GRN and KLRB1 define a shared peripheral-blood transcriptomic signature linking SLE and IPF.

BACKGROUND

Systemic lupus erythematosus (SLE) and idiopathic pulmonary fibrosis (IPF) share immune-inflammatory features, yet their convergent peripheral-blood transcriptomic signatures remain incompletely defined. We sought to identify shared blood gene programs linking SLE and IPF, prioritize robust cross-disease markers, and evaluate parsimonious diagnostic models with experimental and external assessments.

METHODS

Peripheral-blood transcriptomes were analyzed in GEO discovery cohorts (SLE: GSE49454; IPF: GSE33566). Differential expression (limma) and weighted gene co-expression network analysis (WGCNA) were performed separately per disease, and concordant shared signals were integrated to form a shared candidate pool. Consensus feature selection combined LASSO logistic regression, nested cross-validated SVM-RFE, and random forest to derive parsimonious gene panels for SLE and IPF. Logistic-regression models were trained in discovery cohorts and externally validated in independent cohorts (SLE: GSE65391, GSE72509; IPF: baseline samples from longitudinal GSE93606). Experimental validation was conducted in an independent hospital cohort (60 SLE, 30 healthy controls) using PBMC RT-qPCR and serum GRN ELISA, with correlation and covariate-adjusted association analyses. Fixed models were additionally applied without refitting to non-target inflammatory cohorts (RA: GSE93272; ICU sepsis/non-infectious critical illness: GSE134347).

RESULTS

Discovery analyses identified 389 SLE and 248 IPF DEGs and yielded 43 concordantly regulated shared DEGs; WGCNA identified 43 shared module genes, producing a non-redundant shared candidate pool of 78 genes enriched for B-cell and myeloid programs. Consensus selection generated a 6-gene SLE panel (EIF2AK2, GRN, ASGR2, KLRB1, LGALS9, KLF13) and a 4-gene IPF panel (GRN, ARG1, KLRB1, FCMR). The SLE model achieved AUC 0.996 in discovery and validated at AUC 0.888 (GSE65391) and 0.761 (GSE72509); the IPF model achieved AUC 0.906 in discovery and 0.722 in baseline validation. In the hospital cohort, RT-qPCR confirmed dysregulation of the six-gene panel, and serum GRN was markedly elevated in SLE (median [IQR] 43.58 [38.44-54.42] vs. 14.26 [12.79-15.26] ng/mL). Within SLE, serum GRN correlated with SLEDAI and inversely with C3/C4 and WBC; after covariate adjustment, associations with WBC, ESR, C3, and C4 remained significant, whereas associations with hs-CRP and SLEDAI were attenuated. In non-target cohorts, the SLE model showed moderate discrimination for RA (AUC 0.73) but limited discrimination for ICU sepsis (AUC 0.64) and none for non-infectious critical illness (AUC 0.50), while the IPF model showed minimal discrimination for RA (AUC 0.51) but high discrimination for ICU groups (AUC 0.99 and 0.96).

CONCLUSIONS

GRN and KLRB1 anchor a shared peripheral-blood transcriptomic signature linking SLE and IPF, enabling parsimonious diagnostic models with multi-cohort validation and clinical experimental support. External in silico applications to other inflammatory contexts indicate context-dependent model behavior, underscoring the importance of cohort-appropriate interpretation and validation.