identifying the Cell Types

Integrating the Datasets

Differential Expression Analysis

Machine Learning models

Comparison of Module Scores

Pathway Enrichment Analysis

Calculation of Ligand-Receptor Interaction Scores Between Platelets and Other Cell Types

Trajectory Inference of Transition in Platelets Subclusters

Results

Platelets Amplify Endotheliopathy and Disseminated Intravascular Coagulation in Fatal Patients

Platelet Subpopulations Associated with Disease Severity

Characteristics of Platelet Subpopulations

Pseudotime Trajectory Analysis Identifies Platelet Signature Dynamics in Survival or Fatal Disease Outcomes

Unique and Shared Gene Expression Changes in Platelets from COVID-19, SSH, Sepsis, and SLE Samples

Pathway Enrichment Related to Disease Severity in Platelets

The Platelets’ Crosstalk with Monocytes and Lymphocytes

Discussion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1