For years, companies have offered mail-in genetic tests that promise personalized health insights. With one simple swab, consumers are told they can learn about a host of traits, including their genetic predisposition to certain diseases. Yet, for many complex traits, the story isn’t so straightforward. Untangling the roles of genes and environment remains a challenge, and new research from Arslan Zaidi, an assistant professor in the Department of Genetics, Cell Biology and Development, reveals that the standard models for assessing heritability don’t work as well as assumed, especially in genetically diverse populations.
“The problem is that a lot of the methods that estimate heritability make various assumptions about the population,” says Zaidi, whose team recently published the findings in Genetics. “Primarily, the models assume that the population is homogeneous and randomly mating, but that’s not how humans behave.”
Geneticists rely on a concept called SNP heritability to study complex traits that are influenced by many, often unknown, genes. Short for single-nucleotide polymorphism, SNP refers to a single-letter difference in DNA that serves as a genetic marker. Even if a particular SNP doesn’t directly cause a trait, it might sit near a mutation that does. By analyzing hundreds of thousands of these markers across the genome in large groups of people, scientists can estimate how much of the variation in a trait, such as diabetes risk or high blood pressure, can be attributed to genetic differences rather than environmental factors.
Zaidi, his former graduate student Jinguo Huang, who has since earned her doctorate at Pennsylvania State University, and his current graduate student, Nicole Kleman, a doctoral student in Bioinformatics and Computational Biology at the University of Minnesota, encountered problems with three leading computational models used to estimate SNP heritability. Because the techniques work indirectly, in cases when underlying genes have not been identified, the models can miss important signals and produce biased results. The team found their performance was particularly unreliable when the models were applied to recently mixed communities, such as African American and Latino populations.
At first, they suspected the discrepancies might stem from errors in their own analyses, but after two years of work, combining complex theoretical mathematics with large-scale computer simulations, they confirmed that the flaws are inherent to the models. “These methods were not doing what they’re supposed to be doing in some situations,” Zaidi says. “For example, it’s very difficult to say that differences between populations in the trait average are driven by a difference in average genetic risk.”
Working with collaborators Saonli Basu, a professor in the University of Minnesota’s School of Public Health, and Mark Shriver of Penn State, the team also identified a key reason for this breakdown. Most heritability models assume that each piece of DNA can be inherited independently. In reality, mutations often cluster in nonrandom patterns, which geneticists call directional linkage disequilibrium. The effect is especially pronounced in populations with mixed ancestry. These hidden correlations can throw off heritability estimates, sometimes pushing them too high and other times too low. “This matters when we want to study traits like health disparities, which vary between populations,” Zaidi explains. “Disparities might be driven by social factors, but they also may be driven by genetic differences.”
For researchers, this finding underscores the need for caution when applying heritability models. If the models systematically get heritability wrong, they could distort how scientists design large-scale studies meant to uncover the genetic factors behind certain traits and cause researchers to overlook the role of environmental variables in explaining differences in disease prevalence.
Zaidi’s team is now pursuing methods that can account for the complexities of real human populations, where migration and mixing have shaped genetic variation in ways that current models cannot capture. “Most populations are highly complicated,” Zaidi says. “It has huge implications for research on health disparities, as well as for applying human genetics more equitably across populations.” - Jonathan Damery