Challenges In Bringing Personalized Cancer Vaccines To The Clinic
Contributed Commentary by Amitabha Chaudhuri
June 21, 2018 | Cancer vaccines originate from proteins expressed by tumor cells but not normal cells. These tumor-associated antigens are recognized by our immune system as foreign and are eliminated, protecting us from the disease. The vaccination approach to cancer treatment originated with Coley’s vaccine, first used by William Coley in 1891 to treat inoperable cancers by injecting heat inactivated Streptococcus pyogenes, demonstrating that systemic activation of the host immune system can inhibit tumor growth.
Experiments on mice provided more direct evidence of tumor-specific antigens. A panel of tumors created by treating mice with chemicals or radiation was used to immunize different groups of mice. Remarkably, each group immunized with a specific tumor was protected only when they were challenged by the same tumor. This suggested that each tumor is uniquely recognized by the immune system and, by extension, carried tumor-specific antigens. It took 40 years after these seminal studies to identify the first human melanoma-specific tumor antigen MAGE1. However, discovering tumor-specific genetic alterations—the source of cancer vaccines—has become easy, cheap, and scalable with the introduction of next-generation sequencing technologies, ushering in the era of personalized cancer vaccines.
Successful deployment of cancer vaccine therapies in the clinic is not an easy development path. The complex process of immune recognition involves multiple steps:
Step 1: Proteins produced in the cells are degraded into peptides as part of their housekeeping chore by the proteasomal machinery, and a subset of these peptides enter into the endoplasmic reticulum (ER).
Step 2: Peptides are loaded onto the human leukocyte antigen (HLA) class-I proteins (also referred to as MHC in mice) in the ER. Peptides with certain features bind HLA and reach the cell surface. The stability of the HLA-peptide complex is further regulated by a group of aminopeptidases–endoplasmic reticulum peptidase-1 and 2 (ERAP-1 & 2) that trim HLA-bound peptide into its correct length. For HLA class-I, 9-11-mer peptides are preferred; for class-II binding, 14-17-mer peptides.
Step 3: The interaction between HLA-peptide complex with CD8 T cell receptor. The binding strength of interaction, the occupancy time, and the copy number of antigenic peptides presented on the surface of cells determine T cell response, which is critical to the success of the vaccine therapy.
Identifying tumor-specific mutated proteins has become routine through the use of whole genome/whole exome and whole transcriptome sequencing. But there is a greater challenge to predicting vaccine candidates from the cancer mutations. As mentioned above, a peptide must pass the filters imposed by steps 1 through 3 to activate T cells and evoke an anti-tumor immune response. Many algorithms have modeled steps 1 and 2 to select peptides presented on the surface of cells. However, presentation of a peptide bound to HLA solves only half the problem; the second half is predicting whether that peptide will interact with the T-cell receptor (TCR) and if the quality of interaction will activate the T cells and empowering them to eliminate the tumor cells.
Converting Predicted Vaccine Candidates Into A Cocktail Of Selected Vaccines For Therapy
The cancer vaccine industry faces a challenge: How to transition from prediction to effective vaccine therapy? For cancers such as melanoma, non-small cell lung cancer, and bladder cancer with high mutation burden (over 100 mutations/tumor on an average), the challenge is to develop a process to whittle down the vaccine? candidates to a number 10 or less that can be formulated and given to patients. Most cancer vaccine platforms can reduce the number of predicted candidates to around 25% using filters such as expression of the antigen at the RNA level, the HLA-binding affinity of the peptide, the binding affinity of the peptide to the TAP transporters, and the presence of proteasomal processing sites in the peptide for optimal processing.
Unfortunately, algorithms are far from accurate, reducing the overall accuracy of the process. Highly sensitive mass spectrometry approaches can identify peptides present on the cell surface in complex with HLA, cutting through most of the predictions associated with steps 1 and 2. However, mass spectrometry is difficult to integrate in clinical practice primarily because of fresh/frozen tissue requirements, limited scalability, cost and access to sensitive instruments, and lack of open-source data analysis pipelines. The industry is working toward striking a balance between the use of genomics-driven prediction technologies and mass spec-based methods to increase the accuracy of predictions.
The second half of the overall problem—whether the HLA-peptide complex will interact with the TCR to activate and convert the T cells into a tumor-killing machine—is a bigger challenge. Foremost, the surface presentation does not guarantee TCR binding, and few available tools model the interaction between HLA-peptide complex and TCR. Research companies can use machine learning approaches for predicting TCR binding by scoring for the enrichment of physicochemical properties of amino acids that favor TCR binding in the target peptide. The prediction method can be further bolstered by performing an ex vivo T cell activation assay by co-culturing dendritic cells with T cells in the presence of added peptides. Such assays are time-consuming, expensive, and low throughput but provide definitive evidence of whether the peptide is immunogenic.
What Approach Is Suitable In A Clinical Trial Setting?
Vaccine companies have taken two approaches. One set has focused on the mode of vaccine delivery without significant efforts in designing the vaccine cocktail for therapy—RNA (BioNTech, Moderna), DNA (Geneos), or peptides (Neon Therapeutics). These companies are using their genomics-based prediction pipelines without performing downstream T cell activation assays. Two 2017 Nature papers, using RNA (doi: 10.1038/nature23003) or peptide vaccines (doi: 10.1038/nature22991), reported clinical evidence of control of tumor growth and metastasis, although the immune mechanism is not fully understood.
Other companies are focusing on selecting the vaccine candidates, including Gritstone Oncology, which uses mass spectrometry to profile peptides presented on the surface of tumor cells. Genocea Biosciences’ ATLAS platform uses a bacterial expression system to screen predicted neoantigens for T cell activation to select their vaccine cocktail. MedGenome, my company, combines its TCR-binding algorithm with cell-based assays and single-cell sequencing to identify vaccine cocktails.
Personalized cancer vaccines are expected to enhance the efficacy of checkpoint inhibitors by priming and expanding naive antigen-specific T cells for the immune-mediated elimination of tumors. Challenges in selecting optimum vaccine cocktails are being overcome by additional assays to increase the specificity and efficacy of treatment without increasing the adverse effects. The field is rapidly evolving with technological innovations, use of novel adjuvants and delivery systems, smart clinical trial designs, and rational combinations. I believe, cancer vaccines will become a mainstream therapeutic choice in immuno-oncology in the future.
Amitabha Chaudhuri is the VP of Research & Development at MedGenome. Chaudhuri has over 15 years of experience in oncology target discovery and validation at Genentech, Inpharmatica, and CuraGen Corporation. He holds a Ph.D. in Biochemistry from the Indian Institute of Science, Bangalore and did his Post Doctoral research at Massachusetts General Hospital, Harvard Medical School and at the Department of Molecular & Cellular Biology, Harvard University. He can be reached at amitc@medgenome.com.
