Contrasting signals of positive selection in genes involved in human skin-color variation from tests based on SNP scans and resequencing

Applying all commonly used neutrality-test statistics for allele frequency distribution to the newly generated sequence data provided conflicting results regarding evidence for positive selection. Previous haplotype-based findings could not be clearly confirmed. Although some tests were marginally significant for some populations and genes, none of them were significant after multiple-testing correction. Combined P values for each gene-population pair did not improve these results. Application of Approximate Bayesian Computation Markov chain Monte Carlo based to these sequence data using a simple forward simulator revealed broad posterior distributions of the selective parameters for all four genes, providing no support for positive selection. However, when we applied this approach to published sequence data on SLC45A2, another human pigmentation candidate gene, we could readily confirm evidence for positive selection, as previously detected with sequence-based and some haplotype-based tests.Overall, our data indicate that even genes that are strong biological candidates for positive selection and show reproducible signatures of positive selection in SNP scans do not always show the same replicability of selection signals in other tests, which should be considered in future studies on detecting positive selection in genetic data.Large-scale genotyping projects using genome-wide single-nucleotide polymorphisms (SNPs) have provided large amounts of data describing the genetic diversity of human populations [1-6]. Several statistical methods have been developed and used for detection of signatures of selective processes from genome-wide SNP data, which we refer to as 'SNP scans' [7]. All these approaches try to recover fingerprints of selective sweeps by detecting signals in the haplotypic variation of a genomic region and/or the spectrum of the variation of the genetic diversity [8-15]. However, the results obtained with the different test statistics usually show limi