Immunogenomic pathways associated with cytotoxic lymphocyte infiltration and survival in colorectal cancer

The data shown here is based upon data originally generated and organized by FireCloud from the Broad Institute (Fig. 1). Full permission access transcriptomic data was obtained from dbGAP. We downloaded CRC patients’ open access, pre-processed mRNA expression data (level 3 data, [18]) from both platforms, Illumina-HiSeq and Illumina-GA; as well as mRNA RNA-Seq by Expectation Maximization (RSEM) normalized data; and patients’ clinical data from the cohorts TCGA_COAD_ControlledAccess and TCGA_READ_ControlledAccess by gsutil Tool.

We integrated pertinent clinical data (age, gender, microsatellite status, anatomical location, pathologic stage, Tumor Node and Metastasis (TNM) classification, days to last follow up, and vital status), and RNA-seq by participant ID. Each CRC patient has pre-identified microsatellite status labeled as “microsatellite instability test results”. Thirty-four patients with colon cancer and 18 patients with rectal cancer were excluded due to missing information including indeterminate microsatellite status, unclear anatomical location or pathologic stage, and unmatched RNA-seq data. Then for every patient, we implemented the R package Microenvironment Cell Populations-Counter on the RNA-Seq by Expectation Maximization (RSEM) normalized RNA-seq data to create cell type abundance scores [4]. There are 10 cell populations simultaneously quantified in the tumor microenvironment, including 8 immune cell populations (T cells, CD8 T cells, Cytotoxic lymphocytes, NK cells, B lineage, Monocytic lineage, Myeloid dendritic cells, Neutrophils), endothelial cells and fibroblasts. Specifically, the gene set for cytotoxic lymphocytes includes the genes: CD8A, EOMES, FGFBP2, GNLY, KLRC3, KLRC4, and KLRD1. We used the median value of the cytotoxic lymphocytes (CL) score to group the patients into high (≥ median value, CL-High) and low (< median value, CL-Low) groups. We then grouped patients by anatomical location and stratified by cytotoxic lymphocyte score and pathologic stage. For consistency in clinical intervention while increasing group size, we assigned pathologic stage I-II to “early” stage, stage I-III as “localized”, and stage IV as “metastatic” (Fig. 1). Patients were grouped by tumor subsite: tumors located in the cecum, ascending colon, hepatic flexure, and transverse colon were categorized as having right-sided colon cancer (Fig. 1, RSC); tumors located in the splenic flexure, descending colon, sigmoid colon or rectosigmoid junction were categorized as having left-sided colon cancers (Fig. 1, LSC); patients with tumors located in the rectum were kept in this group (Fig. 1, REC).

RNA-seq differential gene expression analysis was performed with the edgeR package using the raw data downloaded from the Illumina- HiSeq and Illumina-GA platforms [19]. Differentially expressed genes were defined as genes with an absolute fold change > 1 between patients with high and low cytotoxic lymphocyte scores with a p value < 0.05 [19]. Genes with Benjamini-Hochberg adjusted False Discovery Rate (FDR) < 0.05 were considered to be significantly differentially expressed for further steps. For each cohort, we identified 20,531 total genes by RNA-seq raw counts. The Reactome online browser was used to identify immune functional differentially expressed genes [20]. Figure 1 outlines this process.

Pathway enrichment analysis was performed to evaluate the pathways associated with differentially expressed genes. The genes included in the MCP-counter cytotoxic lymphocyte gene panel (CD8A, EOMES, FGFBP2, GNLY, KLRC3, KLRC4, and KLRD1) were excluded from pathway enrichment analysis. Dotplot was used to illustrate the comparison of enriched Reactome pathways among differentially expressed genes in each location and stage. These results were analyzed by clusterProfiler, DOSE, and ReactomePA R packages. Next, we performed survival analyses using the identified differentially expressed genes. Patients were organized by stage and location as outlined above. The normalized RNA-seq data of differentially expressed genes used for survival analysis was processed using the Survival R package [20]. For each differentially expressed gene, if the normalized gene expression value was more than the median level, we labeled it as “high,” and otherwise as “low.” The Kaplan–Meier survival curves generated were assessed by the Cox regression model for each immune functional differentially expressed gene using the Survminer R package [21, 22]. The survival curves of patients with high gene expression and low gene expression were compared by log-rank test. For each patient, overall survival (OS) was used as the endpoint, either the days from diagnosis to death, or to the last follow-up (Fig. 1).

The Reactome pathway online browser was used to identify differentially expressed genes associated with survival in sub-pathways of “immunoregulatory interactions between a lymphoid and a non-lymphoid cell pathway” (Fig. 1).

In the most recently updated TCGA dataset (June 01, 2016), there are 461 colon cancer (COAD) and 172 rectal cancer (READ) cases (Fig. 1, [17]). Thirty-four patients with colon cancer and 18 patients with rectal cancer were excluded due to missing information. The RNA-seq data with matched clinical data were integrated from COAD patients (Nc = 427) and READ patients (Nc = 154). Cytotoxic lymphocyte abundance scores were generated using the MCP-counter method [4]. Patients were then separated based on the median cytotoxic lymphocyte score (26.72; 95% CI: 24.1–30.1, Additional file 1: Table S1, values and analysis included in Additional file 2). Patients with cytotoxic lymphocyte scores ≥ the median were classified as cytotoxic lymphocyte-high (CL-High) and those with scores < the median were classified as cytotoxic lymphocyte-low (CL-Low). We then confirmed that there was a significant difference in cytotoxic lymphocyte scores between CL-High and CL-Low groups (73.66 ± 64.2 v 14.07 ± 6.69, p < 0.0001, Additional file 1: Figure S1). Colorectal cancer patients were then separated by anatomical location, cytotoxic lymphocyte score, and stage (Fig. 1). The demographic, clinical, and pathologic characteristics of each patient cohort is summarized in Table 1. Microsatellite status composition of patients with CL-High and CL-Low tumors was significantly different in the right-sided colon cancer (p < 0.0001) and rectal cancer (p = 0.0264) groups with more MSI-H patients among the CL-High patients at both locations (Table 1). Additionally, cytotoxic lymphocyte scores correlated significantly with pathologic tumor stage in right-sided colon cancer (p = 0.0278) and rectal cancer (p = 0.0279) patients, but not in left-sided colon cancer patients (Table 1). This analysis demonstrates that there are significant differences based on tumor location, suggesting that this variable is an important consideration when analyzing patient data [15, 23].

For each cohort, we next performed RNA-seq differential gene expression analysis. Expression of 20,531 genes was determined for each tumor sample from the TCGA. Then gene expression was compared between patients with CL-High and CL-Low tumors in each cohort based on tumor location and stage using the edgeR package (analysis included in Additional file 2). In right-sided early, localized, and metastatic colon cancer patients, 1882, 1781, and 1054 differentially expressed genes were observed, respectively. In left-sided patients, 805, 925, and 1255 genes were differentially expressed in each stage. And in rectal cancer patients, 888, 1316 and 150 genes were differentially expressed at each stage (Fig. 1). In the left-sided group, differentially expressed genes were highest in the metastatic cohort; however, in right-sided and rectal cancer patients, the metastatic cohort had the lowest number of differentially expressed genes. This again suggests that both tumor location and stage are important considerations when analyzing alterations in gene expression. Differentially expressed genes found in the above analysis were subsequently imported into the Reactome Pathway Browser to determine involvement in immune related functional pathways (Fig. 1, [15]). Interestingly, despite significant variation in the number of differentially expressed genes, the ratio of genes associated with immune function was similar in all sites and stages.

To further determine whether there were overlapping pathways associated with cytotoxic lymphocyte infiltration in colorectal cancer, we then compared the Reactome pathway enrichment analysis at each location based on stage. Using the p value adjusted for false discovery rate and the ratio of differentially expressed genes in each pathway, we found that the “immunoregulatory interactions between a lymphoid and a non-lymphoid cell” was the most highly enriched pathway in early and local patients at all tumor locations. Additionally, this was the most highly enriched pathway in patients with metastatic right-sided cancer. This pathway was also among the top pathways enriched among patients with metastatic left-sided colon cancer and rectal cancer (Fig. 2, data and analysis included in Additional file 2). This suggests that despite significant heterogeneity among subjects, redundant pathways of deregulation may be conserved across stage and location.

To further understand the potential for targetable genes within this pathway, we then took differentially expressed genes in the “immunoregulatory interactions between a lymphoid and non-lymphoid cell” pathway and performed a survival analysis using the Kaplan-Meier method and Cox-Proportional Hazards Model based on differentially expressed gene, location, cytotoxic lymphocyte score, and pathologic stage (Fig. 3, Table 2, remaining survival curves and analysis included in Additional file 2). There are a total of 297 genes included in this pathway, and we found 21 (7.1%) unique genes associated with survival in this pathway. As Fig. 1 demonstrates, the number of differentially expressed genes were variable with most genes associated with survival in the right-sided colon cancer group. Additionally, the positive and negative impact of differentially expressed genes on survival depended on cytotoxic lymphocyte abundance scores. The majority of genes associated with a positive impact on survival (bold in Table 2) were in the CL-Low group whereas the majority of genes with a negative impact on survival (italicized in Table 2) were found in the CL-High group. In CL-High right-sided colon cancer patients with metastatic disease, all differentially expressed genes had a negative impact on survival; however, all immune functional differentially expressed genes in the CL-Low group had a positive impact on survival. Often, patients with rectal cancer and left-sided cancer are considered to have similar disease biologically. While we found few differentially expressed genes in this pathway associated with survival in the left-sided and rectal cancer groups, differentially expressed genes in the left-sided colon cancer group primarily had a positive impact on survival with the converse being true in patients in the rectal cancer group. Together this data demonstrates that even within conserved immune pathways, there is significant heterogeneity in the impact on patient survival. This further suggests the importance of a patient centered approach for the application of immune based therapy in colorectal cancer.