Abstract 1563: Prediction of immune-checkpoint blockade response in gastrointestinal cancer patients using a machine learning-based RNA signature

Abstract

Abstract Background: Currently, the objective response rate of immune checkpoint blockade (ICB) immunotherapy in gastrointestinal (GI) cancers was still unsatisfying, underscoring the urgent need for the development of novel biomarkers. To date, candidate biomarkers in GI cancers include microsatellite instability (MSI), tumor mutational burden (TMB), program death ligand 1 (PD-L1) expression, and RNA signatures associated with T cell-inflamed tumor microenvironment. Here, aiming to precisely capture the pre-existing “responsive feature” in GI cancers, we adopted a machine learning strategy based on RNA expression profiling data. Methods: This retrospective study included 96 metastatic GI patients treated with ICBs, who were randomly assigned into the discovery (75%) and validation (25%) cohorts. All formalin-fixed paraffin-embedded (FFPE) tumor specimens were subjected to the 395-plex immune oncology (IO)-related gene target sequencing. Based on the RNA profiling data of patients with durable clinic benefit (DCB) or no durable benefit (NDB), a linear support vector machine (linearSVM) strategy was applied for predictive model construction. Results: Notably, the linearSVM strategy helps construct a 24-gene RNA signature (IO-score) for discriminating DCB and NDB, which comprises genes involving in tumor antigen, tumor suppressor/oncogene, lymphocyte markers, interferon and checkpoint signaling pathways. In specific, patients with higher IO-score demonstrated stronger DCB rates than lower IO-score (discovery cohort: 92.0% [23 of 25] vs. 4.3% [2 of 47], P < 0.001; validation cohort: 85.7% [6 of 7] vs. 17.6% [3 of 17], Fisher's exact test, P =0.004). Moreover, compared with IO-score low subgroup, IO-score high subgroup displayed improved overall survival (discovery cohort: hazard ratio (HR), 0.29; 95% CI, 0.15-0.56; P =0.003; validation cohort: HR, 0.32, 95% CI, 0.10-1.05, P =0.04) as well as progression free survival (discovery cohort: HR, 0.22; 95% CI, 0.13-0.38; P <0.001; validation cohort: HR, 0.23, 95% CI, 0.09-0.60, P =0.007). Importantly, IO-score reveals a higher predictive value in both cohorts (discovery: area under the receiver operating characteristic curve (AUC)=0.97, 95% CI, 0.93-1.0; validation: AUC=0.74, 95% CI, 0.51-0.97), when compared with TMB-High (AUC=0.69, 95% CI, 0.57-0.80), MSI-High (AUC=0.59, 95% CI, 0.45-0.73) and PD-L1 positivity (AUC=0.52, 95% CI, 0.37-0.67). Conclusion: The machine learning methodologies offer a promising approach for capturing the pre-existing “responsive feature” in both tumor and infiltrating immune cells. The novel RNA signature, IO-score, may help facilitate individualized management in immunotherapy in GI cancer patients. A larger cohort study would be useful to optimize the cutoff value of IO-score. Citation Format: Lin Shen, Henghui Zhang, Zhihao Lu, Huan Chen, Xi Jiao, Wei Zhou, Shuang Li, Ying Yang, Dandan Liang. Prediction of immune-checkpoint blockade response in gastrointestinal cancer patients using a machine learning-based RNA signature [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 1563.