Tissue-adjusted transcriptomic pathway differences between class IV and class V lupus nephritis

Abstract

Introduction: Lupus nephritis (LN) is molecularly heterogeneous, and transcriptomic comparisons of histological classes can be confounded by renal tissue compartment and repeated sampling of the same patient. Materials and methods: We reanalyzed GSE127797 to compare class IV with class V LN using 44 Affymetrix Human Transcriptome Array 2.0 profiles from 25 patients; 19 patients contributed paired glomerular and tubulointerstitial profiles. A limma model included tissue compartment and estimated within-patient correlation (consensus correlation, 0.356). Results: No individual feature was significant after Benjamini–Hochberg (BH) correction. Using an explicitly exploratory threshold of nominal p < 0.05 and |log2 fold change| ≥ 0.5, 39 candidate genes were identified (24 higher in class IV and 15 higher in class V). No class-by-tissue interaction survived false-discovery-rate correction, and all 39 candidates had concordant effect directions in both tissue compartments. Excluding five quality-control-flagged profiles retained the direction of all 39 candidates, with 32 remaining above the original exploratory threshold and an all-feature log-fold-change correlation of r = 0.893. The log-fold-change and moderated- t rankings, both derived from the same limma fit, showed high within-fit concordance but were not treated as independent replication. A correlation-adjusted mean rank gene set test using a pre-ranked statistic (CAMERA-PR) sensitivity analysis supported the direction of all 10 displayed Gene Ontology (GO) and all 10 displayed Kyoto Encyclopedia of Genes and Genomes (KEGG) terms; 9 displayed GO terms and all 10 displayed KEGG terms were BH-significant under CAMERA-PR. Class IV showed greater B-cell/immunoglobulin, complement and phagocytosis-related signals, whereas class V showed greater amino-acid and xenobiotic-metabolism signals. A Search Tool for the Retrieval of Interacting Genes/Proteins (STRING)-derived protein association network contained 24 nodes and 51 associations at the primary threshold. C-C motif chemokine ligand 2 ( CCL2 ) ranked first by maximal clique centrality in the primary and broad networks and second in the strict network. Conclusions: Thus, pathway-level contrasts were supported by an inter-gene-correlation-aware sensitivity analysis, whereas individual candidate genes and network hubs remain exploratory and require external validation.