Baseline mucosal transcriptomic endotypes stratify healing capacity and biologic therapy outcomes in ulcerative colitis: an integrative multi-cohort study

Numerous baseline mucosal transcriptomic signatures have been proposed to stratify therapeutic outcomes in ulcerative colitis (UC), but cross-cohort transferability is frequently undermined by platform heterogeneity, endpoint/timepoint differences, and drug-specific biases. The resulting lack of a biologic-agnostic molecular baseline has limited clinical translation and prospective trial design.

We integratively re-analyzed endpoint-linked baseline mucosal transcriptomes from seven public UC GEO series (778 unique baseline samples), with three additional public accessions used for disease-activity and contextual analyses. The primary response/healing synthesis used five study slices spanning anti-TNF, anti-integrin, and etrolizumab-treated cohorts. Unsupervised decomposition identified baseline endotypes, and endotype-outcome associations were estimated within each cohort (OR per 1-SD increase) and summarized by random-effects cross-cohort pooling with heterogeneity reporting, rank/seed sensitivity analyses, endpoint-restricted summaries, and negative controls.

Across five primary response/healing study slices, an epithelial barrier endotype (EPITHELIAL_BARRIER/NMF_C3) was associated with higher odds of response/healing (random-effects cross-cohort pooling OR = 3.171, 95% CI 1.491-6.744), whereas macrophage-inflammation and lymphoid/complement-stromal axes (MACROPHAGE_INFLAMMATION/NMF_C1 and LYMPHOID_COMPLEMENT_STROMAL/NMF_C2) were associated with lower odds of response/healing (OR = 0.359, 0.175-0.735; OR = 0.378, 0.220-0.651). The epithelial estimate was largest in GSE73661 but remained significant in the five-study leave-one-out summary after excluding that cohort (OR = 2.547, 95% CI 1.271-5.105). Cross-mechanism evaluation included UNIFI ustekinumab week-8 mucosal healing (UST arm n = 358; EPITHELIAL_BARRIER OR = 4.483, 2.263-8.881) and directionally consistent effects in etrolizumab. Endotype axes were supported by independent single-cell localization (FDR = 0.0026). Treatment-interaction tests were imprecise and did not establish drug-class-specific prediction. In cohort-heldout prediction, AUC ranged from 0.717 to 0.758, but calibration was variable. A network-based therapeutic prioritization was enriched for UC-approved agents (AUC = 0.741; permutation p = 6.0 × 10^{- 4}) and is presented as hypothesis-generating.

Baseline mucosal endotypes stratify healing capacity and therapeutic outcomes across public cohorts, with orthogonal support from single-cell localization. These findings support prospective evaluation as biologic-agnostic outcome stratifiers rather than established drug-specific predictive biomarkers.