Cancer Data Science Pulse
Using Bioinformatics to Solve the Neoantigen Puzzle
You’ll be discussing the topic, “Bioinformatics Approaches for Neoantigen Identification and Prioritization,” in an upcoming webinar. Can you tell us what first interested you in this topic?
I was always interested in computers. Growing up in the digital age, I was probably one of the first generations with access to a computer since I was a toddler. So I began tinkering with our home computer from a very early age.
Then, in college, I was studying biology and genetics and molecular genetics and first learned about large-scale high-throughput technologies. Instead of running an experiment and seeing what happens with 10 mice or a petri dish, researchers were starting to generate microarrays and produce millions of measurements, all at once. Those data quickly became intractable to manage without the aid of computers. My interest in computers and genetics seemed to lead naturally to bioinformatics, so I began using computational approaches to do big data analysis in biology.
My interest in neoantigens came much later, evolving over the past 15 years to where I am today. My Ph.D. focused on RNA expression and using next generation and sequencing approaches to study gene expression patterns. After coming to Washington University, I started working primarily on DNA analysis and identifying somatic mutations. This led to looking for variants that are unique to a particular patient’s tumor, which, in turn, helped us develop personalized assays to survey the disease burden over time. Now we’re using bioinformatics, not only to identify variants underlying cancer, but to monitor for response to therapy and to predict if a cancer will reoccur.
Neoantigens enable us to look at these mutations in a new way—from an immunological perspective. Neoantigens are proteins that result from mutations in cancer cell genes. Because these aberrant antigens are found in tumor cells, but not in normal cells, they offer a new way of alerting the immune system that these are foreign cells. When presented on the cell surface by major histocompatibility complex molecules, neoantigens can be recognized by T cells, encouraging or stimulating an anti-tumor response.
Who should attend this seminar?
My previous seminars have ranged from crash courses in genomics and biological techniques for medical students, to “in the weeds” descriptions of the tools and approaches for the bioinformatics community.
For the Data Science Seminar, I plan to offer a high-level introduction to the biology of neoantigens for those who may not be familiar with this area of study. I’ll then describe some of the bioinformatic and genomics challenges we face in trying to identify neoantigens. I’ll also discuss the difficulties in interpreting these proteins so we can accurately predict the best neoantigens to leverage for future therapies.
I’m hoping this will have a broad appeal—for attendees interested in the translational aspect of this technology, as well as people who want to better understand the complexities and difficulties of doing the analysis, along with the nuances and caveats to keep in mind when interpreting results.
Have there been any surprises or things you didn’t expect in your work with neoantigens?
The findings on shared neoantigens have been surprising, and we’re still not sure how that will turn out. Neoantigens arise from hundreds of mutations found in exons from many different genes. Yet a handful of common mutations have continued to crop up over and over again. These “hot spot” mutations are known to occur in particular types of cancer, and we can anticipate their presence even before we sequence those patients’ tumors. Initially the hope was that we could apply neoantigen-based therapies to target these tumors, creating a sort of “off the shelf” type therapy, as opposed to a more personalized approach. This would save time and resources.
Unfortunately, this line of research hasn’t gone as expected. To be effective, a neoantigen needs to do more than reflect a change in the tumor; it also needs to be detected by the immune system’s surveillance. But the mechanisms that go into making a neoantigen discoverable, and which allow the immune system to launch an attack, vary from individual to individual. As a result, using a one-size-fits-all approach to shared neoantigens hasn’t been very effective.
A common neoantigen-based therapy may still be a possibility. But in my work, we haven’t found that to be the case. Each patient is so unique, and most of the targets we’ve identified are “passengers,” not “drivers” of these hot-spot mutations, meaning they’re not primarily involved in driving cancer but can still allow the immune system to recognize tumor cells. Could these targets someday help the immune system recognize that tumor? We don’t yet know. But this does underscore how uniquely tailored neoantigens are to each patient’s tumor.
Were there particular challenges that you had to overcome?
From start to finish, it’s a complicated process with a lot of challenges. It starts with a piece of a patient’s tumor, followed by isolating DNA, generating huge amounts of data, and analyzing those data through highly complex pipelines. That analysis involves many different algorithms, each of which address a different neoantigen characteristic. At the end of this process, we’re able to generate a final list of candidates that might be useful for therapy using a neoantigen vaccine. Add to this process that all this work must be done quickly, within 2–4 weeks, if we’re to develop a personalized therapeutic that fits within a clinical workflow. That takes quite a team of people, all working together, along with significant resources.
Can you describe that team?
Our team at Washington University includes specialists in software development, pipeline analysts, members of our immunotherapy tumor board (with training in bioinformatics, genomics, immunology), as well as people with expertise in specific types of cancer (such as pathologists, oncologists, surgeons). In addition, we have a whole cadre of unsung heroes who help with logistical, regulatory, and administrative requirements. These kinds of translational efforts, where you’re taking broad genomics data and applying the findings in a clinical setting, requires a diverse and highly skilled team. My contribution is on the bioinformatics side of things, but I’m just a small piece of a much larger puzzle. At times it can be hard to see how, or even if, it all will fit together. But when it all works together, it’s a true example of what “Team Science” can accomplish.
Have you encountered any other challenges?
Predicting and developing neoantigen vaccines requires a lot of data. We’re at the very early stages (i.e., Phase 1 clinical trials) so there just isn’t much data—both before and after treatment—to allow us to fully understand how neoantigens stimulate an immune response. For example, is the gene underlying the neoantigen highly expressed? That’s important because it would lead to more copies of the new neoantigen, increasing the chance that it will be recognized by T cells. Then, we need to determine how well the T cells bind to the neoantigen and how stable that bond is, as well as how the neoantigen is processed on that presenting molecule. At the end of the day, what we really care about is how well the patient’s immune system recognized the neoantigen, and whether it led to an effective T cell antitumor response.
There also are bioinformatic challenges. The sheer size and the complexity of the pipelines that we use to analyze these data often make them difficult to maintain, or even to fully understand how they work. Each of those tools have many settings or parameters that need to be adjusted. And the overall workflow often becomes a huge, interconnected graph of dependencies—one little piece out of place can dramatically change the result (in an undesirable way). So it creates a need for best practices in bioinformatics reproducibility. That’s something we’re hoping to address in our work through NCI’s Informatics Technology for Cancer Research (ITCR).
You mention your work to develop best practices with NCI’s ITCR. Can you elaborate on that?
NCI’s ITCR is supporting labs to develop open, robust software to move cancer research forward. Given the complexity of our analyses, we need tools that are well engineered, well documented, and well maintained if we’re to be successful. That led us to develop a suite of tools called personalized cancer vaccine tools(pVACtools). In many cases, we’re leveraging existing algorithms and pulling all the pieces together into a toolbox to allow for neoantigen analysis.
Here, we’ve mostly been discussing the translational side of research, but there are many uses for neoantigens in basic science. They can be used to design an experiment in the lab or to better understand how T cells work in a tumor system, or in developing novel therapies. Likewise, other labs have expertise in the immune system or specific diseases but lack basic bioinformatics analysis support. The goal of the pVACtools suite is to enable those researchers to tackle this complex bioinformatics work.
Where do you see this topic headed in the next 5–10 years?
My ultimate hope is that this neoantigen technology is of benefit to patients, but we’re still in Phase 1 clinical trials so we don’t yet know how well this therapy will work. We’ve had compelling results in mice showing that neoantigen-based therapies, including vaccines, can be used to treat tumors, and we know from other types of therapies (e.g., CAR T-cell and immune check point inhibitors), that T cells have useful antitumor effects. But it remains an open question whether a more targeted or personalized approach to stimulate specific T cells with neoantigens will work.
It’s also unknown whether the vaccination approach produces a T cell response that’s strong enough to treat a patient with a high tumor burden, or what the response needs to be to get the largest therapeutic effect.
Maybe someday this therapy will be used along with checkpoint therapies. That would enable us to broadly stimulate the immune system to unleash an overall response, while at the same time using neoantigens to direct a more targeted tumor-specific attack. Or this therapy may prove useful as a vaccine to prevent recurrence of disease.
It’s hard to predict the future of this research. In the meantime, we’re systematically curating findings from existing clinical trials in one place to track the progress. This resource pulls together approaches (such as vaccines, personalized T cell therapies), as well as delivery mechanisms, agents, settings, and more. The website summarizes 130 clinical trials, offering a broad view of the research from groups around the world.
What satisfies you the most about this work?
What drew me to cancer research, in general, was the hope of creating successful therapeutic outcomes for patients. I especially like the personalized nature of this, using what’s unique to each patient to tailor individualized treatment. Neoantigen vaccines are the pinnacle of this type of this uniquely tailored therapy. If they’re successful, they have the potential to treat cancer without the side effects, which would be on par with that of a flu shot.
From a genomics and bioinformatics perspective, it’s building on everything I’ve worked on in the past 15 years—transcriptome analysis, genomic analysis, bioinformatics tooling. There’s so many intricate factors in addressing neoantigen prioritization. I’m still tinkering, in a very intricate way, with computational approaches to analyze this complex data—that’s what I love to do.