Research

Cancer ecology, evolution and treatment.

Tumors are ecosystems. Cancer cells compete and cooperate with one another and with the stromal, immune and bone cells around them, and treatment reshapes those interactions and the selection they impose. Over more than fifteen years, our work in cancer ecology and evolution has developed along three main themes: the bone marrow as an ecosystem in which cancers grow, relapse and resist therapy; stromal niches that shelter tumor cells from treatment; and the interactions that determine what selection favors as tumors evolve.

Treatment is not a separate theme but the thread that runs through all three. In each, we ask how therapy changes the ecology and evolution of the disease, and how that understanding can be used to treat it better.

Research themes

Three questions about tumors as evolving ecosystems

Cancer in the bone marrow ecosystem

Bone is a living tissue, constantly remodeled by bone-resorbing osteoclasts and bone-forming osteoblasts and supported by mesenchymal stromal and immune cells. Cancers that grow in bone, such as multiple myeloma, leukemias and bone-metastatic prostate cancer, co-opt this ecosystem. Together with experimental collaborators over more than a decade, we have built spatial, hybrid models of bone and bone marrow to understand how tumors establish, how they disrupt bone homeostasis, and how the marrow environment shapes relapse and resistance.

Increasingly we model immune cells as well: whether a T cell finds, engages and kills a tumor cell depends on where cells are and how long they stay in contact.

Treatment. In myeloma, our model predicted that targeting environment-mediated drug resistance deepens responses by eliminating sensitive cells sheltered near stroma and bone, a prediction supported in vivo. We also use these models to optimize standard-of-care therapy in bone-metastatic prostate cancer and to rethink treatment-free intervals for T-cell engagers.

Stromal niches and residual disease

Targeted and cytotoxic therapies often shrink tumors dramatically but rarely eliminate them: a subset of cells survives and eventually gives rise to resistant disease. With experimental and pathology collaborators, we combine spatial analysis of tumor tissue, experiments and agent-based models to show how proximity to stroma shapes that survival.

Peristromal niches shelter lung cancer cells from targeted therapy, an ecological rescue that buys time for evolutionary escape. In breast cancer, stroma-enhanced proliferation lets residual tumors recover between chemotherapy cycles, and in prostate cancer the reactivity of the stroma shapes how tumor cells evolve.

Treatment. Because stromal protection combines many signals, blocking any single one has limited value. Instead, residual cells can be targeted through their collateral sensitivities: in lung cancer, adaptive HER2 upregulation made them vulnerable to the antibody-drug conjugate T-DXd.

Selection in an evolving ecosystem

Mutations generate variation, but which variants succeed depends on selection, and selection depends on how a cell's traits shape its interactions with other tumor cells, with stromal cells and with the physical environment. We have used evolutionary game theory since our earliest work to study these interactions, from the emergence of glycolytic and invasive phenotypes to cooperation over angiogenesis, and with experimental collaborators we have measured the games that cancer cells actually play.

We also study where heritable variation comes from: spontaneous cell fusions that let tumor cells recombine, hypoxia that accelerates the tempo of evolution, and the adaptive state of a population, which determines whether new mutations help or harm.

Treatment. Therapy is a selective pressure like any other. Fibroblasts and targeted therapy can switch the evolutionary game a tumor plays; raising the mutation rate helps populations under therapy but not at baseline, which points to restricting mutational variability as a strategy; and drug combinations can be designed as evolutionary double binds.

How we work

Integrated mathematical oncology

The method follows the biological question, but four commitments run through all three themes.

Built with experimentalists and clinicians

Our models are developed in long-standing collaborations with experimental, pathology and clinical colleagues, so that predictions are matched to experiments that can test them.

Spatial, hybrid models

Individual cells interact with one another and with diffusing factors in realistic tissue. We build these models with the Hybrid Automata Library, an open platform we co-developed.

Evolutionary game theory

Frequency-dependent interactions among cancer phenotypes and the cells around them, from theoretical games to games measured in co-culture.

Calibrated, open and reproducible

We fit models to longitudinal and spatial data, quantify what the data can and cannot determine, and release models and code. See our guide to spatial agent-based modelling of tumour evolution.