Performance of a priori and a posteriori calibration strategies in divergence time estimation

Abstract

Relaxed molecular clock methods allow the use of genomic data to estimate divergence times across the tree of life. This is most commonly achievedin Bayesian analyseswhere themolecular clock is calibrateda priori throughthe integrationof fossil information. Alternatively, fossil calibrations can be used a posteriori, to transform previously estimated relative divergence times that were inferredwithout considering fossil information, into absolute divergence times.However, as branch length is the product of the rate of evolution and the duration in time of the considered branch, the extent towhich a posteriori calibrated, relative divergence time methods can disambiguate time and rate, is unclear. Here, we use forward evolutionary simulations and compare a priori and a posteriori calibration strategies using different molecular clock methods and models. Specifically, we compare three Bayesian methods, the strict clock, uncorrelated clock and autocorrelated clock, and the non-Bayesian algorithm implemented in RelTime. We simulate phylogenieswithmultiple, independent substitution rate changes and show that correct timescales cannot be inferred without theuse of calibrations.Under our simulation conditions, a posteriori calibration strategies almost invariably inferred incorrect rate changes and divergence times. The apriori integrationof fossil calibrations is fundamental in these cases to improve the accuracy of the estimated divergence times. Relative divergence times and absolute timescales derived by calibrating relative timescales to geological time a posteriori appear to be less reliable than a priori calibrated, timescales.