Assessing model fit in phylogeographical investigations: an example from the North American sandbar willow Salix melanopsis

Abstract

Aim Coalescent models enable the direct estimation of parameters with clear biological relevance (i.e. divergence time, migration rate and rate of expansion), but they have typically been applied to phylogeographical research without a priori assessment of their fit to the empirical system. Here we explore the extent to which phylogeographical inference can be misled by evaluating the fit of several population genetic models to empirical data collected from the sandbar willow, Salix melanopsis. Location The Pacific Northwest mesic forest of North America. Methods We collected sequence data from five loci in 145 individuals. We assessed model fit in: (1) models delimiting previously proposed races within S. melanopsis; (2) historical biogeographical models, each describing the timing and pattern of diversification; and (3) coalescent models that correspond to those implemented in popular software packages such as IMa, lamarc, and Migrate‐n. Results We found little evidence for previous hypotheses of cryptic races delimited by habitat type (mesic, lowland or subalpine); rather, our results suggested that these variants originated from the same source population. Historical biogeographical models demonstrate that S. melanopsis has recently expanded from a single refugial population, probably located in the northern Rocky Mountains. An analysis using approximate Bayesian computation indicated that the single population expansion model implemented in lamarc is a better fit to the data than multi‐population models incorporating migration and/or divergence as implemented in Migrate‐n and IMa, suggesting that the parameters estimated from the latter are potentially misleading for this system. Main conclusions Our research highlights the importance of assessing model fit in addition to estimating parameters to understand evolutionary processes. Taken together, they allow us to infer the historical demography of S. melanopsis in a manner that is not biased by previous work in the system.

Model selection as a tool for phylogeographic inference: an example from the willow Salix melanopsis

“…Rather than expecting to find a single refuge located in either of these drainages, it might be better to envision either multiple refugia in individual river canyons in both drainages or a refuge that encompassed regions of both drainages. It is also worth noting that in contrast to previous findings (Tsai & Carstens ), the ∂a∂i analysis did not find support for models that included population expansion. This finding could result from the smaller number of sampled individuals in this study, from the conservative SNP calling that may eliminate true SNPs that are present at low frequency, or be indicative that previous studies were errant.…”

“…Additionally, the Mantel test did not show a significant signal for isolation by distance among the samples ( P = 0.483), suggesting that factors other than simple geographic distances are influencing genetic structure in Salix melanopsis . This result is in contrast with the isolation‐by‐distance results in the study by Tsai & Carstens (), suggesting that this analysis is sensitive to differences in the number of samples, loci or polymorphism. The clustering algorithm implemented in Structure revealed genetic structure at several levels of K (Fig.…”

Multilocus tests of Pleistocene refugia and ancient divergence in a pair of Atlantic Forest antbirds (Myrmeciza)

“…; Beaumont ; Csilléry et al . ; Tsai & Carstens ). We focused on nuclear variation in the ABC analyses to avoid analytical difficulties related to the divergent rates of evolution of mitochondrial and nuclear sequences.…”

Estimating the temporal and spatial extent of gene flow among sympatric lizard populations (genusSceloporus) in the southern Mexican highlands

“…The modelled population histories are often generated based on inferences obtained from geological data (Carstens et al . ), paleoclimatic data (Spellman & Klicka ), or based on previous genetic studies (Tsai & Carstens ), and the parameterizations used in the models can be derived from estimates from empirical data (Carstens et al . ).…”