Bayesian Divergence-Time Estimation with Genome-Wide Single-Nucleotide Polymorphism Data of Sea Catfishes (Ariidae) Supports Miocene Closure of the Panamanian Isthmus

Abstract

The closure of the Isthmus of Panama has long been considered to be one of the best defined biogeographic calibration points for molecular divergence-time estimation. However, geological and biological evidence has recently cast doubt on the presumed timing of the initial isthmus closure around 3 Ma but has instead suggested the existence of temporary land bridges as early as the Middle or Late Miocene. The biological evidence supporting these earlier land bridges was based either on only few molecular markers or on concatenation of genome-wide sequence data, an approach that is known to result in potentially misleading branch lengths and divergence times, which could compromise the reliability of this evidence. To allow divergence-time estimation with genomic data using the more appropriate multispecies coalescent (MSC) model, we here develop a new method combining the single-nucleotide polymorphism-based Bayesian species-tree inference of the software SNAPP with a molecular clock model that can be calibrated with fossil or biogeographic constraints. We validate our approach with simulations and use our method to reanalyze genomic data of Neotropical army ants (Dorylinae) that previously supported divergence times of Central and South American populations before the isthmus closure around 3 Ma. Our reanalysis with the MSC model shifts all of these divergence times to ages younger than 3 Ma, suggesting that the older estimates supporting the earlier existence of temporary land bridges were artifacts resulting at least partially from the use of concatenation. We then apply our method to a new restriction-site associated DNA-sequencing data set of Neotropical sea catfishes (Ariidae) and calibrate their species tree with extensive information from the fossil record. We identify a series of divergences between groups of Caribbean and Pacific sea catfishes around 10 Ma, indicating that processes related to the emergence of the isthmus led to vicariant speciation already in the Late Miocene, millions of years before the final isthmus closure.

Figure 1.

Comparison of true and estimated node ages (experiment 1). Results are based on 100 species trees and 300 to 3 000 SNPs generated per species tree. a) Node ages estimated with an age constraint on the root. b) Node ages estimated with an age constraint on a node that is approximately a third as old as the root. Mean age estimates of constrained and unconstrained nodes are marked with red and gray circles, respectively, and vertical bars indicate 95% HPD intervals.

Figure 2.

Estimates of node ages, the clock rate, , and the population size, with and without ascertainment bias (experiment 4). Results are based on data sets of 1 000 SNPs generated for each of 100 species trees, analyzed with and without SNAPP’s ascertainment-bias correction or after adding invariant sites to the data sets. Gray circles indicate mean estimates and 95% HPD intervals are marked with vertical bars. The visualization of node-age estimates in a) is equivalent to the illustration in Fig. 1, except that only unconstrained nodes are shown. Note that logarithmic scales are used for estimates of the clock rate (a) and (b).

Figure 3.

Error in node-age estimates obtained with the MSC or with concatenation (experiment 5). Results are based on analyses of 100 data sets of 1 000 SNPs, simulated with population sizes , , and . Gray and red dots indicate node-age estimates obtained with the MSC implemented in SNAPP and with BEAST analyses of concatenated data, respectively. Node-age error is measured as the ratio of the estimated node age over the true node age. Solid lines represent mean node-age errors in bins of 0.2 myr. Only nodes with true ages up to 10 myr are shown to highlight differences between the two methods. Note that a logarithmic scale is used for node-age error.

Figure 4.

Time-calibrated species tree of Neotropical sea catfishes. a) Map of Panama and north-western South America with sampling locations of specimens used in this study. Colors of circles indicate genera of specimens sampled at a location: Ariopsis, red; Sciades, purple; Bagre, blue; Notarius, green; Cathorops, orange. b) Posterior distribution of time-calibrated species trees inferred with SNAPP, with fossil taxa added a posteriori (images of otoliths and partial skulls are from Aguilera and de Aguilera 2004b and from Aguilera et al. 2013, ; see Supplementary Table S6). Branch color indicates reconstructed geography: Caribbean; green, or Tropical Eastern Pacific (TEP); dark blue. Posterior densities of divergence times between Caribbean and Pacific lineages within Notarius (green), Bagre (blue), and Cathorops (orange) are shown below the species tree. Note that two divergence events around 10 Ma have nearly identical posterior density distributions: the divergence between N. grandicassis and N. biffi and the divergence between B. panamensis and the ancestor of B. bagre and B. marinus. Pie charts on nodes corresponding to divergences between Caribbean and Pacific lineages indicate posterior probabilities of ancestral distributions. All posterior estimates of node support, divergence times, and ancestral geography are summarized in Supplementary Table S7.