A Blood Transcriptomic Signature for Distinguishing Ulcerative Colitis and Crohn's Disease Informed by Non-Inflamed Colorectal Mucosa

Background Developing blood-based transcriptomic biomarkers to distinguish ulcerative colitis (UC) from Crohn's disease (CD) is clinically desirable because blood sampling provides a minimally invasive alternative to tissue biopsy. However, identifying UC/CD-discriminatory transcriptional signals directly in blood remains challenging because the two diseases show substantial similarity in their peripheral-blood transcriptomic profiles against a background of shared inflammatory and immune signals. A tissue-informed strategy may provide a rational approach to blood biomarker discovery by first identifying UC/CD-discriminatory transcriptional signals in disease-relevant tissue and subsequently evaluating their detectability in blood. However, how tissue-informed prioritization compares with direct blood-based selection remains unclear. Because the colorectum is a major site of involvement in both UC and colonic CD, colorectal mucosa provides an appropriate tissue source for identifying disease-associated signatures. Non-inflamed colorectal mucosa may be particularly informative because the reduced influence of overt inflammation may facilitate the detection of UC/CD-specific transcriptional differences. Accordingly, this study evaluated whether transcriptional signatures derived from non-inflamed colorectal mucosa could be detected in peripheral blood and support blood-based discrimination of UC and CD. Aims To identify UC/CD-discriminatory transcriptional signatures in non-inflamed colorectal mucosa, evaluate their translation to peripheral blood, and compare tissue-informed prioritization with direct blood-only feature selection and a published mucosal signature for blood-based UC/CD discrimination. Methods We developed a sequential tissue-to-blood transcriptomic framework for UC/CD discrimination. The blood cohort was first partitioned into training and held-out internal test sets. Because the blood and tissue datasets included overlapping participants, tissue samples from held-out blood-test participants were excluded from tissue discovery. Differential-expression analyses were then performed independently in the remaining non-inflamed rectal and colonic mucosa, excluding candidates with reversed UC/CD differential-expression directions in the corresponding inflamed tissue. Within the blood training set, tissue-derived candidates were sequentially filtered for blood testability, tissue-to-blood directional concordance, and blood differential-expression evidence to generate a tissue-informed candidate set. A matched-size blood-only candidate set was derived independently from blood differential-expression analysis. Candidate genes from both approaches were ranked using resampling-based stability selection with LASSO and elastic-net models, followed by repeated five-fold cross-validation to determine panel size. Logistic regression models based on the tissue-informed panel, blood-only panel, and published PI3/ANXA1/VDR mucosal signature were fitted in the complete blood training cohort and subsequently evaluated in the held-out test cohort. Results Non-inflamed colorectal mucosa showed more extensive transcriptional differences between IBD and controls than between UC and CD, and most UC/CD-discriminatory candidates retained concordant differential-expression directions in corresponding inflamed tissue. Tissue-to-blood filtering yielded 51 tissue-informed candidates, and the top 51 blood-only genes were retained as a matched-size comparator set. Stability-based prioritization and panel-size optimization identified a five-gene tissue-informed panel comprising GLIS3, SLC4A10, HKDC1, ITGB8, and CRIP1 and a nine-gene blood-only panel comprising ENPP3, SORCS3, MTMR9LP, ABCG1, GALNTL6, POLB, DHRS9, DISC1FP1, and LINC00402. In the held-out test cohort, the tissue-informed panel achieved an AUC of 0.69 (95% CI, 0.61–0.77), compared with 0.67 (95% CI, 0.59–0.76) for the blood-only panel and 0.55 (95% CI, 0.46–0.64) for the PI3/ANXA1/VDR mucosal signature. Conclusions UC/CD-associated transcriptional signals identified in non-inflamed colorectal tissue were selectively retained in peripheral blood. Tissue-informed prioritization yielded a compact five-gene panel with moderate held-out discrimination, similar to that of the larger nine-gene blood-only panel, and a higher observed AUC than the published mucosal signature. Although tissue-informed prioritization did not improve discriminatory performance over direct blood-based selection, it provided a biologically constrained approach to blood biomarker discovery that yielded a more compact feature set while retaining similar observed discrimination. External validation is required to assess the generalizability of these findings.