NEW YORK — Polygenic risk scores derived from African Americans work better in sub-Saharan African individuals than scores developed in Europeans, a new study has found, but their performance varies between African populations.
Most genome-wide association studies have relied on cohorts of largely European ancestry, so polygenic risk scores developed from those studies do not perform as well in genetically diverse populations, including those in Africa.
"Lack of precise [genetic risk scores] in Africans hinders risk stratification and targeted treatments essential for precision medicine and may exacerbate health disparities," co-senior author Segun Fatumo from the African Computational Genomics Research Group in Uganda and colleagues wrote in their new paper, appearing Thursday in Nature Medicine.
In their new analysis, the researchers examined how polygenic risk scores for lipid traits derived from African-American or multiethnic populations performed among sub-Saharan Africans, as compared to scores generated from populations of European ancestry. While the scores developed from African Americans performed better overall, how well they worked differed between South African Zulu and Ugandan cohorts.
For their study, the scientists developed polygenic risk scores for various lipid traits using summary statistic data from the Million Veteran Program, a research effort launched in 2011 to collect genetic and other data from 1 million US veterans. They generated risk scores for lipid traits using African, European, and multi-ancestry populations, and then applied those scores to two cohorts of sub-Saharan African individuals.
Within the South African Zulu cohort, the best-performing risk score for low-density lipoprotein cholesterol was derived from the African-American population, followed by the one derived through the multi-ancestry population — which included individuals of African, European, and Hispanic ancestry —and then the European population-derived score.
The African American-derived risk score for LDL-C, though, performed worse in the Ugandan cohort — but still better than the other scores. Among the South African Zulu cohort, the score explained 8.14 percent of the variance seen in that trait, but it only explained 0.026 percent of the variance seen among Ugandans.
Using total cholesterol levels as an example, the researchers examined the transferability of the risk scores. Similar to their LDL-C findings, the African American-derived risk score — consisting of 286 SNPs — performed poorly in the Ugandan cohort, but better in the South African Zulu cohort.
They compared the allele frequencies of the SNPs in the score between the Ugandan and South African Zulu cohorts and found differences in their frequency, as well as in age and body mass index between the cohorts that may affect transferability. The researchers also noted that the South African Zulu cohort was recruited in an urban setting, while the Ugandan cohort was mostly rural, suggesting that both genetic and environmental factors could influence polygenic risk scores performance.
"The poor performance of [genetic risk scores] within the same ancestry population hinders the implementation of [genetic risk scores] in preventative healthcare. It may lead to inaccurate results when applied to different ethnic groups within sub-Saharan Africa," Fatumo and colleagues wrote. "This further suggests the need for more efforts to optimize polygenic prediction in Africa."