Before the Human Genome Project, we naïvely thought that variants in single genes could account for major mental illnesses and complex behavioral traits. It was a classic case of looking for your lost car keys only under the lamppost. In 1988, Nature published a British claim for the mapping of a schizophrenia gene that didn’t hold up. Science trumped that by publishing evidence for a “gay gene” on the X chromosome. The evidence was perilously thin, collected from fewer than fifty gay couples, and never replicated.
A quarter century later, Benjamin Neale’s team at the Broad Institute performed a state-of-the-art genome-wide analysis involving 1 million DNA markers on a database of nearly 500,000 people. The results painted a vastly more complex picture of same-sex behavior, one in which gene variants explained less than half the variance in the trait. The top five gene “hits” made up less than 1 percent of the variance. Nevertheless, even highly polygenic diseases and traits may have simple genetic switches. “Just because it is polygenic doesn’t mean it doesn’t have a monogenetic solution,” Church says. For example, stature is a very polygenic trait, but many patients with short stature can be treated with human growth hormone.
Would a musician want to ensure their child had the mysterious gift of perfect (or absolute) pitch? My father had perfect pitch, which propelled him to a successful career in the West End as musical director of hit shows including Cabaret and Fiddler on the Roof. As a boy, I’d be ushered backstage after a Saturday matinee to meet a young Judi Dench or Topol. If there’s a gene for perfect pitch, I didn’t inherit it.IV My former colleague Alissa Poh recounted her daily experience living with absolute pitch: her car horn hovers between an E and F, her cell phone rings in A minor, while her refrigerator hums in B-flat.25 Studies by Jane Gitschier among others support a nature and nurture model—the trait is manifest by inheriting an as yet unidentified gene along with early musical training.26 But perfect pitch doesn’t make a musician, nor are all great musicians born with perfect pitch.
In the next fifty to one hundred years, it might become possible to apply genome surgery on artistic or mathematical behavior. And, as Church predicts, once we can do it for a single gene, we will develop safe methods to extend this in parallel, multiplexing edits at multiple genes simultaneously.
In early 2019, I was invited to attend an unusual conference at the Ditchley Estate, a quintessential English stately home reminiscent of Downton Abbey just outside Oxford, ostensibly to discuss the intersection of gene editing and artificial intelligence (AI). During World War II, Winston Churchill spent weekends there (Checkers, the official retreat of the prime minister, was too recognizable for the Luftwaffe). After a welcoming reception with tea and biscuits, forty of us took our seats at a long boardroom table in the old library. The first speaker to be introduced was Stephen Hsu, a theoretical physicist at Michigan State University. This seemed like a peculiar choice—until Hsu started talking.
Hsu’s interest in genetics traces back to his childhood, avidly watching Star Trek and pondering Kirk, Khan, and the Eugenics Wars. “If I get to be one of the scientists who makes real some amazing trope from science fiction, that would be the most awesome thing in the world,” Hsu told Radiolab.27 Although he gravitated toward physics, Hsu remained fascinated by the link between genetics and intelligence. He was formerly an advisor to BGI’s controversial Cognitive Genomics project, since aborted. Now he believes he can apply AI to the prediction of complex polygenic traits including cognitive ability. Once, when asked to give his view of a superior human intelligence, Hsu offered as an example John von Neumann, the 20th-century polymath, developer of game theory, and computer science, who was capable of total recall and a photographic memory. “In my opinion,” Hsu says, “genotypes exist that correspond to phenotypes as far beyond von Neumann as he was beyond a normal human.”
Hsu cofounded a PGT clinic called Genomic Prediction, located in an unremarkable office park off the New Jersey Turnpike, a short drive from Manhattan. Dressed in a T-shirt and torn genes, Nathan Treff, the company’s chief medical officer, met me in his small office decorated with framed posters of Pearl Jam and Iron Man.28 Genomic Prediction offers the usual menu of PGT services—tests for chromosomal abnormalities and genetic diseases such as CF, Tay-Sachs, and Huntington’s disease. But Genomic Prediction goes further, offering couples tests for polygenic conditions including heart disease, obesity, diabetes, and short stature. Also on the menu: low cognitive ability.
Since a landmark paper from researchers at the Wellcome Trust Sanger Institute in 2007, researchers have identified thousands of gene variants that influence our risk for hundreds of complex traits.V We can’t point to a solitary genetic risk factor for type 2 diabetes or obesity, but we can confidently circle dozens or hundreds of specific DNA variants that influence our susceptibility to these and other disorders.
More recently, government-funded databases such as the UK Biobank have made available full genomic and medical data on some 500,000 (mostly European) volunteers. This allows researchers to run machine learning programs to “train” on the data, looking for genetic predictors for medical and behavioral traits. Kathiresan’s group at Harvard Medical School developed polygenic risk scores (PRS) for five complex diseases including heart disease, type 2 diabetes, and breast cancer.29 Hsu’s team extended this work, but he doesn’t stop at merely medical disorders. By identifying some 20,000 DNA variants that influence height, Hsu claims he can build an algorithm to calculate a PRS and predict someone’s height to plus or minus one inch.30
Whereas most investigators are studying PRS in patients, Hsu is courting controversy by insisting he can calculate a PRS before birth in an individual embryo. Hence Genomic Prediction’s menu of available polygenic risks includes short stature and low cognitive ability. Hsu scornfully dismisses criticisms, shocked that human geneticists have virtually no idea what their colleagues in livestock or corn breeding are doing.
The genetics of intelligence is a controversial and fraught issue. A recent Canadian study examined the effect of DNA deletions or insertions (copy number variants) in more than 24,000 people. The authors concluded that the number of genes that influenced intelligence was around 10,000—fully half the genes in the human genome.31 But that hasn’t deterred Hsu. Projects like the UK Biobank haven’t been conducting IQ tests but have asked volunteers to list their level of education. Using that as a proxy for IQ, Hsu’s team mined these data for DNA markers associated with cognitive ability. He says he can predict cognitive ability with a correlation of about 30–40 percent. In the same way that a college dean would look askance at a student who had underperformed on the SAT, Hsu says, “parents may deserve a warning if we find that an embryo has super elevated risk of intellectual disability.”32
Championing couples’ “reproductive liberty,” Treff’s team has already reported IVF embryo screenings for breast cancer and type 1 diabetes.33 Hsu says they can go further to predict which embryo would develop with low cognitive ability—that is, it carries an excess of DNA variants that are predicted to depress IQ below a clinical threshold. The company is not offering clients the option to select for an extra-intelligent embryo—not because the technology isn’t there, Hsu says, just that society isn’t ready for it. Suppose you want to start a family and you learn that embryo #4 was predicted to be in the top percentile of cognitive ability. What would you do? Would you rather rely on the embryologist judging the health of each embryo by shape and morphology (as happens now) or by analysis of its DNA?