Mismatch of the morphology model is mostly unproblematic in total-evidence dating: insights from an extensive simulation study

Abstract

Calibrating the molecular clock is the most contentious step in every dating analysis, but the emerging total-evidence dating approach promises increased objectivity. It combines molecular and morphological data of extant and fossil taxa in a Bayesian framework.Information about absolute node ages stems from the inferred fossil placements and associated branch lengths, under the assumption of a morphological clock. We here use computer simulations to assess the impact of mismatch of the morphology model, such as misspecification of character states and transition rates, non-stationarity of the evolutionary process, and extensive variation of evolutionary rates among branches. Comparisons with published datasets suggest that, at least for evolutionary rates typically observed in discrete morphological characters, the total-evidence dating framework is surprisingly robust to these factors. We show that even with relatively low numbers of morphological characters sampled, extensive model mismatch is mostly irrelevant for the performance of the method.The only exception we found are cases of highly asymmetric state frequencies and thus transition rates, but these can be accounted for by appropriate morphology models. In contrast, we find that the temporal scope of fossil sampling has a major impact on divergence time estimates, with the time signal quickly eroding if only rather young fossils are included in an analysis. Our results suggest that total-evidence dating might work even without a good understanding of morphological evolution and that study design should instead focus on an adequate sampling of all relevant fossils, even those with highly incomplete preservation.

Sphenodontian phylogeny and the impact of model choice in Bayesian morphological clock estimates of divergence times and evolutionary rates

“…The latter study found that most of the analyzed morphological datasets do not have strong levels of state frequency heterogeneity, therefore assuming a symmetric state frequency provides an adequate substitution model. However, ongoing research suggests that in datasets where there is heterogenic state frequencies in the characters, this model violation may result in strongly biased estimates of divergence times [76].…”

“…This may be corrected by an extensive sampling of fossil species, although unpractical under many different circumstances [124]. On the other hand, ongoing research using the uniform tree prior (which does not include sampled ancestors) indicates that it is actually advantageous to focus fossil taxon sampling on the oldest known taxa [76]. The discrepancy between these results suggest the tree model, and not fossil sampling strategy, may be the cause of overestimated divergence times, an explanation that is supported by our findings herein.…”

Ignoring fossil age uncertainty leads to inaccurate topology and divergence times in time calibrated tree inference

“…Similarly, we did not examine the impact of morphological model violations such as rate heterogeneity among characters or the effects of non-uniform missing character data. A recent study suggested that even large deviations from the true model may have limited impact on divergence times estimates using total-evidence dating under the uniform tree model (Klopfstein et al, 2019). However, none of their simulation scenarios excluded molecular data and thus these findings may not be applicable to fully extinct clades.…”

Molecular and morphological clocks for estimating evolutionary divergence times

“…Moreover, in the fossilized birth-death (FBD) process [17,18], the age of the fossil species provide the calibration information that translates the morphological distances into absolute times and rates, which are propagated to the other nodes on the phylogeny for which molecular data provides time structure information. Nevertheless, utilizing morphological characters in a dating analysis is complicated [11,16,[19][20][21][22][23]. To begin with, morphological traits are driven by natural selection and adaptation, they occasionally experience convergent evolution, and rarely evolve in a clock-like fashion [24,25].…”

“…Therefore, morphological datasets consider mainly traits that are shared by multiple lineages, with a relatively small sampling of autapomorphies and invariant characters, which is known to have an impact on divergence time estimates [22]. Lastly, morphological datasets present extensive variation of evolutionary rates among branches in the phylogeny and generally relatively low numbers of characters sampled compared to phylogenomic datasets [21].…”