What’s in the junk?

Most women​ who undertake IVF will have their embryos screened for genetic abnormalities. Clinics in some parts of the world also offer to select an embryo for implantation based on genetic markers for everything from eye and hair colour to behavioural, emotional and cognitive traits. The next, far more consequential, step is the genome editing of human embryos. As the geneticist Peter Visscher and bioethicist Julian Savulescu put it earlier this year in an article for Nature, genome editing has the potential to create ‘human phenotypes that have never been previously observed’. Stephen Hsu, founder of the embryo screening company Genomic Prediction, believes that genetic engineering ‘will one day create the smartest humans who have ever lived’.

This statement would have seemed absurd to a previous generation of biologists and geneticists, led by Stephen Jay Gould and Richard Lewontin, who forcefully opposed arguments for the genetic determination of intelligence. IQ tests attracted particular criticism, with Gould devoting a considerable proportion of The Mismeasure of Man (1981) to dismissing IQ scores as artefacts of flawed methodology. This was among the criticisms made of The Bell Curve (1994), Richard Herrnstein and Charles Murray’s analysis of the correlations between IQ test data and race, unemployment, poor parenting and crime in the US. More recently, the historian of science Jessica Riskin and biologist Marcus Feldman excoriated Kathryn Paige Harden’s The Genetic Lottery (2021) for repeating Herrnstein and Murray’s mistakes: ‘Harden revives central features of the earlier, now discredited biological theories of intelligence: the presentation of interpretive opinions as objective facts … spurious reduction to a biological mechanism that is not only hypothetical but unspecified; and a claim to be writing in the interest of social progress.’

On this view, attempting to select ‘intelligent’ embryos based on their genetic make-up, or to produce them by altering their genomes, is to make a category error. It can’t work on what Lewontin described as a ‘historically contingent mental construct’. Writing in 1982, he was even more direct: ‘There are no “genes for handsomeness” or “genes for intelligence” any more than there are “genes for saintliness”.’ In this, he was correct. In the 25 years since the human genome was first sequenced, genetic studies have overturned many ideas about the ways in which genetic variation relates to human attributes, including that there are ‘genes for’ things. This idea, which many people still hold, is based on the genetics of monogenic human disease, where mutations in a single gene directly cause, or at least increase the risk of, a particular disease. In the early 2000s, researchers tried, and failed, to apply this logic to diseases such as diabetes and hypertension, as well as personality. Media reports on genes for depression, the ‘IQ gene’ and the ‘warrior gene’ took little notice of the poor quality of the science underlying these studies. Mutations in a single gene can’t explain susceptibility to common complex diseases, mood, personality or intelligence. Geneticists, recognising the failure of these experiments, turned to an alternative model, which considered the combined effect of many different mutations plus the influence of the environment.

The human genome has around twenty thousand genes. Suppose that fifty of them are involved in a disease – for example, the genes involved in making insulin. A mutation in any one of them will have so small an effect that it won’t do you much harm, but if you are unlucky enough to possess mutations in 25 such genes, you have a high chance of developing diabetes (environmental factors, such as poor diet, would further increase the risk). It’s also possible that a particular mutation could have a large effect, large enough that the possession of very few such mutations would be enough to affect your insulin production, but these mutations are so rare that very few people will have them.

Both these possibilities obtain for many traits and diseases. In the early 1990s, the geneticist Mary-Claire King discovered that mutations were causing breast cancer in some families. These mutations were tracked down to a gene called BRCA1; subsequent work discovered mutations in a second gene, BRCA2. Mutations in either BRCA1 or BRCA2 confer a high risk of developing cancer, but the majority of families in which breast cancer occurs do not carry mutations in either of these two genes. There must be additional cancer susceptibility genes, perhaps less significant individually but no less dangerous in combination.

When it first became possible, around 2007, to screen the entire genome for common DNA sequence variants, some were discovered that contributed to disease, altered height and weight, and influenced other traits. Disconcertingly, very few turned out to be within genes. They were almost all found in parts of the genome whose function, if any, was unknown. Our genes make up only about 2 per cent of our DNA. Together with another 8 per cent or so of the genome, they encode important biological functions. The other 90 per cent is junk DNA (that’s the proper term). It doesn’t do anything bad – the geneticist Sydney Brenner compared it to the stuff we keep, as opposed to the rubbish we throw away – but until recently there was no evidence that it did anything important either. In these genetic deserts, sequence variants could hardly be expected to matter.

It turns out that they do matter, not individually but in aggregate. A typical variant makes a miserably small contribution to disease risk, increasing it by 1 per cent or less above baseline (adding a millimetre to your height, say, or a gram to your weight). This makes them hard to find. Genetic studies with tens or hundreds of thousands of subjects are required. And there are many variants for each disease. In 2007, fifty risk variants were considered a lot, but as sample sizes for genetic studies increased, that number kept growing, first to hundreds then to thousands. One recent study of high blood pressure recruited a million people and found 2103 independent genetic signals. We still don’t know the full number: according to a recently proposed omnigenic theory, a given trait might be the product of variations affecting virtually every gene.

The response to these discoveries has been mixed. Some researchers are trying to work out what these sequence variants do, starting from the assumption that they somehow control the function of genes. Even if the effect of any individual variant on disease is small, linking it to a gene might lead to a drug target. This onerous path has made some progress, but it has been slow: almost twenty years after the first genetic studies of inflammatory bowel disease, the identification last year of the role of a sequence variant in a gene desert led to the creation of a new drug. Others have taken the position that the multitude of effects, and their small scale, makes biological interpretation of junk DNA meaningless, but that we can still gain information from it, for example by adding up the effects of all the variants carried by an individual in order to predict their disease risk. This approach is a central feature of Harden’s book, and it is the premise of the screening tests that predict intelligence from a genetic analysis of embryos.

Some researchers have continued to seek the relatively rare but easier to study mutations within genes, to see if any of them cause (or prevent) disease. Finding mutations like this requires the sequencing of all twenty thousand genes in vast numbers of people. But the expense is worth it: mutations have been found in genes that affect the risk of obesity, liver disease, cholesterol levels and diabetes, and mutations in ten genes have been found that increase the risk for schizophrenia. Some genes are promising targets for editing: a clinical trial is currently underway that involves editing a gene to lower LDL cholesterol.

By 2015, it had become clear that the genetic basis of common diseases comprises a multitude of sequence variants outside genes, each altering risk by a minimal amount, together with a few rare, hard-to-find mutations within genes that have larger effects. The picture is the same regardless of which disease is examined, even for psychiatric illness: the genetic bases of hypertension, obesity, diabetes, autism, schizophrenia and bipolar disorder all look roughly the same. And that is also true of variation in other traits, for example height, weight and intelligence. Say there are five hundred genes involved in brain function. Might some of these be candidates for genome editing to alter intelligence? Not necessarily, though to understand why that’s the case, we need to understand the results from experiments examining genetic data on an even larger scale, using data from biobanks.

The UK Biobank is perhaps the single most important experiment in human genetics, though this was never the intention behind it. At its inception in 2003, the epidemiologists, cardiologists and oncologists involved had little interest in genetics; genetic analysis was then of unproven value. However, epidemiologists do care greatly about any risk factor that can be established as a cause of disease, not simply correlated with it. The history of epidemiology is replete with examples of risk factors that turned out not to be risk factors after all (such as the antioxidant vitamins C and E, once thought to reduce the risk of cardiovascular disease and cancer). In principle, genetic analysis solves this problem because association with a genetic variant, fixed at the point of conception, must precede the onset of disease.

The UK Biobank holds the complete genome sequences of half a million people. In 2023 this data was made available to all qualified researchers. It is an enormously valuable research resource. Other countries are attempting to catch up: Finland now has its own biobank, also with half a million participants, as do Estonia, Taiwan, Japan, Denmark and, in the US, the Department of Veterans Affairs and the National Institutes of Health. The largest repositories of genetic data are owned by consumer companies such as 23andMe and AncestryDNA.

Biobanks provide a vast source of material for those who investigate the importance of genetic effects. Any trait recorded can be, and usually has been, subject to genetic analysis: how much coffee you drink, the time of day you get up, whether you nap or not, how much money you earn, sexual orientation and, of course, intelligence. What is unique about these studies is that they are no longer restricted to testing one disease, one trait, in one set of conditions. Instead, they have greatly expanded the picture of what genetic variation does. And the information gained makes it hard to sustain the idea that we will ever be able to edit the genome for traits such as intelligence.

First, the hope that genetics would solve the problem of causation has proved somewhat illusory. For example, in many parts of the world people are genetically more closely related to people who live nearby than they are to those who live further away. And since where you live can also affect your behaviour, a correlation between behaviour and genes could be explained by your birthplace rather than your genes. These effects are small, but then so are the effects that we are looking for. They can also run both ways. A UK Biobank analysis of migration following the demise of the British coalmining industry suggested that the people who left might have a predisposition to higher educational attainment than those who stayed. Does that mean people who are more intelligent tend to have the opportunity or motivation to move out of economically depressed areas, or is it that people who move out of those areas tend to have better access to education?

In a second blow to the idea that there are genes ‘for intelligence’, effects from a person’s own genome seem to be only part of the story. Geneticists now recognise the importance of what are called indirect genetic effects: genetic variants carried by your parents, but not inherited by you. One explanation for this phenomenon is that the variants not passed on affect parental behaviour, creating environments that influence their child’s intellectual and social development. For example, a variant that made your parents more likely to read to you might increase your intelligence, even if you don’t inherit it. Such effects can be counterintuitive; a genetic variant in your parents’ DNA may make you more intelligent when you don’t inherit it, but have the opposite effect when you do.

As researchers have come to better understand these issues, they have developed experimental designs to estimate the contribution of the various factors in play. Applied to height or breast cancer or cholesterol the results are reassuring: direct effects seem to be in the majority and the effect of a genetic variant will be similar from one person to the next. But for educational attainment (taken as a proxy for intelligence) the evidence is shakier: indirect genetic effects are similar in magnitude to the direct effects. What we measure reflects some average of direct effects, indirect effects and ‘confounding’ (misleading correlations), with no guarantee that these would be the same from one individual to the next, and certainly no guarantee that they would be the same when transmitted – or edited – in the next generation.

Third, these variants affect multiple systems. A gene that contributes to intelligence is going to do many other things as well. We can see this in the relationship between intelligence and other traits. For example, genetic variants that increase educational attainment also tend to increase the genetic predisposition to autism spectrum disorder and anorexia, and – though the picture isn’t clear – to influence measures of personality traits. They will undoubtedly turn out to be correlated with many other traits that haven’t yet been assessed. Genome editing under these circumstances is a risky proposition. The genetic basis of cognitive functioning arises from very many DNA sequence variants, which can act both directly and indirectly, often in different directions, depending on the context in which they are found. Possession of the same genetic variant may contribute to an increase in intelligence in one person, decrease it in another and do nothing in a third.

None of this means that intelligence is without biological basis or isn’t heritable – or that it doesn’t evolve. But it does mean that we still only partly understand the relationship between genes and cognitive abilities. Over the last fifty years or so, geneticists have gone from asserting that there is no evidence for the heritability of intelligence to claiming that mutations in a single gene can increase or decrease intellect, to believing that genetic variants in many genes affect the trait, to realising that the important variants aren’t in genes but in junk DNA. Now they think that these genetic effects are subject to a series of confounds no one had fully taken into account. We know more than we did fifty years ago, but we also have a better sense of how little we know.