Why using genetic risk scores on embryos is wrong

Due to the unwise announcement of using polygenic risk score selection embryo selection by Genomic Prediction, a US company, and its exposure in articles such as this by the Economist, the wisdom of selecting embryos by polygenic risk scores has become a concrete question. I do not think it is medically justified, nor scientifically sound, nor do I think it is ethically wise. In this blog post I justify this position.

Despite my disquiet and skepticism on this usage, I do support the long standing use of embryo selection via pre-implantation genetic diagnosis (PGD) of embryos for selected rare diseases, and I also support active research and development of polygenic risk scores in adult screening. However, I think the combination of these two technologies is wrong, though without saying it is clearly wrong for all time.

A brief reminder. Preimplantation genetic diagnosis involves making a set in-vitro fertilized human embryos from prospective parent’s egg and sperm, and extract a single cell from each embryo. The removal of a single cell at the right stage does not affect the embryo’s health. The single extracted cell has a full genome, and so can be used to assess the genome of this particular embryo. In the case of a rare recessive disease such as Cystic Fibrosis for example, where if both parents are heterozygous carriers, 1 in 4 of the embryos will go onto develop the disease. Here PGD can test a series of embryos formed from the parent’s egg and sperm, and only embryos which do not have the disease are implanted into the mother.

For Polygenic risk scores (PRS), a genome-wide model of the genetic component of a trait (for example, heart disease risk or breast cancer risk) is developed by looking at hundreds of thousands of individuals with (partially) known outcomes and a high density (though not necessarily full genome) sequence. These statistical models, once validated and tested, can be used to provide a risk of new individuals potentially before outcomes (if they fit the baseline criteria of the model; usually due to the population they come from). Both the research and the testing of this technologies is on adults, and the most near term areas being explored are existing population wide screening, such as heart attack and breast cancer risk.

So – if one is comfortable with both of these technologies, then the argument is that combining them to choose embryos (children) who have lower “bad” risk (for example, lower risk of heart disease) could be a good thing. As I mentioned at the start, I disagree on safety, science and ethical grounds.

The first is a straightforward safety aspect. Polygenic risk scores are currently being discussed in a screening process, where the polygenic score either changes the resources assigned to different individuals (for example, bringing forward breast cancer screening visits for a subset of the population) or adding to the overall threshold for clinical intervention along with other factors (for example, recommending statins along with other factors such as cholesterol, age and weight). These interventions are relatively low risk (though certainly not 0 due to the real danger of overdiagnosis in the first case and adverse drug reactions in the second) and the genetic information is blended with other factors. Using polygenic risk scores alone for embryo selection is a large intervention for a uncertain gain (see below) with likely other unintended consequences. For example, selecting embryos on the basis of PRS scores is likely to increase the chance of recent homozygous haplotypes, and any deleterious recessive mutation will increase. Both the fundamental procedure of PGD is risky and the PRS approach will provide more, untested risk for the children.

The second is the soundness of the science. Polygenic risk scores are statistical models of the genetic component of the variation in traits in individuals observed in the population; they are not models of which bits of the limited possibilities of a genomes will impact a trait given particular parents. Of course, these two models are expected to be related, but they are not the same. Most trivially, in the case of embryo selection one is constrained by the haplotypes of the parents. This sounds like a small change but it is not; on a population scale for example, the underlying linear model of alleles at a locus is a reasonable approximation to the truth of often more complex recessive/dominant behaviour, but when one is constrained to just four possibilities per locus, this local non linear behaviour will almost certainly have an effect, and potentially a substantial one. This is the science-modelling inverse of the safety issue I raised above. It is worth noting the big gains in animal breeding genetics use the genetics almost exclusively for mate selection (for example which bull’s semen is used to impregnate which cow), and very rarely for embryo selection (the only cases I know is in the context of deliberately genetic engineered cases, which is a very different, and in anycase, animal breeding is clearly ethically different than human embryo selection). Furthermore the genome is a very big place, and there are very many traits (risks for different diseases etc) which one might want to “optimise” for. However, the big size of the genome coupled with the very many places on the genome means that the underlying model drives towards uncorrelated trait distributions; this means as the number of traits one wants to simultaneously optimise on so the number of embryos one needs to select from increases.

Finally this is unlikely to be ethical. The ethics of these procedures needs to be determined by a societal process, not a scientific process; ethical discussions should be informed by the science, but no scientist has a privileged position in the discussion. So, now speaking a citizen, I think it is mistake in society to place such fundamental choices of attributes of the children into the hands of parents without an extremely good rationale. Although parents make many choices on behalf of their children, starting with the desire to have a child, and their upbringing, to deliberately select an unchangeable genetic feature for their child from birth must pass the highest level of scrutiny and societal acceptance. In addition there are seductive sounding traits, such as educational performance, or facial symmetry, or hair or skin colour which are (broadly) as ammeanable to genetic analysis as heart disease risk or breast cancer risk – to have parents try to weigh both the science but also the fundamental rights or wrongs of what to choose as unchangeable genetic components is, in my view, fundamentally changing the relationship between parents and children.

These are the reasons I strongly believe this current procedure is not safe, not scientifically sound and not ethical. But could such a procedure sometime in the future meet these criteria? How would one assess it? Certainly I can imagine in the future more research that extends the definition of severe genetic disease that one could test from one locus to multiple loci, and it is possible that considerations of genome-wide “background” should be used. This is in some sense extending from the rare disease cases now. This is quite different in the specifics from the current PRS models, and I would be arguing for considerable research on (for example) local non-linear interactions (recessive / dominance) in this modelling. In these scenarios one could be more certain of the severe consequences of certain genotype combinations, and formally at this point I don’t see this to being ethically fundamentally different from current practice on rare disease. However, we are a long way from this, and the broad, population wide views of genetic risk is not fit for purpose for this individual risk. Furthermore this will correctly bring up the more complex “slippery slope” arguments towards broad trait selection.

How does one assess all of this, in particular the ethical dimension? Here I am a big proponent of the UK’s Human Fertilization and Embryo Authority, which is how this is regulated in the UK. A procedure has to pass medical, scientific and ethical bars, and these are considered carefully with evidence and discussion. Critically the ethics is a societal process informed by science but without scientists taking a privileged position. In contrast, the mainly self-regulation of the medical practice in the US I think is dangerous (though I am less of an expert on US regulation).

Until there is good regulation of this technology worldwide I believe we will have more of these stories and more unsavoury applications.