In recent years, we have made great strides in our understanding of the complex interaction between the immune system and cancer which have led to significant advances in cancer immunotherapies (CI). Despite the successful application of CI across a wide range of indications, only a minority of patients experience significant improvement in survival from these therapies. There are several obstacles that still exist for the field including:
- The inability of current biomarkers to accurately predict treatment efficacy and patient response and the need for additional biomarkers
- Need for identification of targetable antigens (neoantigens)
- Determining the dominant drivers of cancer immunity and understanding the molecular and cellular drivers of primary vs secondary immune escape
- Understanding the development of resistance to cancer immunotherapies
- More effective assessment of cancer immunotherapy combinations (1,2)
Below, we expand upon a few of these areas and discuss particular considerations for resolving the multifaceted issues facing the field of immuno-oncology.
One of the major impediments of widely implementing immunotherapies is the lack of cancer-specific, targetable antigens, also known as “neoantigens”. Targeting tumor-associated antigens which are expressed by both tumor and normal tissues may be an option, however, it can lead to off-target toxicities and low efficacy. Therefore, it is important to develop an approach that can both effectively target these neoantigens and enhance T cell reactivity. Thus, the challenge to building an efficacious neoantigen-based cancer vaccine is to not only accurately identify, but also select, and prioritize the putative immunogenic neoantigen candidates. This may be achieved partly through whole exome and transcriptome sequencing of the tumor to detect various source mutations including single nucleotide variants, insertions, deletions, and fusions. It also requires accurate human leukocyte antigen (HLA) typing from whole exome sequencing of the germline sample to derive the patient’s HLA genotype for neoantigen predictions. This is followed by complex computational methods to characterize the neoantigens and synthesize the top candidates for vaccine development. At Personalis®, we have been actively refining a machine learning model to improve neoantigen identification and predictions. The model is well aligned with the major findings from the recent TESLA Consortium study (3) with the integration of similar features into the design including the presentation features as well as foreignness metric. Our prediction algorithm, Systematic HLA Epitope Ranking Pan Algorithm (SHERPA™), is developed and trained on a proprietary machine learning model using high-quality immunopeptidomics data generated from genetically engineered monoallelic K562 cell lines for predicting putative patient-specific neoantigens.
Another challenge is to effectively determine the driver of cancer immunity. Initially tumors were identified as inflamed based on high expression of PD-L1 by tumor cells or tumor-infiltrating immune cells and/or the prevalence of tumor-infiltrating lymphocytes (TILs) (2,4).
While inflamed tumors tend to respond to checkpoint inhibitors (CPIs), more recent data has shown that additional biomarkers beyond PD-L1 play a very critical role in inflamed tumors (Figure 1). These are IFNγ signatures, B cells, and genomic instability as defined by MSI or high tumor mutational burden (TMB).
Figure 1: Tumor Immunity and response to checkpoint inhibitors. Image from Top 10 Challenges in Cancer Immunotherapy, Immunity Perspective, Priti S. Hegde and Daniel S. Chen, Immunity 52, January 14, 2020.
A study from Griss, et al. (5) has shown that tumor-associated B cells are vital in melanoma response and that the depletion of B cells decreases tumor-related inflammation and CD8+ T cell numbers. Moreover, tumors containing an abundance of plasmablast-like B cells are associated with an enhanced survival benefit to CPI therapies (6).
Studies have shown that tumors such as bladder, colon, and non-small cell lung cancer (NSCLC) with high TMB and MSI respond well to CPI. However, inflamed tumors can undergo loss of heterozygosity (LOH) in the HLA locus (5). Thus, there is a need to identify these complex biomarkers when determining cancer drivers as this can impair the ability of T cells to recognize neoantigens. To this regard, Personalis has developed a multi-dimensional approach modeling both tumor and immune-related molecular mechanisms that resulted in a composite neoantigen presentation score (NEOPS™) which takes into account antigen presentation impairments, such as HLA allele-specific LOH while calculating neoantigen burden (7). However, there is also still an existing need for highly accurate HLA LOH algorithms to allow clinical utility, but the field lacks robust, allele-specific validation approaches. Moreover, algorithms of unknown sensitivity and specificity have led to significant discrepancies in the estimated occurrence of HLA LOH as an immune escape mechanism across tumor types (8,9). To address this obstacle, Personalis has also developed a machine learning algorithm to detect HLA LOH (DASH™ – Deletion of Allele-Specific HLAs) and established the accuracy of the algorithm with an allele-specific PCR validation strategy. Further, investigation of the frequencies of HLA LOH across 23 tumor types in a cohort of approximately 800 patients (Figure 2) demonstrated the prevalence of HLA LOH in many cancer types using this optimized DASH algorithm (9).
Figure 2: HLA LOH prevalence across a large pan-cancer cohort profiled by ImmunoID NeXT.
Even though high TMB has shown some promise for being predictive of response to CPIs, studies have also shown that PD-L1 expression is a better biomarker of survival benefit compared to TMB in Phase III trials in NSCLC (10,11). Moreover, a number of cancers are characterized by the presence of inflammation without high TMB. These include renal cell carcinoma (RCC), hepatocellular carcinoma (HCC), triple-negative breast cancer, gastric cancer, and, to an extent, head and neck cancers. Conversely, diseases such as small cell lung cancer (SCLC) represent tumors characterized by high TMB in the absence of other markers associated with inflamed tumors. Thus, utilizing the tumor mutational landscape as a predictive biomarker is complex and seems to have very unique indication-specific considerations (Figure 3).
Figure 3: TMB and association with Immune Phenotype Image from Top 10 Challenges in Cancer Immunotherapy, Immunity Perspective, Priti S. Hegde and Daniel S. Chen, Immunity 52, January 14, 2020.