Modelling and Inferring Large-scale Demographic Fluctuations in Structured Populations Through Simulations and PSMC-based Methods.

The climatic oscillations of the Quaternary have likely affected the demographic history of many species, and PSMC (Pairwise Sequentially Markovian Coalescent) has been widely used to investigate these histories. However, it is increasingly acknowledged that PSMC trajectories are difficult to interpret. First, they are influenced by connectivity changes, even without population size changes. Second, most PSMC curves exhibit a few humps when tens of cycles occurred during the Pleistocene. Finally, responses to ancient habitat change have been shown to be species-specific. To address these issues, we simulated structured populations where connectivity (or population size and connectivity) varied according to successive interglacial and glacial periods during the last 2.6 million years. We computed the IICR (Inverse Instantaneous Coalescence Rate), the function that PSMC estimates, and ran PSMC. We further varied the generation length and assumed that some species were positively or negatively affected by glacials. We found that the IICR carries information regarding the demographic oscillations, but that PSMC fails to recover it for times older than 300 ky. For the last 200 ky, PSMC was often able to reproduce qualitatively the demographic oscillations. We also tested SNIF (Structured Non-stationary Inferential Framework), which produced good results using the IICR curve as an input but not when using the PSMC curve. Altogether, our study suggests that the humps older than 300 ky in PSMC histories are unlikely to represent trends of population size or connectivity. However, improving the estimation of the IICR could potentially help reconstruct some of these past demographic changes.