Advancing lupus nephritis research through multi-omics and predictive modeling.

Abstract

IntroductionLupus nephritis (LN) is characterized by significant heterogeneity and a complex pathophysiology, which traditional methods struggle to fully resolve. Advanced multi-omics approaches are essential to disentangle its cellular and molecular drivers.MethodsWe employed an integrative strategy combining single-cell RNA sequencing (scRNA-seq) profiling of LN biopsies with large-scale bulk RNA-seq cohorts. We applied non-negative matrix factorization (NMF) to scRNA-seq data to define robust immune meta-programs and utilized CellChat to decode cell-cell communication networks. Leveraging these insights to overcome sample size limitations, we prioritized key pathways and developed 399 machine learning predictive models using bulk transcriptomics, validated on independent cohorts.ResultsScRNA-seq analysis revealed a distinct cellular landscape, including a rare population of plasmacytoid dendritic cells (pDCs) and an expanded population of CD56dimCD16⁺ natural killer (NK) cells expressing high levels of IFN-γ and perforin, suggesting a role in inflammatory pathology. Macrophage subpopulation CM2 emerged as a central pro-inflammatory hub, potentially driving fibrosis via autocrine signaling and epithelial activation. We observed reduced Treg-B cell interactions, suggesting a regulatory collapse. Our machine learning models, based on innate immunity, circadian rhythms, apoptosis, and NF-κB signaling, achieved high diagnostic accuracy (AUC = 0.929 for innate immunity). Hub genes, including CYBB, CSF2RB, and IRF8, were confirmed to be upregulated in LN and correlated with clinical severity in external validation datasets. Molecular docking simulations suggested a potential structural basis for CYBB-dexamethasone interaction, providing a hypothesis for future verification.DiscussionThis study identifies CM2 macrophages and dysregulated pDC-NK axes as key drivers of LN. By bridging cellular interactomes with clinical predictive modeling, we provide a robust roadmap for precision detection and identifying potential therapeutic targets in LN.