Multi-model inference in comparative phylogeography: an integrative approach based on multiple lines of evidence

Abstract

Comparative phylogeography has its roots in classical biogeography and, historically, relies on a pattern-based approach. Here, we present a model-based framework for comparative phylogeography. Our framework was initially developed for statistical phylogeography based on a multi-model inference approach, by coupling ecological niche modeling, coalescent simulation and direct spatio-temporal reconstruction of lineage diffusion using a relaxed random walk model. This multi-model inference framework is particularly useful to investigate the complex dynamics and current patterns in genetic diversity in response to processes operating on multiple taxonomic levels in comparative phylogeography. In addition, because of the lack, or incompleteness of fossil record, the understanding of the role of biogeographical events (vicariance and dispersal routes) in most regions worldwide is barely known. Thus, we believe that the expansion of that framework for multiple species under a comparative approach may give clues on genetic legacies in response to Quaternary climate changes and other biogeographical processes.

Keywords: Quaternary climatic changes; biogeography; coalescence; palaeodistribution modeling; vicariance.

FIGURE 1

Schematic representation for inferring phylogeographic processes using ecological niche modeling (ENM), coalescence simulation, and relaxed random walk (RRW) model. (A) Uncertainties from ENM methods, climatic simulations (AOGCMs), and other source of evidence (e.g., fossil record) are explored to set multiple distribution dynamics through time, from which alternative demographic hypotheses are inferred. Colored circles represent demes with distinct sizes and locations through time; these demographic scenarios (demographic growth, shrink, and spatial shift) may be simulated using coalescence analyses. LIG: last interglacial (∼125,000 years ago); LGM: last glacial maximum (∼21,000 year ago); Pres: present; N_LGM and N₀: effective population size at LGM and present, respectively. (B) An ensemble from multiple ENMs provides the environmental context to infer range shifts and perform spatial analyses considering genetic variation across populations (pie charts representing haplotype distributions). Here, maps are representing a geographic range shift of suitable areas toward the northeastern South America from LGM to present-day, which may explain the current spatial pattern of genetic diversity; i.e., increases in climatic suitability across last glacial cycle support higher haplotype diversity at present-day. (C) Coalescent framework used to select the most likely demographic scenario matching the empirical genetic parameters. Demographic hypotheses in a spatially explicit context through time (H1, H2, and H3; colored circles represent the population dynamics trough time) are simulated using coalescent framework to investigate their consequent population genetic structure. Models are selected from multiple criteria, such as likelihood based on posterior predictive distribution and akaike information criterion (AIC). (D) Relaxed random walk (RRW) modeling used to predict historical dispersal routes at the time scale of molecular sequencing data. Saving the time scaling from each model, the dispersal patterns are comparable to the ENM predictions for range shifts (see B). Considering the distribution map from figure, the RRW predicts intermittent dispersal routes through the time in a similar direction of range shift predicted from ENMs. The evidence from multiple models indicates that the dispersal routes predicted by RRW could be the result of climatic forcing across sequential glaciations. Moreover, RRW provides support to explore other features of population dynamics (e.g., source and sink of migrants) and biogeographic processes (e.g., dispersal barriers). Representation of Markov Chain was adapted from Professor Peter Beerli Lecture Notes (http://evolution.gs.washington.edu).

FIGURE 2

Schematic view of the generalized framework for coupling ecological niche modeling, coalescent simulation, and diffusion model in a statistical comparative phylogeography approach. Palaeodistribution maps resulting from ENMs are used to generate demographic hypotheses (see Figure 1) in a spatially explicit context through time (hypotheses H1 and H2; colored circles represent the population dynamics), according to the range shifts in species distributions in the past (e.g., range retraction or range stability). Other hypotheses may also be set based on fossil records (hypothesis H3) or a priori biogeographic scenario. Uncertainties in setting alternative hypotheses can be incorporated into the framework using several ENM methods and projecting distributions through different AOGCMs. In a next step, simulated coalescence structures are compared with observed data in a model selection approach, allowing selecting among the demographic hypothesis (H1, H2, or H3) the most likely to generate the current phylogeographic structure derived from molecular data. At the same time, phylogeographic diffusion models allow reconstructing colonization routes that are compared with palaeodistribution maps. Finally, this framework can be expanded into a multi-species comparative approach allowing inferring how whole assemblages responded to the interplay between climate changes, geographic barriers, and demographic processes, shaping the current patterns of species distribution, and biodiversity. Coalescent time for each species can be compared using, for instance, Approximate Bayesian Computation implemented in MTML-msBayes. Representation of Markov Chain was adapted from Professor Peter Beerli Lecture Notes (http://evolution.gs.washington.edu).