Combining evidence of natural selection with association analysis increases power to detect malaria-resistance variants

Abstract

Statistical power to detect disease variants can be increased by weighting candidates by their evidence of natural selection. To demonstrate that this theoretical idea works in practice, we performed an association study of 10 putative resistance variants in 471 severe malaria cases and 474 controls from the Luo in Kenya. We replicated associations at HBB (P=.0008) and CD36 (P=.03) but also showed that the same variants are unusually differentiated in frequency between the Luo and Yoruba (who historically have been exposed to malaria) and the Masai and Kikuyu (who have not been exposed). This empirically demonstrates that combining association analysis with evidence of natural selection can increase power to detect risk variants by orders of magnitude--up to P=.000018 for HBB and P=.00043 for CD36.

Figure 1.

Principal-components analysis of samples from four different populations genotyped for 1,454 SNPs. The first eigenvector clusters the Yoruba, Luo, Kikuyu, and Masai. This is contrary to the expectation based on linguistics (the Luo and Masai both speak Nilotic languages) or geography. The F_ST values estimated between pairs of populations are as follows: Yoruba-Luo, 0.008; Yoruba-Masai, 0.021; Yoruba-Kikuyu, 0.015; Luo-Kikuyu, 0.008; Luo-Masai, 0.011; and Kikuyu-Masai, 0.0012. The top eigenvector is highly statistically significant by principal-components analysis (P≪10^-12). The second eigenvector is not significant (P=.09).