P656 Plasma Inflammation-related Proteins Predict the Therapeutic Effect of 5-ASA in Patients with Ulcerative Colitis

Abstract

Abstract Background 5-aminosalicylate acid (5-ASA) is the first line therapy for mild to moderate ulcerative colitis (UC). However, up to 56% of patients on 5-ASA treatment could not achieve mucosal healing, which is the therapeutic target of UC. At present, predictors for patients’ response to 5-ASA are lacking. Our aim was to identify circulating markers to predict the therapeutic effect of 5-ASA in patients with UC. Methods Plasma samples were collected from patients with active UC at baseline and analyzed using the OLINK panel (Target 96 Inflammation). Effective therapy was defined as achieving clinical remission within 8 weeks or mucosal healing within 1 year of 5-ASA treatment. Orthogonal Projections to Latent Structures Discriminant Analyses (OPLS-DA) were implemented to correlate patient groups to plasma proteins in linear multivariate models. Machine learning models (ML) were constructed to predict ineffective therapy in UC patients. Results Among the 36 UC patients involved (media age 38.5 years old, men 61.6%), 18 patients belonged to the effective group and the ineffective group, respectively. In multivariate analysis, circulating inflammatory protein profiles could distinguish the patients who did not respond well to 5-ASA treatment (Figure 1A). Elevated levels of Axin 1 (p=0.012), sulfotransferase 1A1 (ST1A1) (p=0.032) and interleukin 15 receptor subunit alpha (IL-15RA) (p=0.022), and decreased level of interleukin 2 (IL-2) (p=0.036) were observed in UC patients refractory to 5-ASA therapy (Figure 1B). Receiver operating characteristic curves (ROC) for predicting the therapeutic efficacy of 5-ASA were constructed. The areas under ROC (AUC) of Axin 1, ST1A1, IL-15RA and IL-2 for the prediction of ineffective 5-ASA therapy were 0.744 (p=0.012, sensitivity 83.3%, specificity 72.2%), 0.710 (p=0.031, sensitivity 94.4%, specificity 55.6%), 0.704 (p=0.037, sensitivity 88.9%, specificity 50%) and 0.704 (p=0.037, sensitivity 77.8%, specificity 61.1%), respectively (Figure 1C). Combination of the four protein markers improved the performance of prediction (AUC 0.860 to 1.000, sensitivity 72.2%-100%, specificity 83.3%-100% across all machine learning classifiers) (Figure 1D). Conclusion We identified four plasma inflammation-related proteins associated with the therapeutic effect of 5-ASA in UC. The ML models demonstrated excellent performance in predicting the therapeutic effect of 5-ASA, which could help clinicians identify UC patients who are refractory to 5-ASA in the early course of disease and timely initiation of treatment escalation.