Polygenic scores and PGT-P: Lost in translation?

Preimplantation genetic testing (PGT) is an ethically sensitive reproductive technique used in conjunction with IVF to test embryos for genetic and chromosomal disorders before they are transferred to the womb. PGT is regulated in the UK by the Human Fertilisation and Embryology Authority (HFEA), and is only permitted where there is a risk of specific monogenic conditions (PGT-M) or chromosome structural rearrangements (PGT-SR), or to screen for abnormalities in the number of chromosomes (PGT-A).

These applications are in sharp contrast to preimplantation genetic testing for complex traits, known as PGT-P (also sometimes referred to as 'polygenic embryo screening'). PGT-P uses polygenic scores in a PGT framework, to offer prospective parents an insight into their embryos' likelihood of developing complex traits or conditions based on their genetics, so that they can use this information to prioritise which embryo(s) to transfer.

Complex traits emerge from the interplay of thousands of DNA variants (each with a small individual effect) and environmental influences throughout a person's life. They include medical disorders (such as diabetes and cardiovascular disease), physical traits (such as height) and mental health and neurodevelopmental conditions (such as schizophrenia and attention deficit hyperactivity disorder), as well as social and behavioural traits such as years spent in education (educational attainment).

PGT-P is illegal in the UK (see BioNews 1135), but the regulatory landscape varies across other European countries, making its use theoretically possible in some of these countries. Meanwhile, PGT-P is permitted in the USA, where a commercial market is emerging. At the time of writing, it is offered by three companies – Genomic Prediction (see BioNews 976 and 1064), Orchid Health (see BioNews 1238) and Nucleus. Notably, Nucleus is marketing PGT-P for non-medical traits such as intelligence and height, perhaps reflecting a degree of US public acceptance and willingness to use it (see BioNews 1178).

The growing interest in and availability of PGT-P in the USA has prompted major scientific bodies – including the European Society of Human Genetics, the European Society of Human Reproduction and Embryology and the American College of Medical Genetics and Genomics – to issue statements urging caution regarding its implementation in clinical practice. The concerns of these bodies revolve primarily around the scientific validity, clinical utility and wide-ranging ethical implications of PGT-P (see BioNews 1130, 1137 and 1232).

Given this clash between US PGT-P offerings and widespread scientific caution, it is perhaps timely to take a step back and consider the science of polygenic scores (the 'P' in PGT-P) to understand what they are, and what they can – and importantly, cannot – tell us about an individual's risk of or resilience to a complex trait.

Polygenic scores: utility in research

A polygenic score estimates an individual's genetic susceptibility to a specific complex trait. It is obtained by combining the effects of thousands (sometimes millions) of DNA variants associated with that trait into a single summary number.

The information used to construct a polygenic score comes from large-scale genome-wide association studies (GWAS). In a GWAS, scientists compare DNA variation data from very large groups of people who differ in the trait of interest. For example, some have a diagnosis of attention deficit hyperactivity disorder (ADHD) while others do not, or they differ in their number of ADHD symptoms. Using this approach, thousands of common DNA variants that correlate with various traits have been identified.

Essentially, a person's set of DNA variants is measured and each of their variants is given a 'score'. These individual scores are then added up to create an overall polygenic score for a person, with respect to the specific trait. Individuals with a higher score are more likely to develop a specific condition (such as schizophrenia) or rank higher for a trait (such as intelligence) than individuals with a lower score.

For many complex traits, polygenic scores can predict a percentage of the variation seen at a population level. For example, the polygenic score for intelligence predicts four percent of the differences in cognitive function in European populations. This increases to ten percent when combined with the polygenic score for educational attainment. This makes polygenic scores a useful research tool for understanding the developmental pathways to conditions, investigating genetic overlap across traits and exploring gene-environment interplay.

Current limitations of polygenic scores

While polygenic scores can be good at accounting for population-level variation, they are probabilistic and not accurate at predicting outcomes for individuals. For instance, many people with a high polygenic score for depression will not develop the condition, and vice versa. This is because polygenic traits result from a complex interplay of both genes and a multitude of environmental factors (like upbringing and lifestyle) across the life course. Polygenic scores only account for a small part of the genetic picture. Changes in a person's environment, and other unmeasured genetic influences, will alter polygenic score predispositions.

A second key limitation is that polygenic scores may be less accurate when applied to populations whose environmental conditions, demographic characteristics or ancestry differ significantly from those in the original GWAS. Given that the vast majority of polygenic scores are currently developed using genetic data primarily from people of European ancestry, often from a narrow sociodemographic background, this raises serious concerns about the potential of polygenic score testing to reinforce or exacerbate inequities in society (see BioNews 1231, 1293, 1297, 1300 and 1302).

Third, a GWAS tells us nothing about why a DNA variant is associated with the trait – it just finds a statistical link. For social and mental health traits, much of the signal that polygenic scores capture isn't coming solely from the biology of the person's DNA, but rather is coming from social and environmental forces that are tangled up with our genes. This highlights the ethical complexities of making decisions based not only on a person's DNA, but also on factors (such as social advantage) that might be captured in the polygenic score. When efforts are made to filter out such factors, the predictive power of the polygenic score typically drops.

Finally, it transpires that common DNA variants often impact multiple traits. As a result, a person's polygenic score for one trait will often also correlate (positively or negatively) with other, untested, traits. For example, the polygenic score for schizophrenia is positively associated with the score for how long a person stays in education. This means that if prospective parents were to select an embryo at lower risk for schizophrenia, they might inadvertently also be choosing an embryo with a lower score for educational attainment. This highlights the sort of complex tradeoffs that need to be navigated and communicated, if using polygenic scores for embryo selection.

So, will any of these issues be solved as science advances? Certainly, it seems likely that the predictive power of polygenic scores will grow – especially when combined with other information – but they will never be a crystal ball. It is also likely that polygenic scores will become more generalisable, as genetic datasets become more diverse across populations and sociodemographic groups.

However, genetic interconnectedness across traits seems to be part of nature's repertoire, and is here to stay. An added layer of complexity when considering commercial PGT-P is that it is often unclear how the polygenic scores are constructed, which specific variants they include, or how well they perform across different populations and contexts. This lack of transparency adds a further layer of uncertainty to an already complex science.

Lost in translation?

The potential for implementing polygenic score testing in healthcare to inform clinical practice and population health management (through risk-stratified screening) for serious medical conditions has only recently begun to be assessed. Its utility is largely untested when applied to embryos, particularly for non-medical traits such as intelligence. In short, PGT-P is an application of a new and rapidly evolving science, used in a different setting and for a different purpose than that for which polygenic score testing is primarily being researched.

Testing the benefits of PGT-P in reducing risk in future children is challenging, as robust long-term follow-up studies will be difficult to achieve. The few published modelling and simulation studies show that expected relative risk reductions for diagnosable conditions may be substantial, while conversely, gains for dimensional traits are likely to be small. Additional factors – such as the selection strategy, predictive power of the polygenic score, and number of available embryos – will impact potential utility.

These factors need to be considered alongside the long list of possible harms to prospective parents, future offspring and wider society.