Immune signatures and targets in human tumors

Cancer is a global problem and causes many deaths every year. Treatment with immunotherapy has, in the past decades, helped decrease cancer mortality. High-throughput sequencing technologies have opened new research possibilities, and the advantages of using omics data in cancer research have improved immunotherapy strategies and, among others, contribute to novel findings regarding biomarker discovery and personalized cancer vaccines. Biomarkers can ideally be used to predict the possibility of a patient obtaining clinical benefit from immunotherapy; however, those currently known account poorly across all patients. Hence, the optimal biomarker has yet to be found. CD8+ T cells are key players in the immune system to combat cancer. They have the ability to specifically recognize personalized mutation-derived epitopes (neoepitopes) on the surface of cancer cells and directly kill cancerous cells. Multiple approaches exist to efficiently predict possible neoepitope candidates that are presented at the surface of cancer cells, but only a small fraction of these is truly immunogenic, meaning that they are, in fact, recognized by CD8+ T cells. This thesis studies potential biomarkers to predict the outcome of treatment with immunotherapy, either through the use of omics data and/or through the characterization and identification of immunogenic neoepitopes. Four Manuscripts are included to address this: Manuscript I investigates predictive biomarkers across a diverse cohort and states that a combination of biomarker is more suited to predict patient survival than observing single biomarkers. The suggested combinations include T-cell signatures and cancer cell signatures. The Manuscript propose a combination of either neoepitope load and programmed cell death ligand 2 (PD-L2) or neoepitope, PD-L2, and cytolytic activity (CYT) as a combined potential biomarker. Manuscript II and Manuscript III study a large pool of neoeptiope candiadtes and validate their immunogenicity potential. Both Manuscripts suggest that the abundance and frequency of neoantigen reactive CD88+ T cells (NARTs) can be used as a predictive biomarker in different setups. The tumor microenvironment (TME) also results in diversity when comparing patients with many detected NARTs to those with few. Furthermore, we observe differences in the characteristics of neoepitope candidates and describe how they can be used to distinguish immunogenic neoepitopes from non-immunogenic ones. The final Manuscript (Manuscript IV) investigates even more broadly the characteristics of immunogenic neoepitopes. A feature-based machine learning approach with random forest modeling using a dataset of above 19000 validated neoepitope candidates are used to predict the immuno-genicity of neoepitopes. From this analysis, the physicochemical properties of the neopeptides are found to be the most important features in predicting immunogenicity, and a high abundance of hydrophobic and aromatic residues are the most essential properties. Additionally, patient-specific TME features are implemented in the neoepitope prediction, resulting in a slight improvement.
Even though the statements in this thesis need further validation with more data, the four Manuscripts of this thesis add novel insight into the field of cancer immunotherapy through the discovery of potential biomarkers that can be used to predict survival probability from immunotherapy treatment and by improving the neoepitope prediction.