A more recently developed suite of methods for quantifying dispersal from genetic data is based on what are known as assignment tests. These assign individuals to their most likely population of origin by comparing their genotypes to the genetic profiles of various populations. This approach differs from the indirect methods of assessing gene flow (Nem) because it identifies individuals that have dispersed from their natal population, as opposed to comparing overall genetic similarities between populations. The original assignment tests use a maximum likelihood method to calculate the probabilities that a given genotype arose from alternative populations based on the allele frequencies in those populations (Paetkau et al.,
1995). Individuals are then assigned to the population from which they have the highest probability of originating, unless all of the probabilities are low, in which case the true natal population may not have been sampled. This approach assumes that all populations are in Hardy-Weinberg equilibrium and that the loci being characterized are in linkage equilibrium.
In recent years, various modifications have been made to improve the accuracy of assignment tests. One refinement has been the application of a Bayesian method (Pritchard, Stephens and Donnelly, 2000; Wilson and Rannala, 2003). The Bayesian approach to statistical inference is based on subjective statements of probability. Data are assumed to be fixed, and prior information is used to test the likelihood that various parameters can explain the data. Unlike classical statistics, which typically provide a single P value, Bayesian statistics often provide multiple probabilities, and this means that numerous scenarios (such as several candidate populations of origin) can be compared simultaneously. Simulation studies and applications to actual data sets have found that the Bayesian approach often outperforms the original frequency-based method (Cornuet et al., 1999; Manel, Berthier and Luikart, 2002), although this approach does assume that in each case the true population of origin has been sampled, which may not be the case.
Assignment tests generally perform better when detailed genetic profiles of populations are obtained from a relatively large number of individuals using multiple (up to 20) highly polymorphic loci. For this reason, assignment tests have been most commonly based on microsatellite data, although they can also be applied to allozyme, microsatellite, RAPD and RFLP data. A number of simulation studies have suggested that these tests work best when candidate populations are genetically distinct from each other (Cornuet et al., 1999), in which case dispersal must be relatively infrequent, although a recent study on the grand skink (Oligosoma grande) demonstrated that these conditions need not always be met. The grand skink is a large territorial lizard that lives in New Zealand in groups of around 20 that inhabit free-standing rock outcrops separated by 50-150 m of inhospitable vegetation. Mark--recapture studies have revealed frequent dispersal among sites, but assignment tests still managed to correctly assign between 65 and 100 per cent of migrant individuals to their natal population (depending on which assignment method was used), even when FST values were as low as 0.04 (Berry, Tocher and Sarre, 2004).
Although assignment tests often provide valuable information, their performance does depend on a number of variables, including the extent of population differentiation, the sample size of individuals and loci, and the variability of markers. Under some conditions they may be rather imprecise, as illustrated by a study of Montana wolverines that used four different methods of Bayesian and frequency-based assignment tests to investigate dispersal. A total of 89 individuals were genotyped and, although 25 of these were classified as migrants according to at least one method, only nine individuals were classified as migrants by all four methods (Cegelski, Waits and Anderson, 2003). Nevertheless, assignment tests have tremendous potential as a method for tracking dispersal and it is likely that future refinements will increase their use. We will return to assignment tests in later chapters when we look at some specific applications of these methods to questions in behavioural ecology and wildlife forensics.
Was this article helpful?