Inferring the Past: How did Your Cancer Grow?

Abstract

Cancer growth, just like phylogeny and human development, requires genome replication, which inevitably produces “passenger” (neutral) replication errors. In practice, DNA methylation patterns from just a few CpG sites can convey information about cancer cell ancestry when sampled from related lineages. We take a genealogical approach to infer tumor growth patterns using variation in DNA methylation patterns. We apply approximate Bayesian computation (ABC), a simulation-based method, to model tumor growth under a variety of evolutionary scenarios, finding models that fit observed DNA methylation patterns. Applying the ABC approach to data from two separate genomic regions in a collection of 12 colon tumors, we find evidence that cancer trees are consistent with Gompertzian growth (rapid initial clonal expansion), which is the growth pattern of most tumors.