In 2014, at Washington University School of Medicine in St. Louis, six melanoma patients received infusions of an anticancer vaccine composed of their own dendritic cells. Our WashU colleagues had extracted immune cells from the patients’ blood two months earlier, cultured them in the lab, and mixed in peptides selected and synthesized based on specific mutations present in the genomes of each patient’s tumor. The cells had then taken up the peptides much as they take up foreign antigens in the body in the course of normal immune patrol. When the clinical team administered the vaccines—each patient received three infusions over several months—they hoped that the dendritic cells would induce activation and expansion of T cells capable of identifying and destroying the cancer cells, while sparing healthy tissue.
This first test of personalized cancer vaccines in people grew out of our collaborative efforts to develop a computational pipeline to identify tumor-unique mutations that could induce immune responses in cancer patients, helping them to fight their diseases. The pipeline’s origin can be traced to the ideas of Bob Schreiber, a cancer immunologist also at WashU. For many years, Schreiber had studied mice that developed sarcomas after exposure to a chemical carcinogen as a model system for characterizing the interactions between cancers and the immune system. In 2011, he approached us about the possibility of sequencing the DNA of these cancer cells to identify unique cancer peptides, or neoantigens, with the potential to stimulate the immune system against cancer. In contrast to cell-based immune therapies, which directly provide the patient with tumor-attacking T cells, the idea was that these neoantigens could be used to create vaccines that stimulate the differentiation of endogenous killer T cells.
This activity in the cancer neoantigen vaccine space is indicative of the excitement around the general concept of personalized cancer vaccines.
At that time, most cancer genomics efforts had focused on finding mutations in cancer genes for which there was a targeted small molecule treatment option. Schreiber’s thought to search for neoantigens seemed to us like an interesting new twist that could lead to a novel type of therapeutic intervention, so we began to build the analytical pipeline he had in mind. Using his mouse model, we identified and validated a neoantigen peptide that was present in the sarcomas of animals that lacked a fully functioning immune system, and that became the target of T cell–mediated elimination when introduced into an immune-capable mouse.1 In a second study, we demonstrated that neoantigens were the targets of T cells activated by checkpoint blockade therapy, and that the introduction of synthesized neoantigens as a vaccine was sufficient to eliminate the sarcoma from mice.2 By demonstrating the efficacy of the neoantigen vaccine in the mouse model, we established the potential for using this approach in cancer patients with demonstrable neoantigens.
As we worked with Bob Schreiber’s lab, we were approached by WashU colleagues Gerald Linette and Beatriz Carreno, a husband-and-wife team now at the University of Pennsylvania’s Perelman School of Medicine who studied melanoma, a type of skin cancer. They had previously developed and tested a dendritic cell vaccine using antigenic peptides from a protein, gp100, that was commonly mutated in melanoma patients. Adding patient-specific neoantigen peptides identified by our pipeline, they wanted to conduct a clinical trial of an enhanced vaccine.
The resulting pipeline, called pVACSeq, evaluates mutated peptide sequences identified from an individual patient’s cancer for their potential to elicit an immune response.3 This analysis includes, for example, the peptides’ ability to be bound by the patient’s particular major histocompatibility complex (MHC) proteins, which are used by both cancer cells and dendritic cells to present the neoantigens on their surface where they can be “seen” by the immune system. We used the pVACSeq pipeline to generate a list of 7 to 10 neoantigens for each of the six patients recruited into the clinical trial. From the blood samples obtained from the patients, our colleagues isolated dendritic cells and created each personalized vaccine by inducing them to take up the synthetic peptides for each set of neoantigens.
After completing the infusions, we evaluated whether the patients experienced an increase in T cells with receptors that bound to the peptides included in the vaccine. We also looked at the diversity of those receptors. Higher receptor diversity is an indicator of the robustness of the T cell populations’ ability to recognize presented neoantigens and elicit cancer cell killing.
Overall, each of the three patients we reported on had seven unique neoantigen peptides added to their vaccine, and of these, three neoantigens had elicited a T cell response, with increased diversity of T cell receptors for these neoantigens compared with pre-vaccine levels, in each melanoma patient.4 These results indicated that it was possible to prime the immune system to recognize nonself, cancer-specific peptides and to elicit an antitumor T cell response that targets cancer cells. In these patients, as well as another three patients whose datasets we completed after publication, we saw no evidence of severe adverse events following vaccine administration, highlighting the cancer cell specificity of the approach. However, our trial results also exposed the need for refining our neoantigen prediction pipeline, as not all of the peptides included in the vaccines elicited a T cell response.
Since the publication of our personalized melanoma vaccine trial, three subsequent studies have demonstrated the potential of neoantigen-based vaccines in treating human cancers. Both academic and industry sponsors are now conducting many more such clinical trials. (See table above.) This activity in the cancer neoantigen vaccine space is indicative of the excitement around the general concept of personalized cancer vaccines, which have thus far demonstrated highly specific immune responses against cancer cells without severe adverse events in patients. This enthusiasm exists in spite of several remaining challenges that must be addressed in clinical trials before taking the concept forward into mainstream cancer care.
CANCER VACCINE BASICS
To create an individualized cancer vaccine, researchers must identify cancer-specific peptides called neoantigens, then use a cell-, protein-, or nucleic acid–based platform to deliver those neoantigens to patients to prime the immune system to attack the tumor. Antigen-presenting cells such as dendritic cells (purple) internalize the cancer-specific peptides (bright green) selected for a personalized cancer vaccine and display them on their surface with the help of major histocompatibility complex (MHC) proteins. This triggers T cells (blue) with receptors that bind those neoantigens to differentiate into effector, or killer, T cells (green) that mobilize an immune reaction against cancer cells (orange).
In designing a vaccine that initiates this process, researchers have several options:
1. DENDRITIC CELL VACCINE
Monocytes are extracted from the blood of patients and cultured with synthetic versions of selected cancer neoantigens to form mature dendritic cells carrying those neoantigens. These cells are then reinfused into the patient’s circulation.
2. LONG PEPTIDE VACCINES
Synthetic peptides containing the neoantigen sequences are injected into the body, where they are picked up by antigen presenting cells.
3. DNA & RNA Vaccines
Nucleic acids encoding the neoantigens are introduced into the body, where they are translated into proteins and picked up by antigen presenting cells.
In the 1980s, a handful of cancer immunology laboratories began exploring the idea that proteins encoded by genes that were mutated in tumors could serve as immune targets on cancer cells. The central hypothesis was that such neoantigens, upon binding by MHC proteins, would be presented on the outer membrane surface of cancer cells where they could be detected by T cells with matching receptors.
Over the past few decades, significant experimental efforts in basic cancer immunology laboratories have laid bare the foundational details of cancer cell–immune system interactions, ultimately validating the neoantigen concept. Researchers including Bob Schreiber have also defined the mechanisms by which cancer cells suppress the immune response to escape detection, explaining why the presence of neoantigens doesn’t always lead to tumor cell destruction: immune suppression sometimes prevents these peptides from being identified and attacked by T cells. These hypotheses have been reinforced by the now-demonstrated T cell–mediated cancer cell killing activated by several checkpoint blockade inhibitor therapies, which are proving to be a powerful weapon in the fight against cancer on their own and in combination with other cancer treatments, including personalized cancer vaccines.
In 2008, just as efforts to identify mutations in cancer genes using polymerase chain reaction (PCR) and Sanger sequencing were escalating, Bert Vogelstein of Johns Hopkins Medicine and James Allison, then at Memorial Sloan Kettering Cancer Center, proposed using genome-wide mutation discovery to identify cancer neoantigens.5, Among a set of 11 breast and 11 colorectal cancers, they predicted neoantigens by identifying missense mutations in cancer-related genes from Sanger sequencing data and using available neural networks to evaluate the binding of the encoded proteins to the most common MHC type. Given their selective approach with a relatively small number of samples, they predicted that these cancers average between 7 (breast) and 10 (colorectal) novel epitopes per patient. Importantly, they suggested the immune system could, by a variety of approaches including personalized vaccines, be stimulated to respond to these novel peptides.
Since that time, next-generation sequencing and advances in computational platforms to support analyses of the data—for example, comparing tumor sequences with those of matched healthy tissue from the same patient, and using the healthy tissue sequence data instead of an expensive clinical test to determine MHC haplotypes—have sparked an explosion in neoantigen research. Coincident with advances in the analysis of next-generation sequencing data, neural network–based approaches have expanded their ability to predict the binding characteristics of novel, tumor-unique peptides to additional MHC haplotypes based on annotated data from the binding kinetics of known peptides. These and other computational advances have brought neoantigen prediction to its current level of sophistication—one that is fundamental to the clinical development of the personalized cancer vaccines.
Before building an individualized cancer vaccine, researchers must determine which neoantigens will be included to elicit an immune response against the tumor cells. In collaboration with scientists around the world, our group has developed a computational pipeline for selecting those cancer-specific peptides that are most likely to drive a robust immune response against the tumor.
DNA sequencing, alignment, and variant calling
Next-generation sequencing data from tumor and normal DNA are aligned and compared to the human reference genome and then to each other to identify tumor-specific alterations. These variants are then evaluated for their resultant changes to the amino acid sequences of the encoded proteins.
The selected sequences are evaluated by computer models that predict the binding of the neoantigens to the major histocompatibility complex (MHC) proteins that would present them on the surface of cells.
Candidate Neoantigens Filter #1
RNAseq data from tumor RNA are evaluated to ensure the predicted alterations are being made into RNA transcripts, and we can further cull the list for other reasons, such as lack of sequence coverage of that region or gene.
Candidate Neoantigens Filter #2
Researchers isolate MHC proteins from patients and evaluate their bound peptides by mass spectrometry to validate that these peptides are being presented by MHC. In aggregate, such data can be used to improve the computer models that predict neoantigens.
Current research and clinical trials
Which of the more than two dozen personalized cancer vaccine approaches currently in early-stage trials will ultimately succeed depends not only on determining the numbers and types of neoantigens needed to best improve patient outcomes, but on which vaccine platforms are the most scalable, cost-effective, and quick to produce. Dendritic-cell vaccines are highly effective in eliciting an immune response because the role of dendritic cells in the immune system is to present antigens.
However, their preparation requires careful culturing of blood-derived immune cells to mature the dendritic cells prior to incubation with selected tumor peptides. Scaling this process would require significant automation of the laboratory-based steps in order to produce vaccines for multiple patients simultaneously.
Another approach is to use a cocktail of synthetic long peptides that are injected intramuscularly. The peptides are typically about 20 amino acids in length, and each contains a nested 8– to 11–amino acid neoantigen sequence. Two safety trials using this approach have been published, one in melanoma patients6 and one in glioblastoma patients.7 Because they are synthesized, the production of long peptides is scalable, but their use as human therapeutics requires specific process controls to ensure good manufacturing practices (GMP). And the costs to produce them are high because the longer the peptide, the more incomplete sequences occur, limiting yield. Also, each patient’s neoantigen peptides must be synthesized in isolation to ensure purity, and certain peptide sequences can be challenging to put into solution prior to injection. All these aspects may limit their widespread use in personalized vaccines in the future.
Computational advances have brought neoantigen prediction to its current level of sophistication—one that is supporting the clinical development of the personalized cancer vaccines that depend on such predictions.
Another way to elicit a neoantigen-evoked immune response is with DNA or RNA vaccines that encode predicted neoantigens. These vaccines have proven efficient in preclinical models and early-stage human testing of RNA vaccines, and they are relatively inexpensive to produce because their scalability of synthesis and GMP/quality control aspects are more straightforward than those of long peptides or dendritic-cell vaccines. The final sequences of the DNA- or RNA-encoded neoantigens are easier to confirm following vaccine synthesis than are peptide sequences, requiring basic sequencing technology instead of less-sensitive mass spectrometry. However, getting the nucleic acids containing the neoantigen-coding sequences into cells to permit their conversion to peptides is challenging.8 (See infographic above.)
For now, all of these vaccine platforms are on the table as researchers continue to study the immune system’s response to specific neoantigens. Questions include the minimum number and types of neoantigens required to elicit a strong cancer-targeted immune response and whether all neoantigens are encoded by all cancer cells (a clonal mutation), or by only a fraction of them (a subclonal mutation).
Meanwhile, we and others have been working to refine the computational tools for neoantigen prediction. The original approaches used the most common and easily detected variants—point mutations that change amino acid sequences—to predict neoantigens that bind to class I MHC molecules and elicit CD8+ T cell responses.
However, choosing peptide neoantigens that elicit not only MHC class I but also class II immunity involving CD4+ T cell responses is thought to be beneficial due to the cooperativity of the two T cell types in killing tumor cells. Furthermore, rarer mutation types that dramatically alter protein sequences, such as those resulting from frameshift insertion or deletion events or from fusion peptides due to structural alterations of chromosomes, will likely yield more-potent neoantigens with strong MHC binding potential because they are more obviously nonself to the immune system. Last December, we introduced new pipeline capabilities to address these types of predictions.9
We also recently showed that patient-specific corrections, the amino acid sequences encoded by single nucleotide polymorphisms (SNPs) unique to the patient’s genome, need to be incorporated into the peptide sequences around each somatic mutation. These corrections can yield significant differences in the predicted MHC binding scores for neoantigens, and including this step will ensure that these predictions are accurate on a per-patient basis.10
Even with improved methods, predicted neoantigens may not be produced by proteolysis or be presented by MHC at all, meaning that including these sequences in a vaccine design will not harm, but will not help drive an anticancer immune response. This realization has led researchers to isolate MHC proteins and their associated peptide neoantigens from cancers, separate the two, and identify them by mass spectrometry. Recently, several large-scale efforts to collect such information from patients has resulted in databases that can be used to train neural networks or other machine learning algorithms to refine neoantigen predictions to those most likely to be processed and presented by MHC.11,12
Jim Heath’s group at Caltech designed a unique microfluidic device and novel screening system to test predicted neoantigens for their ability to activate human T cells isolated from the cancer patient to be vaccinated.13 This highly sensitive approach can be used to detect even rare T cells from the patient that are specific for certain neoantigens. This information can be used to produce personalized vaccines or to engineer patient-derived T cells to express the appropriate receptor(s) to fight that patient’s cancer.
Personalized cancer vaccine trials in the United States
There are currently more than two dozen ongoing Phase 1 and Phase 2 trials using different vaccine platforms such as DNA, RNA, synthetic long peptides, and dendritic cells. Here are a few examples; see this story online for a comprehensive list.
See the full table.
An uncertain future
The promising results of early preclinical and clinical work on neoantigen vaccines have led to a substantial number of clinical studies of personalized neoantigen vaccine–based immunotherapy. How and whether these efforts will pan out depends largely on methodological refinements that ensure the reliability of neoantigen prediction and scalability of the vaccine platform. Most important, of course, is how cancer burden is affected. Studies published to date provide glimmers of hope in this regard, with shared results that show the vaccines do elicit some T cell responses and have few severe adverse side effects. And when cancer vaccines were combined with PD-1 checkpoint inhibitors, trial participants receiving both had more-durable responses than with either therapy alone.6,8
The diverse expertise needed to fully address the many questions about neoantigen-based personalized cancer vaccines makes this a truly multidisciplinary effort. Researchers specializing in cancer genomics, cancer immunology, vaccine development, oncology, and diagnostics have driven early clinical trials. The scalability of cancer vaccine development and quality control will require pharmaceutical-grade production experts and systematic clinical monitoring approaches that evaluate vaccine response, emerging resistance, and severe adverse events—perhaps on an unprecedented scale that involves cooperation between industry and academia.
Even with all hands on deck, there are many challenges ahead. In an already crowded field of immune-based therapies and associated clinical trials, the competition for patient accrual is intense. This may limit thorough testing of neoantigen vaccines in favor of therapies that are less personalized and therefore more amenable to mass production. With existing US Food and Drug Administration–approved immunotherapies in place for many types of cancer, it will be interesting to see whether neoantigen vaccines find a niche and if so, how broad that niche becomes. Regardless, efforts in this space will enhance our understanding and will elucidate next steps toward more-effective cancer immunotherapeutics.
Jasreet Hundal is the Personalized Cancer Vaccines Project Manager at the McDonnell Genome Institute at Washington University School of Medicine in St. Louis, Missouri. Elaine R. Mardis is a co-executive director at the Institute for Genomic Medicine at Nationwide Children’s Hospital in Columbus, Ohio, Professor of Pediatrics at the Ohio State University College of Medicine, and president of the American Association for Cancer Research.
- H. Matsushita et al., “Cancer exome analysis reveals a T-cell-dependent mechanism of cancer immunoediting,” Nature, 482:400–404, 2012.
- M.M. Gubin et al., “Checkpoint blockade cancer immunotherapy targets tumour-specific mutant antigens,” Nature, 515:577–81, 2014.
- J. Hundal et al., “pVAC-Seq: A genome-guided in silico approach to identifying tumor neoantigens,” Genome Med, 8:11, 2016.
- B.M. Carreno et al., “A dendritic cell vaccine increases the breadth and diversity of melanoma neoantigen-specific T cells,” Science, 348:803–808, 2015.
- N.H. Segal et al., “Epitope landscape in breast and colorectal cancer,” Cancer Res, 68:889–92, 2008.
- P.A. Ott et al., “An immunogenic personal neoantigen vaccine for patients with melanoma,” Nature, 547:217–21, 2017.
- D.B. Keskin et al., “Neoantigen vaccine generates intratumoral T cell responses in phase Ib glioblastoma trial,” Nature, 565:234–39, 2019.
- U. Sahin et al., “Personalized RNA mutanome vaccines mobilize poly-specific therapeutic immunity against cancer,” Nature, 547:222–26, 2017.
- J. Hundal et al., “pVACtools: a computational toolkit to select and visualize cancer neoantigens,” bioRxiv, doi:10.1101/501817, 2018.
- J. Hundal et al., “Accounting for proximal variants improves neoantigen prediction,” Nat Genet, 51:175–79, 2019.
- M. Bassani-Sternberg et al., Abstract PL02-02: Immunopeptidomics: Accelerating the development of personalized cancer immunotherapy, Molecular Cancer Therapeutics: Genomics, Proteomics, and Target Discovery, 2018.
- Y. Samuels et al., “Combined analysis of antigen presentation and T cell recognition reveals restricted immune responses in melanoma,” Cancer Discov, 8:1366–75, 2018.
- C. Ma et al., “A clinical microchip for evaluation of single immune cells reveals high functional heterogeneity in phenotypically similar T cells,” Nat Med, 17:738–43, 2011.