The comparative biogeography of Philippine geckos challenges predictions from a paradigm of climate-driven vicariant diversification across an island archipelago

Abstract

A primary goal of biogeography is to understand how large-scale environmental processes, like climate change, affect diversification. One often-invoked but seldom tested process is the so-called "species-pump" model, in which repeated bouts of co-speciation is driven by oscillating climate-induced habitat connectivity cycles. For example, over the past three million years, the landscape of the Philippine Islands has repeatedly coalesced and fragmented due to sea-level changes associated with the glacial cycles. This repeated climate-driven vicariance has been proposed as a model of speciation across evolutionary lineages codistributed throughout the islands. This model predicts speciation times that are temporally clustered around the times when interglacial rises in sea level fragmented the islands. Given the significance and conceptual impact the model has shown, surprisingly few tests of this prediction have been provided. We collected comparative genomic data from 16 pairs of insular gecko populations to test the prediction of temporally clustered divergences. Specifically, we analyze these data in a full-likelihood, Bayesian model-choice framework to test for shared divergence times among the pairs. Our results provide support against the species-pump model prediction in favor of an alternative interpretation, namely that each pair of gecko populations diverged independently. These results suggest the repeated bouts of climate-driven landscape fragmentation has not been an important mechanism of

Insights from a general, full-likelihood Bayesian approach to inferring shared evolutionary events from genomic data: Inferring shared demographic events is challenging

“…Thus, violating the assumption of unlinked sites has little affect on the estimation of the timing and sharing of demographic changes or the effective population sizes. This is consistent with the findings of Oaks (2019) and Oaks et al (2019b) that linked sites had little impact on the estimation of shared divergence times. These results suggest that analyzing all of the sites in loci assembled from reducedrepresentation genomic libraries (e.g., sequence-capture or RADseq loci) is a better strategy than excluding sites to avoid violating the assumption of unlinked characters.…”

“…If this lack of realism is problematic, it should cause the method to overestimate the number of events by misidentifying the temporal variation among species affected by the same process as being the result of multiple events. However, what we see here (e.g., Figure 7) and what has been shown previously (Oaks et al, 2013Oaks, 2014Oaks, , 2019Oaks et al, 2019b) is the opposite; even when we model shared events as simultaneous, methods tend to underestimate (and almost never overestimate) the number of events. We do see overestimates when there is little information in the data and the posterior largely reflects the prior (e.g., bottom two rows of Figure 3).…”

“…We also see that the accuracy and precision of estimates of the timing of a demographic change event do not increase with the number of populations that share the event ( Figure S9). This makes sense for two reasons: (1) it is difficult to correctly identify the sharing of demographic events among populations (Row 2 of Figure S8), and (2) Oaks (2019) and Oaks et al (2019b) showed that the amount of information about the timing of events plateaus quickly as the number of characters increases. Thus, given 500,000 characters from each population, little information is to be gained about the timing of the demographic change, even if the method can correctly identify that several populations shared the same event.…”

“…However, this is only true when the prior distributions match the true underlying distributions that generated the data, and these overestimates would be easy to identify in practice by testing for prior sensitivity and noticing that the posterior probabilities of event models are similar to the prior probabilities (i.e., small Bayes factors). Furthermore, Oaks et al (2019b) showed that even with millions of bases of genomic data from pairs of gecko populations, ecoevolity was only able to detect differences in divergence times between comparisons greater than several thousand years. Thus, it seems unlikely that over-estimating the number of events among taxa (i.e., estimating temporal independence of comparisons that shared the same historical process) is a real problem for these types of inferences.…”

Impacts of the Toba eruption and montane forest expansion on diversification in Sumatran parachuting frogs (Rhacophorus)

“…To better understand the variability in posterior estimates between data sets, we used 'simcoevolity' v.0.2.4 (Oaks et al, 2019b) within the pycoevolity package to simulate 500 data sets using the '-c' option, which simulates alignments that matched our empirical data for numbers of individuals, locus number, locus length, and patterns of missing data within each locus. This strategy allowed us to identify if differences between estimates from the datasets were due to differences in the number of sites or loci sampled, patterns of missing data within each locus.…”