The prognostic impact of the colorectal cancer microenvironment components and their spatial relationships

Background & Objectives:The tumour immune infiltrate contributes majorly in tumour progression and patient survival outcome. However, the aggressive potential of colorectal cancer cannot be manifested without capturing the interplay between cancer cells and the host’s immune response. In this study, we examined the prognostic value of tumour buds(TBs) reflecting tumour progression, the tumour immune infiltrate and their spatial relationships in stage II colorectal cancer. Methods:Two multiplex immunofluorescence antibody panels comprising CD3+, CD8+, CD68+, CD163+, pancytokeratin and Hoechst were applied in two sequential whole slide images (n=232). Cell quantification and the evaluation of their spatial interactions was performed through automated image analysis. Exported features were processed through machine learning algorithms for the development of a new prognostic risk model. Results:TB (p=0.001) and lymphocytic density (p<0.001) were found to be prognostically significant independently. Though, their integration with their spatial relationship (p<0.001) reported a more significant cohort stratification than the clinical gold standard of pT stage (p=0.003) in a training cohort (n=114) and two independent validation cohorts (n=56 and n=62). Inclusion of the macrophage cell data in a new prognostic model allowed the low-risk patients to be identified with 100% sensitivity. No clinicopathological features were selected for any of the models by the machine-learning workflow. Conclusion:Here we demonstrate how the integration of automated image analysis and machine learning in studying the tumour heterotypic microenvironment can lead to the more accurate stratification of patients with stage II colorectal cancer. Funding for the study was provided by Medical Research Scotland, andIndica Labs, Inc., provided in-kind resource.