Nested clade phylogeographic analysis (NCPA) is a widely used method that aims to identify past demographic events that have shaped the history of a population. In an earlier study, NCPA has been fully automated, allowing it to be tested with simulated data sets generated under a null model in which samples simulated from a panmictic population are geographically distributed. It was noted that NCPA was prone to inferring false positives, corroborating earlier findings. The present study aims to evaluate both single-locus and multilocus NCPA under the scenario of restricted gene flow among spatially distributed populations. We have developed a new program, ANeCA-ML, which implements multilocus NCPA. Data were simulated under 3 models of gene flow: a stepping stone model, an island model, and a stepping stone model with some long-distance dispersal. Results indicate that single-locus NCPA tends to give a high frequency of false positives, but, unlike the random-mating scenario presented previously, inferences are not limited to restricted gene flow with isolation by distance or contiguous range expansion. The proportion of single-locus data sets that contained false inferences was 76% for the panmictic case, 87% for the stepping stone model, 79% for the stepping stone model with long-distance dispersal, and more than 99% for the island model. The frequency of inferences is inversely related to the amount of gene flow between demes. We performed multilocus NCPA by grouping the simulated loci into data sets of 5 loci. The false-positive rate was reduced in multilocus NCPA for some inferences but remained high for others. The proportion of multilocus data sets that contained false inferences was 17% for the panmictic case, 30% for the stepping stone model, 4% for the stepping stone model with long-distance dispersal, and 54% for the island model. Multilocus NCPA reduces the false-positive rate by restricting the sensitivity of the method but does not appear to increase the accuracy of the approach. Three classical tests-the analysis of molecular variance method, Fu's Fs, and the Mantel test-show that there is information in the data that gives rise to explicable results using these standard approaches. In conclusion, for the scenarios that we have examined, our simulation study suggests that the NCPA method is unreliable and its inferences may be misleading. We suggest that the NCPA method should not be used without objective simulation-based testing by independent researchers.
Read full abstract