The analysis of immune cell infiltration in tumor microenvironment of colon cancer
Totally 1576 cases with transcriptome information and clinical data were downloaded from TCGA and GEO database in March 2021. Patient characteristics were shown in table S1. Immune cell infiltration was quantified by CIBERSORT and ESTIMATE algorithms and clustering was performed to classify the patients into different clusters(9, 10).
According to the immune cell infiltration, patients were dived into three ICI clusters: ICI cluster A, ICI cluster B and ICI cluster C. Survival between the three ICI clusters was significantly different. ICI cluster A showed the best survival while ICI cluster B showed the worse survival (p=0.006) (figure 1A). We made a further comparison of the immune cell contents in tumor microenvironment of the three different ICI clusters. It was shown that high infiltration of CD8 T cell, follicular T helper cells, activated memory CD4 T cells, activated NK cells and M1 macrophages in ICI cluster A. ICI cluster B showed a lower infiltration of CD8 T cell, follicular T helper cells, activated memory CD4 T cells, activated NK cells, M1 macrophages, but had a higher level of resting NK cells, M0 macrophages and activated mast cells (figure 1B and 1C). We also made a correlation coefficient heatmap to visualized the immune cell interaction in tumor microenvironment (figure1D). Expression of two important immune checkpoint molecules, PD-L1 and CTLA4, were examined. Both PD-L1 and CTLA4 expression were higher in ICI cluster A than ICI cluster C. But expression of PD-L1 did not show difference between ICI cluster A and ICI cluster B. Expression of CTLA4 did not show difference between ICI cluster B and ICI cluster C (figure 1E and 1F).
Immune gene cluster analysis
To reveal the gene expression characteristics in different ICI clusters, we used the limma packages of R software to analyze the transcriptome variations (table S2). Then an unsupervised clustering of the differentially expressed genes was performed. Patients were classified into three gene clusters: gene cluster A, gene cluster B and gene cluster C. Genes that were positively related to the gene cluster were assigned as ICI signature A, and the rest differentially expressed genes were assigned as ICI signature B(12) (table S3). A heatmap was used to visualized the transcriptomic profile difference between the three gene cluster(14) (figure 2A). Gene ontology analysis was performed to clarify the enriched biological processes in ICI signature A and signature B (figure 2B and 2C). Expression of PD-L1 and CTLA4 in the three gene clusters were examined. Both PD-L1 and CTLA4 expression were lower in gene cluster B than ICI cluster C. But no significant difference was observed between gene cluster B and cluster A (figure 2D and 2E). To explore the prognostic value of gene clusters, we performed the survival analysis. It was shown that patients in gene cluster B had better survival than patients in the other two gene clusters (p=0.023) (figure 2F). Gene cluster C had higher infiltration of CD8 T cells, M1 macrophages, dendritic cells, NK cells than gene cluster B. And immune score and stromal score in cluster C were also higher than cluster B. However, gene cluster B showed a higher level of B cells, plasma cells and lower level of M1 macrophages (figure 2G).
Construction of immune cell infiltration score model
ICI scores of each patient were calculated by principle-component analysis (PCA). The optimal cutoff value was found out using the cut-off package of R software. Patients from TCGA database were assigned to high ICI score group or low ICI score group according to their ICI scores (figure 3A). We compared the ICI scores of different subgroup patients. It was indicated that ICI score was higher in patients who had no lymph node invasion and no metastasis. Patients with T1-2 and stage I-II disease also showed higher ICI score (figure 3B).
We evaluated the expression level of immune-checkpoint and immune-activity associated genes. PDCD1, HAVCR2, CTLA4, CD274, LAG3 were selected as immune-checkpoint associated genes and CXCL9, CXCL10, GZMB, GZMA, TNF, CD8A, TBX2, PRF1, IFNG were selected as immune-activity associated genes(15-17). We found that most of the immune-checkpoint genes and immune-activity associated genes, except PDCD1, were significantly decreased in high ICI score group (figure 3C and 3D.). Gene set enrichment analysis (GSEA) was performed and it was revealed that calcium signaling pathway, leukocyte transendothelial migration pathway, MAPK signaling pathway, TGF β pathway, and WNT signaling pathway were enriched in high ICI score group while base excision repair pathway, cell cycle pathway, homologous recombination pathway, mismatch repair pathway and nucleotide excision repair pathway were enriched in low ICI score group (figure 3E and 3F). To evaluate the prognostic value of the ICI scores, Kaplan-Meier analysis was performed in the TCGA cohort. Patients in high ICI score group showed better survival than patients in the low ICI group (p=0.017) (figure 3G). We did a further validation of the prognostic value of the ICI score in the alll patients form both TCGA and GEO cohort. It was revealed that high ICI group also showed a better OS than low ICI group in total patient cohort (p=0.002) (figure 3H).
The relationship between ICI score and TMB
It has been indicated that tumor mutation burden (TMB) is associated with patients’ response to immunotherapy(18, 19). Patients with high TMB showed an improved response to PD-1 inhibitors and other antitumor therapy. High TMB also predicted better prognosis in colorectal cancer patients (20, 21). Thus, an exploration was performed to analyze the correlation between ICI score and TMB. We compared the TMB level of patients in high and low ICI score groups. It was shown that there was no significant difference on TMB between the two group patients (figure 4A). Correlation analysis revealed that ICI score was not correlated with TMB (figure 4B). Survival analysis demonstrated that patients with high TMB had better OS than patients with low TMB (p=0.040) (figure 4C). To found out whether the combination of TMB and ICI score can better predict the prognosis of colon cancer patients, a stratified survival analysis was performed. Survival differences was observed between high ICI score and low ICI score patients both in high TMB and low TMB subgroups (figure 4D). The above results suggested that ICI score is independent of TMB and might serve as a potential prognostic biomarker in colon cancer patients.
Distribution of somatic mutation in high and low ICI score group
The distribution of somatic mutation of driver genes in colon cancer was evaluated in both high and low ICI score groups. The driver genes were evaluated using the maftools package of R software(13). We further analyzed the top twenty genes with the highest mutation frequency and result was shown in figure 4E and 4F. There were totally 133 genes with significantly different mutation frequencies between high and low ICI score groups (Table S4). The results might provide information for further studying the mechanism of immune cell infiltration and gene mutation in immune therapy.