Multi-omics derivation of a core gene signature for predicting therapeutic response and characterizing immune dysregulation in inflammatory bowel disease

Mingming Wang¹Liping Liang²Zibo Tang³Han Jimin⁴Lele Wu⁵Le Liu^1*Ye Chen^1*

• ¹Southern Medical University, Guangzhou, China
• ²Guangzhou First People's Hospital, Guangzhou, Guangdong Province, China
• ³Shenzhen People's Hospital, Jinan University, Shenzhen, Guangdong Province, China
• ⁴Tsinghua University, Beijing, Beijing, China
• ⁵Shenzhen Hospital, Southern Medical University, Shenzhen, Guangdong Province, China

Background Inflammatory bowel disease (IBD) presents unpredictable therapeutic responses and complex immune dysregulation. Current precision medicine approaches lack robust molecular tools integrating transcriptomic signatures with immune dynamics for personalized treatment guidance. Methods We performed multi-omics analyses of GEO datasets using machine learning algorithms (LASSO/Random Forest) to derive a four-gene signature. Validation employed ten algorithms and nomogram construction. Immune infiltration (CIBERSORT/ssGSEA), single-cell RNA sequencing, and DSS-colitis models characterized immune dynamics, cellular specificity, and therapeutic response modulation. Results We identified 536 differentially expressed genes significantly enriched in IL-17 signaling, TNF signaling, and cytokine-cytokine receptor interactions. WGCNA revealed six co-expression modules with disease-specific correlations: turquoise module strongly correlated with Crohn's disease (r=0.6, P=4×10⁻²⁰) and purple module with ulcerative colitis (r=0.55, P=1×10⁻¹⁶). The four-gene signature (CDC14A, PDK2, CHAD, UGT2A3) demonstrated exceptional diagnostic performance across ten validation algorithms (AUC range: 0.86-0.97), with the integrated nomogram achieving superior accuracy (AUC=0.952) compared to individual genes (CDC14A: 0.934, PDK2: 0.913, CHAD: 0.893, UGT2A3: 0.797). Consensus clustering stratified patients into two distinct molecular subtypes: Cluster 1 exhibited elevated M1 macrophages, activated dendritic cells, and neutrophils with enhanced glycolysis and mTORC1 signaling, while Cluster 2 showed higher signature gene expression, enhanced oxidative phosphorylation, and enrichment in regulatory immune populations including Tregs and M2 macrophages. Single-cell RNA sequencing revealed cell-type-specific expression patterns: PDK2 demonstrated widespread expression across epithelial cycling cells and stem cells, UGT2A3 showed preferential epithelial localization, and CDC14A exhibited selective enrichment in innate lymphoid cells. Nomogram-based risk stratification effectively predicted biologic treatment responses across multiple therapeutic classes using four independent treatment datasets (GSE16879, GSE92415, GSE73661, GSE206285): low-risk patients demonstrated superior response rates to golimumab (59.1%), infliximab (54.8%), and vedolizumab (29% vs. 15% in high-risk group). Connectivity Map analysis identified MS.275 as the top therapeutic enhancer, with experimental validation in DSS-induced colitis confirming synergistic anti-inflammatory effects with TNF-α inhibitors, improving disease activity indices and restoring signature gene expression patterns. Conclusion This mechanistically grounded four-gene signature enables precise IBD patient stratification across distinct immunological subtypes and predicts biologic responses, providing validated molecular tools for precision immunotherapy and personalized treatment optimization.