Analysis of the molecular nature associated with microsatellite status in colon cancer identifies clinical implications for immunotherapy

Abstract

Background: Microsatellite instability in colon cancer implies favorable therapeutic outcomes after checkpoint blockade immunotherapy. However, the molecular nature of microsatellite instability is not well elucidated.

Methods: We examined the immune microenvironment of colon cancer using assessments of the bulk transcriptome and the single-cell transcriptome focusing on molecular nature of microsatellite stability (MSS) and microsatellite instability (MSI) in colorectal cancer from a public database. The association of the mutation pattern and microsatellite status was analyzed by a random forest algorithm in The Cancer Genome Atlas (TCGA) and validated by our in-house dataset (39 tumor mutational burden (TMB)-low MSS colon cancer, 10 TMB-high MSS colon cancer, 15 MSI colon cancer). A prognostic model was constructed to predict the survival potential and stratify microsatellite status by a neural network.

Results: Despite the hostile CD8⁺ cytotoxic T lymphocyte (CTL)/Th1 microenvironment in MSI colon cancer, a high percentage of exhausted CD8⁺ T cells and upregulated expression of immune checkpoints were identified in MSI colon cancer at the single-cell level, indicating the potential neutralizing effect of cytotoxic T-cell activity by exhausted T-cell status. A more homogeneous highly expressed pattern of PD1 was observed in CD8⁺ T cells from MSI colon cancer; however, a small subgroup of CD8⁺ T cells with high expression of checkpoint molecules was identified in MSS patients. A random forest algorithm predicted important mutations that were associated with MSI status in the TCGA colon cancer cohort, and our in-house cohort validated higher frequencies of BRAF, ARID1A, RNF43, and KM2B mutations in MSI colon cancer. A robust microsatellite status-related gene signature was built to predict the prognosis and differentiate between MSI and MSS tumors. A neural network using the expression profile of the microsatellite status-related gene signature was constructed. A receiver operating characteristic curve was used to evaluate the accuracy rate of neural network, reaching 100%.

Conclusion: Our analysis unraveled the difference in the molecular nature and genomic variance in MSI and MSS colon cancer. The microsatellite status-related gene signature is better at predicting the prognosis of patients with colon cancer and response to the combination of immune checkpoint inhibitor-based immunotherapy and anti-VEGF therapy.

Keywords: computational biology; immunotherapy; tumor biomarkers; tumor microenvironment.

Figure 1

Estimation of immune cell populations and immune signatures in TCGA-colon cancer. (A) Heatmap showing the estimation of immune cell populations in the TCGA-colon cancer cohort. (B) The ratio of immune cell populations and microsatellite status. The CYT activity (C), IFNG expression (D), TIS score (E), and CD8:Treg ratio (F) in colon cancer with MSS, MSI-L, and MSI-H. The CD274 (G), PDCD1 (H), and CTLA4 (I) expression and exhausted score (J) in colon cancer with MSS, MSI-L, and MSI-H. (K) UMAP clustering of tumor microenvironment cell populations with information regarding cell types from GSE146771. (L) UMAP clustering of tumor microenvironment cell populations with information regarding microsatellite status from GSE146771.

Figure 2

Molecular nature of colon cancer with MSS and MSI. (A) UMAP clustering of B cells, CD4⁺ T cells, CD8⁺ T cells, and malignant cells with information regarding cell types from GSE146771. (B) UMAP clustering of B cells, CD4⁺ T cells, CD8⁺ T cells, and malignant cells with information regarding microsatellite status and cell types from GSE146771. (C) The expression of stemness-related genes in each subgroup of cells. (D) The gene features in each subgroup of cells. (E) The expression of featured genes in CD8⁺ T cells from MSI colon cancer. (F) The expression of featured genes in CD8⁺ T cells from MSS colon cancer. (G) The expression of featured genes in CD4⁺ T cells from MSI colon cancer. (H) The expression of featured genes in CD4⁺ T cells from MSS colon cancer. (I) The ratio of exhausted CD8⁺ T cells and non-exhausted CD8⁺ T cells in colon cancer with MSS and MSI.

Figure 3

CCIs in colon cancer with MSS and MSI. The directions and number of ligand–receptor interactions in MSS (A) and MSI (B) tumors in GSE146771. The intracellular network of PD1 in CD8⁺ T cells in MSS (C) and MSI (D) tumors. (E) The expression level of MHC-I molecules and MHC-II molecules (HLA-DRA) in MSI and MSS colon cancer in the TCGA colon cancer cohort. (F) The TCGA colon cancer transcriptome data were analyzed using the curated Molecular Signatures Database.

Figure 4

Random forest algorithm on the TCGA colon cancer dataset identifying the most important genes associated with MSI status. The importance ranking of gene mutations and a heatmap showing the association of gene mutations and microsatellite status.

Figure 5

In-house validation of the random forest model. (A) BRAF, ARID1A, RNF43, and KM2B mutation frequency in TMB-low MSS colon cancer, TMB-high MSS colon cancer, and MSI colon cancer. (B) BRAF, ARID1A, RNF43, and KM2B mutation types in TMB-low MSS colon cancer, TMB-high MSS colon cancer, and MSI colon cancer.

Figure 6

Construction of a microsatellite status–related gene signature by TCGA colon cancer dataset. (A) The volcano plot shows the DEGs between MSS and MSI-H. (B) The volcano plot shows the DEGs between MSI-L and MSI-H. (C) The Venn plot shows the total genes and overlapping genes from the volcano plots. (D) Gene ontology analysis by the total genes from the Venn plot. (E) The coefficient calculated by LASSO Cox regression. (F) Kaplan-Meier analysis showed that patients with colon cancer with a higher MSRS exhibited an unfavorable CSS in the TCGA cohort. (G) tROC analysis indicated a higher prediction efficacy of the MSRS than other clinicopathological traits.

Figure 7

Microsatellite status–related gene signature serves as a predictor for clinical outcome and neural network-based machine learning model predicts microsatellite status in colon cancer. (A) Patients in the low-MSRS group had a higher ratio of complete remission. (B) The ratio of immune cell populations and MSRS. (C) The expression profile of costimulatory/coinhibitory immune checkpoints landscape in the TCGA cohort. (D) The association of MSRS and VEGF activities in MSS, MSI-L, and MSI-H tumors. (E) Schematic diagram of the neural network. (F) The loss value in the testing set decreased during the training process. (G) The confusion matrix in the testing cohort.