Abstract

   Plasma extracellular vesicles (EVs) are cell-derived lipid particles
   and reportedly play a role in sepsis pathogenesis. This study aimed to
   identify EV cargo proteins in septic patients and explore their
   association with key sepsis pathophysiology. Plasma EVs were subjected
   to Tandem Mass Tag (TMT)-based quantitative proteomic analysis. We
   identified 522 differentially expressed (DE) EV proteins in septic
   patients (n = 15) compared to the healthy controls (n = 10). The KEGG
   analysis of the DE proteins revealed multiple functional pathways
   linked to sepsis, e.g., complement/coagulation, platelet activation,
   phagosome, inflammation, and neutrophil extracellular trap formation.
   Weighted Gene Coexpression Network Analysis of 1,642 EV proteins
   identified nine unique protein modules, some of which were highly
   correlated with the sepsis diagnosis and diverse endotype markers
   including organ injury, inflammation, coagulopathy, and endothelial
   activation, and mortality. ROC analysis revealed a list of novel EV
   proteins that exhibited strong diagnostic performance. Cell
   type-specific enrichment analysis revealed the cellular origins of EVs,
   including immune and epithelial cells, neurons, and glial cells. Thus,
   the current study discovered complex proteomic signatures in plasma EVs
   that are closely associated with key pathophysiological responses in
   sepsis. These findings support the importance of EV cargo proteins in
   the patients’ immune responses, coagulation, and endothelial activation
   and lay the foundation for future mechanistic study of plasma EVs and
   their clinical application as potential diagnostic and prognostic
   markers.

Supplementary Information

   The online version contains supplementary material available at
   10.1038/s41598-025-06430-x.

   Keywords: Sepsis, Extracellular vesicles (EVs), Mass spectrometry,
   Proteomics, Tandem mass Tag (TMT)

   Subject terms: Biomarkers, Molecular medicine

Introduction

   Sepsis is induced by a dysregulated host response to infection and
   characterized clinically by hyperinflammation, hemodynamic collapse,
   endothelial injury, coagulopathy, and multiple organ
   dysfunction^[41]1,[42]2. Sepsis is the leading cause of in-hospital
   death in the United States^[43]3,[44]4. While antibiotic therapy and
   supportive care such as vasopressor and fluid resuscitation have
   significantly improved sepsis survival in the past decades, further
   progress remains a challenge. Numerous clinical trials aimed at
   modulating the immune response have failed in part due to complex
   sepsis pathogenesis and high heterogeneity of septic patients in
   demographics, causative pathogens, organ involvement, and immune
   profiles^[45]5,[46]6. To address these challenges, omics- and systems
   medicine-based approaches have been employed to identify sepsis
   endotypes based on gene expression patterns, immune response profiles,
   and pathophysiological outcomes^[47]5.

   Extracellular vesicles (EVs) are cell-derived lipid particles with
   heterogeneous and biologically active cargo molecules containing
   various proteins and nucleic acids and have emerged as key mediators of
   intercellular communication under various physiological and
   pathophysiological conditions^[48]7,[49]8. Given their multifaceted
   biological and pathophysiological effects, uncovering the protein
   composition of EVs could potentially reveal mechanisms of underlying
   diseases, promote the discovery of novel biomarkers, help risk
   stratification, and refine therapeutic targets^[50]9,[51]10. In sepsis,
   EVs have been studied for their roles in pathophysiological responses
   such as inflammation, coagulation, and various organ
   dysfunctions^[52]11–[53]14, and as potential biomarkers^[54]11,[55]15.
   Identifying EV cargo proteins is a critical first step toward fully
   understanding the contents and functions of plasma EVs in human sepsis.

   In the current study, we conducted mass spectrometry-based proteomic
   profiling of EV cargo using Tandem Mass Tag (TMT) technology for 15
   septic patients admitted to the ICU and 10 healthy participants. The
   TMT-based proteomic method enables precise proteomic quantification and
   offers the advantages of high sensitivity, accuracy, and reliability of
   detection with minimal technical variability^[56]16. We identify a long
   list of proteins differentially expressed (DE) in septic EVs. Employing
   various bioinformatics tools such as the Kyoto Encyclopedia of Genes
   and Genomes (KEGG) pathway and Gene Ontology (GO), we systematically
   analyze the functions of various differentially expressed proteins and
   their specific functional pathways. Using Weighted Gene Co-expression
   Network Analysis (WGCNA), we also investigate the module-trait
   relationship between the EV cargo proteins and various traits, such as
   multiple plasma endotype markers linked to sepsis pathophysiology and
   pathogenesis, as well as clinical laboratory and outcome data. The
   revelation of the functional pathways associated with the EV proteins
   offers a mechanistic insight into sepsis pathogenesis. Finally, ROC
   analysis identified a list of EV cargo proteins, some of them newly
   identified, that exhibited strong diagnostic performance for sepsis.

Results

Patient cohort

   The enrollment criteria include (1) with a clinical, surgical, or
   radiological identified source of infection, (2) having organ
   dysfunction with SOFA score ≥ 4, and (3) between the ages of 18 and 80
   years. A cohort of 15 septic patients with a mean age of 53.7 (23–78)
   years was enrolled for this study, and 53% of them were male.
   Similarly, a cohort of 10 healthy controls with a mean age of 40.4
   (25–66) years was included. The average SOFA score for the 15 septic
   patients at admission was 11.3 (range 6–18) and the 28-day mortality
   rate was 26.7%. A detailed summary of the demographic and clinical
   laboratory data can be found in Table [57]1.

Table 1.

   Demographic and clinical data of septic and healthy subjects.
                                  Healthy (n = 10) Sepsis (n = 15)
   Age (years)                    40.4 (25–66)     53.7 (23–78)
   Sex (M/F)                      2/8              8/7
   Ethnicity
    Black                         2 (20%)          4 (26.7%)
    Asian                         0 (0%)           1 (6.7%)
    Hispanic                      0 (0%)           0 (0%)
    White                         8 (80%)          10 (66.7%)
    Other                         0 (0%)           0 (0%)
   SOFA score                                      11.3 (6–18)
   ICU stay (Day)                                  18.1 (4–54)
   28-day Mortality                                4 (26.7%)
   Hematology Labs
    INR (≤ 1.2)                                    1.6 (1.1-3)
    PTT (30–40 s)                                  37.2 (23–66)
    Platelets (150–300 × 10^3/µL)                  195.7 (41–445)
   Chemistry Labs
    Lactate (< 2 mmol/L)                           2.8 (1-11.5)
    Creatinine (0.7–1.3 mg/dL)                     1.7 (0.38–3.34)
   [58]Open in a new tab

   Demographics with average (range) or count (percent).

   INR, international normalized ratio; PTT, partial thromboplastin time.

Plasma EV characterization

   As illustrated in Fig. [59]1, we isolated plasma EVs from 10 healthy
   control (HC) individuals and 15 septic patients using
   ultracentrifugation. While the method has a limitation of EV purity, it
   has been widely used and accepted with a higher yield than other
   methods. To enhance EV purity, we optimized the EV isolation steps by
   increasing the number of washing and ultracentrifugation cycles using
   control plasma samples. Three rounds of washing and ultracentrifugation
   enabled us to identify the highest percentage of exosome proteins,
   according to the comparison with the ExoCarta Top 100 list, enriching
   exosome proteins to 7.33% with the 409 total identified proteins
   (Supplementary Table [60]S1). Thus, we used the three times of
   ultracentrifugation and washing to enrich EVs using plasma samples from
   sepsis patients and HC individuals. After the EV enrichment, we
   conducted a nanoparticle tracking analysis (NTA) to evaluate the
   density and size of the isolated EV samples. The numbers of plasma EVs
   were 5.8 ± 3.8 × 10^9/mL and 16.6 ± 11.3 × 10^9 /mL (mean ± SD) in the
   HC individuals and septic patients, respectively (Fig. [61]2A). The
   number of septic EVs showed a significant increase compared to HC EVs,
   with P value of 0.0031. The average sizes of the isolated EVs were
   98.9 ± 9.4 nm in HC individuals and 100.2 ± 10.1 nm in septic patients
   (Fig. [62]2B), with overall even size distributions between the sepsis
   patients and HC individuals (Fig. [63]2C). Additionally, we examined
   the relative abundance of CD9 and Alix (PDCD6IP), markers of endosomal
   EVs by immunoblotting (Fig. [64]2D and Supplementary Figure [65]S1).
   Alix is reportedly crucial for exosome formation and secretion, while
   CD9 is a tetraspanin protein expressed on the exosome surface^[66]17.
   Across all samples from HC individuals and septic patients, Alix
   (PDCD6IP) and CD9 were found on the expected sizes. Of note, both Alix
   (PDCD6IP) and CD9 were confirmed in the EV-enriched proteomes, but not
   in the corresponding plasma proteomes (Supplementary Tables S3 and
   S15).

Fig. 1.

   [67]Fig. 1
   [68]Open in a new tab

   TMT-based quantitative proteomic analysis of plasma EVs in septic and
   healthy subjects. Human plasma samples were prepared from 15 septic
   patients and age/sex-matched 10 HC individuals. Plasma extracellular
   vehicles (EVs) were purified using ultra-centrifugation. Cargo proteins
   were extracted from EVs using ultra-sonication and subsequently
   subjected to reduction, alkylation, and enzymatic digestion. The
   peptides were then labeled with TMT reagents and separated into 24
   fractions before being subjected to mass spectrometry and statistical
   and bioinformatic analyses.

Fig. 2.

   [69]Fig. 2
   [70]Open in a new tab

   Characterization of plasma EVs from septic patients and health
   controls. The concentration of EVs (A) and the mean EV size (B)
   isolated from the plasma of Healthy control (Control) individuals and
   septic patients (Sepsis) were quantified using a nanoparticle tracking
   analysis (Viewsizer 3000, Horiba Scientific). Comparative analysis was
   conducted based on EV samples from 15 septic patients and 10 HC
   individuals. The bar in the middle of the dots indicates the mean
   value. An unpaired t-test was used for the statistical analysis (**:
   P ≤ 0.01, and ns: not significant). (C) The representative size
   distribution for two EV samples is presented. (D) Western blot of Alix
   (PDCD6IP) and CD9 in the EVs isolated from a group of HC individuals
   and septic patients. Alix: PDCD6IP (Programmed cell death 6-interacting
   protein), CD9: tetra-spanin membrane protein.

Proteomic profiling of EV cargo proteins

   To profile plasma EV cargo proteins, we conducted the TMT-based
   quantitative proteomic analysis of the EV proteins. We acquired
   1,918,398 MS/MS spectra with 154,249 spectra matched to peptides. This
   resulted in the identification of 18,348 peptides and 2371 proteins. A
   total of 1642 proteins were quantified across all 25 samples without
   any missing values (Supplementary Table [71]S2 and S3). The statistical
   analysis revealed 522 proteins that were differentially expressed with
   a q-value < 0.05, which represents a False Discovery Rate (FDR), in
   septic patients as compared to healthy individuals (Fig. [72]3A). Out
   of the 522 differentially expressed proteins, 301 proteins were
   upregulated, which included neutrophil defensin 1 and 3 (DEFA1, DEFA3),
   ADP-ribosylation factor 5 (ARF5), nicotinamide phosphoribosyl
   transferase (NAMPT), histone H4 (H4C1), mitogen-activated protein
   kinase 4 (MAPK4), C-reactive protein (CRP), serum amyloid A-1 protein
   (SAA1), histone HIST2H3PS2 (H3-2), histone H2A type 2-C (H2AC2). Among
   the 221 downregulated EV proteins from the lowest q-value were CD5
   antigen-like (CD5L) proteins, haptoglobin-related protein (HPR),
   Immunoglobulin heavy constant mu (IGHM), Immunoglobulin heavy variable
   2–5 (IGHV2-5), Immunoglobulin heavy variable 3/OR16-13 (IGHV3OR16-13),
   Immunoglobulin lambda variable 3–27 (IGLV3-27), and apolipoprotein L1
   (APOL1) (Supplementary Table S4).

Fig. 3.

   [73]Fig. 3
   [74]Open in a new tab

   Proteomic analysis of differentially expressed EV proteins in septic
   patients compared to HC individuals. (A) A volcano plot for the plasma
   EV proteins from 15 septic patients and 10 HC individuals. The x-axis
   represents the fold-change of septic patients/HC individuals in the Z
   score scale (a standardized measure that indicates how far a data point
   is away from the mean of a distribution. Z score of 1 presents 1
   standard deviation). The y-axis represents the P value of statistical
   analysis in -Log[10] scale. The curved lines indicate the boundary for
   a q-value of 0.05. Proteins with q < 0.05 are differentially expressed
   in septic patients, and representative proteins among them are
   highlighted in red. (B) Principal Component Analysis (PCA) for the 522
   differentially expressed EV proteins between septic (n = 15) and HC
   individuals (n = 10). (C) The enriched pathways in the KEGG Pathway
   analysis^[75]18 were displayed using the GO plot package. The GO plot
   shows the up/down-regulation of differentially expressed proteins in
   sepsis, displaying the –Log[10] (P value) and z-score values for each
   pathway. The red circles represent upregulated proteins in each
   pathway, while the blue circles represent downregulated proteins. The
   size of the trapezoids in the inner circle of the GO circle reflects
   the –Log[10] (P value) of the enriched pathway. The color of the
   trapezoids corresponds to the z-score. The z-score in the package is
   calculated as follows: z-score = (the number of upregulated proteins –
   the number of downregulated proteins)/square root (the number of
   proteins). (D) GO analysis of Cellular Components (GO-CC) describes
   where the differentially expressed EV proteins originate from. GO-CC
   enrichment analysis selected the top 11 as representative lists based
   on –Log[10] (P value). The count represents the number of enriched
   proteins in GO terms. The % represents the proportion of genes in the
   input list associated with the specific GO term.

   To assess potential sex differences in EV proteomes, we compared EV
   proteomic profiles between male and female subjects in both the
   combined cohort and the sepsis group alone. The statistical analysis
   results indicated no difference in EV proteomes between male and female
   subjects in both comparisons (Supplementary Figure [76]S2).

   To evaluate the discrimination capability of the differentially
   expressed proteins between the sepsis and control groups, we conducted
   principal component analysis (PCA) of the 522 differentially expressed
   EV proteins, which revealed a clear separation between the two groups
   with no overlap (Fig. [77]3B). Of note, five of the fifteen sepsis
   samples were positioned outside the 95% confidence interval in the PCA
   plot. Since we used two TMT experimental batches, we examined whether
   the batch effect affected this variation within the group, but did not
   observe any obvious batch effect (Supplementary Figure S3). Therefore,
   the observed distribution of the five samples likely reflects the
   patient heterogeneity, rather than technical artifacts.

   To determine functional pathways associated with the differentially
   expressed proteins, we conducted KEGG pathway^[78]18 analysis and
   identified the multiple enriched pathways that include the complement
   and coagulation cascades as the most enriched pathway with the lowest P
   value (56.08 of -Log[10] (P value)), followed by phagosome pathway,
   malaria infection-related pathways, neutrophil extracellular trap
   formation, platelet activation pathway and a number of other pathways
   (Fig. [79]3C, Supplementary Table S5). Supplementary Figure S4
   illustrates 30 out of 138 proteins in the complement and coagulation
   systems, which were the most enriched according to the KEGG analysis
   and their functionally associated pathways. To investigate further the
   key proteins in the enriched pathways of the differential EV proteins,
   we conducted protein-protein interaction (PPI) analyses using the
   STRING database (Supplementary Figure S5A ~ S5D). Among the proteins
   enriched in the complement and coagulation cascades pathway, proteins
   such as Fibrinogen alpha chain (FGA), Vitronectin (VTN), and Plasma
   protease C1 inhibitor (SERPING1) were the key proteins (Supplementary
   Figure S5A). Among the proteins enriched in the phagosome pathway,
   which is most enriched after the complement and coagulation cascades,
   Platelet glycoprotein 4 (CD36), Integrin alpha-M (ITGAM), and Integrin
   beta-2 (ITGB2) were shown as the key proteins (Supplementary Figure
   S5B). In the Malaria pathway, which implies a response to an infection,
   ICAM1, CD36, and Platelet endothelial cell adhesion molecule (PECAM1)
   were among the important proteins (Supplementary Figure S5C). In the
   neutrophil extracellular trap formation pathway, ICAM1, ITGB2, and
   Low-affinity immunoglobulin gamma Fc region receptor II-a (FCGR2A) were
   among the important proteins (Supplementary Figure S5D). Subsequently,
   we conducted a Gene Ontology Cellular Components (GO-CC) analysis to
   determine the subcellular origins of the differentially expressed
   proteins in the EVs (Fig. [80]3D and Supplement Table S6). The
   extracellular space was the most enriched cellular compartment,
   followed by extracellular region, extracellular exosome, extracellular
   vesicle, extracellular organelle, and extracellular membrane-bound
   organelle in the analysis. These GO-CC results suggest that the
   differentially expressed EV proteins in septic patients are primarily
   derived from extracellular compartments, confirming their origin within
   the EV proteome and highlighting the extracellular space and
   vesicle-related compartments as major sources. In addition to GO-CC, GO
   Molecular Function (GO-MF) analysis of the differentially expressed EV
   proteins revealed enrichment for antigen binding and various protein
   binding activities. GO Biological Process (GO-BP) analysis of the
   differentially expressed proteins revealed enrichment for immune system
   processes, including adaptive immune response, immune effector
   processes, and leukocyte-mediated immunity (Supplementary Figure S6).

EV protein clusters are closely correlated with sepsis diagnosis

   To further understand the potential role of the EV cargo proteins in
   sepsis pathophysiology, we conducted WGCNA with the total EV proteins
   that have been quantified – with or without differential expression –
   and obtained protein clusters (modules) that have similar expression
   patterns across the plasma samples (Supplementary Figures S7-S8),
   representing a group of proteins potentially from the same
   pathways^[81]19. The WGCNA was conducted using all quantified EV
   proteins, regardless of statistical significance. This approach allows
   for unbiased detection of protein co-expression modules, capturing
   biologically relevant patterns that might be overlooked when
   restricting the analysis only to differentially expressed proteins. The
   WGCNA yielded 9 modules (Supplementary Table S7). Among them, the M3,
   M4, and M6 modules exhibited the highest correlation with the sepsis
   diagnosis (P value ≤ 0.001) followed by M5 and M7 with 0.01 < P ≤ 0.05
   (Fig. [82]4A, Supplementary Table S8). Interestingly, M3 and M4 modules
   were negatively correlated with the sepsis diagnosis trait, showing
   down-regulation of the eigenprotein (meta-expression profile of the
   proteins in the module) in sepsis (Figs. [83]4B and D), while the M5,
   M6, and M7 modules were positively correlated with the sepsis
   diagnosis, showing up-regulation of eigenprotein (Figs. [84]4F, H, and
   J). To find key proteins, we examined module membership (MM) and
   protein significance (PS) values in each module. For the M3 module,
   Haptoglobin-related protein (HPR), Hepatocyte growth factor activator
   (HGFAC), and CD5L were the key proteins, exhibiting high correlation
   with sepsis diagnosis and strongly representing the M3 module.
   (Fig. [85]4C). These proteins are considered as key proteins involved
   in the sepsis pathogenesis process. For the M4 module, Apolipoprotein
   L1 (APOL1), Apolipoprotein A1 (APOA1), and Myeloid and Erythroid
   Nuclear Termination Stage-Specific Protein (MENT) were the key
   proteins, exhibiting high correlation with sepsis diagnosis and
   strongly representing the M4 module. (Fig. [86]4E). For the M5 module,
   H4C1, MAPK4, and Histone 3.1 (H3C1) are the key proteins, exhibiting
   high correlation with sepsis diagnosis and strongly representing the M5
   module. (Fig. [87]4G). For the M6 module, Defensin A1 (DEFA1),
   High-affinity immunoglobulin epsilon receptor subunit gamma (FCER1G),
   and Leukosialin (SPN) were the key proteins, exhibiting high
   correlation with sepsis diagnosis and strongly representing the M6
   module (Fig. [88]4I). Similarly, in the M7 module, CRP,
   Lipopolysaccharide-binding protein (LBP), and Ceruloplasmin (CP) were
   the key proteins, exhibiting a high correlation with sepsis diagnosis
   and strongly representing the M7 module. (Fig. [89]4K). It is
   noteworthy that the M8 module was significantly associated with
   sepsis-related mortality, showing higher eigenprotein expression in
   non-survivors than in survivors (P < 0.01; Fig. [90]4L). Although the
   M8 module did not show a significant correlation with sepsis diagnosis,
   MM-PS analysis revealed a moderate but significant correlation
   (cor = 0.5, P = 0.00064; Fig. [91]4M). SERPINA7, A1BG, and THBS2 were
   identified as key proteins in the M8 module, showing strong correlation
   with sepsis mortality.

Fig. 4.

   [92]Fig. 4
   [93]Open in a new tab

   The module–trait relationships by the WGCNA analysis of plasma-derived
   EV proteome data. (A) A heatmap illustrating Pearson correlations
   between eigenprotein expression levels of 9 WGCNA modules and the
   values of the sepsis-related traits of the plasma samples. The meanings
   of abbreviations for the traits in clinical lab data are as follows:
   BUN: blood urea nitrogen, Cr: creatinine, SOFA: sequential organ
   failure assessment, PLT: platelet count, INR: international normalized
   ratio, PTT: partial thromboplastin, PT: prothrombin time, WBC: White
   blood cell count, Hg: hemoglobin, Hct: hematocrit, K: potassium, and
   Na: sodium. The correlations were color-coded on a scale ranging from 1
   (indicating a positive correlation, red) to -1 (indicating a negative
   correlation, blue). The size of each circle corresponds to the P value,
   while the color indicates the correlation value. (B, D, F, H, J, L)
   Relative eigenprotein abundances for the modules that showed a
   significant correlation with the sepsis diagnosis were shown on the box
   plots; M3 (B), M4 (D), M5 (F), M6 (H), and M7 (J). (L) Relative
   eigenprotein abundance for the M8 module that showed a significant
   correlation with the sepsis-related mortality was shown on the box
   plot. (C, E, G, I, K, M) MM-PS plots show the relationship between the
   module membership (MM) for each module and protein significance (PS)
   for sepsis diagnosis; M3 (C), M4 (E), M5 (G), M6 (I), and M7 (K). (M)
   MM-PS plots show the relationship between MM for M8 and PS for the
   sepsis-related mortality. The key driving proteins were marked by black
   arrows in the plots. An unpaired t-test was used for the statistical
   analysis of eigenprotein expression (*, P < 0.05; ***, P < 0.001; ****,
   P < 0.0001).

EV protein clusters are associated with pathophysiological markers of sepsis

   The pathophysiology of sepsis is complex and heterogeneous from patient
   to patient. To capture this, we designed a set of plasma markers known
   to be linked to sepsis pathophysiology, including organ injury,
   coagulation, inflammation, and endothelial cell (EC) activation. As
   illustrated in Fig. [94]4A, using a Luminex multiplex platform and
   clinical laboratory tests, we measured the plasma samples from the two
   groups of participants for their organ injury markers (S100B, enolase2,
   Tie-2, angiopoietin-1, angiopoietin-2, BUN, Cr, SOFA), coagulation
   markers (vWF-A2, thrombomodulin/BDCA-3, P-selectin/CD62P, ADAMTS13,
   coagulation factor III/tissue factor, serpin C1/AT-III, serpin
   E1/PAI-1, CXCL4/PF4, D-dimer, Platelet, INR, PTT, PT), inflammation
   markers (IL6, IL1β, TNFα, IL8, CXCL2, WBC), EC activation markers
   (syndecan1, VCAM-1, ICAM-1, E-selectin), hemogram (Hg, Hct, Hg, K, Na,
   Lactate), and the clinical outcomes (length of ICU stay, 28-day
   mortality).

   We discovered that all the plasma markers in the Luminex panel, except
   for CXCL2 and Tie2 (Angiopoietin-1 receptor), were markedly upregulated
   in the septic cohort compared with the healthy controls (Supplementary
   Table S9). Moreover, as shown in Fig. [95]4A, the modules highly
   related to sepsis diagnosis, such as M3, M4, M5, M6, and M7, were
   significantly correlated to sepsis-related pathophysiological markers.
   Among the markers, M3 module were highly correlated with
   Angiopoietin-1, ADAMTS13, CXCL4/PF4, IL1β, TNFα, and E-selectin,
   whereas M4 module was closely correlated with angiopoietin-1,
   thrombomodulin, Coagulation factor/tissue factor, serpin C1/AT-III,
   CXCL4/PF4, IL1β, syndecan-1, and VCAM-1. The correlated traits with the
   M5 module were S100B, Enolase2, Serpin E1/PAI-1, and D-dimer. For M6
   module, thrombomodulin/BDCA-3, P-selectin/CD62P, ADAMTS13, IL1β, and
   TNFα are significantly regulated. In M7 module, there were no traits
   correlated with P ≤ 0.01. Interestingly, in the right panel of
   Fig. [96]4A, the WBC traits exhibited a high correlation with the M6
   module, which, not surprisingly, may suggest that M6’s correlation with
   the coagulation and inflammation traits are probably linked to
   circulating WBC or vice versa.

   Another noticeable result in WGCNA was that the M8 module showed a high
   correlation with 28-day mortality even though it was not correlated to
   sepsis diagnosis. Since the M8 module is highly correlated with the
   traits related to coagulation (vWF, coagulation factor III, and PTT)
   and EC activation (syndecan1, VCAM1, and ICAM1), it suggests that these
   traits may be connected to the 28-day mortality in these septic
   patients. To investigate mortality-associated proteomic changes, we
   compared plasma EV proteins between sepsis non-survivors (n = 4) and
   healthy controls (n = 10). This analysis revealed 245 significantly
   altered proteins (q < 0.05). In contrast, the comparison between
   survivors (n = 11) and controls revealed 623 significantly altered
   proteins (Supplementary Figure S9). In addition, the M8 module was
   significantly correlated with 28-day mortality, as determined by
   comparing sepsis survivors and non-survivors. Out of 43 proteins in the
   M8 module, 15 proteins were significantly elevated in sepsis patients
   who died within 28 days compared to survivors (Supplementary Figure
   S10). Among these, PKM, A1BG, HPX, and SERPINA7 not only showed
   increased abundance in the mortality group but were also significantly
   different between healthy controls and sepsis patients, highlighting
   their potential as both diagnostic and prognostic biomarkers for
   sepsis-related mortality. These findings also suggest distinct EV
   proteomic signatures associated with sepsis survival outcomes.

   Figure [97]5 shows the top player proteins in each of these EV protein
   modules that showed a close correlation with the physiological traits
   of sepsis. Specifically, we plotted PS and MM of each protein in the
   modules to determine significant proteins from M3 to M6 modules in
   relation to their respective top 4 traits beside sepsis diagnosis trait
   with significant correlation with P ≤ 0.01 (Figs. [98]5 and
   Supplementary Table S8). Other traits that exhibited significance at
   P < 0.01 in the M3 (A), M4 (B), and M6 (C) modules, but not presented
   in Fig. [99]5, are displayed in Supplementary Figure S11.

Fig. 5.

   [100]Fig. 5
   [101]Open in a new tab

   The Module membership (MM) and protein significance (PS) plots in the
   M3-M6 modules. The traits in Luminex that correlate to M3 ~ M6 modules
   in P ≤ 0.01 were selected to analyze MM-PS plots and find significant
   proteins in sepsis pathophysiology. (A) MM-PS plots show the
   relationship between module membership (MM) for the M3 module and
   protein significance (PS) for the top 4 traits except diagnosis (TNFα,
   ADAMTS13, CXCL4/PFA4, IL1β). The key driving proteins were marked by
   black arrows. The protein list in the plot is presented in
   Supplementary Table S8. (B) MM-PS plots show the relationship between
   MM for the M4 module and PS for the top 4 traits except diagnosis
   (Syndecan1, Angiopoietin-1, Serpin C1/AT-III, CXCL4/PF4). (C) MM-PS
   plots show the relationship between MM for the M5 module and PS for the
   top 4 traits except diagnosis (D-dimer, SERPINE1, S100B, and Enolase2).
   (D) MM-PS plots show the relationship between MM for the M6 module and
   PS for the top 4 traits except diagnosis (TNFα, P-selectin/CD62P,
   thrombomodulin/BDCA3, ADAMTS13).

Cell-type and pathway analyses of the WGCNA module proteins

   Since the M3, M4, and M6 modules were identified as critical modules
   showing strong correlations with the sepsis diagnosis (P ≤ 0.001), and
   to a lesser degree with M5 and M7 modules (0.001 < P ≤ 0.01), each of
   the five modules was subjected to cell-type as well as KEGG pathway
   enrichment analysis to identify the specific cell types from which the
   proteins are originated and the functional pathways they belong to. We
   utilized the DISCO database, which contains information on cell-type
   marker genes annotated through single-cell RNA sequencing, and the top
   10 enriched cell types were selected based on -Log[10] (P value)
   (Fig. [102]6). The M3 proteins were enriched mainly for epithelial
   cells, immune cells, glial cells, and fibroblasts (Fig. [103]6A). The
   most enriched pathways for these proteins were the complement and
   coagulation cascades by KEGG pathway analysis (Fig. [104]6B,
   Supplementary Table S12). Since the M3 proteins were mostly
   downregulated in sepsis, these data suggest that the M3 proteins
   related to the pathways were likely downregulated during sepsis, such
   as coagulation factor V (F5) in the coagulation cascade and C4b-binding
   protein (C4BP) in the complement cascade (Supplement Figure [105]S1 and
   Supplement Table S4). The M4 proteins were enriched for fibroblasts,
   immune cells, epithelial cells, and neuronal cells, and functionally
   linked to the proteasome pathway (Fig. [106]6C-D). The M5 module was
   enriched for immune cells and mainly related functionally to proteasome
   and metabolic pathways (Fig. [107]6E-F). The M6 proteins were enriched
   for hematopoietic stem cells and arterial endothelial cells and
   functionally related to actin cytoskeleton, leukocyte trans-endothelial
   migration, and platelet activation (Fig. [108]6G-H). The M7 module was
   enriched for neuronal cells, melanocytes, epithelial cells, and
   fibroblasts (Fig. [109]6I). The complement and coagulation cascades
   were the most enriched pathway for the M7 module (Fig. [110]6J). The
   module proteins are related to the upregulation of the pathways, such
   as CD59 in the complement cascade and coagulation factor X (F10) in the
   coagulation cascade (Supplement Figure [111]S1 and Supplement Table
   S4). Finally, to access key proteins in each related pathway, we
   conducted interactome analysis for the modules using STRING PPI
   (Supplementary Figure [112]S12- S16, Supplementary Table S11). While
   C1Q family proteins related to the complement and coagulation cascades
   are noticeable in the M3 module, the apolipoprotein family in the M4
   module was examined significantly. In the M5 module, proteins related
   to metabolic pathways, such as the Elongation factor family (EEF),
   Alcohol dehydrogenase 6 (ADH6), and histone proteins, were key
   proteins. In the M6 module, Cell division control protein 42 homolog
   (CDC42) and Profilin-1 (PFN1) in the actin filament organization and
   proteins of some integrin beta (ITGB) family in the extracellular
   matrix organization were significantly detected, while in the M7
   module, Antithrombin-III (SerpinC1) and Complement factor I (CFI) were
   key proteins.

Fig. 6.

   [113]Fig. 6
   [114]Open in a new tab

   Cell-type enrichment and pathway analysis of the EV proteins in the M3,
   M4, M5, M6, and M7 modules. (A, C, E, G, I) Cell-type enrichment
   analysis for proteins in the modules using the DISCO database,
   highlighting the top 10 enriched cell types for each module, based on
   -Log[10] (P value). Each bar color represents a specific cell type. (A)
   is for M3, (C) for M4, (E) for M5, (G) for M6, and (I) for M7. (B, D,
   F, H, J) KEGG pathway analysis for the modules was performed and the
   top 10 significant pathway terms were selected for each module based on
   P value, showing upregulated pathways in red and downregulated pathways
   in blue. (B) is for M3, (D) is for M4, (F) is for M5, (H) is for M6,
   and (J) is for M7.

Selection of EV proteins highly correlated with the sepsis diagnosis

   Finally, to test the discrimination abilities of EV candidate markers
   for sepsis, we performed ROC analyses on the 522 differentially
   expressed proteins in the sepsis cohort (Supplementary Table S13). We
   selected the top highly discriminative proteins in AUCs > 0.8 with a
   significant Z score as shown in Table [115]2. With the Z score, we
   narrowed down the proteins more correlated with the sepsis diagnosis
   among many strong AUC proteins. The candidates, all closely associated
   with sepsis diagnosis, are grouped in the four functional pathways −
   complement and coagulation cascade, immune responses, endothelial
   activation and metabolic pathways. Integrin beta-2 (ITGB2), von
   Willebrand factor (VWF), and Inter-alpha-trypsin inhibitor heavy chain
   H4 (ITIH4) are the top three of the upregulated markers in the
   complement and coagulation pathways. ADP-ribosylation factor 5 (ARF5),
   Golgi-associated plant pathogenesis-related protein 1 (GLIPR2), and
   Sphingosine 1-phosphate receptor 4 (S1PR4) are the top three of them in
   the immune responses, while Neutrophil defensin 1 (DEFA1), DFEA3, and
   S100-A9 (Protein S100-A9) are the top three in the endothelial
   activation. H4C1 (Histone H4), MAPK4 (Mitogen-activated protein kinase
   4), and SLC16A3 (Monocarboxylate transporter 4) are the top three in
   the pathways related to metabolism. While these proteins seemed highly
   correlated to sepsis diagnosis in a positive way, HPR
   (Haptoglobin-related protein), HGFAC (Hepatocyte growth factor
   activator), TNN (Tenascin-N), APOL1 (Apolipoprotein L1), and PF4
   (Platelet factor 4) are correlated to sepsis diagnosis in a negative
   way.

Table 2.

   ROC analysis of selected EV cargo proteins correlated highly with
   sepsis diagnosis.
   Gene name Gene description AUC Z score References