Genomic Characterization of Six Virus-Associated Cancers Identifies Changes in the Tumor Immune Microenvironment and Altered Genetic Programs

Abstract

Viruses affect approximately 20% of all human cancers and induce expression of immunogenic viral oncoproteins that make these tumors potent targets for immune checkpoint inhibitors. In this study, we apply computational tools to The Cancer Genome Atlas (TCGA) and other genomic datasets to define how virus infection shapes the tumor immune microenvironment and genetic architecture of 6 virus-associated tumor types. Across cancers, the cellular composition of the microenvironment varied by viral status, with virus-positive tumors often exhibiting increased infiltration of cytolytic cell types compared with their virus-negative counterparts. Analyses of the infiltrating T-cell receptor repertoire in these patients revealed that Epstein-Barr virus infection was associated with decreased receptor diversity in multiple cancers, suggesting an antigen-driven clonal T-cell response. Tissue-specific gene-expression signatures capturing virus-associated transcriptomic changes successfully predicted virus status in independent datasets and were associated with both immune- and proliferation-related features that were predictive of patient prognosis. Together, the analyses presented suggest viruses have distinct effects in different tumors, with implications for immunotherapy.Significance: This study utilizes TCGA and other genomic datasets to further our understanding of how viruses affect the tumor immune response in different cancer types.Graphical Abstract: http://cancerres.aacrjournals.org/content/canres/78/22/6413/F1.large.jpg Cancer Res; 78(22); 6413-23. ©2018 AACR.

Figure 1:. Differences in immune infiltration levels between virus-infected and non-infected samples.

A, Boxplots depicting the distribution of immune cell infiltration scores across samples from six cancer types stratified by virus infection status. Dark colors indicate virus-infected samples and light colors indicate non-infected samples. Each box spans quartiles with the lines representing the median infiltration score for each group. Whiskers represent absolute range excluding outliers. All outliers were included in the plot. Significant associations are marked (* P < 0.05). B, Heatmap marking significant differences in immune infiltration scores between samples infected with noted viruses and non-infected samples. All viruses infecting more than one patient in the denoted tumor type are shown. Red color indicates significant increases in infected samples (P < 0.05), green indicates significant decreases (P < 0.05) and grey indicates no significant difference (P > 0.05). All p-values were calculated using the Wilcoxon sum-rank test.

Figure 2:. CD8+ T cell infiltration after adjusting for microsatellite instability and virus infection status.

Boxplots depicting the distribution of CD8+ T cell infiltration scores across COADREAD and STES samples stratified by MSI status and virus infection status. Dark colors indicate virus-infected samples and light colors indicate non-infected samples. Each box spans quartiles with the lines representing the median CD8+ T cell infiltration score for each group. Whiskers represent absolute range excluding outliers. All outliers were included in the plot. P-values were calculated using the Wilcoxon sum-rank test. Significant associations are marked (^ P ≤ 0.10, *** P < 0.01).

Figure 3:. T cell receptor repertoire diversity between samples infected with different viruses.

A, Boxplots depicting the distribution of unique CDR3 calls (clonotypes) per 1,000 TCR reads in samples from six tumor types infected with different viruses. Each box spans quartiles with the lines representing the median clonal diversity for each group. Whiskers represent absolute range excluding outliers. All outliers were included in the plot. P-values were calculated using the Wilcoxon sum-rank test. B, Meta-z-score absolute values indicating associations between infection of a given virus and reduced TCR clonal diversity across 6 tumor types. Viruses were ranked by unweighted meta-z-score. Green bars indicate an unweighted meta-z-score < −1.96 (significantly lower TCR clonal diversity, two-tailed p-value < 0.05) while grey bars indicate an unweighted meta-z-score whose absolute value is < 1.96.

Figure 4:. Performance and characterization of the virus infection gene expression signature.

A, ROC curves illustrating the accuracy of using the virus gene expression signature to classify infected samples from non-infected samples. Plots on the left depict the signature’s performance in training data while plots on the right depict the signature’s performance in test datasets. Gray lines and numbers in parentheses in the left plots indicate the 95% bootstrap confidence intervals of the ROC curves and their respective AUCs. From top to bottom, test datasets were obtained from GEO under accession numbers GSE40774, GSE49288, and GSE62232. B, Heatmap of AUCs for signatures trained in one tissue type (columns) and applied to another (rows). To show contrast, all AUCs < 0.5 were trimmed to 0.5. C, Boxplots depicting the weight of the gene MKI67 (black diamond), CD274 (large black circle) and the weight distribution of genes comprising the ESTIMATE immune gene expression signature (boxes). Dotted lines at 0.13 indicate threshold at which weights correspond to P < 0.05. Each box spans quartiles with the lines representing the median signature weight in each group. Whiskers represent absolute range excluding outliers. All outliers were included in the plot.

Figure 5:. Association between the virus infection gene expression signature and survival of head and neck cancer and bladder patients.

A, Kaplan-Meier plot depicting the survival probability over five years for samples with high (red) and low (blue) virus infection signature scores in the Thurlow et al head and neck cancer dataset. B, Boxplots depicting the difference in MKI67 expression (left) and ESTIMATE immune score (right) between signature low and signature high samples in the Thurlow et al dataset. C, Kaplan-Meier plot depicting the survival probability over five years for samples with high (red) and low (blue) virus infection signature scores in the Kim et al bladder cancer dataset. D, Boxplots depicting the difference in MKI67 expression (left) and ESTIMATE immune score (right) between signature low and signature high samples in the Kim et al dataset. For all Kaplan-Meier plots, samples were stratified into high and low groups using the median virus infection score. P-values were calculated using the log-rank test and indicate difference between the survival distributions of the full dataset. Vertical hash marks indicate censored data. In all boxplots, boxes span quartiles with the lines representing the median expression or score for each group. Whiskers represent absolute range excluding outliers. All outliers were included in the plot. P-values were calculated using the Wilcoxon sum-rank test.