Integrative plasma proteomics and myeloid- interferon profiling reveal an AI-validated vascular- endothelial stress signature distinguishing SLE flare from remission in an Indian cohort a discovery – phase study
Frontiers in Immunology, 2026
Karmakar A., Mishra S., Kumar U., Kamath R., Ravindran V., Suryakanth V., Prabhu S., Nagaraju S., Prabhu M., Karmakar S.
| Disease area | Application area | Sample type | Products |
|---|---|---|---|
Immunological & Inflammatory Diseases | Patient Stratification | Plasma |  Olink Target 48 |
Abstract
Background
Systemic lupus erythematosus (SLE) is a heterogeneous autoimmune disorder characterized by unpredictable flares and variable clinical quiescence. Despite validated clinical indices like the British Isles Lupus Assessment Group (BILAG) score, reliable molecular biomarkers for monitoring disease activity remain limited, particularly in underrepresented South Asian populations. Weaimed to identify arobust molecular framework to distinguish SLE flares from remission in an Indian cohort.
Methods
We conducted a discovery-phase study in an Indian cohort (n=16) stratified by Easy-BILAG scoring. Plasma proteomic profiling via LC-MS/MS was integrated with targeted cytokine quantification using the Olink Proximity Extension Assay (PEA). Differential expression and network analyses delineated immune-regulatory, hypoxic-vascular, and myeloid-activation pathways. A Random Forest classifier was trained on selected biomarkers and evaluated using leave-one-out cross-validation (LOOCV), permutation testing, and bootstrapped AUROC confidence intervals, with model interpretability assessed by SHAP values. Data are available via ProteomeXchange with identifier PXD075349.
Results
Proteomic comparison identified a compact panel of proteins distinguishing flare from remission, characterized by a molecular polarization; flare states exhibited upregulation of COL18A1 and CSF1 (vascular and myeloid activation), while remission showed sustained expression of cytoskeletal scaffolding and immunoregulatory components, including FLNA, SH3BGRL3, and IGHG4. Cytokine analyses identified coordinated chemokine modules (CXCL9, CCL2, CCL3, and CCL13) preferentially upregulated during flare. The machine-learning model achieved robust internal discrimination with a mean AUROC of 0.96. Notably, a COL18A1 normalized protein expression cut-off yielded 100% specificity and 87.5% sensitivity, acting as an objective ‘rule-in’ adjunct for active disease. Normalized protein expression (NPX) cut-off yielded 100% specificity and 87.5% sensitivity, acting as an objective “rule-in” adjunct for active disease.
Conclusions
This study establishes a parsimonious 5-protein biosignature of candidate leads (COL18A1, HYOU1, IGHG4, FLNA, and SH3BGRL3) that effectively captures the multifactorial pathophysiology of SLE flare. By anchoring discovery in a systematically under sampled Indian population, this work enhances global diversity in lupus biomarker research and establishes a scalable, AI-driven framework for precision assessment of disease activity.