DOP70 An integrated multi-omics biomarker predicting endoscopic response in ustekinumab treated patients with Crohn's disease. (25th January 2019)

• Record Type: Journal Article
• Authors: Verstockt, B; Sudahakar, P; Creyns, B; Verstockt, S; Cremer, J; Wollants, W-J; Organe, S; Korcsmaros, T; Madgwick, M; Van Assche, G; Breynaert, C; Vermeire, S; Ferrante, M
• Is Part Of: Journal of Crohn's and colitis. Volume 13(2019)Supplement 1
• Page Start: S072
• Page End: S073
• Publication Date: 2019-01-25
• Subjects: Inflammatory bowel diseases -- Periodicals; 616.344005
• DOI: 10.1093/ecco-jcc/jjy222.104
• Languages: English
• ISSNs: 1873-9946

Abstract

Abstract: Background: Ustekinumab (UST), an anti-IL12/23p40 monoclonal antibody, has been approved for Crohn's disease (CD). The aim of this study was to identify baseline predictors of response using several omics layers, which ultimately may result in a multi-omics panel allowing individualised UST therapy. Methods: Inflamed colonic ( n = 25) and ileal ( n = 22) biopsies were retrieved prior to first UST administration in patients with active CD, in addition to sorted circulating CD14+ monocytes and CD4+ T cells ( n = 39). RNA was extracted from both lysed biopsies and sorted cells, and RNA sequencing performed. Proteomic analysis was performed on baseline serum samples ( n = 86) using OLINK Proseek inflammation. Genotyping data were generated using Immunochip ( n = 38). The genetic risk burden was determined for every patient using the SNPs which overlap with genes encoding functional proteins or RNAs. The six above-described layers of omics data were integrated and analysed using Multi-Omics Factor Analysis (MOFA). The strongest omic layers in terms of variance contribution to the latent factors explaining endoscopic response (≥50% in SES-CD by w24) were identified. Dimensionality reduction and feature extraction from the strongest -omic layers were performed followed by predictive modelling on the top-ranked features. Cross-validation using distinct test and training sets was performed for the ensemble and individual classifiers, as an internal validation to avoid over-fitting. Results: MOFA identified 19 latent factors (LF, minimum explained variance 2%), with 3 LFs correlating with endoscopic response at w24 ( r = −0.24, r = 0.27, r = −0.25; p = 0.03, p = 0.01, p = 0.02). The genomic and CD14 transcriptomic layers contributed significantly to the prediction of endoscopic response. Predictive modelling based on the results of the most dominant omic layers revealed a 10-feature panel predicting endoscopic response at w24 with an accuracy of 98%. In contrast, classification performance based on 10 randomly selected features resulted in a drastic drop in accuracy (66%). Only 2 of the 10 features exhibited significant correlation with baseline faecal calprotectin, and 1 with CRP, suggesting that this panel is not a simple surrogate of baseline inflammation. From the genetic risk burden, we identified a 15-gene panel which could classify (accuracy 96.6%) the patients based on endoscopic response. Conclusions: Through multi-omic data integration, we discovered pathways contributing to UST response, and identified a 10-feature transcriptomic and 15-feature genomic panel predicting endoscopic response to UST standard dosage. Further validation in larger and independent cohorts is warranted, as well as its UST specificity.