Bayesian phylodynamics for developmental biology: incorporating age-dependence

Abstract

As novel technologies for single-cell lineage tracing emerge, phylogenetic and phylodynamic tools are increasingly being used to study developmental processes. However, traditional phylodynamic methods, which were originally developed to study viral evolution, rely on assumptions that are difficult to justify in developmental contexts. Notably, due to cells dividing after characteristic generation times rather than after exponential waiting times--as assumed by the traditionally used birth-death model--empirical cell lineage trees deviate from birth-death phylogenies. Here, we present a non-trivial extension of the birth-death phylodynamic model that captures this characteristic feature of development. By applying our method to a public dataset of stem cell colonies, we show how previous estimates of the underlying population-dynamic parameters were biased by the choice of a birth-death tree prior. Beyond developmental biology, our framework provides an approach for analyzing systems where classical birth-death assumptions may be violated or where empirical tree shapes are poorly captured by those expected under standard phylodynamic models. Our method is available as a BEAST2 package. SignificanceApplying phylodynamic inference methods to data from developmental biology requires reassessment of the foundational assumptions underlying these tools. We show that cell population dynamics can be captured by an age-dependent branching process, as opposed to the widely used birth-death process. We develop computational methodology for efficient phylodynamic inference under this age-dependent model, thus providing a tool for connecting cell population dynamics to lineage trees. Our method is furthermore, to our knowledge, the first performant implementation of an age-dependent phylodynamic likelihood, and may be more generally applicable to systems which are ill-characterized by traditional birth-death models.