Purifying selection and measures of population differentiation

From the paper:

Because allele frequencies are known to be lower at positions under stronger purifying selection (Subramanian and Kumar 2006), we examined the relationship between population differentiation estimates and minor allele frequencies (MAF). They are also highly positively correlated (Fig. 3A; P less than 10-15). 

... 

Our results suggest that interpreting and comparing results from population genomic studies now should consider this dependence of FST on the frequency of the allele as well as the functional importance (evolutionary rate) of the position. For example, estimates of FST at sites on the Y chromosome and at sites in the mitochondrial genome are sometimes compared to detect the difference in male vs. female migration rates (e.g., Seielstad et al. 1998). In such examinations, we now need to compare FST at sites with similar MAF across populations to detect the difference in migration rates. It is also important when we compare FST at sites among different populations. For example, African populations are known to have higher MAF across populations than non-African populations (Tishkoff and Kidd, 2004). Without consideration of MAF before comparing FST estimates at different positions in the two may lead to incorrect inference of higher degree of population differentiation among one set of populations as compared to the other set of populations.   

I can't say that I am very comfortable with the technical details of this paper, but from my first reading, it would appear that African populations have higher minor allele frequency (MAF) than non-African ones, so measures of population divergence for Africans may be inflated relative to non-Africans, and be reflective of lower levels of purifying selection in African populations.

This would of course have monumental implications for the reconstruction of human prehistory, as it would suggest that high Fst values between different African populations and between Africans and Eurasians may be, at least partially, driven by lower levels of purifying selection in the former. It is also possible, as I've suggested, that excess variation (manifesting itself as higher minor allele frequency) may be the result of admixture between divergent Homo populations.

Biology is a mess, but hopefully statistical geneticists are up to the task!

Mol Biol Evol (2012) doi: 10.1093/molbev/mss187

Purifying selection modulates the estimates of population differentiation and confounds genome-wide comparisons across single nucleotide polymorphisms

Takahiro Maruki et al.

An improved understanding of the biological and numerical properties of measures of population differentiation across loci is becoming increasingly more important because of their growing use in analyzing genome-wide polymorphism data for detecting population structures, inferring the rates of migration, and identifying local adaptations. In a genome-wide analysis, we discovered that the estimates of population differentiation (e.g., FST, θ, and Jost’s D) calculated for human single nucleotide polymorphisms (SNPs) are strongly and positively correlated to the position-specific evolutionary rates measured from multispecies alignments. That is, genomic positions (loci) experiencing higher purifying selection (lower evolutionary rates) produce lower values for the degree of population differentiation than those evolving with faster rates. We show that this pattern is completely mediated by the negative effects of purifying selection on the minor allele frequency at individual loci. Our results suggest that inferences and methods relying on the comparison of population differentiation estimates (FST, θ, and Jost’s D) based on SNPs across genomic positions should be restricted to loci with similar minor allele frequencies and/or the rates of evolution in genome-scale surveys.

Link