Immunotherapy is rapidly becoming a well-founded form of cancer treatment as it employs and strengthens a patient’s immune system to attack tumors.
Chimeric antigen receptor (CAR) T cells – genetically altered T cells used in immunotherapy that can locate and destroy cancer cells – show great promise. Yet they still struggle in broader applications, particularly in attacking non-tumor cells and persistence in killing tumor cells.
Natasa Miskov-Zivanov, principal investigator and assistant professor of electrical and computer engineering at the University of Pittsburgh Swanson School of Engineering, received a $299,986 EAGER Award from the National Science Foundation to develop an automated approach to design CAR T cells. The two-year project, “Development of Hybrid Knowledge- and Data-Driven Approach to Guide the Design of Immunotherapeutic Cells,” received the maximum amount of funding for an EAGER Award.
“There is an urgent need to design better receptor systems to finely tune T cell activation, enhance tumor specificity and control signaling pathway and cell types,” Miskov-Zivanov said. “Successfully completing this project means a revolutionary change in cancer treatment.”
Miskov-Zivanov’s approach is a novel one.
“There hasn’t been much work with computational modeling behind CAR T cells,” Miskov-Zivanov explained. “We’ll be using machine learning and artificial intelligence methods that utilize both data from experiments and knowledge about pathways from literature to significantly reduce the combinatorial complexity of CAR design.”
It takes time to design the right T cell – time a patient doesn’t have
A patient’s outcome when treated with CAR T cells is reliant on the cell’s ability to expand and persist after infusion.
But, how does CAR T cell therapy work?
CAR T cells bind to a specific antigen on a targeted cell. Once the antigen is bound, the CAR T cell transmits an activation signal which spreads through the receptor’s intracellular signaling domain to its pathways inside. CARs then link intracellular behavior and phenotype decision, leading to cell proliferation and targeting the tumor cell. CARs are almost like little bounty hunters that an oncologist infuses into the patient’s body. These cells find their target, lock onto it, and keep it from spreading. Tumors, however, don’t make this an easy task.
“This process gets complicated with solid tumors,” Miskov-Zivanov said. “Their structure prevents infiltration by the T cells and their heterogeneity overcomes the T cells’ antigen specificity. So, the main goal when creating T cells is to generate phenotypes with increased antitumor cytotoxicity and persistence that can help penetrate a durable and stubborn tumor cell.”
Designing and testing new CARs requires a significant amount of time and resources. Researchers and synthetic biologists must choose from a large number of candidate receptor domains and then decide the order of selected domains on the receptor.
Opening the black box
Creating a more efficient process to engineer T cells isn’t new, but Miskov-Zivanov said these approaches all suffer the same fate: they treat the cell as a black box.
“A lot of these newer methods rely on experimental data and machine learning based only on data to draw conclusions,” she said. “The computational methods they use can’t provide mechanistic explanations for their predictions, which is a large setback in biological and clinical research. More than anything, it’s an incredibly slow process – taking weeks at a time.”
Miskov-Zivanov’s approach diverges from these previous strategies by integrating experimental data and knowledge sources, leading to the development of components for a larger computational framework that can recommend and explain high-performing synthetic and natural systems for effective, reliable therapies.
“Creating this framework will help oncologists determine the best treatment for cancer patients all while improving a patients’ outcome with these deadly diseases,” Miskov-Zivanov said.
Recently, Miskov-Zivanov worked alongside UPMC researchers to develop a mathematical model that can help oncologists determine the effective dose and timing for a patient.
“I have been so impressed by the success of CAR T cell therapy ever since I heard about it more than ten years ago,” Miskov-Zivanov said. “The work with UPMC was one way for me to assist oncologists and cancer researchers by using computational modeling. This EAGER project, on the other hand, takes a different and more broad approach that relies on machine learning and artificial intelligence to assist in CAR T cell design, and the hope is that it could be applied in the future when treating different diseases with immunotherapy.
According to the NSF, the EAGER funding mechanism may be used to support exploratory work in its early stages on untested, but potentially transformative, research ideas or approaches. This work may be considered “high risk-high payoff” in the sense that it, for example, involves radically different approaches, applies new expertise, or engages novel disciplinary or interdisciplinary perspectives.
