Recurrence after surgery for concurrent metastatic colorectal cancer: The perspective of bioinformatics and machine learning.

Abstract

4043 Background: Recurrence of concurrent metastatic colorectal cancers (mCRCs) after surgery is still a challenge. But mCRCs’ outcomes are heterogeneous, and no clinicopathological methods can predict its recurrence and guide postoperative treatment from an intrinsic cell activities and extrinsic immune microenvironment perspective. We aimed to identify such gene models. Methods: Gene expression analysis on CRCs. Based on metastasis-related genes, a metastatic evaluation model (MEM) was developed, dividing mCRCs into high and low recurrence risk clusters. Machine learning tested MEM’s importance to predict recurrence. Further investigating MEM’s two clusters made an immune prognostic model (IPM) with immune genes differentially expressed between MEM clusters. The predictive performance of MEM and IPM on prognosis was comprehensively analyzed and validated. The mechanism of IPM on the immune microenvironment and response to immuno/chemotherapy was analyzed extensively. Results: RNA data of 998 CRCs were analyzed. High postoperative recurrence risk in mCRCs was owing to immune response’s down-regulation, which was influenced by 3 MEM genes ( BAMBI, F13A1, LCN2) and their related 3 IPM genes ( SLIT2, CDKN2A, CLU). MEM and IPM were developed and validated on 239 mCRCs to differentiate a low and high recurrence risk (AUCs > 0.7). Functional enrichment analysis showed immune response and immune system diseases pathway represented the major function and pathway related to IPM gene. IPM high-risk group (IPM-high) had higher fractions of Tregs ( P= 0.04), lower fractions of resisting memory CD4+ T cells ( P= 0.02) than IPM-low. And stroma and immune cells in IPM-high samples were scant ( P= 0.0002, 0.001, respectively). In IPM-high, MHC class II molecules all down-expressed, and DNA methylation disordered. TIDE algorithm and GDSC analysis discovered IPM-low was more promising to respond to both anti-CTLA4 therapy ( P= 0.005) and common FDA targeted drugs ( P< 0.05), while IPM-high had nonresponse to both of them. But anti-CDKN2A agent with activation of MHC class II response might reverse the dilemma of this refractory mCRCs subgroup. Conclusions: Postoperative recurrence of mCRC is strongly related to immune microenvironment. Our two relative gene models could identify subgroups of mCRC with different recurrence risk, and stratify mCRCs sensitive to immune/chemotherapy, even highlight the ignored importance of MHC class II molecules on immunotherapy in mCRCs for the first time.