Update on Tumor Neoantigens and Their Utility: Why It’s Good To Be Different

Abstract

Anti-tumor rejection by the immune system is a complex process that is regulated by several factors. Among these factors are the quality and quantity of mutational events that occur in cancer cells. Perhaps one of the most important types of mutations that influence anti-tumor immunity is the neoantigen, i.e. a non-self-antigen that arises as a result of somatic mutation. Recent work has demonstrated that neoantigens hold significant promise for developing new diagnostic and therapeutic modalities. Therapeutic targeting of neoantigens is important for achieving benefit following therapy with immune checkpoint blockade agents or for cancer vaccines targeting mutations. Here, we review our understanding of neoantigens and discuss new developments in the quest to use them in cancer immunotherapy.

Mutagenesis as a driver of tumorigenicity and immunogenicity

A fundamental role of the immune system is to discern the difference between foreign and self entities. Cancer can be viewed as an insidious type of foreign entity originating from self; hence, it is teleologically attractive, in principle, to enlist the immune system to combat cancers by targeting elements within tumors that can be recognized as foreign. Neoantigens, which are peptide antigens derived from mutated genes and presented on major histocompatibility complexes (MHC), are perhaps the most widely recognized of these foreign elements in cancers. With the approval of several checkpoint inhibitor immunotherapies for multiple malignancies, such as anti-CTLA-4 and anti-PD-1, understanding the nature of anti-tumor rejection and how immunotherapies can augment this process has taken on new importance. Yet, despite the strides that have been achieved in this field, our mechanistic understanding remains far from complete.

Studying tumor antigen generation and recognition in cancers has been an area of intense focus in the quest to understand how immune systems interact with cancer cells. In isogenic mouse models, it was observed that mutagen-associated fibrosarcomas possessed immunogenicity that was not seen in spontaneous tumors and normal tissues [1]. However, immunogenicity and tumorigenicity appear to be parallel processes that occur as a consequence of mutagenesis. Due to different selective pressures, it was also observed that immunogenicity could be lost through serial transplantation, while these tumors maintained their malignant potential. These fundamental observations formed a basic mechanistic underpinning for cancer immunotherapy – that cancer cell growth and the host reaction to tumor growth are two related but distinct processes that could potentially be uncoupled. Work over the past several decades has built upon these observations.

Tumor-associated antigens (TAA) can be broadly classified into self and non-self- antigens. Self-antigens, which are genetically present in both tumor and normal cells, arise from non-mutated proteins that are aberrantly expressed or overexpressed in tumor cells. One class of self-antigens are the cancer-testis antigens (CTA), which includes MAGE-A, NY-ESO-1, SSX-2, and others [2–5]. Because trophoblasts and male gametes generally lack expression of class I MHC molecules, CTAs are thought to be selectively presented to the immune system by tumor cells, at least in part. A second class of self-antigens is the lineage-specific differentiation markers, which include Melan-A/MART-1, gp100, tyrosinase, etc. These antigens usually have tissue specific expression and have been found to exhibit some level of enriched expression in various cancers. Both CTAs and lineage-specific markers have been used for diagnostic and monitoring purposes. However, since they are wild-type (non-mutated) sequences, they can be subject to immune tolerance. Although CTAs and lineage-specific antigens are known to elicit CD8+ T cell responses in tumors, the basis for such T cell responses is not fully clear, as it may be expected that T cells recognizing self-antigens would have undergone negative selection in the thymus. Recent work has shown that negative selection and thymic depletion may be less complete than previously thought and that T cell clones recognizing self-antigens may be selected against but not totally eliminated [6, 7].

Over the last several decades, a large number of clinical trials have been conducted to test the efficacy of cancer vaccines targeting shared cancer antigens. These trials have used a wide array of peptide targets coupled with many different adjuvants. For example, Baumgaertner et al. vaccinated 19 melanoma patients in a phase I trial with long NY-ESO-1 specific peptides and CpG. They showed that vaccination was able to elicit CD8+ and CD4+ T cell responses [8]. Adams et al. showed that vaccination with NY-ESO-1 protein with the TLR7 agonist imiquimod was able to elicit humoral and cellular immune responses [9]. Many other similar trials have been conducted. While measureable immune reactions have been generated by these vaccines, these treatments have generally not been effective in reducing tumor burden or bringing about a meaningful regression of disease. On occasion, responses have been noted, but it is not clear what caused these rare responses. It should be noted that these trials were carried out in the absence of concurrent immune checkpoint blockade treatment and therefore, the number of tumor-specific T cells induced by the vaccine was low. The lack of clinical efficacy from vaccine treatment alone has been attributed to weak antigen delivery modalities that induced low T cell titers, as well as immune checkpoints remaining intact, which ultimately prevented tumor cell killing. A new generation of vaccines targeting self-antigens in combination with anti-PD1 or anti-CTLA4 is being currently tested. Results from these efforts are pending, but these studies highlight the promising impact of neoantigens in the immunotherapy armamentarium. Here, we review their role in checkpoint blockade therapies as well as the potential for targeted therapies based on neoantigen identification and selection.

Role of neoantigens in checkpoint blockade therapies

It has been long postulated that non-self-antigens generated by somatic alterations – neoantigens - could be differentially recognized by the immune system as these sequences would be unique to the tumor. Until the advent of advances in high throughput sequencing technologies, this hypothesis was difficult to test. However, multiple lines of investigation now point to neoantigens as an important class of cancer antigen that underlies immune rejection of tumors. In particular, recent studies have linked mutational load with clinical benefit in patients treated with immune checkpoint blockade therapy. The first study to show that mutational burden predicts response to immune checkpoint blockade was reported by our group in Snyder et al.[10]. In this study, the authors sequenced 64 patients with advanced melanoma and showed that somatic mutation burden was strongly associated with clinical response to anti-CTLA4. Similarly, in Rizvi et al., the authors demonstrated that mutation burden was a strong predictor of clinical response in non-small cell lung cancer patients treated with anti-PD1 therapy [11]. Thereafter, the link between mutation burden and clinical benefit following checkpoint blockade immunotherapy was validated multiple times and in multiple tumor types [12–17]. Mutational burden varies across different types of human cancers [18] and efficacy of immune checkpoint therapy is concentrated in the tumors with high mutational load [10, 19–21] (Fig. 1). Importantly, the utility of mutational load as a predictor of response has been validated in a phase III trial examining the efficacy of nivolumab as a first-line treatment for advanced non-small cell lung cancer patients. In this study, high tumor mutational load, but not PD-L1 levels, identified patients who went on to benefit from nivolumab treatment [22]. In all these studies, the recurrent theme that has emerged is that increased mutation burden from carcinogen exposure or inactivating mutations in DNA damage repair genes is associated with improved anti-tumor immunity, clinical outcomes, response rate, duration of clinical benefit, and survival. Increased mutational load leads to increased formation of neoantigens, which can be recognized and eliminated by the immune system [10, 11, 17, 23, 24]. Successful treatment of tumors with anti-PD1 results in immunoediting, depletion of neoantigens, and concurrent clonal expansion of T cells [17].

Figure 1. Tumor mutational load, neoantigen generation, and therapeutic efficacy.

Tumor mutational load is shown in various tumor types. Likelihood of neoantigen generation is shown along the right axis. Green boxes show some tumor types that are associated with significant response rates with immune checkpoint blockade therapy. Tumors that are generated by genotoxins (UV, smoking) are noted. Adopted from Lawrence et al. Nature (2013).

Cancer immunoediting, the process in which the immune system modifies cancer growth and immunogenicity, occurs in 3 phases - elimination, equilibrium, and escape. In mouse models, it has been demonstrated that mutagen-induced tumors can be highly antigenic. Whole exome sequencing in combination with in silico epitope prediction can be used to identify candidate neoantigens which may be targets of immunoediting [25]. Examination of the mutome and empiric testing of predicted peptides demonstrate that not all peptides are immunogenic, however. In the subgroup of immunogenic peptides, mutations induced a preferential immune response in comparison to wild-type counterparts [10, 26, 27].

Identifying which mutations are targeted by the immune system is currently a subject of intense interest. At the time of clinical presentation, tumors have already been subjected to significant immunoediting. In fact, the spectrum of oncogenic mutations that are found in individual tumors is heavily influenced by patient HLA genotype [28].

Indeed, poor MHC-I presentation of mutations correlates with higher frequency of those mutations among tumors. Additional neoantigens can remain in the tumor because the immune system, in the absence of treatment, is unable to mount an effective immune response against these antigens. Interestingly, inhibition of immune checkpoint blockade therapy with anti-CTLA4 and/or anti-PD1 can re-activate immunogenicity against these types of neoantigens in mice [29] and in humans [10, 11, 17]. These data support the concept that checkpoint inhibitor immunotherapy may help unmask less immunogenic neoantigens or enable exhausted T cells recognizing neoantigens to become reinvigorated and functionally active again. Since only a subset of mutated peptides are presented on MHC class I and only a fraction of these are immunogenic, predicting which mutations form effective neoantigens from sequencing data represents a formidable challenge [30]. Current computational approaches are being refined to improve neoantigen identification accuracy and are discussed in detail in a subsequent section.

Interestingly, looking across patients within high mutation load malignancies, few neoantigens are shared across patients - at least ones that are an exact match (ie. 9 out of 9 amino acids identical). In a cohort of 110 melanoma treated with ipilimumab, 77,803 neoantigens were predicted; only 28 (~0.04%) were found in more than one patient who experienced prolonged clinical benefit, but were absent in patients with no clinical benefit or long-term survival. The neoantigens that were shared in responders in this cohort did not seem to track with response in other independent cohorts such as the one described by Snyder et al. [10] or Hugo et al.[12]. In other studies, it is clear that neoantigens encoded by recurrent driver mutations such those in IDH1 and KRAS can be shared between patients [27, 31]. Therapeutic targeting of these shared neoantigens may represent an efficient treatment modality for few patients only since antigens are expressed at a relatively low frequency and the immunogenic peptides are restricted by particular HLA molecules. More data and larger study cohorts, as well as cross tumor comparisons, are necessary to determine what proportion of patients may potentially benefit from therapies targeting shared neoantigens.

In a 2015 study, our group first showed that mutations in genes that are essential for DNA damage repair associate with good clinical response to anti-PD1 therapy [11]. Included among these tumors are those with mismatch repair deficiency, which lie at the extreme end of the mutation load spectrum, harboring very high levels of mutations. In a phase 2 study of pembrolizumab (an anti-PD-1 antibody) in 41 metastatic carcinoma patients with or without mismatch deficiency, treatment resulted in improved progression-free and overall survival across multiple tumor types [13]. These results were validated in a cohort of 86 patients across 12 different tumor types with mismatch repair deficiency, with a 53% objective response rate and a 21% complete response rate. Mismatch repair deficient cancers had a mean number of 1783 somatic mutations as compared to 73 mutations in mismatch repair sufficient tumors. This corresponded to a mean of 578 and 21 candidate neoantigens, respectively. Similar results were obtained with nivolumab, another anti-PD1 drug [32]. These results have led to the FDA approval of immune checkpoint inhibitor therapy for the treatment of advanced mismatch repair deficient malignancies [24]. Importantly, this is the first FDA approval for a cancer drug that is pathway-related and applicable across cancer types.

Responses to immune checkpoint therapy have not been restricted to high mutation load malignancies. For example, renal cell carcinoma (RCC) has been considered a moderate to low mutation frequency malignancy with a 10-fold lower mutation rate than melanoma; however, it is clear that RCC is an immunotherapy responsive malignancy, with the historic use of cytokine therapy, high dose IL-2, and the more recent approval of anti-PD-1 monotherapy for treatment [33]. Given the clear clinical benefit of immune modulating agents but low mutation rate in RCC, it is likely that mutation load alone cannot explain responses to checkpoint inhibitor immunotherapy. In malignancies with high mutation rates such as melanoma, NSCLC, and bladder cancer, mutations are dominated by non-synonymous single nucleotide changes, which form neoantigens at low rates. Interestingly, in a pan-cancer analysis of 19 cancer types, renal cell carcinoma was associated with the highest level of insertion and deletion type (indel) mutations, both by proportion and by absolute number. Indels can frequently create a new open reading frame and a higher proportion of neoantigens can theoretically result [34]. However, this remains to be validated as indel mutations still remain poorly characterized in most patient cohorts and the accuracy of identification of these types of mutations are highly subject to variations in computational approaches.

Treatment with anti-PD-1 therapy could also be associated with increased neoantigen-specific T-cell expansion. Using deep sequencing of the T cell receptor (TCR) repertoire, analysis of lymphocytes in responders noted a rapid increase in a neoantigen-specific activated effector T cell clone, which was associated with tumor regression; however, as tumor regression plateaued, the clonal population returned to baseline [11]. Using methods to predict mutant peptides with significant binding affinity for patient-specific human lymphocyte antigen (HLA) class I alleles, positive clinical outcomes following immune checkpoint blockade therapy were correlated to the absolute number of neoantigens, but not the frequency of neoantigens per nonsynonymous mutation, suggesting that not all mutations carry equal weight and are equally subject to immunoediting. Using the T cell populations expanded by PD-1 therapy, responses to neoantigen peptides could also be experimentally validated. The neoantigens associated with T cell expansion post immune checkpoint therapy may represent a cohort of highly potent neoantigens, which may be useful for predicting response to immune checkpoint therapy [24].

Recent Developments and Challenges in Neoantigen Prediction

The current state of algorithm and tool development have been nicely summarized elsewhere [35, 36]. Here, we focus on approaches and considerations that have seen the most use in recent developments. A key set of tools in the immune-oncology field addresses computational prediction of which mutations give rise to neoantigens, primarily focusing on HLA class I antigen presentation, which is commonly found on solid tumor cells. Prediction is useful across a number of applications, including prioritization of candidate neoantigens for in vitro T cell analyses [37], characterization of neoantigens identified by comprehensive screens [38, 39], and selection of epitopes for personalized therapeutic vaccines [40–42]. And, as summarized above, neoantigen prediction is also relevant for understanding immune modulator (i.e. immune checkpoint inhibitor) responses [10, 11], as well as studies of cancer evolution under immune checkpoint blockade [17, 43, 44].

Current neoantigen prediction approaches have leveraged prior efforts in epitope prediction from autoimmunity and infectious diseases, focusing on in silico estimation of HLA binding affinity using neural network models trained on large datasets of in vitro binding of peptides to various HLA alleles [45]. This toolset is now relatively mature and widely used [46], and is expected to capture HLA/peptide binding adequately for directly characterized and relatively more common HLA allele [47, 48]. For instance, the latest tool in the popular NetMHC series, NetMHC-4.0 [49], has reported an AUC of ~0.9 for prediction of HLA/peptide binding across >80 HLA class I alleles and peptides of several possible lengths.

Recently, however, studies of antigen presentation in human tissue cell lines [50, 51] and murine cancer models [30] have been carried out by sequencing HLA-presented peptides by liquid chromatography-mass spectrometry (LC-MS/MS) directly [52]. Although limited by the sensitivity of mass-spectrometric methods, these studies have revealed the rather low positive predictive value of HLA binding-affinity based predictions for actual HLA peptide presentation, with <5% of predicted bound peptides found presented on the surface of cells. These observations were consistent with a deep mass-spectrometric analyses of primary human melanoma specimens [50, 51], where <1% of mutations were found as HLA-presented neoantigens. Additionally, the presented neoantigens were not top-ranking candidates by binding affinity for alleles where robust predictions could be made. Given that a similarly small fraction of mutations (~1–2%) have been shown to give rise to immunologically relevant neoantigens by exhaustive TIL-based assays [39], and the requirement of tumor antigen presentation for T cell recognition, it may be expected that current HLA binding-based methods will prove insufficient for neoantigen prediction for clinical use. Therefore, active efforts are now underway to improve the accuracy of prediction algorithms.

Attempts to improve antigen presentation prediction beyond HLA binding have been made through the incorporation of steps of the antigen processing pathway occurring before and after HLA/peptide binding, including proteasome processing and TAP transport[53], and the stability of the complexed peptide and HLA[54] (Figure 2). Although modest improvements in predictive performance have been achieved by early modeling of these additional elements, the datasets available for training may not be sufficient for significant progress; for instance, the NetMHCstab tool trained on stability measurements for ~5,500 peptides across 10 HLA alleles improved HLA ligand and T cell epitope prediction over binding affinity (NetMHCcons, trained on >150K data points) by an AUC increase of only ~0.1–1%. Further training data may improve upon these approaches.

Figure 2. Representative tools and workflow for neoantigen prediction.

Diagram shows commonly used tools for neoantigen prediction. Tool name and references are listed in the table below the diagram. TSNA, tumor-specific neoantigen.

Given these challenges, several investigators have recently sought to advance epitope prediction by training models using LC-MS/MS HLA peptide sequences directly, reasoning that these data may better reflect endogenous antigen processing and can now be generated in sufficient volume (~25K −95K peptides in recent studies) to capture the biological complexity accurate prediction demands. Improvements in MS technology (such as that utilized in the new Orbitrap mass spectrometers) and utilization of comprehensive genomic analyses have enabled the identification of key variables impacting HLA peptide presentation. These factors include gene expression at the mRNA level [55], sequence context upstream and downstream of presented peptides, and presentation hotspots within genes [50]. Initial neural network models trained on HLA peptide sequences from 16 engineered single HLA allele cell lines already offered a ~2X improvement in positive predictive value for HLA ligands over binding affinity prediction, demonstrating promise for the approach [56]. Future work will extend these lines of investigation to examine tumor-specific antigen presentation features and expand coverage to additional HLA alleles.

Overall, antigen prediction is poised to advance from its historical foundation in binding affinity estimation to the more comprehensive, integrative modeling of antigen processing offered by evolving HLA peptide sequencing and other tools. These will capture both canonical and emerging epitope classes such as post-translationally modified and spliced peptides [57, 58] and, where relevant, class II HLA [59]. For neoantigens, prediction should, thus, become more reliable through higher sensitivity and positive predictive value, and help bridge the gap between current practices where hundreds of candidate neoantigens are predicted for each tumor sample versus the small number of actual functional neoantigens found on comprehensive patient screens. More accurate antigen prediction will enable progress not only in the study of neoantigens, but cancer immunotherapy more broadly, perhaps accelerating target discovery in shared tumor antigens such as viral proteins and cancer testis genes (CTAs)[60]. For neoantigen identification, integrated pipelines will need to be developed beginning with tumor genomic characterization, variant analysis, and accurate prediction of which mutations are likely to give rise to tumor-specific neoantigens [61].

Recent work has used genetic and computational modeling approaches to identify common or shared features of candidate neoantigens that selectively occur in the genomes of patients that have benefited from immune checkpoint therapy. In a study of over 1500 patients, Chowell et al. found that patient germline diversity at the main HLA class I loci strongly associate with response and survival in patients that reap benefit from immune checkpoint immunotherapy [62]. Interestingly, patients with alleles in the B44 superfamily responded very well to immune checkpoint therapy, often experiencing complete responses and living disease-free of many years. HLA superfamilies are characterized by shared consensus sequences in the peptide ligands they bind and Chowell et al. showed that some B44 consensus binding sequences were often present in the predicted neoantigens of responding patients. Interestingly, shared features in candidate neoantigens were first observed in an exploratory analysis of the exomes of melanoma patients who responded to CTLA4 treatment. Three recent studies have expanded on this concept. Kim et al. used a machine learning approach to develop the Neopepsee algorithm [63]. They examined independent sets of melanoma, leukemia, and stomach cancer and modeled immunogenicity features (including sequence and amino acid immunogenicity information) from candidate neoantigens to construct a machine learning classifier. Strikingly, they also found that sequence homology to known pathogenic epitopes was a strong feature driving classification. In a separate study, Luksza et al. used a neoantigen fitness model to model neoantigens and predict response to immune checkpoint blockade. Here, the investigators developed a model using likelihood of peptide presentation by MHC class I, non-linear dependence of sequence similarity to known antigens and pathogen epitopes, and clonality of the neoantigenic alleles. The model could predict survival in three separate cohorts of patients treated with immune checkpoint blockade therapy. Third, in another study on long-term survivors of pancreatic cancer following surgical resection, the investigators found that the tumors of long-term survivors harbored neoantigens that were extremely similar to known epitopes from pathogens [64]. Using T cell assays and TCR sequencing, they showed that the same T cell clones recognized and targeted neoantigens and their homologous pathogen counterparts. These studies and others begin to uncover shared sequence and amino acid features that help determine the functional consequences of potentially neoantigenic peptides.

Considerations for the Development of New Cancer Vaccines

Perhaps one of the most exciting therapeutic uses of neoantigens are for constructing cancer vaccines. Therapeutic cancer vaccines have been shown to provide some level of therapeutic benefit as monotherapy in patients with premalignant disease or for the prevention of recurrence after treatment of the primary tumor [65, 66]. In patients with established tumors, however, cancer vaccine approaches have, to date, been less successful. This stands in contrast to the recent clinical successes observed with immune checkpoint blockade, focused on reactivating the patient’s existing immune responses, which has demonstrated promise for the treatment of patients with advanced malignancies [20, 21]. However, as monotherapy, immune checkpoint blockade currently provides long term clinical benefit to only a minority of patients [67].

The past failures of cancer vaccines can, in part, be attributed to suboptimal selection of tumor antigens, such as tumor-associated self-antigens that require breaking of immune tolerance to elicit T-cell responses [68]. In part, low response rates were also due to the selection of poorly immunogenic antigen delivery platforms as well as a lack of understanding of the highly immunosuppressive microenvironments that characterize tumors [69]. With the identification of neoantigens as the functional targets of immune checkpoint blockade in clinical responders, the advancement of a better understanding of vaccine technologies, and a growing toolbox of reagents to counter the immune suppressive tumor microenvironment, the future of therapeutic cancer vaccines now looks promising. To build highly efficacious cancer vaccine regimens for patients with established cancers, three primary elements in the design of vaccine regimens need to be addressed and improved: (i) the selected tumor antigens should be exclusively expressed and presented by the tumor cells and need to be foreign to the patient’s immune system, (ii) highly immunogenic vaccine delivery platform(s) are required to activate, boost and maintain high levels of functional CD4 and CD8 T-cell titers against the selected tumor-antigens, and (iii) immune checkpoint blockade needs to be employed effectively to activate potent and high titer tumor-specific T-cells and to counter the immune suppressive tumor microenvironment to keep the vaccine-induced T-cells active long-term. With our improved understanding of the requirements to achieve these three key elements, there now exists the potential to develop therapeutic cancer vaccine regimens that specifically and potently target tumors in the majority of patients with advanced stage disease.

Tumor-Specific Neoantigens as Ideal Cancer Vaccine Targets

Most therapeutic cancer vaccines evaluated in the past have targeted tumor-associated self-antigens, which have been shown to elicit T cells of low avidity [68]. The T cell repertoire is limited for self-antigens due to both central and peripheral tolerance mechanisms, namely the expression of these antigens in the thymus and the suppression by regulatory T cells, respectively[70, 71]. Therefore, from the perspective of the T cell repertoire, neoantigens are highly attractive for therapeutic cancer vaccine approaches [72]. As discussed above, neoantigens can arise from somatic mutations that result in the production of a novel peptide presented on the tumor cell surface by MHC-I molecules or from viral peptides in virally-induced cancers (EBV- or HPV-induced tumors). Evidence that neoantigens recognized by the immune system can lead to effective destruction of tumors comes from preclinical cancer vaccine studies and patients treated with immune checkpoint modulators [10, 11, 30]. Similarly, Zhou and Rosenberg adoptively transferred expanded autologous tumor-infiltrating T cells specific to mutated forms of GAS7 and GAPDH into a patient who achieved a complete response,[73] and recently, Stevanović et al. demonstrated that immunodominant T cell reactivities against mutated neoantigens can be found in patients in complete regression of human papillomavirus-associated metastatic cervical cancer after tumor-infiltrating adoptive T cell therapy [60]. Also, higher mutational and neoantigen loads predict for clinical benefit in patients treated with adoptive T cell transfer [74]. Together, these data suggest that a potent cancer vaccine that induces and boosts T cell responses against tumor-specific neoantigens is likely to increase the responder frequency of patients on immune checkpoint blockade.

Vaccine Technologies

Although signs of activity have been observed in patients with early stage disease treated with a variety of cancer vaccines [75, 76], the many failures of vaccines in late stage disease suggest that improved or more potent vaccine platforms are needed to drive higher CD4 and CD8 T cell titers against tumor-specific antigens (Fig. 3). The requirements for a potent cancer vaccine platform are multifold. First, the vaccine platform must drive high expression of tumor-specific antigens, especially if the antigens are not directly being expressed in antigen presenting cells (APCs), in which case T-cell priming relies on cross-priming. Second, the tumor antigens need to be expressed in a highly immunogenic environment that effectively leads to (i) recruitment of professional APCs to the site of tumor antigen expression, (ii) uptake of the antigens by APCs, and (iii) maturation, activation and trafficking of APCs to the vaccine draining lymph nodes where T-cell activation occurs [77]. Many of the cancer vaccine platforms that did not meet primary endpoints in late stage clinical trials such as peptide-based vaccines [78], whole cell vaccines [79] and protein vaccines [80] failed to induce potent CD8 T cell responses in patients post-vaccination, suggesting that not all of the above listed requirements for a potent cancer vaccine platform were met. While several approaches are currently being pursued, viral-based delivery platforms represent a promising solution that has demonstrated efficacy in the context of infectious disease [81, 82] (Box 1)

Figure 3. Cancer vaccines, checkpoint blockade, and immune activation.

Top panel shows the effects of combined vaccine and immune checkpoint blockade therapy on T cell activity. Bottom panel displays the critical locations and immune activities that need to be enhanced in order to elicit long-term anti-tumor immunity.

Text Box 1: Viral-Based Vaccines

Many effective infectious disease vaccines utilize viral-based vaccine platforms [83], and recombinant viral vectors are a powerful vaccine platform that possess many positive features of other vaccine modalities with minimal disadvantages. Viral vectors effectively enter target cells, express target antigens to high levels and have intrinsic adjuvant properties, as they express pathogen-associated molecular patterns (PAMPs) that induce strong innate immunity, thus, providing a “’danger environment” which attracts and activates APCs. Targeted gene deletions offer the opportunity to reduce/eliminate the ability of viral vectors to replicate, which increases safety without impacting potency. The major drawback in using viral vector-based vaccines is the pre-existing or de novo induced immune response against immunogenic components of the viral vector itself (eg, structural proteins) which limits the effectiveness of the prime vaccination or boost vaccinations (if de novo induced post initial vaccination). Strategies to overcome this limitation at the prime vaccination include the use of viral vectors for which humans do not have pre-existing immunity like simian or chimpanzee adenoviral vectors (AdV) [84, 85] and heterologous prime/boost regiments where an alternative vaccine platform is being utilized for the boost vaccinations which itself is not being recognized and neutralized by the immune system [82] Repeat vaccinations are likely to be essential to boost the tumor-specific T-cells, which otherwise would contract within a couple of weeks [86].

Text Box 2: Immune Modulators to Increase Vaccine Potency and Counter the Immune Suppressive Tumor Microenvironment

Although prophylactic vaccines against infectious disease targets have been shown to provide protection against infection in humans, the vaccine approaches in those cases might not be sufficient to develop an effective cancer vaccine for patients with advanced stage disease due to the many mechanisms utilized by the tumor to escape immune-mediated destruction. However, with the advent of drugs that can alter immune checkpoint function, our toolbox of reagents is now rich for countering the immune suppressive tumor microenvironment and to enhance vaccine potency by blocking immune inhibitory molecules induced by vaccination and expressed on recently activated T cells [87]. Evidence that the combination of cancer vaccines with immune checkpoint blockade strongly increases vaccine-induced immune responses and improves therapeutic benefit has been provided from preclinical studies and early clinical trials [88–91]. These combination approaches will be necessary for the effective treatment of patients with advanced stage malignancies where immune checkpoint monotherapy has shown to be less effective than in patients with earlier stage of disease. Without abrogation of immune checkpoints, it is likely that even neoantigen-based vaccines that strongly induce immunity would be rendered ineffective by upfront or feedback mechanisms that shut down T cell activity. A proof of principle study demonstrating this concept was recently published by Ribas et al.[92]. In this study, the combination of anti-PD1 therapy with an oncolytic virotherapy (talimogene laherparepvec) resulted in a much higher overall and complete response rate in patients with metastatic melanoma. While talimogene laherparepvec is not a personalized therapy and does not utilize neoantigens, it likely boosts cross-presentation, which then enhances the activity of cells that are re-invigorated by anti-PD1.

Unless tumors are eradicated rapidly by a potent vaccine, long-term immune control requires an effective vaccine boost strategy to keep the immune pressure on the tumor and control tumor progression. Thus, heterologous prime/boost approaches utilizing a potent viral vector based prime vaccination and multiple boost vaccination with an alternative vaccine platform is anticipated to activate and maintain high T-cell titers to the selected tumor antigens in humans and have the potential to provide therapeutic benefit to patients with advanced stage cancers. In addition to vaccination parameters, an effective immune response must overcome the suppressive tumor micro-environment. This approach has of course been pioneered by checkpoint blockade therapies but novel tools are likely to further potentiate immune efficacy (Text Box 2)

Concluding Remarks

In summary, neoantigens are a fascinating and useful class of tumor antigens that we are beginning to understand. With the identification of neoantigens as the functional targets of immune checkpoint blockade in responders, the better understanding of the requirement to activate and maintain high CD4 and CD8 T-cell responses against tumor antigens, the advancement of vaccine technologies, and the advent of a rich and growing toolbox of reagents that counter the immune suppressive tumor microenvironment and increase vaccine-induced immune responses, the future of therapeutic cancer vaccines looks bright. There are clearly important questions that need to be answered (see Outstanding Questions). With the understanding that therapeutic benefit provided by immune checkpoint blockade is currently limited to patients with pre-existing tumor-specific T-cell responses, combinatorial approaches of immune checkpoint blockade with potent cancer vaccines specific to tumor neoantigens are anticipated to increase responder frequency of patients treated with immune checkpoint blockade.

Outstanding Questions.

• How many neoantigens are required for elimination of any given tumor cell?

• What are the features of neoantigens that most strongly associate with the ability to be recognized and targeted by T cells?

• What immune checkpoint therapies or adjuvants best stimulate neoantigen immunity?

• How many neoantigens should be included in neoantigen-based vaccines to achieve optimal anti-tumor activity?

Highlights.

• Neoantigens are antigens generated by somatic mutations that can be recognized by the host immune system.

• Neoantigens in tumors can be targeted by T cells.

• Neoantigens can be identified and used in cancer therapies, such as cancer vaccines.