Madagascar’s Central Highlands are largely composed of grasslands, interspersed with patches of forest. The pre-human extent of these grasslands is a topic of vigorous debate, with conventional wisdom holding that they are anthropogenic in nature and emerging evidence supporting that grasslands were a component of the pre-human Central Highlands vegetation. Here, we shed light on the temporal dynamics of Madagascar’s vegetative composition by conducting a population genomic investigation of Goodman’s mouse lemur (Microcebus lehilahytsara; Cheirogaleidae). These small-bodied primates occur both in Madagascar’s eastern rainforests and in the Central Highlands, which makes them a valuable indicator species. Population divergences among forest-dwelling mammals can serve as a proxy for habitat fragmentation and patterns of post-divergence gene flow can reveal potential migration corridors consistent with a wooded grassland mosiac. We used RADseq data to infer phylogenetic relationships, population structure, demographic models of post-divergence gene flow, and population size change through time. These analyses offer evidence that open habitats are an ancient component of the Central Highlands, and that wide-spread forest fragmentation occurred naturally during a period of decreased precipitation near the last glacial maximum. Models of gene flow suggest that migration across the Central Highlands has been possible from the Pleistocene through the recent Holocene via riparian corridors. Notably, though our findings support the hypothesis that Central Highland grasslands predate human arrival, we also find evidence for human-mediated population declines. This highlights the extent to which species imminently threatened by human-mediated deforestation may be more vulnerable from paleoclimatic changes.

Implementation of the coalescent model in a Bayesian framework is an emerging strength in genetically based species delimitation studies. By providing an objective measure of species diagnosis, these methods represent a quantitative enhancement to the analysis of multilocus data, and complement more traditional methods based on phenotypic and ecological characteristics. Recognized as two species 20 years ago, mouse lemurs (genus Microcebus) now comprise more than 20 species, largely diagnosed from mtDNA sequence data. With each new species description, enthusiasm has been tempered with scientific scepticism. Here, we present a statistically justified and unbiased Bayesian approach towards mouse lemur species delimitation. We perform validation tests using multilocus sequence data and two methodologies: (i) reverse-jump Markov chain Monte Carlo sampling to assess the likelihood of different models defined a priori by a guide tree, and (ii) a Bayes factor delimitation test that compares different species-tree models without a guide tree. We assess the sensitivity of these methods using randomized individual assignments, which has been used in bpp studies, but not with Bayes factor delimitation tests. Our results validate previously diagnosed taxa, as well as new species hypotheses, resulting in support for three new mouse lemur species. As the challenge of multiple researchers using differing criteria to describe diversity is not unique to Microcebus, the methods used here have significant potential for clarifying diversity in other taxonomic groups. We echo previous studies in advocating that multiple lines of evidence, including use of the coalescent model, should be trusted to delimit new species.