In 2015, a study in Nature Genetics introduced a surprising new possibility: Perhaps weight and education are so intimately connected because they share some of the same genetic roots. Using enormous collections of genetic data, the study’s authors searched for pairs of traits that were correlated with the same genes. For each pair they calculated a metric called “genetic correlation,” which quantifies just how similar the whole set of genes linked to one trait is to that linked to another trait. A smattering of trait pairs popped out as having significant genetic correlations, among them body mass index (BMI) and years of education—as well as more obvious pairs, like depression and anxiety, or type 2 diabetes and blood glucose levels. (Researchers have since tried to explain the apparent genetic link between weight and education by suggesting that people who are genetically predisposed to be better decision makers, and are presumably successful in the classroom, are more likely to adopt healthy lifestyles.) Compared to simpler, behavioral explanations, such genetic explanations might sound far-fetched. But the data would seem to offer few other alternatives. Genes, after all, have an unquestionable primacy. If the same genes are associated with both education and BMI, it stands to reason that those traits must have intertwined biological roots. Now, a new study in Science shows that this idea is illusory. It suggests that geneticists must also consider what comes before people’s genes: their parents. Even if two traits are statistically associated with the same genes, they might not have any true genetic overlap: That same pattern can also appear if people with those traits tend to mate with each other. (This is called “cross-trait assortative mating.”) For example, people with many years of education, who are likely to be of a higher social class, tend to seek out partners who display markers of social standing like a low BMI, and vice versa. Their children will then have genes linked to both high education and low weight. If this happens repeatedly across a population, the two traits will appear to share some of the same genetic causes, because the traits and genes will co-occur so frequently. In reality, they will have been inherited from different sides of the family. Genetic correlations have become a popular tool because of what they seem to suggest about the underlying biology of a pair of traits, says Richard Border, a postdoctoral scholar in neurology and computer science at the University of California, Los Angeles and the study’s lead author. But cross-trait assortative mating challenges such inferences. “It is basically a way of breaking that logic,” Border says. Border and his colleagues are not the first to raise the possibility of spurious genetic correlations. When designing studies, geneticists can control for the effects of factors like parental traits and childhood environment by comparing people who have those things in common—that is, siblings. Earlier this year, statistical geneticist Laurence Howe and a team of researchers did just that. When Howe compared siblings with each other, he observed no genetic correlation between BMI and years of education. Somehow, it was parents, and not genes themselves, that had made weight and education seem genetically connected. Border and his colleagues found that evidence. Studying cross-trait assortative mating in detail requires knowing how much it actually happens in the real world. It seems reasonable that depressed people might end up with anxious people due to their shared experience of living with a mental illness, or that educated people would tend to marry people who got high scores on IQ tests, but Border needed to put numbers on those trends. The team was able to find the information they needed in the UK Biobank, an enormous dataset that comprises genetic, medical, and demographic data about hundreds of thousands of UK residents. They found that the more often people who had a particular pair of traits tended to couple up, the more those traits seemed to be genetically correlated. It was reasonable to suspect, then, that assortative mating was in fact making some genetic correlations appear stronger than they would otherwise be. Still, this observation didn’t prove that assortative mating could create the illusion of a genetic link where none existed. So Border and his team turned to a computational approach: Following the marital trends they had observed in the real world Biobank data, they simulated a population of people who paired off into couples. These imaginary couples reproduced, and their children found mates, and their children’s children—and so on. The scientists tracked the genes and traits of all these simulated individuals, and, using that information, they were able to calculate genetic correlations across each generation. What they found confirmed their suspicions—even if two traits were totally genetically unrelated in the first generation, if people who had those traits tended to mate with each other, the genes eventually started to seem correlated. Based on the simulations, they estimated that assortative mating alone could explain as much as half of the genetic correlation between BMI and education. But assortative mating didn’t go as far toward explaining some of the other apparent correlations they simulated. It appears to play a smaller role in the genetic correlations between some pairs of psychiatric conditions, like bipolar disorder and schizophrenia, or major depression and anxiety. Because each pair of conditions shares so many genetic similarities, some scientists have wondered whether they should even be considered separate conditions at all. Even taking assortative mating into account, that argument would still appear to hold water. That’s a particularly good piece of news given the current state of psychiatry, says Verneri Anttila, a genetics researcher at the University of Helsinki. “Genetic analysis has been a rare ray of hope in the recent decade,” he says. Whereas some approaches to understanding and treating psychiatric illness seem to have stalled, genetic studies have continued to provide new insight and could one day play a role in the creation of new treatments. The most meaningful impact of this work may take place outside of the lab, Border says. As genetic technologies advance, and scientists get better at predicting how people will look and behave just by inspecting their genomes, some people are trying to bring those tools out into the real world. For example, if you spit into a plastic tube and ship it to their lab, 23andMe will use your genes to tell you how likely you are to develop type 2 diabetes. And because it is possible to predict how far someone will go in school just by looking at their genes, some scientists have suggested that schools should use genetic analysis to better target resources to their students. But if social phenomena like cross-assortative mating can meaningfully affect the results of genetics studies, it’s hard to say for sure that genes can tell you something about a child’s intrinsic capacity to succeed in school. “Our results,” Border says, “suggest that this is kind of a terrible idea.” Currently, he is investigating how familial features other than cross-trait assortative mating—like the passing down of class and wealth—can create the illusion of genetic relationships. This research, and his work on assortative mating, have convinced him that genetic prediction tools are not ready for real-world deployment in domains like education. “Especially with behavioral traits,” he says, “we really need to really understand what we’re actually measuring before we start to move forward with this stuff.”
