Differences in immunogenicity between cancer mutation signatures shed light on immunoediting

Abstract

Immunoediting is a process through which the immune system plays a role in shaping the mutational landscape of cancer and, consequently, in cancer progression. One critical aspect of immunoediting is the phenomenon known as neoantigen depletion. Neoantigens are mutated peptides that may arise from somatic mutations in cancer cells presented on the cell surface by the MHC molecules. Theoretically, these neoantigens can mark the cancer cells to be identified and consequently eliminated by immune cells, such as cytotoxic T-cells. Accurate neoantigen predictions allow researchers to identify which mutations generate immunogenic peptides to initiate an effective immune response against cancer cells. This has significant implications for the development of personalized immunotherapies and cancer vaccines. Cancer immunoediting not only occurs during tumor progression but also in pa tients receiving anticancer immunotherapies. Neoantigen depletion during cancer progression contributes to innate resistance to immunotherapies, resulting in incon sistent results across patients and cancer types. Patients also acquire resistance to immunotherapy during treatment, leading to treatment ineffectiveness. Therefore, to effectively harness the power of the immune system against cancer and to fully understand cancer progression, a thorough understanding of cancer immunoediting is crucial. This thesis aims to gain a deeper understanding of some of the sources of variation in immunogenicity, as well as potential mechanisms to escape from immune responses. Ultimately, we are to use these findings to enhance our understanding of the impact of immunoediting on the mutational landscape of human cancers. Two recent high profile studies have reported that recurrent driver mutations occur in the gaps in the capacity of MHC molecules to present neoantigens. This implies that the immune system selects against driver mutations that can potentially give rise to neoantigens. These findings have important implications in studying can cer progression and the role of immune system in determining how cancers develop. Interestingly, although depletion of driver mutations predicted to be immunogenic has been reported the same was not observed for passenger mutations. Therefore, in Chapter 2 we tested if the passenger mutations that are predicted to be immunogenic occur preferentially on lowly expressed or non-expressed genes which may help to explain this observation. When we controlled for gene length and sequence context, we found no evidence to support this hypothesis. Consequently, we re-evaluated the results reported by and found that these results are based on unjustified sta tistical assumptions. Our analysis found no link between MHC genotype and the occurrence of driver mutations. Consistent with this, we also found no relation ship between cancer risk in individuals from the UK Biobank and the coverage of common driver mutations predicted from their MHC genotypes. In Chapter 3, we performed an analysis to predict immunogenicity of somatic mutations that arise from different cancer mutation signatures. The study found that mutated peptides resulting from specific mutation signatures were more likely to be presented by certain HLA alleles compared to peptides originating from other mutation signatures. Notably, the median activity of the mutation signatures in a given cancer could be used to predict the average number of mutations inferred to be immunogenic with high accuracy (R2 = 0.87). Our results revealed that variations in the immunogenicity of mutations in tumors can be attributed to the differences in immunogenicity of mutation signatures and their activities. The limited variability in mutation signature immunogenicity and activity across different types of cancer resulted in small variation in the expected immunogenicity of various cancer types. Our findings also highlighted that the MHC-I genotype is the major determinant of the predicted immunogenicity of tumors. It was also discovered that mutation signature 20 yielded the highest proportion of immunogenic mutations, based on the HLA allele frequencies in the TCGA cohort. When comparing different types of cancer in the TCGA cohort, CESC had the highest expected number of immuno genic mutations, while PRAD had the highest observed proportion of immunogenic mutations. Recent studies have reported that patient MHC-I genotype plays a role in deter mining immunotherapy responses. However, the extent of this influence appears to be inconsistent, and the underlying reasons for this inconsistency remain unclear. For instance, in the case of melanoma, the B44 HLA supertype has been linked to a better response. Interestingly, non-small cell lung cancer (NSCLC) has a simi lar somatic mutation burden and immunotherapy response as melanoma, but the B44 supertype has not been found to have an impact on the immunotherapy response in NSCLC. This divergence has been attributed to underlying differences in mutational processes between melanoma and NSCLC. We performed mutation sig nature analysis for two ICB treated melanoma cohorts. The findings of this analysis revealed a significant enrichment of C > T mutations, which is consistent with pre vious studies. Furthermore, we used a combination of mutation signature activity and patient-specific HLA genotype to estimate the expected proportion of immuno genic mutations for these cohorts. A higher expected proportion of immunogenic mutations was associated with a tendency towards improved overall patient survival. To gain insights into the role of immune selection in shaping the somatic mu tation landscape and consequently the progression of cancer, we must consider the types of mutations occurring in a cancer, and the underlying mutational processes driving them. In Chapter 4, we developed a method that considers the mutational and evolutionary processes involved in tumor growth to identify and quantify the immunoediting signal. The MHC-I restricted immunoediting signal was weak and inconsistent across cancer types in the TCGA cohort. Moreover, the weak immu noediting signal persists even when we use the randomized HLA alleles. Finally, we estimated that fewer than 1% of mutations inferred to be immunogenic, were removed through immunosurveillance. In summary, firstly, we investigated the relationship between the occurrence of driver mutations in a tumor and the MHC genotype of the patient. Then, we assessed the predicted immunogenicity of mutations arising from different somatic mutation signatures. We also examined the variation in tumor immunogenicity based on the activity of mutation signatures. We used the predicted immunogenicity of samples in the TCGA cohort to evaluate the contribution of immunoediting to the mutational landscape in cancer. We also used this method to estimate an upperbound on immunoediting signal.