Immunotherapy has revolutionized cancer treatment by harnessing the power of T cells to attack tumors. But, even when these treatments are successful, scientists often struggle to understand exactly how they work.
To monitor the body, T cells use unique T cell receptors (TCRs) that interact with peptides displayed on cell surfaces via major histocompatibility complex (MHC) proteins, or human leukocyte antigens in humans. Each TCR has high cross-reactivity, so it recognizes thousands of different peptide-MHC (pMHC) combinations.1 This diversity makes it difficult to determine which cancer antigens T cells recognize. “It’s one of the most challenging problems in immunology, if not biology,” said Brian Baker, a biochemist at the University of Notre Dame. Sequencing a patient’s TCRs doesn’t directly reveal their antigen targets, and mapping these relationships is particularly difficult because antigen-specific T cells are rare.2
Michael Birnbaum, an immunologist at MIT, leads MATCHMAKERS, an international group of scientists who received funding from Cancer Grand Challenges to tackle the problem of how T cells recognize cancer cells.
MIT Biological Engineering
Patients often have dozens or even hundreds of T cells reacting to their tumor, alongside an equally vast number of potential antigen targets. Experimentally testing all these combinations is an incredibly time-consuming process. “The combinatorial space gets super big,” said Michael Birnbaum, an immunologist at the Massachusetts Institute of Technology. “Compared to anything [like] normal tests in the clinic, it's just not feasible.”
Recognizing the scale of this problem, Cancer Grand Challenges, a global research initiative funded by Cancer Research UK and the National Cancer Institute, awarded Birnbaum and his international team a grant to fund their efforts to crack the TCR recognition code. Their goal is to transform the labor-intensive process of identifying TCR targets by leveraging AI and computational advances.
Over the next five years, the team will generate sequence and structural data to better understand TCR-pMHC interactions. Structural data, which reveals how TCRs bind to antigens in three-dimensional space, is harder to collect but essential for improving predictive models. Using high-throughput biochemical methods, they aim to refine machine learning algorithms to predict antigen recognition more accurately. “Prediction and design are very tightly interwoven,” said Birnbaum, who hopes to use these predictions to develop de novo TCRs for immunotherapies.
Birnbaum envisions their research extending beyond cancer, impacting fields such as infectious diseases and autoimmunity. “Anywhere T cells are involved in disease, we’ll be able to make predictions,” he said.
- Attaf M, et al. Cell Mol Immunol. 2015;12(4):391-399.
- Joglekar AV, Li G. Nat Methods. 2021;18(8):873-880.