Today, nearly every newborn in a U.S. hospital gets a heel prick soon after birth. The blood test screens for rare but serious physical disorders, such as sickle cell anemia and hyperthyroidism, whose worst effects may be mitigated with early treatment.
What if, in addition, a small sample of each baby’s saliva was sent out to a lab, where—for just a few dollars—the baby’s DNA was analyzed and a multitude of “risk scores” returned? These would not be diagnoses but instead prognostication: This baby is at elevated risk for developing heart disease in 50 years. That baby is more likely than average to suffer from depression or schizophrenia someday. This baby may have a very high IQ—or a low one.
Robert Plomin, PhD, a psychologist and geneticist at King’s College London, thinks that future is just around the corner, and he welcomes it. In academic papers and in a new book, “Blueprint: How DNA Makes Us Who We Are,” he makes the case that in the very near future, we will be able to know, at birth, something about our risk for developing nearly every imaginable psychological and physical trait and illness—our “polygenic risk scores”—and that this knowledge will help researchers develop new treatments and interventions, and will help all of us live lives better in keeping with our individual natures.
“In 10 years, it will be seen as unethical not to [genotype everyone at birth],” he predicts. “Knowledge is power; forewarned is forearmed.”
Critics, meanwhile, see that vision as both hyperbolic and horrifying—scientifically implausible but also a morass of ethical problems. University of Virginia psychologist Eric Turkheimer, PhD, for instance, objects to the crystal ball comparisons and argues that polygenic risk scores do not really tell us more than we could already learn from simply looking at a person’s parents’ traits.
What’s indisputable, though, is that as the cost of DNA genotyping has plummeted over the past half decade or so, research on the genetics of psychological traits has grown exponentially. Researchers have discovered thousands of genetic variations that each contribute a tiny drop to our genetic propensity to a multitude of traits. And, whether these DNA tests can tell us much about an individual newborn’s destiny, they are already a useful research tool that is providing new insights into how genes and environments interact, new avenues for understanding how mental illnesses (and other illnesses) develop and new pathways to explore potential treatments.
A history of heredity
The road to polygenic risk scores was winding. Scientists have been studying heredity for more than a century, ever since Sir Francis Galton proposed using twins to help untangle the mystery of nature versus nurture. Decades of those twin, adoption and other family studies pointed to one overall conclusion: “Everything is heritable,” as Turkheimer put it in his “first law of behavioral geneticsopens in new window.”
The amount of heritability varies by trait, but for most psychological traits and disorders, it is substantial. Schizophrenia is about 50 percent heritable—that is, genes account for about 50 percent of the variance in the trait across a population. IQ is also about 50 percent heritable. Autism is about 70 percent heritable. And heritability of the Big Five personality traits ranges from about 40 to 60 percent.
But knowing that a trait or disorder is partly heritable only tells you about population-level transmission; it doesn’t tell you whether a particular person will inherit it. No psychological trait is 100 percent heritable—after all, identical twins share exactly the same genetic code, but they are not the same person. And knowing a trait’s heritability doesn’t tell you anything about the actual genes—or the environmental mechanisms—that influence it. Researchers could only begin to probe those questions in the 1990s and early 2000s, when DNA genotyping became available.
First, a quick primer: The human genome is made up of about 3 billion base pairs (made up of chemical building blocks called A, C, T and G) on 23 pairs of chromosomes. The vast majority of the genome is identical from person to person, but what’s interesting to researchers are the differences—spots called single-nucleotide polymorphisms (SNPs) where, for example, one letter that is usually a G has been switched for a T, or a C has been switched for an A—that contribute to our diversity as a species.
In its early days, DNA genotyping was very expensive. So, researchers in psychology and other fields turned to a seemingly promising strategy: Rather than try to look for interesting SNPs across a person’s entire genome (which would cost too much), they looked at “candidate genes”—genes they had good reason to think might be related to the trait they were studying. A researcher interested in depression might look at SNPs on one or two genes, for example, that were involved in the serotonin system.
The hope was that these studies would quickly identify “the gene” for depression, attention-deficit hyperactivity disorder (ADHD), IQ and many other traits and disorders.
“The assumption was that single genes would be interesting to work on,” says Terrie Moffitt, PhD, a psychologist at Duke University who worked on candidate gene studies in the 1990s and early 2000s.
The strategy didn’t work out. Gradually, as most candidate gene studies failed to find anything interesting (or failed to replicate), researchers realized that the problem was that each psychological trait or disorder wasn’t linked to just one or two or a dozen genes—but to hundreds or even thousands, each of which contributed only a tiny amount to the trait’s heritability. To find all those SNPs, you couldn’t just look on candidate genes, you’d have to scan the whole genome. And because the amount of variance that each individual SNP contributed to the trait was so tiny, you’d have to do so across huge subject pools—hundreds of thousands of people—to locate the relevant SNPs.
“From about 2003 to 2012, everyone was just waiting for these large GWAS [genome-wide association studies],” says Moffitt.
Finally, as the cost of genotyping began to fall, those studies became possible, then plentiful. Over the past five years or so, researchers have conducted ever-more and ever-larger GWAS, identifying thousands of SNPs linked with personality, intelligence, depression and a host of other psychological traits and disorders (and, outside of psychology, with physical traits and illnesses, like obesity and heart disease, too).
Today, genotyping a person’s DNA, and looking for hundreds of thousands of SNPs, costs less than $100, and millions of people have sent spit swabs to commercial companies like 23andMe to do personal genetic testing. In one of the largest GWAS to date, researchers examined the genomes of more than 1.1 million people—including 23andMe customers as well as people in another huge DNA database, the UK Biobank—to look for genes related to educational attainment. With this huge pool, they identified 1,271 SNPs related to how many years people had spent in school. Each SNP contributed a minuscule amount on its own, but when added together, they explained 13 percent of variance in educational attainment in the sample (Nature Geneticsopens in new window, Vol. 50, No. 8, 2018). Also last year, a meta-analysis of GWAS on depression, with nearly 150,000 people with depression and 350,000 controls, found 44 genes associated with major depressive disorder (Nature Geneticsopens in new window, Vol. 50, No. 5, 2018). And a study with 135,000 people found 35 genes associated with lifetime cannabis use (Nature Neuroscienceopens in new window, Vol. 21, No. 9, 2018). These are just a small sample of the mushrooming number of GWAS.
Polygenic risk scores, then, are a way to take the information from GWAS and apply them to an individual. “Polygenic” means “many genes,” and that’s what these risk scores include. Once you genotype a person’s DNA, you can comb through it for SNPs that have—through large GWAS—been associated with a particular trait. Then you simply add up the number of these SNPs in the DNA, while weighting them appropriately because some SNPs are more strongly associated with a trait than others. The resulting number is the person’s polygenic score for that trait. It’s usually expressed as a percentile—as in, this person is in the 70th percentile of genetic risk for developing schizophrenia, for example, or the 90th percentile for academic achievement.
An ethical minefield
So, how much do these scores matter and what do they tell us? The answer depends on who you ask. Turkheimer, the skeptic, is not impressed. Right now, he points out, the largest GWAS—the one on educational attainment—can explain 13 percent of population variance in that trait. Other GWAS on psychological traits and disorders explain less—about 7 percent of the variance in schizophrenia and 3 percent in depression, for example. While that’s significant for a single variable, Turkheimer points out that it’s far less than the actual heritability of these traits, which we already knew from twin and adoption studies was high.
“If anything, what we’ve found is smaller than we’d have expected in, say, 1990 [before researchers began working with DNA],” he says. “They don’t predict very well.”
Plomin, on the other hand, who calls himself a “cheerleader” for polygenic scores, takes an expansive view. He believes that as GWAS continue to get larger and researchers refine their techniques for calculating risk scores, their predictive power will continue to improve, right up to the limits of heritability itself.
That’s a technical disagreement, but the larger debate is philosophical and ethical as well.
Plomin believes these scores will be invaluable to parents looking for guidance in raising their children, and for people looking for insight into their own strengths and weaknesses. Parents with a child at high risk for dyslexia, for example—a disorder that’s often not discovered until children are already having trouble in school—could instead get them early reading help and head off the worst consequences. People who know that they are at high risk for alcohol or other substance use disorders could take more care to avoid alcohol and drugs early on.
But Turkheimer and others worry about a dangerous endgame to these predictions. Take the dyslexia example, for instance. For every child correctly identified as dyslexic via their DNA, several other children might be misidentified—after all, polygenic risk scores only indicate risk, they are not diagnoses. What are the consequences of identifying children as “high-risk” for disorders they never develop? And, in a more extreme example, what if, as some have suggested, polygenic risk scores for intelligence became part of the criteria used to track children into different educational paths—college versus trade school? “The suggestion to put kids in schools because of their polygenic risk scores is exactly the same as assigning them to schools based on their parents’ IQ scores, except not as good, because the predictions aren’t as good. To me, that’s a transparently terrible outcome,” Turkheimer says.
More broadly, focusing on what genetics can tell us about individual differences plays into a view of the world that dangerously discounts the importance of environment, says Jonathan Kaplan, PhD, a philosopher of science at Oregon State University who studies the ethics of behavioral genetics research. For example, focusing on what GWAS tell us about IQ and educational attainment in an individual can discount the importance of attending a safe, functional and well-funded school.
“That’s the worry. It’s not that there’s anything wrong with the research, but it tends to crowd out other explanations in ways that are deeply problematic,” he says.
Research implications
These deep—and important—societal implications aside, most researchers do agree that GWAS and polygenic risk scores are increasingly useful research tools.
First, for medical research, locating thousands of new genes related to mental health and other disorders could give scientists new avenues to look for novel medications and other treatments. Of course, that was once the hope behind candidate gene studies—that by finding the one or two genes responsible for a disorder and looking at the systems those genes were involved in, we’d learn more about how to treat the disorder. The fact that most mental health disorders turned out to be influenced by many genes significantly complicated that picture, but it didn’t erase it.
“It helps us understand the biology behind these disorders,” says Gerome Breen, PhD, a psychiatric geneticist at King’s College London. “Recent studies are helping us broaden our thinking and approach, in depression for example, and think about different biological processes than we had before. It could get us away from the serotonin-dominant approach.”
It could also influence the way psychologists and others conceptualize and categorize mental illness. Recent GWAS, for example, have shown that there is significant overlap in the genes involved in many distinct mental disorders, including schizophrenia, bipolar disorder, ADHD and depression (Scienceopens in new window, Vol. 360, No. 6395, 2018). More results like these could lead psychologists, psychiatrists and other researchers to rethink the diagnostic distinctions between these disorders.
Finally, polygenic risk scores could help researchers who want to bring personalized medicine to mental health treatment—helping tailor treatment to individuals. For example, Breen is interested in how polygenic risk scores might help predict schizophrenia patients’ response to treatment. In one study, he found evidence that among schizophrenia patients experiencing first-episode psychosis, those with higher polygenic risk scores for the disorder were more likely to have depressive symptoms and lower global functioning before treatment; but also, they tended to show more improvement in symptoms after treatment compared with patients with lower polygenic risk scores (Translational Psychiatryopens in new window, Vol. 8, No. 1, 2018).
Steven Hollon, PhD, a psychologist at Vanderbilt University in Tennessee who has spent decades studying depression treatment, is excited by these possibilities. His background is not in genetics, but he is teaming up with Breen and King’s College London psychologist Thalia Eley, PhD, to devise a study that will look at how polygenic risk scores might predict depression patients’ responses to a behavioral treatment versus medication. Previous research, he says, has shown that some patients do better with behavioral treatment and others with medication, but right now, there are few good ways to predict which patients will fall into which category. Their grant proposal is not yet funded, but Hollon has high hopes for the research method.
“Twenty years ago, we couldn’t have imagined this,” he says.
In another line of research, polygenic risk scores have also—perhaps counter intuitively—caught the attention of psychologists and other social scientists who want to understand more about how we are shaped by our environments as well as our genes.
For decades, the main way to conduct such research was through twin and adoption studies. If you wanted to know how parenting affected a particular outcome in children, for example, you could look for identical twins raised in different families and see how they differed on that outcome, or you could compare identical to fraternal twins. It’s an effective method, but with a limited participant pool.
Polygenic risk scores, in theory, mean that you could run similar studies in the general population, by using the risk scores as a covariate to control for genetics.
For example, in one study, Moffitt, her colleague and spouse Avshalom Caspi, PhD, and postdoc Jasmin Wertz, PhD, are combining polygenic risk scores for educational attainment with long-running cohort studies in New Zealand and the United Kingdom to look at how parenting affects children’s lifetime risk for antisocial behavior. (Many researchers are interested in using these educational attainment scores to study other areas because the educational attainment GWAS is the largest to date, and thus has the best predictive power.
And because educational attainment is related to so many traits, it can be used as a proxy to study many factors that are statistically linked to education level, including criminal behavior, longevity and more.)
“We take the mother’s DNA and calculate the educational attainment genetic score,” Moffitt explains. “Then we look at measures of what she’s doing—we do home visits and interview parents about how often they read books, that sort of thing.” Then they look at the children’s DNA and outcomes, such as whether they have a criminal record. “Behavioral geneticists would say, sure, smart kids do well because they’re born to smart parents. But what we’re able to do by controlling for polygenic risk scores is to say that reading books matters, music lessons matter—quite apart from genetics,” Moffitt says.
Studies like this one point to the broad reach of polygenic risk scores as a research tool. To skeptics like Turkheimer, that’s where their importance lies. “There’s all kinds of interesting social science you can do when you have these genetic estimates,” he says. To champions like Plomin, meanwhile, they’re yet more evidence that all psychologists—even those who have never considered integrating genetics into their work—should be paying attention to the field.
“All psychologists should be taking the opportunity to incorporate DNA into your research,” Plomin says. “It costs what, $100? fMRI costs maybe $500 an hour. Any sample a psychologist studies, if you don’t collect DNA, you’re doing yourself a real disservice.”