Research

Cancer is an evolutionary process that typically spans multiple decades before it causes symptoms and is detected.

We and others have shown that due to this relatively slow growth (often <1% per day) there is a large time window for an earlier detection of cancer. The survival probability of patients with a tumor diagnosed early is five to ten times higher than when diagnosed at an advanced stage. The last stage of cancer progression, metastasis, is responsible for 90% of cancer-related deaths. Moreover, in a tumor with billions of cells virtually any point mutation is expected to be present in a few cells. Hence, at a genetic level, not only is every cancer type different, but also every tumor of the same type and every cell of the same tumor are different. This enormous heterogeneity of advanced cancers poses a major barrier to drug development and long-term disease control.

Our research

Our research in the Translational Cancer Evolution Laboratory focuses on the stochastic evolutionary dynamics of cancer with the goal to improve the detection and treatment of tumors. We develop computational methods to learn from large-scale biological data sets and design mathematical models to predict a cancer's trajectory, generate novel hypotheses, and explain observations on a mechanistic level. We apply these methods to genomic data from clinically-annotated patient cohorts to advance precision medicine.

Mathematical Methods

We have a track-record in developing mathematical methods of cancer evolution to illuminate cancer detectability, tumor progression, intratumoral heterogeneity, and treatment response (Reiter et al, Nature Reviews Cancer 2019, Bozic et al, eLife 2013). We implemented publicly available computational tools to reconstruct the evolutionary history of cancers from DNA sequencing data which have been used in >50 studies (Reiter et al, Nature Commun 2017). These algorithms allowed us to refine the step-wise progression of pancreatic cancers and demonstrated the distinct and common origin of lymphatic and distant metastases in colorectal cancer patients (Naxerova et al, Science 2017; Makohon-Moore et al, Nature 2018). We discovered that mutated cancer driver genes that are heterogeneous within primary tumors or among untreated metastases lack predicted function and therefore single biopsies typically capture the essential information for initial therapeutic decision making (Reiter et al, Science 2018; Reiter et al, Nature Reviews Cancer 2019).

Statistical Framework

We developed a statistical framework to show that distant metastases cluster together much more often than expected by chance as well as more often than lymph node metastases indicating a stricter evolutionary bottleneck for distant metastases (Reiter et al, Nature Genetics 2020). Most recently, we developed the first stochastic model of circulating tumor DNA shedding to approximate the potential of liquid biopsies for lung cancer early detection (Avanzini et al, Science Adv 2020). We closely collaborate with many other research labs, physician-scientists, and clinicians at Stanford, Harvard, Johns Hopkins, Memorial Sloan Kettering Cancer Center and others to tackle the most pressing scientific and clinical challenges.