Spatial autocorrelation analysis and ecological niche modelling allows inference of range dynamics driving the population genetic structure of a Neotropical savanna tree

Many empirical studies have assessed fine-scale spatial genetic structure (SGS), i.e. the nonrandom spatial distribution of genotypes, within plant populations using genetic markers and spatial autocorrelation techniques. These studies mostly provided qualitative descriptions of SGS, rendering quantitative comparisons among studies difficult. The theory of isolation by distance can predict the pattern of SGS under limited gene dispersal, suggesting new approaches, based on the relationship between pairwise relatedness coefficients and the spatial distance between individuals, to quantify SGS and infer gene dispersal parameters. Here we review the theory underlying such methods and discuss issues about their application to plant populations, such as the choice of the relatedness statistics, the sampling scheme to adopt, the procedure to test SGS, and the interpretation of spatial autocorrelograms. We propose to quantify SGS by an 'Sp' statistic primarily dependent upon the rate of decrease of pairwise kinship coefficients between individuals with the logarithm of the distance in two dimensions. Under certain conditions, this statistic estimates the reciprocal of the neighbourhood size. Reanalysing data from, mostly, published studies, the Sp statistic was assessed for 47 plant species. It was found to be significantly related to the mating system (higher in selfing species) and to the life form (higher in herbs than trees), as well as to the population density (higher under low density). We discuss the necessity for comparing SGS with direct estimates of gene dispersal distances, and show how the approach presented can be extended to assess (i) the level of biparental inbreeding, and (ii) the kurtosis of the gene dispersal distribution.

Population genetic theory predicts that plant populations will exhibit internal spatial autocorrelation when propagule flow is restricted, but as an empirical reality, spatial structure is rarely consistent across loci or sites, and is generally weak. A lack of sensitivity in the statistical procedures may explain the discrepancy. Most work to date, based on allozymes, has involved pattern analysis for individual alleles, but new PCR-based genetic markers are coming into vogue, with vastly increased numbers of alleles. The field is badly in need of an explicitly multivariate approach to autocorrelation analysis, and our purpose here is to introduce a new approach that is applicable to multiallelic codominant, multilocus arrays. The procedure treats the genetic data set as a whole, strengthening the spatial signal and reducing the stochastic (allele-to-allele, and locus-to-locus) noise. We (i) develop a very general multivariate method, based on genetic distance methods, (ii) illustrate it for multiallelic codominant loci, and (iii) provide nonparametric permutational testing procedures for the full correlogram. We illustrate the new method with an example data set from the orchid Caladenia tentaculata, for which we show (iv) how the multivariate treatment compares with the single-allele treatment, (v) that intermediate frequency alleles from highly polymorphic loci perform well and rare alleles poorly, (vi) that a multilocus treatment provides clearer answers than separate single-locus treatments, and (vii) that weighting alleles differentially improves our resolution minimally. The results, though specific to Caladenia, offer encouragement for wider application.