Some 18-months later, the team is presenting the first of its data in a poster session at the American Association for Cancer Research (AACR) annual meeting, which kicked off in Washington, D.C., on Saturday.
It’s the company’s first big reveal. So what have they got?
Enough to shoot for an IND filing and human trials in the middle of next year, said Cofounder, President, and CEO Andrew Allen.
The poster outlines three concepts, each with supporting data. Combined, Allen said they close the loop on what the company is trying to achieve: A model for predicting tumor-specific neoantigens that can be used to trigger a robust T-cell immunotherapy response. The initial target is lung cancer.
What does that all mean?
Like Neon Therapeutics and The Parker Institute for Cancer Immunotherapy, Gritstone is targeting neoantigens. These are mutations that arise “de novo” in a given cancer — they’re not otherwise found in the human genome.
As an immunotherapy target, they offer two major benefits: They’re foreign to the immune system and they’re not found in healthy tissue.
By comparison, traditional lung cancer targets such as ALK or EGFR have been present in the body since early development. They may be overexpressed in cancer cells, but the immune system has over time learned to tolerate them as “self.” That’s not a good platform for triggering a T-cell attack. Conversely, if the immune system was activated against those receptors, some healthy tissue would be hit.
Neoantigens are next-level personalized medicine, with next-level logistical challenges.
Predicting cancer neoantigens
Given the recurrent failures in cancer vaccine development, Gritstone is taking a two-pronged approach that verifies that specific neoantigens are truly being expressed on cancer cells.
“For success here, you’ve got to do two things well,” Allen said. “You’ve got to predict neoantigens well because that’s a big part of the problem. And then you’ve got to deliver them in a way that is going to drive large numbers of highly active T-cells.”
The first half of the puzzle is being pieced together by a team of around twenty, working in a facility in Cambridge, Massachusetts. The resulting data also make up the first findings in Gritstone’s AACR poster. The company asked whether the predictive modeling can be out-sourced.
Lung cancer has a high mutational burden, Allen explained — there are on average around 300 genomic changes. Of those, only around 1 percent will be truly novel neoantigens. It’s a drop in the ocean that is easily missed when a generalized approach to tumor profiling is deployed. Third-party labs that look for standard receptor targets typically omit between 20-25 percent of the mutations, Allen said. In some patients, 50-60 percent of the mutations are lost. Scientists need better data to build a cancer vaccine that works.
Zooming in on lung cancer, the Cambridge crew have extensively characterized hundreds of real tumor samples using DNA and RNA sequencing, mass spectrometry and deep learning.
Deep learning fast-tracks the process and removes the limitations of current thinking, Allen explained. It’s pure mathematics: it doesn’t apply the researcher’s biases and hypothesis and it’s not limited by our imagination.
“You’re saying, let me look for associations in a completely unconstrained way,” he said.
Those associations are then iteratively tested, to be rejected or strengthened. It eventually leads to a model that can predict from the sequence alone, which of those mutations will create peptides or antigens that will be presented on the tumor cell surface.
“Our estimate, when we test ourselves on fresh data, is that we’re operating at something like ten-fold better than the public domain approach that many of our competitors are using,” Allen said.
Therein lies the second segment of findings in the AACR poster, which asked if Gritstone’s in-house approach is effective. It seems it is. In the future, its scientists can sequence fresh tumor biopsies to accurately predict what mutated peptides could be targeted.
Rallying the immune response
The second challenge with cancer vaccines is learning how to weaponize the neoantigens to ensure the immune response doesn’t fall flat.
“Our model doesn’t necessarily predict antigens, it predicts whether a peptide will be presented by an HLA class 1 molecule on the cell surface,” Allen noted. “To be an antigen, you also have to stimulate a T-cell response.”
A West Coast team of around 30 is working on this problem in Gritstone’s headquarters in Emeryville, California.
Lessons on how to make a successful cancer vaccine, Allen said, could not be found in the cancer vaccine field. Not a lot has worked. Instead, Gritstone looked to the field of infectious diseases. Certain viruses, such as Malaria, are able to bury themselves deep within cells, he said. That necessitates a robust CD8 T-cell response — the kind Gritstone is hoping to produce.
“What was striking to us was that so many people were using viruses as a vector for delivering the antigens, in order to get these really effective T-cell responses. And so that’s the path that we’ve pursued,” he explained.
It led to the third component of the poster. The company took the isolated peptides and some HLA-matched T-cells and asked; can they prime a T-cell response to a given neoantigen in vitro. Can they show that it does register an immune response?
“So that’s really closing the loop and obviously suggesting that, were this to be a patient that we were predicting and making a vaccine,” Allen said. “We have identified an antigen that should in principle be able to make good T-cells in response to the vaccine that may have the potential to kill the tumor.”
The company can connect a DNA mutation to an altered protein and show that it is processed and presented as an altered peptide. Gritstone may be the first to connect those dots in lung cancer, he said.
An eventual vaccine would be given in combination with an immune modulator, such as a PD-1 inhibitor , setting the immune system up for an optimal anti-tumor response.
It’s all theory for now, but Gritstone’s integrated use of deep learning and bioinformatics is broadening the basic theories the human mind can generate.
Photo: Esben_H, Getty Images