Cerebrospinal fluid proteome reveals novel biomarkers for differentiating central NPSLE from central nervous system infection

BACKGROUND: Central neuropsychiatric systemic lupus erythematosus (cNPSLE) is a major cause of mortality in systemic lupus erythematosus (SLE). Distinguishing cNPSLE from central nervous system (CNS) infections remains a critical clinical challenge due to overlapping clinical manifestations, often leading to diagnostic delays and inappropriate therapeutic interventions. We aimed to identify and validate cerebrospinal fluid (CSF) biomarkers to resolve this diagnostic dilemma. METHODS: We conducted the largest cNPSLE CSF proteomic investigation to date using high-throughput data-independent acquisition mass spectrometry (DIA-MS) in a discovery cohort comprising 20 patients with cNPSLE, 8 with CNS infection (cNIC), and 11 non-diseased controls (NDC). Differentially expressed proteins were analyzed for pathway enrichment. Candidate biomarkers were subsequently validated using enzyme-linked immunosorbent assay (ELISA) in two independent cohorts: an internal validation cohort (46 cNPSLE vs. 29 cNIC) and an external validation cohort (40 cNPSLE vs. 13 cNIC). Robustness of discovery-stage protein selection was assessed by stratified bootstrap resampling and leave-one-out analysis. The diagnostic panel was fitted and locked in the internal cohort, and applied to the external cohort without refitting. RESULTS: Our proteomic profiling quantified 2175 CSF proteins, revealing enrichment in immune and metabolic pathways. We identified three proteins, matrilin-3 (MATN3), tyrosine-protein kinase Mer (MERTK), and V-set and immunoglobulin domain-containing 4 (VSIG4), that were consistently upregulated in cNPSLE compared with both cNIC and NDC groups. In the internal and external validation cohorts, MATN3 demonstrated high diagnostic accuracy (AUC = 0.875 and 0.868, respectively), followed by MERTK (AUC = 0.801 and 0.775) and VSIG4 (AUC = 0.724 and 0.752). Notably, a composite three-protein panel achieved discriminatory performance, yielding AUCs of 0.921 in the internal cohort and 0.892 in the external cohort, when the locked model trained in the internal cohort was applied without refitting. CONCLUSION: We identified and validated a three-protein CSF panel (MATN3, MERTK, and VSIG4) that differentiates cNPSLE from CNS infections across multi-center cohorts. This diagnostic signature holds potential for clinical translation to facilitate early diagnosis and precise management of NPSLE.