Demographic inference under the coalescent in a spatial continuum

Abstract

Understanding population dynamics from the analysis of molecular and spatial data requires sound statistical modeling. Current approaches assume that populations are naturally partitioned into discrete demes, thereby failing to be relevant in cases where individuals are scattered on a spatial continuum. Other models predict the formation of increasingly tight clusters of individuals in space, which, again, conflicts with biological evidence. Building on recent theoretical work, we introduce a new genealogy-based inference framework that alleviates these issues. This approach effectively implements a stochastic model in which the distribution of individuals is homogeneous and stationary, thereby providing a relevant null model for the fluctuation of genetic diversity in time and space. Importantly, the spatial density of individuals in a population and their range of dispersal during the course of evolution are two parameters that can be inferred separately with this method. The validity of the new inference framework is confirmed with extensive simulations and the analysis of influenza sequences collected over five seasons in the USA.

Coalescent inferences in conservation genetics: should the exception become the rule?

“…Allowing for local bottlenecks and longdistance jumps, the spatial-L-coalescent can recover both small local N e and long-distance correlated genealogies deriving from long-distance dispersal events [27,28]. Without needing to assume discrete demes or homogeneous population distribution, this new framework has been shown to predict very well local and global N e values when classic F ST measures otherwise are largely uncorrelated to observed values [26][27][28][29].…”

“…At the state of the art, some MMCs maximum-likelihood estimators have been developed and are available to infer the effective population size and skewness of the offspring distribution of marine species [11,25,30], such as MetaGeneTree [17] (table 1). A recent software based on spatial-L-coalescent (PhyREX) by Guindon et al [29] estimates global N e values in continuous space as an alternative to classic F ST estimates. Moreover, two MMCs simulators are currently available: algorithms by Kelleher et al for continuous space evolution [28] and Hybrid-Lambda for species evolution [31], which could be used to fit evolutionary hypotheses to observations using simulation approaches (table 1).…”

Disentangling genetic structure for genetic monitoring of complex populations

“…Currently, the only model with immediate potential to address most of the requirements for long‐term genetic monitoring is the spatial Λ‐Fleming‐Viot (SLFV) model (Barton, Etheridge, & Véber, ; Guindon, Guo, & Welch, ; Joseph, Hickerson, & Alvarado‐Serrano, ; Kelleher, Barton, & Etheridge, ). The SLFV is a spatially explicit extension of the ‐Fleming–Viot model which is itself an extension of the Fleming–Viot model (Fleming & Viot, ).…”

“…In the simplest case (Kelleher et al., ), the family of distributions E ( x ) for a d + 1 dimensional landscape L is composed of uniform distributions within d ‐spheres of radius r centered at points e . Alternatively, a Gaussian distribution for the selection has been used (Guindon et al., ). Nonhomogeneity in the landscape can be incorporated with different families of E ( x ), which might, for example, depend on the distribution of habitats, land use patterns, other environmental characteristics, or the state (genetic or demographic) of the individuals. Select a set of individuals based upon the spatial distribution E ( x ).…”

“…() compares the efficiency of alternative algorithms for accomplishing this. In the simplest cases, is uniform across the d ‐sphere defined by E ( x ) (Kelleher et al., ) or may only depend on the distance between individuals (Guindon et al., ). However, more complex distributions can depend on the locations of individuals in , on environmental characteristics across L , or on individual states.…”