Exploring Patterns and Extent of Bias in Estimating Divergence Time from Mitochondrial DNA Sequence Data in a Particular Lineage: A Case Study of Salamanders (Order Caudata)

Abstract

In the practice of molecular dating, substitution saturation will bias the results if not properly modeled. Date estimates based on commonly used mitochondrial DNA sequences likely suffer from this problem because of their high substitution rate. Nevertheless, the patterns and extent of such expected bias remain unknown for many major evolutionary lineages, which often differ in ages, available calibrations, and substitution rates of their mitochondrial genome. In this case study of salamanders, we used estimates based on multiple nuclear exons to assess the effects of saturation on dating divergences using mitochondrial genome sequences on a timescale of ~200-300 My. The results indicated that, due to saturation for older divergences and in the absence of younger effective calibration points, dates derived from the mitochondrial data were considerably overestimated and systematically biased toward the calibration point for the ingroup root. The overestimate might be as great as 3-10 times (about 20 My) older than actual divergence dates for recent splitting events and 40 My older for events that are more ancient. For deep divergences, dates estimated were strongly compressed together. Furthermore, excluding the third codon positions of protein-coding genes or only using the RNA genes or second codon positions did not considerably improve the performance. In the order Caudata, slowly evolving markers such as nuclear exons are preferred for dating a phylogeny covering a relatively wide time span. Dates estimated from these markers can be used as secondary calibrations for dating recent events based on rapidly evolving markers for which mitochondrial DNA sequences are attractive candidates due to their short coalescent time. In other groups, similar evaluation should be performed to facilitate the choice of markers for molecular dating and making inferences from the results.