December 02, 2011

Natural selection in African Americans pre- and post-admixture

I have mentioned before that African Americans should not be used to generalize about Africa, not only because of their ~20% European admixture, but also because they live in an environment completely different from the one their African ancestors adapted to: different climate, different set/intensity of pathogens, different social position, different physical requirements and workloads. It is nice to see a paper which attempts to quantify pre- and post-admixture signals of selection in this population; I think this may be a fertile area of future research, and it may also illuminate some of the specificities of the AA population.

Genome Research doi:10.1101/gr.124784.111

Genome-wide detection of natural selection in African Americans pre-and post-admixture

Wenfei Jin et al.

It is particularly meaningful to investigate natural selection in African Americans (AfA) due to the high mortality their African ancestry has experienced in history. In this study, we examined 491,526 autosomal SNPs genotyped in 5,210 individuals and conducted a genome-wide search for selection signals in 1,890 AfA. Several genomic regions showing excess of African or European ancestry, which were thought as the footprints of selection since population admixture, were detected based on a commonly used approach. However, we also developed a new strategy to detect natural selection both pre-and post-admixture by reconstructing an ancestral African population (AAF) from inferred African components of ancestry in AfA and comparing it with indigenous African populations (IAF). Interestingly, many selection-candidate genes identified by the new approach were associated with AfA specific high-risk diseases such as prostate cancer and hypertension, suggesting an important role these disease-related genes might have played in adapting to new environment. CD36 and HBB, whose mutations confer a degree of protection against malaria, were also located in the highly differentiated regions between AAF and IAF. Further analysis showed that the frequencies of alleles protecting against malaria in AAF were lower than that in IAF, which consists with the relaxed selection pressure of malaria in the New World. There is no overlap between the top candidate genes detected by the two approaches, indicating the different environmental pressures AfA experienced pre-and post-population-admixture. We suggest that the new approach is reasonably powerful and can also be applied to other admixed populations such as Latinos and Uyghurs.



Andrew Oh-Willeke said...

A small point. "AA" as an abbreviation for "African-American" is ambiguous, because it is used at least as commonly as an abbreviation for "Asian American." Hence, the author's preference for "AfA."

On the merits, this is a very good sample size for this very specific subpopulation (almost as big as the entire global number of human autosomal genomes ever sequenced just a few years ago), which we know from prior studies doesn't have much substructure within its AAF component with a rich number of data points per individual within the sample, so it has very good statistical power and the sample is large enough that even a big increase in the sample size, would increase the statistical significance of the results only marginally.

Further, the study design is really quite clever and elegant. It is measuring genotype change in a very direct way and as well understood context. The differences in the environments that would give rise to new selective pressures are well documented. The parameters of the demographic history and source populations can be determined accurately from historical records and can be coroborated with prior ancestry determination oriented studies of uniparental and autosomal markers.

Arguably, this is one of the best designed studies of evolutionary selection at a genotype level in humans over a time frame of a few hundred years ever conducts - there are good studies for much longer time frame, and there are good studies for single generations (although without nearly the sample size), but I'm not aware of any other really well designed studies over this intermediate time frame.

In addition to the parallel studies of populations admixed in the historic era which are suggested in the abstract, the other obvious way to extend this very solid piece of work on genotypes would be to assemble a data base on phenotypes that can be compared with, particularly, the SNPs where we see selective effects. That methodology could yield a real Rosetta stone for the human genome.

Pascvaks said...

"It is particularly meaningful to investigate natural selection in African Americans (AfA) due to the high mortality their African ancestry has experienced in history. In this study, we examined 491,526 autosomal SNPs genotyped in 5,210 individuals and conducted a genome-wide search for selection signals in 1,890 AfA."

If you're happy, I'm happy. But... is such genetic information so common and readily available to these authors? Can't access the paper to see where they were (Shanghai?) for the study or where they got all this genetic data on AfA and AAF, IAF (some UN Agency?). I guess if they were in China and saying something about Chinese genetics I wouldn't lift an eyebrow. And, oh yes, the opening sentence, it smacks of some sociology intro on "slavery" and "the American historical guilt" for it all. Anyway, I bow to the experts here at this blog.

PS: "the high mortality..experienced in history"??? -- I was under the impression we all died, eventually.

Anonymous as well said...

Diekenes, I'm surprised you let this one slip through. While selection has been established within Africa at the HBB locus, there are separate and distinct haplotypes carrying the causal locus (SNP rs334, protective against malaria, which is not included on the genotyping panel studied). West African populations have incredibly differentiated haplotypes in this region, since rs334 arose independently multiple times, and within each region, increased the associated haplotype in frequency locally.

Hence, using the wrong (or insufficient) ancestral population for local ancestry estimation will not capture the ancestry at this locus correctly, leading to a mis-estimation of African/European ancestry proportions here. The authors only use Yoruba for local ancestry estimation, which is a good, but insufficient proxy for this level of precision. Such ancestry estimation must be exact and unbiased to detect the *extremely* small levels of differentiation resulting from selection after admixture.

I suspect that all the other results the authors obtain are, likewise, spurious signals resulting from mis-specification of ancestral populations. Mostly, I am saddened that the authors were never asked to show that their method is unbiased on any test example where ancestral proxies were inexact. Though the conclusions sound really cool, I'm afraid they aren't providing any real evidence that can't be explained by poor local ancestry estimation.