Clinical indicators and transcriptomic changes in coagulation and complement cascades in COVID-19 patients

To begin our investigation into immune-coagulation changes in COVID-19, we first assessed the standard clinical markers of inflammation (CRP) and coagulation (Platelet counts and Fibrin-D-dimer) using available clinical laboratory parameters from our patient cohort, which included 21 COVID-19-negative controls (HC), 10 convalescent, 26 patients with mild disease and 11 with severe disease described in our earlier study¹⁵. Fibrin-D-dimer and platelet counts were significantly elevated in patients with severe COVID-19 compared to mild cases (p = 0.048 and 0.005, respectively), while CRP levels showed no significant difference (p = 0.51) (Fig. 1A). These clinical laboratory parameters reflect downstream coagulation and inflammatory processes and prompted further molecular analysis to identify the upstream regulatory signatures involved.

Clinical and transcriptomic profiling of complement-coagulation related pathways in COVID-19 and control groups. (A) Dot plots showing clinical laboratory measurements of inflammation (CRP), coagulation (Fibrin-D-dimer), and platelet count in patients with mild (n = 26) and severe (n = 11) COVID-19. Statistical significance was assessed using Mann-Whitney U tests, with p-values indicated for each comparison. A p-value of <0.05 was considered statistically significant. While CRP levels were not significantly different (p = 0.51), both D-dimer and platelet counts were significantly elevated in severe cases (p = 0.048 and p = 0.0052, respectively), indicating enhanced thromboinflammatory activity. (B) Principal Component Analysis (PCA) displaying the sample distribution based on the expression data of 233 curated genes within the ‘Complement-Coagulation related pathways’ across 21 COVID-19-negative controls (HC), 10 convalescent, 26 mild, and 11 severe COVID-19 cases. The PCA illustrates disease-severity-associated separation, with PC1 and PC2 explaining 36.3% and 13.11% of the total variance, respectively. (C) Heatmap displaying z-score normalized expression of selected coagulation and complement-related genes across patient groups, stratified by demographic variables (age, gender, BMI). Color intensity of the heatmap encodes z-score normalized expression levels, accentuating marked upregulation in severe cases compared to others. Comparative analysis includes mild vs. HC, severe vs. HC, and severe vs. mild, with expression fold changes annotated. Comparative log2 fold-change values (LFC) for mild vs. HC, severe vs. HC, and severe vs. mild are shown. Statistical significance was assessed using DESeq2 in R, with p-values adjusted for multiple comparisons using the Benjamini-Hochberg adjustment.

We subsequently performed transcriptomic profiling to explore upstream molecular alterations linked to thromboinflammation in COVID-19, reanalyzing data from our previous study¹⁵. To comprehensively capture coagulation and complement dysregulation, we curated a gene set by integrating components from multiple pathway databases, including KEGG, Reactome, WikiPathways, MSigDB, Panther and Gene Ontology (see Supplementary Table S1). Analysis was performed on curated list of 233 genes. Principal Component Analysis (PCA) of significantly altered complement and coagulation-related genes revealed clear separation between clinical groups (Fig. 1B). Severe cases and healthy controls formed distinct clusters, while mild and convalescent individuals showed intermediate or overlapping profiles. This distribution reflects a progressive transcriptomic shift in coagulation pathways with disease severity and recovery, supporting the context-specific regulation of these genes during COVID-19. The heatmap of z-scaled expression shows that gene expression levels were elevated in COVID-19 patients, with a particularly pronounced increase observed in those with severe disease. As shown in Fig. 1C, 15 genes related to the complement-coagulation cascade, were significantly upregulated and DCN1 to be downregulated in severe disease as compared to mild disease state (Supplementary Table S2). This underscores the central role of coagulation dysregulation in the pathogenesis of severe disease and aligns with previous studies that have highlighted the hypercoagulable state as a hallmark of severe COVID-19, leading to thromboembolic events and multi-organ failure^1,2. To further resolve the heatmap and provide a more detailed view of gene-specific variability, we generated boxplots of the 52 significant differentially expressed genes (Supplementary Figure S1). These visualizations confirm significant upregulation of several prothrombotic genes in severe COVID-19 while also highlighting notable inter-patient variability.

Interestingly, when correlating gene expression levels with clinical laboratory parameters, CRP but not Fibrin-D-dimer or platelet counts was significantly associated with the expression of multiple coagulation genes, including GP9, SERPINE1, ITGB3, MMRN1, and F2RL3 in the severe group (FDR < 0.05; Supplementary Table S3; Supplementary Figure S2). No significant association was noted in the mild group. This suggests that while Fibrin-D-dimer and platelet elevations likely reflect systemic activation at the protein level, CRP levels may align more closely with transcriptional regulation of platelet and endothelial pathways during inflammation.

The heterogeneity observed in the gene expression pattern, even among individuals with severe disease, highlights variability in the response and raises the possibility that additional factors, such as immune memory or autoimmunity, may modulate the response to SARS-CoV-2 infection and contribute to disease variability. Given that CRP, but not Fibrin-D-dimer or platelet counts showed correlation with a subset of coagulation-related genes, systemic inflammation may partially explain the transcriptional variation. We therefore next performed autoantibody profiling against coagulation-related antigens to investigated whether such immune signatures could underlie or complement transcriptional landscape.

Autoantibody profiling in COVID-19 patients

The observed variability in gene expression patterns led us to investigate the presence of autoantibodies against coagulation factors, as they may contribute to the differential responses observed in COVID-19 patients. We focused on a predefined panel of coagulation-related autoantibodies selected based on their hypothesized or documented involvement in COVID-19-associated thromboinflammation and vaccine-induced coagulopathies. Using EDTA-Plasma samples and an in-house Luminex bead-based autoantibody-screening assay, we assessed the presence of circulating autoantibodies to eight coagulation-related factors (ADAMTS13, Factor V, Protein S, SERPINC1, Apo-H, PROC1, prothrombin, and PF4) and their relationship to the expression of genes associated with complement-coagulation related pathways and to clinical disease severity. Figure 2 shows the mean fluorescent intensity (MFI) of each of the specific autoantibodies in different disease categories. The antigen reactivities detected for coagulation-related proteins were low compared to signals typically observed for viral proteins or established antigens in this assay format. To define positivity, we applied a dual threshold approach, defined by either as the mean MFI of the healthy control (HC) group plus seven times their standard deviation (marked by the red dotted line in Fig. 2) or an arbitrary cutoff of 1000 MFI (marked by blue dotted line in Fig. 2), whichever was higher. Our analysis did not reveal autoantibody positivity for any of the tested COVID-19 patients or controls. Therefore, we consistently refer to these reactivities as unvalidated sub-threshold antigen signals or reactivities. We acknowledge that biological interpretation of these subthreshold signals is exploratory, and they may reflect either very low-level binding or non-specific background noise. Moreover, potential limitations of the multiplex-platform such as inter-assay variability and antigen and/or antibody cross-reactivity cannot be excluded. We were limited by any further validations using more sensitive assays and orthogonal confirmatory methods to clarify if there is any biological relevance of these observations. Nonetheless, we consider this a resource-efficient and data leveraging exploratory approach, especially in settings where multiple antigens can be included at virtually no additional cost, allowing identification of putative autoantibody candidates for future mechanistic validation. Future efforts may benefit from higher-sensitivity approaches or epitope-resolved techniques to clarify the significance of low-level signals. Despite the shortcomings, in pairwise comparisons without multiple testing correction of the subthreshold MFIs of the tested antigens across groups, convalescent patients exhibited significantly higher reactivity to Factor V compared to both the active COVID-19 groups and healthy controls (HC) (Mann-Whitney U test, p < 0.05; Fig. 2B). Patients with severe disease exhibited higher anti-protein S reactivity compared to the patients with mild disease (Mann-Whitney U test, p = 0.027; Fig. 2C) despite the MFIs remained below threshold levels. Interestingly, sub-threshold reactivities to SERPINC1, PF4 and prothrombin were numerically elevated in convalescent individuals compared to healthy controls (p-values of 0.011, 0.045 and 0.06, respectively; Fig. 2G, 2H and 2F respectively). This observation, although it cannot be reliably distinguished from background noise, aligns with prior studies suggesting an enrichment of these autoantibodies following SARS-CoV-2 infection¹⁰.

Coagulation related autoantibody profiling in COVID-19 patients. Analysis of autoantibodies against ADAMTS13 (A), Factor V (B), Protein S (C), Apo-H (D), PROC1 (E), Prothrombin (F), SERPINC1 (G), and PF4 (H) depicted through dot plots for four groups: Healthy Controls (n = 21), Convalescent (n = 10), Mild (n = 26), and Severe (n = 11) COVID-19 patients. Positivity thresholds are marked by red dotted lines, set at either the mean fluorescence intensity (MFI) plus seven standard deviations of the healthy control group (MFI + 7SD) or an arbitrary cutoff of 1000 MFI denoted by blue dotted lines, whichever is higher. All samples are subthreshold for all antigens. Autoantibody reactivities were determined using a custom-developed multiplex bead-based immunoasssay (Luminex, Luminex corp., Ausdtin, TX, USA). Statistical significance was evaluated using Mann-Whitney U tests, with p-values indicated for each comparison. Lower-range MFI plots are included as zoomed insets inside each antigen panels to provide a more resolved view of sub-threshold distributions across clinical groups.

Differences in the applied statistical tests indicate the need for further investigations into autoantibodies in these groups, and these screening assay results does not contradict previous reports suggesting that autoantibodies can exacerbate the inflammatory and coagulation responses in COVID-19, potentially leading to more severe outcomes^7,9. These observations underscore the potential for further exploratory studies to investigate autoantibodies and prevalence post-infection, emphasizing the need for a detailed analytical approach across different patient groups.

Correlation between autoantibody reactivity data and gene expression

We subsequently analyzed correlations between the expression of significantly altered genes in the complement-coagulation pathways and MFIs of the autoantibody signals across different patient groups, despite no patient meeting the established positivity threshold. This exploratory approach aimed to identify subtle patterns of immune-coagulation interactions that might be biologically informative even in absence of classical autoantibody positivity. To enhance sensitivity in this exploratory context, we applied a relaxed false discovery rate (FDR) threshold of 0.25, consistent with prior exploratory immunogenomic studies that prioritize hypothesis generation over strict statistical significance^16,17,18. For all correlation analyses involving subthreshold autoantibody reactivities, including those with gene expression, we adjusted p-values using an FDR approach based on the number of unique antigen targets tested in this study, rather than the full 96-analyte bead panel, which contains technical replicates and controls not representing distinct antigens¹⁹ (see Methods). This allowed us to identify directional trends and nominate biologically plausible candidate associations for future validation. As shown in Fig. 3, only the severe COVID-19 group demonstrated gene-autoantibody correlations that passed the adjusted FDR threshold (< 0.25), while the healthy controls (HC), convalescent and mild groups showed no such significance and visually denoted by cross-marked bubbles in the figure (Supplementary Table S4). The lack of correlation is consistent with the expected stability and regulatory effectiveness of their immune systems. In contrast, the severe COVID-19 group demonstrated a distinct pattern of widespread negative correlations between several coagulation-associated subthreshold autoantibody candidates and key regulatory genes. The observed significance-limited to the severe group (adj. p-value < 0.25; Fig. 3D) underscores the importance of context-specific immune dysregulation in driving these interactions. These correlations involved both immune-modulatory genes and endothelial or platelet-associated markers, and while modest in strength (R typically between − 0.65 and − 0.85), they consistently met the adjusted FDR threshold. Genes such as CD40LG, PTX3, FPR1, ADM, TXN, SERPING1, SERPINE1, and CFB showed reproducible associations with subthreshold reactivates targeting ADAMTS13, Factor V, SERPINC1, PROC1, Prothrombin, ApoH, and PF4, all adjusted p-value < 0.25 (Supplementary Table S4). For instance, CD40LG expression negatively correlated with five distinct candidate antigens, including ADAMTS13 (R = − 0.79, adj. p-value = 0.236), SERPINC1 (R = − 0.85, adj. p-value = 0.236), Prothrombin, Apo-H, and PF4, suggesting it may be a convergence point in immune-coagulative dysregulation in severe disease. Similarly, PTX3, a marker of vascular inflammation and predictive marker of COVID-19 mortality²⁰, was negatively associated with both Factor V and PROC1 subthreshold reactivities, while FPR1, involved in neutrophil chemotaxis and linked to pulmonary pathology in severe COVID-19²¹, exhibited strong inverse correlations with Factor V, SERPINC1, and PROC1, all within the exploratory significance threshold. These patterns reinforce a broader observation that severe disease may potentially be characterized by immunological interference with transcriptional regulation of vascular and hemostatic genes, even when classical thresholds for autoantibody positivity are not reached. Importantly, no similar correlations were observed in the other clinical groups, supporting the idea that these transcription-autoantibody reactivity associations might be more than merely reflective of background immune activation, but rather reflect a disease-specific breakdown in regulatory control during severe SARS-CoV-2 infection.

Correlation analysis between unvalidated sub-threshold autoantibody candidates and expression of significant differentially expressed Complement-coagulation-related genes. (Panels: A-D) Spearman correlation heatmaps showing the relationships between subthreshold autoantibody reactivities (x-axis) and expression levels of significantly differentially expressed genes (from Fig. 1) within the coagulation and complement pathways (y-axis) across different clinical groups: healthy controls (HC; n = 21) (A), convalescent (CONV; n = 10) (B), mild COVID-19 (MILD; n = 26) (C), and severe COVID-19 (SEVERE; n = 11) (D). Only genes identified as significantly dysregulated in the transcriptomic analysis (Fig. 1) were included. Each bubble represents a gene-autoantibody pair, with color intensity indicating the correlation coefficient (blue = positive, red = negative), and bubble size reflecting correlation strength. Analyses were conducted using the psych package in R, and significance was evaluated using a relaxed false discovery rate (FDR) threshold of 0.25, suitable for exploratory analysis. Non-significant correlations are marked with a white X mark inside the bubbles.

Interestingly, although Protein S reactivity levels (below threshold) were higher in severe cases than in mild cases (Fig. 2C), correlations with the transcriptional profile were absent (Fig. 3D). This lack of correlation suggests that the pathogenic role of these autoantibodies, if any, may be independent of direct transcriptional changes in coagulation-related genes, potentially acting through post-translational modifications or interactions with other immune pathways that were not captured in this analysis.

To test whether these candidate gene-autoantibody associations reflected general markers of systemic inflammation or coagulation activity, we next examined correlations between subthreshold autoantibody MFI values and routine clinical laboratory parameters CRP, Fibrin-D-dimer, and platelet counts in both mild and severe COVID-19 cases. Notably, no significant associations were observed with any of these markers, even using a relaxed FDR threshold of 0.25 (Supplementary Table S5). This potentially suggests that the observed transcriptional correlations are not merely surrogate reflections of systemic inflammatory load or coagulative imbalance. Instead, they may represent a more specific immunotranscriptional interaction unique to the severe disease state.

The disparate correlation patterns observed between mild and severe COVID-19 cases leads us to speculate unique underlying pathogenic mechanisms tailored to the intensity of the disease. In severe cases, the predominantly negative correlations might reflect a tipping point where homeostatic mechanisms are overwhelmed, leading to uncontrolled coagulation and systemic inflammation. This scenario indicates a possible over-activation of immune responses that could drive the severity of clinical manifestations. Conversely, the milder cases display a pattern of more favorable or neutral correlations, hinting at a more regulated and balanced interplay between the immune and coagulation systems. This controlled response potentially curtails the escalation of the disease, helping to maintain physiological stability and prevent the progression to severe forms of the illness. The lack of adjusted p-values < 0.05, despite noticeable R values, suggests that while trends in the data exist, they do not reach classically accepted statistical significance. This highlights the need for cautious interpretation of these correlations and further investigation into their biological impact on disease progression in COVID-19.

To summarize these findings, severe COVID-19 patients uniquely exhibited reproducible, FDR-filtered correlations between coagulation gene expression and subthreshold autoantibody reactivities, tempting us to speculate that even subthreshold autoantibody reactivities may interact with the host transcriptome in a context-dependent manner. While these associations remain exploratory, they highlight the potential for autoantibody-transcriptional crosstalk as a contributing factor to thromboinflammation in severe disease. This underscores the importance of further mechanistic studies especially integrating transcription, protein regulation, and translational dynamics to unravel the immune-coagulation interface in COVID-19 and similar inflammatory syndromes.

Multi-omics correlations across antigens, transcriptomics, proteomics, and clinical laboratory parameters

Building on the observed associations between subthreshold autoantibody reactivities and transcriptional alterations in severe COVID-19, we next evaluated whether these relationships extended to the plasma proteome and clinically relevant inflammatory markers. Given that several genes associated with autoantibody profiles encode secreted immune or vascular proteins, we hypothesized that these transcriptional and humoral signatures may also manifest in systemic proteomic changes. To investigate this, we performed correlation analyses integrating previously generated and published Olink plasma proteomics data²² with both subthreshold autoantibody MFI and RNA-seq expression profiles across patient subgroups. This approach aimed to uncover potential links between serologic dysregulation, gene expression, and immune-coagulatory markers at the protein level. In addition, we assessed whether these molecular interactions aligned with conventional clinical indicators of inflammation and thrombosis.

Focusing first on autoantibody-protein correlations, we analyzed relationships between subthreshold median fluorescence intensity (MFI) values of coagulation-related antigens and levels of 92 plasma proteins measured using the Olink Immuno-Oncology panel. Applying an exploratory FDR threshold of 0.25 appropriate for hypothesis-generating analyses, we identified several significant correlations in the severe disease group (Fig. 4A–D; Supplementary Table S6). For instance, subthreshold anti-Factor V antigen reactivates correlated negatively with IL-6 (R = − 0.94, adj. p-value = 0.016) and CXCL10 (R = − 0.88, adj. p-value = 0.081) (Fig. 4A and B), cytokines previously implicated in cytokine storm and adverse COVID-19 outcomes²³. A positive correlation was also observed between subthreshold anti-Factor V and CCL17 (R = 0.85, adj. p-value = 0.15) (Fig. 4C), a chemokine reported to be suppressed in severe COVID-19 and proposed as an early triage marker for pneumonia risk²⁴. While the functional impact of these associations remains to be clarified, these findings suggest that even low-level autoantibody reactivities may reflect or contribute to broader inflammatory chemokine modulation in severe disease. Similarly, subthreshold anti-SERPINC1 autoantibodies correlated positively with MMP-12 (R = 0.92, adj. p-value = 0.025) (Fig. 4D), a macrophage-derived metalloproteinase linked to tissue remodeling and lung injury²⁵, indicating potential crosstalk between humoral immunity and extracellular matrix degradation pathways. These associations were not observed in the mild group and were sparse in convalescent individuals, although in the latter, a strong correlation was noted between subthreshold anti-ADAMTS13 and MCP-2 (R = 0.95, adj. p-value = 0.022) as shown in Supplementary Figure S3, possibly reflecting late-phase myeloid or endothelial remodeling.

Correlation between antigen reactivity, age, and circulating proteins in COVID-19. (Panels: A-D) Representative scatter plots highlighting significant correlations between unvalidated subthreshold autoantibody candidates (MFI; x-axis) and Olink-measured plasma protein abundance (NPX values; y-axis) in the COVID-19 severe group. In the represented graphs, Factor V reactivity is showing negative correlation with plasma protein IL-6 (A) and CXCL10 (B), and positive correlation with CCL17 (C), and for SERPINC1 reactivity showing positive correlation with plasma protein MMP-12 (D). Each dot represents one individual across the COVID-19 cohort. Trend lines and shaded confidence intervals are shown. All correlation coefficients (R) and adjusted p-values (padj) are indicated. An exploratory false discovery rate (FDR) threshold of 0.25 was applied to define significance and only features with FDR-adjusted p < 0.25 are presented. (Panels: E–I) Scatter plots showing Spearman correlations between age and subthreshold MFI levels of IgG autoantibodies against Prothrombin (E), Apo-H (F), SERPINC1 (G), PF4 (H), and ADAMTS13 (I) in COVID-19 patients. Each dot represents an individual; best-fit trend lines with 95% confidence intervals are overlaid. Data are color-coded by disease severity (Mild: yellow; Severe: orange). Correlation coefficients (R), p-values, and Benjamini-Hochberg adjusted p-values (padj) are indicated.

We next assessed correlations between RNA-seq gene expression and Olink proteomic markers applying a stringent significance threshold of adjusted p-value < 0.05 to prioritize robust and mechanistically interpretable associations, focusing on genes previously identified as differentially expressed and enriched for complement and coagulation-related functions. Unexpectedly, the mild disease group exhibited a greater number of significant gene-protein correlations than the severe group (adj. p-value < 0.05; Supplementary Table S7) suggesting a more coordinated regulation of transcription and systemic protein expression in less advanced disease. As shown in Supplementary Figure S4A, several genes, including SERPING1, TXN, F5, ITGA2B, and C3AR1, demonstrated robust positive correlations with Olink proteins involved in immune signaling, vascular homeostasis, and leukocyte recruitment. For example, SERPING1 expression strongly correlated with IFN-γ (R = 0.90, adj. p-value < 0.0001), IL-10 (R = 0.78, adj. p-value = 0.0019), and CXCL10 (R = 0.68, adj. p-value = 0.015), implicating its potential involvement in modulating interferon-stimulated and regulatory cytokine axes. Similarly, TXN expression was associated with VEGFA, CCL23, IL-6, and CASP-8, highlighting its connection to redox regulation and inflammatory vascular remodeling. Genes such as C3AR1 and TLR2 also displayed coordinated expression with VEGFA, CSF-1, and IL-10, reinforcing the interplay between complement signaling, myeloid activation, and cytokine regulation. In contrast, gene-protein correlations in severe disease were fewer and less robust with only plasma TNFRSF4 showing negative correlations with C1RL, SELL, SERPINA1, and TLR2 genes (adj. p-value < 0.05; R <– 0.9) (Supplementary Figure S4B), possibly reflecting disrupted transcriptional-proteomic coupling due to immune exhaustion, compartmentalization, or systemic dysregulation in critical illness.

Finally, to relate these molecular layers to clinical inflammation and coagulation markers, we examined correlations between Olink proteins and clinical laboratory parameters including C-reactive protein (CRP), Fibrin-D-dimer, and platelet count (Supplementary Table S8). While most associations did not reach significance, two notable and severity-specific relationships emerged. In the mild group, CRP negatively correlated with TRAIL (R = − 0.69, adj. p-value < 0.05) (Supplementary Figure S5A), a TNF superfamily member involved in apoptosis and antiviral responses. Prior studies have mostly linked TRAIL to protective immune regulation during viral infections²⁶, and its inverse relationship with CRP here suggests a higher TRAIL levels may contribute to attenuated inflammation in mild disease contexts. In severe disease, a strong positive correlation was observed between Fibrin-D-dimer and MUC16 (R = 0.96, adj. p-value < 0.05) (Supplementary Figure S5B). MUC16 (CA125) is a mucin-like glycoprotein primarily known for its role in epithelial integrity and has been implicated in immune modulation in cancer and inflammatory contexts²⁷. Notably, while prior studies have reported higher MUC16 mRNA expression in mild compared to severe COVID-19²⁸, our data show a strong positive correlation between plasma MUC16 protein levels and D-dimer specifically in severe cases. This apparent discrepancy may reflect post-transcriptional regulation, increased epithelial shedding, or systemic release of MUC16 in response to endothelial or mucosal barrier disruption during hyperinflammatory and hypercoagulable states.

Together, these multi-omic correlations across transcriptomics, autoantibody profiling, proteomics, and clinical markers support the hypothesis that subthreshold humoral immune responses, despite remaining below conventional diagnostic cutoffs, may nonetheless associate with coherent immuno-coagulatory signatures in severe disease. In mild cases, more synchronized transcript-protein interactions may reflect preserved immune homeostasis, while severe disease appears characterized by fragmented but selective cross-layer associations involving endothelial stress, inflammation, and coagulopathy. These findings highlight the utility of integrative, quantitative approaches in uncovering latent dimensions of immune dysregulation in COVID-19, and underscore the importance of context and disease phase and severity in interpreting autoantibody and proteomic profiles.

Age-dependent autoantibody reactivity

Our study further explored the relationship between age and autoantibody reactivities across different severities of COVID-19 at subthreshold MFI signals, focusing on potential trends (using an exploratory FDR < 0.25) rather than diagnostic positivity. In all the subthreshold autoantibodies tested in our panel we observed strong positive correlations between age and the subthreshold MFI of candidate reactivities against Prothrombin (R = 0.68, p = 0.02, adj. p-value = 0.07), Apo-H (R = 0.68, p = 0.021, adj. p-value = 0.07), SERPINC1 (R = 0.64, p = 0.035, adj. p-value = 0.07), PF4 (R = 0.66, p = 0.03, adj. p-value = 0.07),and ADAMTS13 (R = 0.57, p = 0.07, adj. p-value = 0.11) in severe cases (Fig. 4E–I). These correlations were consistently stronger in severe patients than in mild cases, where age-dependent reactivities to the same antigens were weaker and statistically non-significant (adj. p-value > 0.42, R < 0.3). Interestingly, when analyzing all COVID-19 patients together, despite lower correlation coefficients, certain subthreshold autoantibody candidates such as Prothrombin, Apo-H, ADAMTS13, and SERPINC1 approached conventional significance thresholds, (adj. p-value < 0.06, R ≥ 0.37) (Supplementary Table S9). This indicates that while individual correlations of subthreshold reactivities may appear modest, collectively they may reveal significant age-related trends across the COVID-19 spectrum.

Although these findings emphasize the complex influence of age on immune system behavior, suggesting that older patients might have distinct immune profiles that could impact their response to COVID-19, it is important to emphasize that these MFI values remained below established positivity thresholds, and while adjusted p-values suggest statistical associations, the biological implications remain exploratory. These results should therefore be interpreted as hypothesis-generating and not diagnostic. This observation aligns with previous studies suggesting that aging is associated with an increase in autoantibody production due to immunosenescence and inflammaging, contributing to the exacerbated clinical manifestations of diseases like COVID-19⁸. For example, older COVID-19 patients frequently exhibit a more severe disease course, characterized by enhanced inflammatory responses and a propensity toward autoreactivity, which may enhance hypercoagulability and increase the risk of thromboembolic events⁹. The correlation of these autoantibodies with disease severity underscores the need for age-stratified analysis and potentially also future therapeutic strategies that address both immune dysregulation and the heightened coagulation state observed in the elderly²⁹. The higher prevalence of autoantibodies in older patients³⁰ accentuates the need to integrate immune modulation and anticoagulation therapies into COVID-19 management strategies specifically designed for this demographic. While our findings do not suggest clinical utility of these low-reactive subthreshold autoantibody candidates in their current form, they point to age-related immune patterns worth investigating further. Future research should delve deeper into how aging affects autoantibody production including new-onset ones that emerges following infection³¹ and its implications for disease progression and therapeutic outcomes in elderly COVID-19 patients.

Our research contributes to the growing body of knowledge on the critical role of autoantibodies in the pathophysiology of COVID-19, especially in severe cases. We observed that while several subthreshold antigen reactivities were negatively correlated with coagulation-related gene expression, they also exhibited both negative (IL-6 and CXCL10) and positive (MMP-12 and CCL17) correlations with plasma proteins linked to inflammation, endothelial stress, and matrix remodeling. While these associations may suggest potential immuno-coagulatory perturbations, the detected reactivities were below established positivity thresholds and remain unvalidated. These inverse associations may also reflect non-causal relationships, such as immune exhaustion or broader epiphenomena associated with severe disease, rather than indicating a direct immunomodulatory role. Therefore, we interpret these findings as exploratory and hypothesis-generating, warranting further mechanistic investigation in targeted follow-up studies. This finding highlights the complexity of immune interactions in COVID-19 and supports the need for further studies to elucidate the mechanisms by which autoantibodies affect disease progression and outcomes, especially in long-term sequelae showing persistent complement dysregulation and thromboinflammation, as recently noted by Cervia-Hasler et al. (2024)³². Understanding these interactions could lead to the development of targeted therapies and exploratory biomarker candidates for disease severity, enhancing our management strategies for both acute and prolonged manifestations of COVID-19. In conclusion, this study underscores the potential of incorporating both autoantibody profiling and gene expression analysis in conjunction with plasma proteomics to dissect the complex pathogenesis of severe COVID-19. While the direct impact of autoantibodies on disease processes requires further validation, our findings hint at a potential relationship between subthreshold antigen reactivities and disease severity, suggesting that they might influence a range of intriguing pathophysiological pathways. The presence of coherent multi-layered correlations despite subclinical autoantibody reactivities supports the hypothesis that these responses may participate in shaping immune and vascular dynamics in a context-dependent manner. These insights may inform future studies and therapeutic strategies aimed at modulating immune responses to improve clinical outcomes in severe COVID-19 cases. Further research will be crucial to validate and establish the clinical relevance of any candidate autoantibodies and their utility in guiding interventions for COVID-19, particularly in severe and long-term scenarios.