Document resource
Introduction Clinical classification frameworks for IBD have recently been revised and acknowledge the pressing need for molecular classification tools, especially if they can be harnessed to map and predict patient trajectories.Methods Mucosal transcriptional modules (TMs) were defined with weighted gene network correlation analysis (WGCNA) in a large population with moderate-to-severe active Ulcerative Colitis (UC, n=358) (UNIFI, Sands et al 2019). TM enrichment scores in baseline biopsies were correlated with key future outcomes, including treatment response and resistance. TM signatures were validated across multiple independent UC datasets and treatments. Functional roles of TMs were inferred with over-representation analysis. Module-trait analyses determined relation to outcomes. Machine learning (ML) and elastic net regression identified the most discriminatory features of TMs associated with outcome. Analyses were performed using R 4.3.2 (Vienna).Results Overall, 23 modules were defined. TM1 (n=325 genes) enrichment associated most highly with ustekinumab resistance (Spearman rank correlation, ρ: -0.26), along with TM16 (n=313, ρ: -0.25) and TM4 (n=168, ρ: -0.23), whilst TM7 (n=439, ρ: 0.25), TM15 (n=139, ρ: 0.24), TM3 (n=714, ρ: 0.23) and TM18 (n=387, ρ: 0.22) correlated most strongly with response. Resistance modules represented neutrophil activity, extracellular matrix dysregulation and cellular stress pathways, respectively. Response modules related to protein glycosylation (TM3) and mitochondrial respiration (TM15). TM1, TM16 and TM4 were highly expressed by inflammatory monocytes, inflammatory fibroblasts and B cells, respectively. All response modules were highly enriched in epithelial cells.Fewer than 5% of patients with the highest TM1, TM16 or TM4 enrichment, or lowest TM7 or TM15 enrichment achieved combined endoscopic and histologic healing at week 8 with ustekinumab, or week 6 with infliximab/golimumab. They predicted treatment resistance/response, with area under the curve (AUC) up to 0.72 for ustekinumab and 0.96 for anti-TNFs. We considered if these inversely related TM groups could classify UC patients. Two new modules were derived: (1) top 50 genes from the resistance modules (n=150) and (2) top 50 genes from the response modules (n=200). Patients were labelled if they belonged to the top tertile of either group. This resulted in full separation along PC (principal component) 1 in four reposited datasets and our original cohorts (figure 1). ML identified the most discriminatory transcripts driving classification and outcomes, which could be refined into feature panels and converted to a clinically tractable biomarker platform.Conclusions Modular analysis of tissue transcriptomics can be harnessed to create a novel molecular classification of UC that encapsulates the predominating biological pathways of disease in individual patients and their likely treatment outcomes. This could unlock precision medicine approaches in UC.Abstract O63 Figure 1