The assessment of patients with ulcerative colitis (UC) remains a challenge despite rapidly evolving medical therapy. Current assessment with endoscopic scoring and histopathology lacks the granularity to correlate strongly with response to therapy. Utilizing a microarray based system for colonic epithelial assessment we looked for stratification of endoscopically similar UC biopsies using molecularly heterogeneous archetypes. Molecular data from 71 UC biopsies (from 61 patients) was obtained from microarray analysis. The top 300 genes correlating with the endoscopic Mayo score (2/3 vs 0/1) were used for an unsupervised analytical method called archetypal analysis, and grouped the biopsies into three distinct clusters. Logistic regression modeling was used to compare archetype scores or cluster membership to endoscopic Mayo score and PC1 in predicting mucosal healing. A contingency table was generated to show evidence of mucosal healing within each of the archetype clusters. We then assessed a subset of biopsies from the original set of 71 (selected due to the availability of biopsies before and after therapy, all on TNF therapy) plus one IBDU case. Patients were classified as ‘responders’ (Mayo 2/3 score to Mayo 0/1 score by last obtained biopsy or scope) or ‘non-responders’ (Mayo 2/3 score that did not decrease to 0/1 by last obtained biopsy or scope). We found three unique groups of biopsies using archetypal analysis (A1: lack of inflammation, A2: inflammation and response to wounding, A3: inflammation). Logistic regression showed that the only models with statistically significant predictive value (p-value < 0.05) were those that contained archetype scores or cluster membership. Response rates differed between archetype clusters with statistical significance (Table 1), while the mayo score distribution within these clusters was not statistically different. In our subset of serial biopsies, the majority of initial biopsies were found to have an A2 archetype, moving to an A1 archetype in follow-up that mirrored response to treatment (Figure 1). This archetypal analysis suggests there is potentially important heterogeneity in UC biopsies that is not accessible by endoscopic Mayo score. Serial biopsies showed dynamic shifts in the archetype composition between biopsies. This may be a useful tool for both initailly prognosticating patients and assessing response to treatment over time with increased granularity and reliablity. UC Patient Response to Therapy Assessed by Initial Archetype Cluster All cases had a Mayo score of 2–3 on initial endoscopy. *score of 0–1 on follow up endoscopy. Pearson’s Chi-Squared = 7.222, df = 2, p = 0.027 UC Patient Response to Therapy Assessed by Initial Archetype Cluster All cases had a Mayo score of 2–3 on initial endoscopy. *score of 0–1 on follow up endoscopy. Pearson’s Chi-Squared = 7.222, df = 2, p = 0.027 Figure 1. Stacked and group bar charts showing archetype composition of biopsies in 10 UC patients taken over time (2–4 serial biopsies/patient). None

Ulcerative colitis (UC) is a chronic, idiopathic inflammatory condition affecting the colonic epithelium, with the inflammasome, T cells, complement activation, and microbiome dysbiosis contributing to pathogenesis. We applied a previously established method of microarray molecular analysis to a set of 71 UC biopsies (from 61 patients), to elucidate the molecular changes associated with active UC, specifically comparing UC to T-cell mediated rejection (TCMR), as a prototype of a sterile, T-cell mediated disease. 71 for-cause UC colonic biopsies were collected at the U of A Hospital in Edmonton, AB and Cedars-Sinai Hospital in LA, California. These biopsies were processed using Affymetrix GeneChip microarrays and the data analyzed in R programming language. Gene expression data was displayed using volcano plots (showing the fold change and association between the genes and endoscopic Mayo score) and heatmaps (showing expression of the top 30 genes in a cell panel). We then analyzed overexpression of the top genes using the DAVID tool1, and compared the results to similar results for top genes overexpressed in TCMR. The volcano plot (Figure 1) showed strong associations between the endoscopic Mayo score and decay accelerating factor (CD55), and moderate associations with calprotectin genes (S100A8/S100A9), with lesser associations for effector T cell transcripts (i.e. CTLA4, interferon gamma (IFNG), and chemokine ligand (CXCL13)), many IFNG inducible transcripts, inflammasome transcripts (CASP1), and toll-like receptors (TLR5). NLRP3 (incriminated in mouse UC models) did not pass the IQR filter and was not significant. Expression of the top genes in a cell panel showed primary expression in monocytes, macrophages, and dendritic cells, with some expression in epithelial and endothelial cells, while minimal expression was found in CD4/CD8 T cells, or in NK cells. Pathway analysis showed major differences between pathway terms. Our findings show that while transcripts associated with cognate T cell inflammatory processes (TCMR) are expressed in UC, there are substantial differences between the pathogenesis of a pure T cell-mediated disease and UC. In our cohort, the top genes associated with the endoscopic Mayo score in UC were not those associated with effector T cells: T cell-associated transcripts were only moderately associated with the Mayo score, and pathway analysis showed significant differences between TCMR and UC, which is much more an inflammatory environment with strong associations to CD55. This suggests that cognate T cell recognition is present in UC but supplemented by a second source of inflammation. 1. Huang et al. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nature Protocols (2009). Figure 1. Molecular landscape of UC by volcano plot. Samples towards upper right have high association and fold change, indicating a significant association with endoscopic Mayo score, while samples towards the middle left are not strongly associated. None