Spatial Relationships Between Immune Infiltrate and Tumour Buds Improves Prognostic Accuracy in Stage II Colorectal Cancer

Purpose of the study: Within the tumour microenvironment, cancer cells coexist and interact with the heterotypic immune system. Tumour budding and the immune infiltrate are established prognostic factors in stage II colorectal cancer (CRC), though the importance of their spatial interaction is less studied. The objectives of this research were to determine the prognostic value of the dynamic interplay between tumour buds (TBs) and infiltrating immune cells.Methods: Multiplexed immunofluorescence for TBs, CD3+, CD8+ lymphocytes and CD68+, CD163+ macrophages was performed across two sequential whole tissue slides (n=232). Automated image and spatial analysis was applied to quantify the distinct cell populations and their spatial interactions. Machine learning was used for the development of a prognostic risk model.Summary of results: TB (p=0.001) and lymphocytic density (p<0.001) were found to be significantly correlated with disease-specific survival. However, a novel prognostic model which also incorporated their spatial relationship was shown to better stratify stage II CRC patients at high risk for disease-specific death (p<0.001) compared with the clinical gold standard of pT stage (p=0.003). The model was developed using data from 114 patients and validated in two independent cohorts (cohort 1: n=56 and cohort 2: n=62). Furthermore, a low ratio of CD68+/CD163+ cells at the tumour core was associated with better survival (p<0.001). A prognostic model integrating the above-mentioned features allowed the identification of patients with low-risk of disease-specific death with 100% sensitivity.Conclusions: We demonstrate an automated machine learning workow that captures the cellular interactions present in the tumour microenvironment and which, may lead to improved and personalised clinical decision making.