Anti‑neutrophil cytoplasmic antibody‑associated glomerulonephritis (ANCA‑GN) and immunoglobulin G4‑related disease (IgG4‑RD) are rare autoimmune disorders with reported clinical overlaps. This study used public transcriptomic datasets to explore cross‑tissue transcriptional associations, identify differentially expressed genes (DEGs) in ANCA‑GN, and search for candidate biomarkers that show consistent expression trends in an IgG4‑RD cohort. The work emphasized machine learning prioritization and multi‑method immune profiling, while acknowledging that findings are predictive and require experimental validation.
Datasets were retrieved from GEO. ANCA‑GN primary analysis used GSE104948 (GPL22945) with 22 ANCA‑GN glomerular samples and 18 healthy glomerular controls. IgG4‑RD analysis used labial salivary gland samples from GSE40568 (GPL570) including 5 IgG4‑RD patients and 3 healthy controls; this cohort was reserved for trend validation because of its small size. An independent ANCA‑GN dataset, GSE108109 (GPL19983; 15 ANCA‑GN, 6 controls), was used for external validation. All analyses were performed in R with batch correction applied where appropriate.
After normalization and batch effect correction (SVA), differential expression testing (limma, adjusted P < 0.05 and |log2FC| > 1) identified 305 DEGs in ANCA‑GN: 209 upregulated and 96 downregulated. The IgG4‑RD dataset produced a larger list of DEGs (519 upregulated, 1,998 downregulated) but, due to limited statistical power, was used mainly to assess expression trend concordance (direction of log2 fold change) rather than to re‑discover DEGs.
Functional annotation of ANCA‑GN DEGs (clusterProfiler) demonstrated enrichment in immune and inflammatory processes. Key pathway enrichments included PI3K–Akt and MAPK signaling among others related to immune activation, cell adhesion, and transcriptional regulation. Detailed GO terms highlighted processes such as regulation of cell adhesion and responses to peptide hormones and xenobiotic stimuli.
ANCA‑GN DEGs were analyzed in STRING with a confidence cutoff ≥0.7, and visualized in Cytoscape. Hub genes were prioritized using topological metrics (Maximal Clique Centrality) to identify highly connected nodes within the PPI network as candidates for downstream machine learning selection.
Two complementary machine learning algorithms were applied to reduce false positives: LASSO regression with 10‑fold cross‑validation to identify genes minimizing prediction error, and the Boruta random forest wrapper with 1,000 iterations to compare feature importance against shadow features. The intersection of genes selected by both methods constituted the candidate gene set for further cross‑tissue assessment.
Because the IgG4‑RD cohort was small, validation emphasized expression trend consistency. DEGs identified in ANCA‑GN were mapped to the IgG4‑RD differential results and log2 fold‑change directions were compared. Genes with the same direction (log2FC > 0 for upregulation or < 0 for downregulation) in both tissues were considered to show consistent trends suggestive of cross‑tissue relevance.
Predictive value in ANCA‑GN was assessed with ROC curves. Internal validation used bootstrap resampling (200 iterations) on GSE104948; external validation used GSE108109. Two candidates emerged with cross‑tissue trends: ITGAM and CXCL1. ITGAM achieved robust discrimination (internal AUC = 0.921; external AUC = 0.844). CXCL1 had high internal discrimination (AUC = 0.937) but substantially lower external performance (AUC = 0.622), indicating limited generalizability.
Single‑gene GSEA (GSEABase) was applied to explore pathway associations of candidate genes in each dataset by ranking genome‑wide correlations and testing KEGG gene sets. Top enriched pathways for target genes were extracted per dataset to identify shared or disease‑specific functional contexts; enrichment results supported immune and inflammatory pathway links for prioritized genes.
Immune profiling used multiple algorithms: CIBERSORT (LM22) for relative proportions of 22 immune cell types, ssGSEA for small‑sample estimation in IgG4‑RD, and MCPcounter for robust cross‑platform cell‑type abundance estimates. Both diseases showed increased B cells, T cells, and monocytes. However, effector compositions differed: neutrophils were prominent in ANCA‑GN, whereas plasma cells and M2 macrophages dominated in IgG4‑RD. Spearman correlations between candidate genes and cell‑type abundances were computed and adjusted for multiple testing.
Regulatory relationships were explored using TRRUST for transcription factors and miRWalk for miRNA–target predictions. Network construction in Cytoscape suggested SPI1 as a potential regulator of ITGAM, while CEBPD, NFKB1, and RELA were implicated as drivers of CXCL1 expression under inflammatory conditions. Predicted miRNA interactions were included to map multi‑layer regulation; these predictions require experimental validation.
The analysis revealed shared transcriptomic trends and immune pathway enrichments between ANCA‑GN and IgG4‑RD at the in silico level. ITGAM and CXCL1 emerged as candidate cross‑tissue biomarkers; ITGAM demonstrated stronger and more consistent diagnostic performance across internal and external ANCA‑GN cohorts, whereas CXCL1’s lower external AUC suggests limited reproducibility.
All findings derive from bioinformatics analyses of public datasets (GSE104948, GSE40568, GSE108109). The IgG4‑RD cohort was small (n = 5 cases), limiting statistical power and necessitating trend‑based validation rather than differential significance. The authors explicitly note that biomarkers and regulatory inferences are predictive and need experimental and clinical validation before translational application.
This study provides a systematic transcriptomic comparison suggesting shared immune‑related signatures between ANCA‑GN and IgG4‑RD and proposes ITGAM (with supporting evidence of robust AUCs) as a promising cross‑tissue biomarker candidate. CXCL1 also shows potential but requires further validation due to inconsistent external performance. These results offer hypotheses for mechanistic study and biomarker development but should not guide clinical decision‑making until validated experimentally.