Don't Use a Polygenic Risk Score to Choose Your Embryo

They don’t work the way you think.

Imagine you are about to start a family. Would you want the ability to choose whether your child would have cystic fibrosis or not? Would you want to be able to choose if that first child would be a boy or a girl?  Would you want to choose her height? Her hair color? Her risk of diabetes?

These questions have been around since the first bioethicist held court in a university classroom, but only now has technology advanced to the point where we need to consider the practical implications of answering them.

This week, in a New England Journal special report, Daniel Benjamin and his team highlight how the use of polygenic risk scores in in vitro fertilization might lead to a literal Brave New World situation.

This isn’t theoretical by the way.  Right now, there are multiple companies out there that will perform genetic sequencing on IVF embryos and give parents a prediction of the likelihood of all sorts of diseases. This is from the public brochure of a company called Genomic Prediction, showing how parents might use information to pick between two potential embryos.

Let’s talk about how the science works.

Screening for monogenic traits – things like Cystic Fibrosis, sickle cell disease, Tay-Sachs, etc, has been around for a long time. And it is common practice for individuals undergoing IVF to screen embryos for those conditions.  Remember that IVF procedures generate multiple embryos, not all of which will be implanted – giving prospective parents an opportunity to make some informed choices.

Most of us don’t see a big ethical issue here – embryos are going to be discarded – it seems reasonable to choose one that has the chance of the life its parents envision for it.

But most traits are not monogenic. I’m not this tall because I have a certain gene that says so. I am my height because of multiple genes acting in complex ways, and because of the environment I grew up in.

Polygenic risk scores combine genetic information from multiple genes – sometimes thousands of genes – to create a single number to predict a complex trait.

Let’s start with a controversial one.  IQ.

In 2018, Nature Genetics published a paper looking at the genomes of around 270,000 people of European ancestry. Gene variants were linked to the individual’s performance on an IQ test. What you see here, known as a Manhattan plot, is how strongly various genetic variants were linked to IQ.

Manhattan plot linking gene variants to IQ

You can tell right away that there’s not a single gene that tells what your IQ will be.  In fact, this study identified roughly 1000 genes linked to IQ. 

The authors could then look at a given individual, see which of those 1000 gene variants they had, and combine them to produce a single number – a genetic estimate of IQ.

That genetic estimate correlated with the actual IQ with a correlation coefficient of 0.23 which is ok, but not great. It means about 94% of the variation in IQ is accounted for by things other than the genetic score.

That said… what if you had to choose between 5 embryos for IVF.  And I gave you 5 polygenic risk scores for intelligence.  Which would you choose? 

Not so fast. The New England Journal manuscript does something critical here which is to say – hold up, this is not what these scores were designed for. They might not even work in the setting of IVF. 

Why? Several reasons.

The biggest, is that all those embryos you are choosing from are related.  Those polygenic risk scores come from a population of unrelated individuals – the genetic diversity of Europe is a hell of a lot bigger than the genetic diversity of my 3 kids.

That picture I showed you had five embryos which spanned one standard deviation of IQ in either direction from the population mean. It is highly unlikely that any couples’ embryos would be that diverse. More likely, you’ll get something like this:

Among genetically related embryos, the marginal benefit of any polygenic risk score is dramatically reduced.

The NEJM authors modeled this using a polygenic risk score for educational attainment.

They compared use of a polygenic score to just randomly picking an embryo under the assumption, in red, that all the embryos were from different families. The use of the risk score, on average, gets you about 1.5 years of extra educational attainment. Within a family though, this is much less – 0.55 years. And that’s for European ancestry, since that is where the score was developed. Its even worse in people of other ancestry.

And look at those standard deviations! Huge.  So, sure maybe if you are of European ancestry, use of a polygenic risk score would, on average, lead to you picking an embryo that would get an extra half-year of educational attainment, but you’ll often do worse – here’s what the distribution actually looks like:

But hey, I get it – more information is better, right? Might as well maximize the odds?

Not necessarily. The New England Journal authors point out another issue with polygenic risk scores – pleiotropy.  The idea here is that genes don’t only do one thing – and neither do polygenic risk scores. For example, the score for educational attainment is not only associated with higher educational attainment, it is also associated with a higher risk of bipolar disorder.

Are the companies disclosing those links when they counsel parents?

We basically have two big issues here.  There is the ethical elephant in the room. If you can pay for the privilege of choosing how smart your baby is, should you be able to? Might that entrench the environmental disparities that exist in our society and convert them into actual genetic disparities? It’s a scary thought.

But, perhaps fortunately, the science isn’t good enough. The application of polygenic risk scores to embryo selection is, at this point, yet another way to offer individuals a sense of control over a situation that, practically, they still have little control over.  Offering control over the uncontrollable is a great business plan, don’t get me wrong, but it’s not an existential threat.

The big question for me is how much better we’ll get at this. I am not a genetic determinist. There is clearly some ceiling on our ability to predict outcomes based on genetics alone. But we haven’t reached it yet. When we do – what will the world, and our children, look like?

A version of this commentary first appeared on medscape.com.