Ignoring stratigraphic age uncertainty leads to erroneous estimates of species divergence times under the fossilized birth-death process

Abstract

Fossil information is essential for estimating species divergence times, and can be integrated into Bayesian phylogenetic inference using the fossilized birth-death (FBD) process. An important aspect of palaeontological data is the uncertainty surrounding specimen ages, which can be handled in different ways during inference. The most common approach is to fix fossil ages to a point estimate within the known age interval. Alternatively, age uncertainty can be incorporated by using priors, and fossil ages are then directly sampled as part of the inference. This study presents a comparison of alternative approaches for handling fossil age uncertainty in analysis using the FBD process. Based on simulations, we find that fixing fossil ages to the midpoint or a random point drawn from within the stratigraphic age range leads to biases in divergence time estimates, while sampling fossil ages leads to estimates that are similar to inferences that employ the correct ages of fossils. Second, we show a comparison using an empirical dataset of extant and fossil cetaceans, which confirms that different methods of handling fossil age uncertainty lead to large differences in estimated node ages. Stratigraphic age uncertainty should thus not be ignored in divergence time estimation and instead should be incorporated explicitly.

Keywords: Bayesian phylogenetics; divergence times estimation; fossil data; fossilized birth–death; stratigraphic age uncertainty.

Figure 1.

Representation of the age intervals obtained from PBDB for North American mammals sampled during the Mesozoic and Cenozoic. Intervals are ordered by the maximum age of the range, from youngest to oldest.

Figure 2.

Representation of the age uncertainty simulation process. Phylogenies with fossils are simulated according to a birth–death–fossilization process. The correct age of each fossil is used to draw an age interval for that fossil from the set obtained from PBDB. This age interval is then used as the basis for the median and random age assignment. A symmetric age interval is also drawn from the correct age.

Figure 3.

Average relative error of the median estimate (column 1) and 95% HPD coverage (column 2) achieved by different age handling methods for the following parameters: (a) divergence times of extant species, (b) diversification rate and (c) turnover. Ages sampled as part of the MCMC are marked by (*).

Figure 4.

MCC trees inferred for the Cetacea dataset using the FBD process with fossil ages fixed to (a) median ages, (b) random ages or (c) sampled within the known interval of uncertainty. The major clades and the clade shown in figure 5 are highlighted.

Figure 5.

Posterior density obtained for the most recent common ancestor of the family Phocoenidae in the Cetacea dataset using the FBD process with fossil ages fixed to median ages, random ages or sampled within the known interval of uncertainty. The densities were scaled to the interval (0; 1).

Figure 6.

Estimates of the (a) diversification rate, (b) turnover and (c) sampling proportion obtained for the Cetacea dataset using the FBD process with fossil ages fixed to median ages, random ages or sampled within the known interval of uncertainty.