Using computational tools, researchers from the Johns Hopkins Kimmel Cancer Center and the Johns Hopkins University School of Medicine have developed a method to predict which patients with a primary liver cancer called hepatocellular carcinoma (HCC) would most benefit from combination treatment using immunotherapy and a targeted therapy that blocks the growth signals that the tumor depends on. The work, supported in part by the National Institutes of Health, was published online July 14 in the Proceedings of the National Academy of Sciences .
"Many cancers have a very fast progression time, and doctors may not necessarily have time to try surgery or different treatments, so our idea was to create a computational model where we could simulate trying different doses or combinations of cancer therapies, and it could help guide physicians toward the best options for patients," says senior study author Atul Deshpande, Ph.D. , an assistant professor of oncology at the Johns Hopkins University School of Medicine. The tool needs to be validated further before it could be incorporated into clinical treatment, he says.
The spatial QSP modeling platform, developed in the laboratory of Aleksander Popel, Ph.D. , a professor of biomedical engineering and oncology at Johns Hopkins, combines two tools. One is quantitative systems pharmacology (QSP), a mathematical model that uses equations to capture the whole body's response to a treatment and can simulate tumor progression and drug effects. The second is an agent-based model that tracks how individual cells behave in a given scenario. Together, they map not just how many cells are present but where they sit, predicting the dynamics of the tumor and the surrounding microenvironment.
In this study, Deshpande and colleagues expanded the platform to model fibroblasts — a cell type previously associated with resistance to immunotherapy in liver cancer — and developed a machine-learning calibration workflow that tunes the simulation to data from real clinical trials, generating virtual patients whose predicted responses can be checked against actual outcomes.
In a type of "war planning," Deshpande says, "we generate a virtual tumor to see what happens in the microenvironment. Do the cancer cells resist? If you change the architecture of the tumor, does that help the cancer cells or the immune cells?"
One advantage of a computational model is scale, he says. From a small, early-phase (phase I) study of 15 patients, it can generate a phase III–sized virtual population, letting researchers estimate how a therapy might perform in a far larger trial — quickly and without risk to anyone. When the team simulated treatment with the targeted therapy cabozantinib and the immunotherapy nivolumab, alone and together, the predicted response rates tracked closely with those reported in real clinical trials — evidence that the virtual patients behave like real ones. The team also validated the model's predicted tumor architectures against real post-treatment tissue and compared the microenvironments of responders and non-responders.
The researchers found that fibroblasts remodeled the tumor microenvironment, creating immunosuppression. Notably, among virtual patients who did not respond to therapy, investigators saw that fibroblast cells formed a type of physical wall, Deshpande says. "Even if immune cells were located near the tumor, the fibroblast would block the immune cells from reaching the tumor," he says.
Ideally, modeling could get to the point where if multiple treatments were available for a particular cancer, researchers could help determine which of the treatments would be most effective, or which to avoid, Deshpande says. Because such architectural features — both the fibroblast barrier the model flags and structural patterns the team measured in patient tumors — are visible before treatment begins, they could eventually help predict who will benefit, the researchers say.
The study was co-supervised by Popel and Elana Fertig, Ph.D., director of the Institute of Genome Science at the University of Maryland School of Medicine; co-authors were Shuming Zhang, Hanwen Wang, Yeonju Cho, Wendy Wong, Mark Yarchoan, Elizabeth Jaffee, Won Jin Ho and Luciane Kagohara of Johns Hopkins; and Heber Rocha of Indiana University.
The work was supported by grants from the National Institutes of Health (grant numbers U24CA284156, U01CA253403 and U01CA212007); the Department of Defense-MEDCOM grant 144517-CA220654P; a Maryland Cancer Moonshot Research Grant; and the Maryland Cigarette Restitution Fund.
