An anonymous incubator space just off the New Jersey Turnpike is the unlikely setting for Genomic Prediction, a genetic testing start-up that is daring to push the boundaries of preimplantation embryo screening, a technique curiously bereft of regulatory oversight. The company’s co-founder and chief scientific officer, Washington State native Nathan Treff, PhD, occupies a small office decorated with Pearl Jam posters, a Seattle Seahawks helmet, and a framed Genomic Prediction logo that features a sigil clearly inspired by the “Game of Thrones” House of Stark.
The lab space is well equipped with some Illumina benchtop sequencers and robotic devices, but there’s nothing to suggest the trail blazing path Genomic Prediction is pursuing. In addition to standard genetic tests—chromosome counts for Down syndrome and other aneuploidies and mutational tests for rare monogenic disorders—the Genomic Prediction menu includes newly developed polygenic risk scores (PRS) which assess the likelihood of developing complex polygenic diseases such as type 1 and 2 diabetes, coronary artery disease, cancer, hypothyroidism, inflammatory bowel disease, schizophrenia and others. Indeed, the company just secured its first formal client for the Expanded Pre-Implantation Genomic Test (EPGT).
Treff, who has published over 100 peer reviewed papers on the genetics of the preimplantation embryo, notes that what excites him is “reducing the risk of common genetic diseases.” However, Genomic Prediction’s inclusion of a test for intellectual disability has proven controversial. While Genomic Prediction’s executives maintain that they will offer prospective parents information on cognitive impairment only, the same machine learning analysis that reveals cognitively impaired embryos reveals outliers at both extremes.
The notion that a company could apply PRS to embryo selection alarms many geneticists and ethicists. Laura Hercher, genetic counselor and director of research at the Sarah Lawrence College Graduate Program in Human Genetics tells GEN, “The race to get this into the prenatal arena and make money off of people is staggering.We should be so cautious and so careful in the prenatal arena because we are operating in a complete black box,” she notes. “There are so many negative consequences you can imagine coming out of it. It’s disturbingly fast when there is so much that we don’t know.”
Until ultrasounds became routine “boy or girl?” was usually the first question asked after a baby’s birth. Today, new parents commonly learn the fetal sex during the 18-week ultrasound. More than 20 years ago, preimplantation genetic testing (PGT) was developed to enable the safe selection of in vitro fertilization (IVF) embryos that have the best chance of a productive pregnancy and a healthy baby. PGT can reveal sex, aneuploidies and hundreds of monogenic disorders (PGT-M) such as cystic fibrosis, hemophilia, Tay-Sachs disease, and sickle-cell disease. These tests, although not without their own controversies, have become routine.
Until now, however, no prenatal genetic testing company has attempted to offer information on risk for more common polygenic disorders such as diabetes and cancer. The PRS is a recently developed tool that can measure the effect of genetic variants to indicate a person’s overall disease risk. Using the genetic variants that differ between large sets of the patient population and a control set, machine learning algorithms can stratify a population, indicating which patients have low or high likelihoods of developing the disease, based on their DNA variants.
“Our results indicate that substantial improvements in predictive power are attainable using training sets with larger case populations,” reads a bioRxiv preprint written by a team of scientists led by Stephen D.H. Hsu, PhD, a co-founder of Genomic Prediction. Hsu, a theoretical physicist, is also the senior vice president for research and innovation at Michigan State University and a member of BGI’s Cognitive Genomics lab, where fellow co-founder Laurent Christian Asker Melchior Tellier, CEO and CTO of Genomic Prediction, serves as co-director. The Cognitive Genomics Lab was created by BGI in 2011 with a focus on the genetics involved in human cognition.
Hsu tells GEN that Genomic Prediction aims to help families have healthier children. He notes that upon reading an article on the steep decline in the cost of genomic sequencing, he “did some back-of-the-envelope calculations” and concluded that if the cost continued to fall for the next 10 years or so, researchers would be able to answer all of these interesting questions that he had been thinking about since he was a kid.
But Genomic Prediction’s methods and goals make many scientists and medical professionals uncomfortable. For some, the science underlying PRS is not solid enough to justify usage in embryo selection. Others argue that, even if Genomic Prediction can accurately predict polygenic risk, it shouldn’t, claiming that doing so would be too close to eugenics. The scientific community knew that this day was coming. But now that it has arrived, are we prepared for it?
Is the PRS ready for prime time?
“There are so many problematic things involved in this product, it’s hard for me to focus on which of the problems I’m more concerned about,” says Hercher. A major concern is whether Genomic Prediction is using PRS appropriately. PRS are typically used to stratify a population into segments, distinguishing the clinically meaningful changes that occur at the two tail ends. The test is meant to “drop people into broad buckets,” Hercher explains, not to make an accurate individual prediction.
Genomic Prediction is using PRS to differentiate between siblings (who share 50% of their common variants), an exercise, she says, of dubious usefulness. “It’s not clear it’s possible to predict which of two sibling embryos is more likely to get these diseases,” Hercher says. “And, if it was, we don’t know how to do it. It’s misleading that they are out there advertising this and, at best, premature.”
“You’d expect most embryos to have a [PRS] for any trait, especially a highly polygenic trait like intelligence, that’s around the average of the two parents,” says Ali Torkamani, PhD, associate professor at Scripps Research Translational Institute and Director of Genomics and Genome Informatics.
But Hsu counters these types of criticisms, telling GEN that the argument against using PRS in embryo selection is driven by a misunderstanding of statistics. He compares Genomic Prediction’s work to other assumptions that are made throughout our lives: “The fact that you got three speeding tickets last year doesn’t imply that you are at higher risk of an accident this year. Surely, we can only conclude that at a population level. But when they issue a policy and charge you 10% more, that’s an individual decision. That is a jump that is made all of the time. That is what statistics is used for.” He adds that before Genomic Prediction introduces a predictor, it is validated and tested in multiple populations in multiple settings.
Arthur L. Caplan, PhD, professor of bioethics and founding head of the Division of Medical Ethics at NYU School of Medicine, questions whether Hsu’s company is choosing certain conditions just because they have a test. Every embryo is going to have some profile of risk. Just because you are looking for your lost keys under a street lamp does not mean that is where your keys are. Rather, it’s just where the light is, Caplan says. Similarly, just because you avoid the diseases on Genomic Prediction’s list does not mean that you will have a healthy child.
Henry T. Greely, JD, director of the Center for Law and the Biosciences at Stanford Law School, notes that literature suggesting PRS works at a population level does not necessarily imply how good they will be at an individual level. “Having lived through various genetics issues for 25 years, [I would say that] in general, [the tests] are never as good as they initially look,” Greely says.
Even if Genomic Prediction can back up its claims, some experts are uncomfortable with the notion of embryo selection based on polygenic diseases. “Some of the listed conditions, even if [they are] diseases, are sliding into eugenics—not health,” warns Caplan.
Last fall, Hercher used an episode of her podcast “The Beagle Has Landed,” to interview Sekar Kethiresan, MD, a proponent of using PRS in the clinical setting. When Kathiresan director of the Center of Genomic Medicine at the Massachusetts General Hospital, was asked about using PRS prenatally and Genomic Prediction specifically, he responded, “I’m very uncomfortable with the use of this kind of information to select embryos for implantation.”
But others believe the results will be underwhelming. “Nobody is going to make super babies this way because we don’t know how to make super babies,” argues Greely. “People think genetics is a lot stronger than it is.” Polygenic traits are complicated: if you have 100 different genes in a trait, keeping slightly better versions of 10 of them isn’t going to get you very much, he insists.
The slope is steep and slippery
Treff asserts that the company is focused on predicting the risk of diseases. That said, the test that has drawn a lot of attention to Genomic Prediction is the predictor for intellectual disability, which indicates outliers at one end of the spectrum as readily as outliers at the other. Selecting embryos for higher intelligence has been a dreaded use of genetics for a long time.
Nathaniel Pearson, PhD, founder of Root, a company that rewards blood and marrow donor volunteers with information about their own genes, notes that “intelligence” is poorly defined. In addition, in the large genome-wide association study (GWAS) that Hsu cites when discussing Genomic Prediction’s intelligence PRS, the main metric was educational attainment (how long someone stays in school). “I don’t believe it for a minute,” declares Hercher. “It doesn’t make any sense scientifically.”
Pearson points to recent work from Calico Life Sciences and Ancestry.com on the impact of genes on lifespan as an important lesson. Surprisingly, they found that the lifespans of spouses were more similar than those of siblings—placing more importance on lifestyle than genetics. By the time most people are having children, Pearson says, they know how long their partners went to school, making educational attainment an inherent piece of the complex mixture of factors that dictate mating choices.
Hsu confirms that the predictor used by Genomic Prediction has been directly compared against IQ scores. But, even if Genomic Prediction could reliably identify and select embryos with higher intelligence, should they? Will they? “We don’t know how it’s going to evolve,” admits Hsu. “The government of Singapore might come to us and say no, we want to alert parents if one of their embryos is likely to be well above average. You can imagine a scenario where we would say, ‘Well OK. I guess in Singapore, it’s OK.’ But, we don’t feel like Americans are ready for it.”
Learning the same lesson over and over again
“You could easily predict that we were going to get into this,” says Caplan. “It would be nice to start to think about how we are going to tackle this, before it gets too far along.”
Caplan urges that action be taken before too many companies start expanding the business: “It’s time for professional societies to say what they think are scientifically based reasons to screen.” He includes in that list the American Society for Reproductive Medicine (ASRM), the American Society for Human Genetics, obstetricians and gynecologists, genetic counselors, religious groups, civic organizations, patient advocacy groups, and disability groups. “Some Joe from company X should not be driving the use of genetics in making the next generation,” Caplan maintains.
But Caplan’s concerns reflect the alarming absence of regulatory oversight in the embryo selection arena. Genomic Prediction’s New Jersey labs have received CLIA certification, and there is no reason to doubt the company’s technical expertise. In the case of an laboratory-developed test (LDT), where the test is developed by a laboratory and specimen samples are sent to that laboratory to be tested, FDAs assessment of their validity is not necessary. Greely would like to see FDA regulations requiring good evidence of efficacy with respect to all sorts of genetic testing, especially PRS. The current law doesn’t require it, and he wishes that the law did. “I think that people will pay a lot of money for something that may be no better than a horoscope,” Greely says. “It might be better than a horoscope, but we don’t know. But I don’t think that Genomic Prediction does either.”
Hsu does not seem comfortable closing doors. “In the future, we may have huge customers from another country and they are really demanding that they want to be able to do cosmetics. They really want to know who has lighter colored skin and darker skin. It will be a tough decision for us. If they are ordering 100,000 tests from us, and they really want this feature—which we can do and which is 100% legal in South Korea—what are we going to do? So, I’m not going to prejudge what we are going to do.”
Hsu adds: “We certainly won’t do things if they are not broadly acceptable in that society and legal. But exactly how we will react? I don’t know.”
Genomic Prediction’s future will eventually be up to the market, says Hercher. For years, she notes, the ASRM said that it was not acceptable to base embryo selection based on sex unless there was a sex-linked disease involved. But IVF centers offered it anyway because their patients wanted it. Eventually, the ASRM threw in the towel because the people voted.
Ronald J. Wapner, MD, vice chair of research in the Department of Obstetrics and Gynecology and the director of Reproductive Genetics at the Columbia University Irving Medical Center, says that this is “fraught with potential for abuse.” After watching how NIPT was introduced over the past decade, his biggest concern is the drift of the technology. NIPT was also offered to the public without going through the FDA—a loophole with an LDT that allows an industry that stands to make a lot of money, to drive a field of medicine.
Wapner is concerned that there isn’t a structure to decide what is ethical. Here the United States lags behind other areas of the world. Wapner and colleagues have formed the reproductive technology council, a group with representatives from the major organizations and experts in the field. They serve to make informed, educated, and ethical decisions, but there is no rule that a group has to consult them.
There is no perfect baby
“It is foolhardy to try to optimize a child,” says Pearson. If we are picking an embryo for one trait, we may skew others in another direction. For example, if parents pick the lowest risk of type 1 diabetes, the child may have a higher risk of cancer than it would have otherwise.
Hercher agrees: “Genomic Prediction is not able to give you a child that is not going to get sick and die. That is a fantasy, and if you are buying into that fantasy, you are going to be angry. I hope that it is at them [Genomic Prediction] and not at your kid.” She adds that there are times when it may make sense to optimize based on a single quality. For example, a family that already has two children with a particularly challenging disease may want to have a third child.
But if parents choose an embryo that has a low risk score of X trait, and that turns out not to be the case, what happens to those expectations? Do the parents agree to live with and love and cherish the child no matter what happens with the other traits? Hercher adds that the conversation about the commercialization of reproduction is an important one to engage in as it raises questions about our sense of acceptance and our values of being a parent.
In the end, our comfort level with screening embryos using PRS does not really matter. The decisions regarding the technology are in the hands of a small company with minimal regulatory oversight. Based on the scientific community’s prior record in this arena, we don’t need an algorithm to predict how this is going to play out.
Our interview with Stephen Hsu can be read at “Polygenic Risk Scores and Genomic Prediction: Q&A with Stephen Hsu.”
Update April 3: The text was revised to make it clearer in paragraph 13 that Dr. Hsu’s quote was addressing multiple criticisms made in the preceding three paragraphs and in paragraph 22 that Genomic Prediction’s predictor used for intellectual disability has been directly compared to IQ scores.