Integrated bioinformatics reveals tubulointerstitial immune microenvironment signatures and machine learning-driven prognostication of clinical treatment response in lupus nephritis.

Lupus nephritis (LN) represents the most severe renal manifestation of systemic lupus erythematosus (SLE), contributing to significant morbidity. While current assessments focus on glomerular pathology, tubulointerstitial lesions may offer critical insights into disease progression and treatment response. This study develops a clinical prediction model integrating tubulointerstitial molecular signatures. We performed bioinformatics analysis using two independent tubulointerstitial gene expression datasets (GSE113342 and GSE200306), applying batch effect correction and principal component analysis (PCA) to identify differentially expressed genes (DEGs). A protein‒protein interaction (PPI) network isolated hub genes, and least absolute shrinkage and selection operator (LASSO) regression defined the novel "Nscore" parameter predictive of treatment response. The Nscore, incorporating seven key genes (EGR1, IL6R, TFRC, CCL19, IFI16, IFI35, and Fra1), showed a significant positive correlation with 24-h proteinuria and effectively distinguished complete-response (CR)/partial-response (PR) from non-response (NR). Immune deconvolution using the CIBERSORT algorithm revealed an increased abundance of T follicular helper (Tfh) cells and M1 macrophages in NR samples. A clinical nomogram integrating Nscore and sex demonstrated excellent discrimination. This model combines molecular biomarkers with clinical parameters to improve personalized therapeutic stratification, advancing treatment strategies beyond traditional glomerulocentric paradigms and identifying immune cell signatures as potential targets for immunomodulatory interventions.