Exploring the Effects of Ecological Parameters on the Spatial Structure of Genealogies

Abstract

Space is a fundamental dimension of evolutionary change, determining how individuals disperse and interact with each other. Consequently, space has an important influence on the structure of genealogies and the distribution of genetic variants over time. Recently, the development of highly flexible simulation tools and computational methods for genealogical inference has greatly increased the potential for incorporating space into models of population genetic variation. It is now possible to explore how spatial ecological parameters can influence the distribution of genetic variation among individuals with unprecedented detail. In this study, we explore the effects of three specific parameters (the dispersal distance, competition distance and mate choice distance) on the spatial structure of genealogies relating individuals across a genome. We carry out a series of in silico experiments using forwards-in-time simulations, to determine how these parameters influence the distance between closely- and distantly-related individuals in these genealogies. We also assess the accuracy of the maximum likelihood estimation of the dispersal distance in a Gaussian model of dispersal from tree-sequence data. Our results showcase the potential for linking phylogeography, population genetics and ecology, in order to answer fundamental questions about the nature of spatial interactions across a landscape.

Inferring the geographic history of recombinant lineages using the full ancestral recombination graph

“…Consistent with previous work (e.g., Ianni-Ravn et al ., 2023; Kalkauskas et al ., 2021), the dispersal estimate from the composite likelihood over trees underestimates the true simulated value (blue curve in Figure 3A), due to habitat boundaries. As we increase the number of trees this estimate, the average dispersal estimate over all trees, asymptotes.…”

“…Further, our simple model of Brownian motion assumes an unbounded space. As a result of the latter, dispersal estimates from trees generated in finite space show a systematic downward bias (Figure 3A and Ianni-Ravn et al ., 2023; Kalkauskas et al ., 2021). Modeling Brownian motion down ARGs includes an additional approximation, the meeting of lineages at recombination nodes.…”

“…There are multiple reasons why we might want to focus on just the recent past. First, long-term movement may not be accurately captured by Brownian motion due to geographic boundaries or large-scale population movements (Ianni-Ravn et al ., 2023). Second, as we move further back into the past, sample lineages can become spatially well mixed (Wakeley, 1999).…”

Estimating dispersal rates and locating genetic ancestors with genome-wide genealogies

“…Our method systematically underestimates simulated dispersal rates (Figure 2A), as expected given that our simple unbounded Brownian motion model allows the samples to be more broadly distributed than the finite habitat (e.g., Ianni-Ravn et al, 2023;Kalkauskas et al, 2021). Ignoring the distant past tends to reduce this underestimate (see T = 1000 in Figure 2A), but ignoring too much of the past increases noise and can lead to even larger underestimates (see T = 100 in Figure 2A).…”

“…A simple estimate of the dispersal rate is then the slope of the regression of the squared pairwise geographic distances, d 2 ij , on the average coalescent times, t ij . This gives a dispersal estimate of roughly 10 km 2 /generation, as does taking the average of d 2 ij /t ij over pairs (Ianni-Ravn et al, 2023). Replacing the pairwise coalescent times, t ij , with pairwise nucleotide diversity divided by the mutation rate (π ij /(7 × 10 −9 )) gives a near identical estimate that is independent of our inferred trees.…”