A common objection to using polygenic risk scores during embryo selection is that there is some risk of pleiotropy—a single gene influencing multiple traits or diseases. The concern is that trying to select against one condition could inadvertently increase the risk of others. Lencz et al. (2022), Turley et al. (2021), and Polyakov et al. (2022) all use the relationship between bipolar disorder (BP) and educational attainment (EA) as one example. If a couple were to select an embryo solely on the basis of EA, they may also get an embryo with an elevated risk of BP. While this is a valid concern for selection on the basis of a single disease or trait, it is not a good argument against all uses of preimplantation genetic testing for polygenic disorders (PGT-P).
It is worth noting that pleiotropy can be either advantageous or disadvantageous. For example, selecting against one disease may inadvertently lower the risk of others. Whether or not pleiotropy is an argument against PGT-P would depend on the pleiotropic relationships of traits being selected for or against. Widen et al. (2022) investigated the genetic relationship between 20 diseases and found “no statistical evidence for strong antagonistic trade-offs in risk reduction” and that the correlations were “mostly positive, and generally mild,” supporting “the folk notion of a general factor which characterizes overall health.” This suggests that pleiotropy is typically somewhat beneficial for disease reduction as PGT-P is currently practiced by companies like Genomic Prediction and Orchid Health. If a couple were to select on the basis of a single disease trait, it is more likely that pleiotropy would be an advantage rather than a disadvantage.
Another important consideration is that the major providers of polygenic screening—Genomic Prediction and Orchid Health—provide information on the risks of many different conditions. If an undesirable pleiotropic relationship is found, this can be incorporated by applying weight to the various outcomes and incorporating it into the cost-benefit analysis. Genomic Prediction provides customers with an Embryo Health Score (EHS) derived from many different PRSs. One might be concerned about potential relationships between reported diseases and unreported traits like personality or intelligence, but this is a reason for an even broader index rather than against PGT-P. Moreover, good outcomes tend to correlate, and genetic correlations mirror phenotypic correlations—a phenomenon known as Cheverud’s conjecture.
Pleiotropy cannot be a sufficient reason for avoiding PGT-P. Imagine that there are two euploid embryos that a couple could choose, and they want to pick the healthiest one, but they are concerned about a potentially negative pleiotropic relationship between health and some unknown trait. They receive their EHS from Genomic Prediction, but they decide it is too risky to even select on the basis of a broad index of health. If they avoid the higher-scoring embryo for reasons of pleiotropy and select the lower-scoring one, they still face a potential risk of an undesirable pleiotropic relationship. If they decide to use embryo morphology or random selection, they could transfer the exact same embryo. The pleiotropic relationship would still exist; they would just be unaware of it. Transferring the same embryo cannot be safe when uninformed and dangerous when informed.
Expecting an unmeasured pleiotropic relationship so strong as to negate the benefits of selection on a broad index would be an unusual view. A person with such an expectation would have to believe that their embryo choice would improve by using Genomic Prediction to get an EHS and then picking a low-scoring embryo—one expected to live a shorter and less healthy life—in the expectation that some unmeasured pleiotropic relationship would make this the better choice. I doubt anyone has ever done this, and no skeptic that I know of advocates for this. I think the reason is that this is a terrible idea since the expected benefits of PGT-P far exceed any potential downside of pleiotropy.
Skeptical articles seem to follow the format of listing numerous concerns about PGT-P and then reaching a conclusion based on the fact that there are many reasons to be skeptical of PGT-P. From this perspective, “pleiotropy could be bad in certain scenarios” counts as a reason, even if this concern is not weighted by likelihood. From a binary perspective where aspects of PGT-P are simply good or bad, pleiotropy should be considered a benefit of PGT-P rather than a drawback.
The small drawbacks of unknown pleiotropy can be accounted for by broadening the index to include many diseases, which is currently practiced by the leading providers—Genomic Prediction and Orchid Health. I would personally advocate going further and incorporating all the aspects of a good life into selection. We should not generally oppose the practice of PGT-P because some possible selection strategies—such as maximizing EA without regard to BP—would be unwise. Instead, we should advocate for a wise selection strategy rather than throwing out relevant information and expecting it to improve decision-making.
The more sophisticated version of this argument is an appeal to the Efficient Market Hypothesis with respect to natural selection: "If trait X is so great, why hasn't natural selection already maximized it? There must be negative second-order effects to increasing it!"
This argument isn't so silly. If I told you that there was a pill that could increase your IQ by fifteen points, would you believe me? What if I showed you a fancy-schmancy table filled with computed Hedge's g's, pristine p-values, and large N sample sizes? How about then? Still probably no. There is no trait more economically valuable than intelligence. If something like a magic intelligence pill existed--at no cost--you wouldn't be hearing about it from me.
You could argue that the same logic applies to increasing intelligence using genetics. Natural selection can be ruthless if lacking a trait proves to be a big enough disadvantage. For example, the trait "having two eyes" shows little variation among humans. And humans who do have fewer than two eyes tend to have them for reasons that aren't terribly heritable (e.g running with scissors in Kindergarten). So we have a case where an obviously beneficial trait is maximized.
But ultimately, the above argument does not go through. The sleight of hand here is a conflation between what's *natural* and what's *good*. There are traits that have only a so-so impact on fitness, but are highly esteemed based on our society's values. Intelligence is arguably one of those traits. The hubris of man be damned: We can do better than natural selection.
(I've been told that intelligence was a boon in the ancestral environment, but that the reason that intelligence isn't maximized is due to an "entropic force" due the brain having such a large mutational target, resulting in our current equilibrium distribution of the trait.)