9

Complexity

At the dawn of the twentieth century, it was already clear that, chemically speaking, you and I are not much different from cans of soup. And yet we can do many complex and even fun things we do not usually see cans of soup doing.

PHILIP NELSON

Speaking of safety … thalidomide is a remarkably useful drug. By suppressing the growth of blood vessels, it effectively treats a common and serious complication of leprosy, and it is active against a type of cancer called multiple myeloma, as well as against certain inflammatory conditions. Unfortunately — as it turns out — it’s also a really good treatment for nausea, and works as a sleeping pill without causing addiction. The latter was a powerful claim when the drug was introduced, because it was a time when barbiturates were widely used and were known to cause death from accidental overdose,81 as well as being highly addictive. Unlike barbiturates, you can swallow a handful of thalidomide tablets and survive the experience without any immediate harm — a handy angle for marketing purposes.

[1 Also deliberate overdose — Marilyn Monroe was only one of many who died this way.]

Advertising for the drug highlighted its safety: an advertisement in the British Medical Journal in 1961 proclaimed the drug ‘highly effective … and outstandingly safe’. One advertisement showed a toddler holding an open medicine bottle, with the message being that, if the bottle had contained barbiturates, the toddler would be in mortal danger. Luckily, the bottle was instead full of nice, safe thalidomide.82

[2 The modern childproof lid for medicine bottles was invented in 1967 by Mr Peter Hedgewick, following efforts over a five-year period to encourage development of such a lid by a Canadian paediatrician, Dr Henri J. Breault. Breault was sick of treating children who had accidentally overdosed and was determined that such overdoses should be prevented. The idea caught on quickly in Ontario, but, despite the obvious benefits of the invention, it took some years for its use to become widespread elsewhere.]

In fact, the drug had hardly been tested for safety at all. A few years after its introduction, it emerged that thalidomide could cause damage to nerves, resulting in chronic limb pain. Then, in late 1961, the first reports of damage to unborn babies started to trickle in. The drug was withdrawn from sale almost immediately in most countries.

Sadly, this was too late to prevent an unprecedented medical catastrophe. Thalidomide’s anti-nausea effects had led to it being marketed as a treatment for morning sickness during pregnancy, starting in the late 1950s. As a result, thousands of children were born with physical malformations. The most striking of these were severe limb deficiencies — being born with missing hands or arms, or with all four limbs reduced to small remnants or being entirely absent. Other problems, such as congenital heart disease and kidney malformations, were also common. Many affected babies did not survive infancy, and many pregnancies were miscarried due to the effects of the drug.

Among developed nations, the United States stood apart in being almost completely spared the ravages of thalidomide. This was thanks to the brilliance of a doctor who was working for the Food and Drug Administration (FDA) at the time of the drug’s introduction elsewhere in the world. Dr Frances Kelsey reviewed the safety information provided by the drug’s manufacturer. She found the evidence unconvincing. Six times applications were brought to the FDA, and six times Dr Kelsey rejected them — saving unknown thousands of children.

Kelsey was a remarkable woman. Born in Canada83 in 1914, she trained in pharmacology at McGill University, receiving her Master of Science degree in 1935. She applied to the newly established pharmacology department at the University of Chicago for a research-assistant position, which came with a scholarship that would allow her to study towards a PhD. She was delighted to receive a letter of offer, but concerned that it came addressed to Mr Kelsey; it appears her name had been confused with ‘Francis’. She asked her professor at McGill if she should send a telegram explaining the point. He told her that would be ridiculous, and advised her to reply accepting the position, but writing (Miss) after her name. Interviewed by the FDA Consumer magazine in 2001,84 Kelsey said, ‘To this day, I do not know if my name had been Elizabeth or Mary Jane, whether I would have had that first big step up. And to his dying day, Professor Geiling [her supervisor in Chicago] would never admit one way or the other.’

[3 Canadians seem to have made a disproportionate contribution to the cause of drug safety.]

[4 Like you, I always keep a stack of these on my bedside table.]

Kelsey’s first big impact on medicine came while she was studying with Geiling. He was asked by the FDA to investigate a series of deaths in people treated with a new antibiotic, sulphanilamide. Kelsey helped to discover that the method used to prepare a syrup version of the drug (which had an unpleasant taste in pill form) involved the use of a poisonous but sweet-tasting solvent, diethylene glycol.85 The syrup tasted good, cured infections, sold very fast … and, because it had not been tested for safety, also killed 107 people, many of them children. This disaster seems to have left a deep impression on Kelsey, and doubtless shaped her thinking in relation to drug testing. As you might expect, it had a similar effect on the country as a whole, and led to legislation, the Federal Food, Drug, and Cosmetic Act of 1938. This law has had a profound and long-lasting impact on the way that the safety of medications is managed. The Act required drug companies to show evidence that medications were safe before they could be marketed, and to provide warnings of potential hazards. This seems an obvious idea now, but it was a major shift from the free-for-all that had prevailed beforehand, and, because of the importance of the US as a centre for drug development and as a market, has had a major and lasting effect on the management of drug safety in the rest of the world as well.

[5 A similar compound, ethylene glycol, is used as antifreeze in car radiators, and is also poisonous.]

In 1960, Kelsey was offered a job at the FDA. The very first task she was given was to review the application for approval of thalidomide. The drug had been used in Germany since 1957 and had rapidly spread through the world after that; it was generally believed to be safe. Kelsey was given the job of reviewing this particular application because it was thought a good idea to give new employees a relatively easy assignment to start with(!). Fortunately for thousands of American children, Kelsey’s training, experience, rigorous approach to the evidence presented, and refusal to be intimidated by the drug company (which put enormous pressure on her to relent), all made her ideally suited to assess this particular application.

For once, the rest of this story is not that of an unsung hero whose work was disregarded during their lifetime. Kelsey received the credit she richly deserved. The Washington Post ran a front-page story entitled ‘“Heroine” of FDA Keeps Bad Drug Off of Market’. In 1962, President John F. Kennedy presented her with the President’s Award for Distinguished Federal Civilian Service, the highest honour the US government gives to its civilian employees. In 2000, she was inducted into the National Women’s Hall of Fame. She retired in 2005, aged 90. In 2010, she received the FDA’s inaugural Drug Safety Excellence Award, which was named after her and is given annually to an FDA employee. Kelsey died in Canada in 2015, aged 101, shortly after being presented the Order of Canada by the Canadian governor-general.

By contrast … I know this is probably unfair on her, given that I am writing from a position of hindsight, but I can’t resist telling you about a letter written to the British Medical Journal by one Paula H. Gosling, published on 16 December 1961. It is a nice example of a longstanding tradition of medical resistance to change. Gosling roundly criticises the decision to withdraw the drug from sale. ‘I must protest against the action of the Distillers Company in withdrawing … thalidomide … from the market’, she says, and then argues that, if this had been done because of the reports of nerve damage, that would have been one thing, but ‘to do so on the grounds of two unconfirmed reports from alien sources of a possible — not proved — association of thalidomide given in early pregnancy is quite irresponsible’. Charmingly, she goes on to point out that because thalidomide is tasteless and odourless, it is the ideal sedative for children … and cats! Lastly, she suggests that ‘any doctor alarmed by these unconfirmed reports’ could prescribe other drugs ‘for pregnant females’ … and concludes, ‘Have we lost all sense of proportion?’

No doubt Gosling came to regret this letter before much time had passed.86

[6 In fairness, I should confess that I have done something rather similar. I wrote a testy letter to the journal Nature, following the early reports that babies whose mothers were infected by zika virus during pregnancy could suffer damage to their brains, causing them to be born with small heads and go on to have neurological problems. The letter mainly complained about sloppy use of terminology, but also pointed out that, so far, there wasn’t really a lot of hard evidence for harmful effects from zika infection. Within a week of its publication, a paper came out in The New England Journal of Medicine that nailed down the relationship, beyond any reasonable doubt. I had vigorously and thoroughly plastered my face with egg, while standing on the most visible platform in science.]

If a drug that is taken during pregnancy causes malformations in the baby, it is called a teratogen.87 Thalidomide is one of the most potent teratogens known — even a single tablet, taken at the wrong time, can have devastating effects. And yet … and yet there were babies whose mothers took thalidomide during the critical time in early pregnancy, when the fetus is most sensitive to its ill effects, and were nonetheless born whole, hale and hearty, somehow unharmed. Very likely, the babies born unscathed by their exposure to thalidomide were a small minority — but there is good evidence that they exist. How could this be?

[7 The word derives, rather unfortunately, from a Greek word meaning ‘monster’. Medications are only one type of teratogen — other drugs, such as alcohol, and maternal infections, such as zika, can also be teratogens.]

*

A few times a year, I give a lecture to medical students about genetics. The idea is to give them an overview of the role of genetics in medicine, reminding them of the fundamental principles they were meant to have learned earlier in the course, and trying to give them a feel for where the field is going. If you’re a medical student today, it’s pretty much guaranteed that some of your future patients will be having genetic tests, and, if you’re going to order a test and receive a report, it helps to have some idea of what the results might mean. Anne Turner used to give the same lecture, which she called ‘genetics in an hour’, and, in the years since I took it over, I find myself constantly realising that some new development should be added, then discovering that the lecture is too long and deleting slides.

Even so, I always take time to make the case that each and every other branch of medicine should be viewed as just a subspecialty of genetics. Pretty much everything that afflicts human beings, and everything about us that is not an affliction, too, has genetics at its core.

Take trauma, for instance. You may not think that being involved in a car accident, or getting punched, is a genetic problem. Consider, though, that there is a major genetic risk factor that strongly influences whether these things will happen to you. It is called the Y chromosome. At every age after 12 months — the age at which most of us are able to get up, about, and into mischief — males are more likely to be injured than females.

You may have some thoughts about why this should be the case. The obvious culprit is testosterone: aggressiveness and impulsiveness are characteristics that will put you in harm’s way, and testosterone is known to promote both.

That may well be a contributing factor, but most likely it’s not as simple as that — there are probably other ways that the Y chromosome influences behaviour. Of course (despite my desire to put genetics at the centre of the medical universe), we should not imagine that the differences in behaviour between men and women are due only to their different and inborn physical and chemical make-up. There are social as well as genetic influences on maleness and the way that men behave. If you are conditioned from birth to think that you should be adventurous and bold, nobody should be very surprised if you do wind up adventurous and bold. If you’re taught that your place is in the home, and that quiet, passive pursuits will suit you best — well, you might cut loose, but, on the whole, there’s a decent chance that this upbringing will affect your choices in life, in ways that tend to keep you out of harm’s way.

Yet to muddy the waters still more, there are other genetic influences on your chance of getting hurt, whether or not you are encumbered by a Y chromosome. Not all men have the same risk of getting hurt as each other, and there is plenty of overlap in behaviour (and risk) between men and women. Both men and women can be more or less impulsive; some men (and women) are more likely to stay home and play video games, while others prefer to go out BASE-jumping. Those differences between members of the same sex are also under genetic control to some degree, but understanding them is not as simple as being able to point at a single errant chromosome and blame that.

In short, the genetics of trauma are complex, and there is an interplay between genes and environment. That balance is not always the same, and there are times when environment can completely overrule genetics. Put the most peaceable, unadventurous woman in Baghdad and she may, by sheer bad luck, be at the market when a car bomb explodes. Put the most (potentially) reckless and aggressive of men in an environment where men are socialised against violence and other opportunities for harm are limited (you’ll note that I couldn’t think of a good example of such an environment, but all things are relative) and he may live, unscathed by any violence, to a ripe old age and then die in his sleep.

This type of interaction between genes and environment has an effect that is easiest to study and understand in a population, rather than at the level of a single individual. It’s a bit like the difference between climate and weather: we know that, on average, men are more likely to experience violence, just as we know that, on average, it’s hotter in summer than in autumn. Nonetheless, it’s not terribly surprising for there to be a cool day in summer or a warm day in autumn; the individual days are like individual people. Just as we can’t predict what the weather will be like a year from today, even perfect knowledge of a baby’s genetic make-up will never tell us exactly what type of person they will become, or even exactly what kind of health problems they may experience.

Virtually all human diseases have some contribution from genes. This ranges from conditions like those that mostly fill this book — in which a single gene is faulty or there is a problem at the level of a chromosome, and that is enough on its own to cause a genetic condition — through to common conditions like stroke, in which genes and environment are both important, and the genetic contribution is not from a single gene but from many. That latter scenario involves many, many different genetic elements, each of which gives a tiny nudge to the chance that you will have a stroke. Some nudge your risk up, some nudge it down. It’s likely that there are hundreds or even thousands of such genetic nudges acting on the risk for each common medical condition; for most people, most of the time, any one of these only has a very small effect. A genetic variation that increased or decreased your chance of having a stroke by 20 per cent would be considered to have a very strong effect. A catalogue of genetic influences on stroke88 identified 287 different places where there was good evidence for such a nudge, most of which only have only a slight impact on risk.

[8 Part of a catalogue kept by the National Institutes of Health, accessible at https://www.ebi.ac.uk/gwas/home]

The main way that we have identified genetic influences on common disease, so far at least, is through genome-wide association studies (GWAS). If you want to run a GWAS,89 you need very large numbers of people — tens to hundreds of thousands90 — about whom you know something. Perhaps you know how tall they are, or what their blood pressure is, or whether they have had a stroke. You get DNA from each of them and look at thousands of places, spread across the genome, where there is known variation between people. Then you compare these genetic results with the known information about the people in your study. The aim is to find a link between the DNA results and the characteristic you are studying.

[9 Personally, I have no desire to do this, but I won’t judge you if you decide to give it a go.]

[10 Putting together data from such large numbers of people often requires huge collaborations. In 2014, a study of the genetics of height published in the journal Nature Genetics had 445 authors, as well as four groups who were not individually named. Yes, I counted them.]

Suppose there is a particular spot where some people have a C whereas others have a T. When we look at people who have not had a stroke, we find that 50 per cent have a C and 50 per cent have a T. Then we look at people who have had a stroke, and find that 60 per cent have a C and 40 per cent have a T. People with stroke are more likely to have a C than the controls.91 The next step is to do the study all over again, in a ‘replication cohort’ — a second group of people — to demonstrate that the link you found the first time around wasn’t a fluke. This is necessary because, in the early days of GWAS, there were numerous GWAS ‘hits’ that turned out to be statistical blips with no relationship to reality. The statistical bar for that second study is a bit lower than for the first, because you’re looking at one particular target rather than scanning the whole genome. This means you don’t need quite as many people in your second group, but it’s still a lot of work. Let’s say that, second time around, you find something similar to your original result. Congratulations, you’ve found a risk factor for stroke!

[11 If the numbers had been the other way round, C would have been protective against stroke rather than a risk factor.]

… but it may not help you all that much. For a start, it may well be that this genetic difference does not, of itself, have any direct bearing on the risk of stroke. It could be an innocent bystander that just happens to be sitting, minding its own business, somewhere close to some other change that is the actual culprit. This means that identifying a variant like this is often only the first part of a long and frustrating search for the actual villain in the piece. The second problem is that the information is pretty meaningless for any one person. If half of the population have a C, and there’s only a small extra risk of stroke in people who have a C, you shouldn’t get too worried if you find that you have it, too. And the information may not be meaningful for people from a different population — the link between this particular C and stroke might only hold true in people from a European background, for example. There is an unfortunate oversupply of studies done in people with European ancestry, unfortunate because of the serious lack of similar studies in people from other populations.

There is a pretty good chance that your GWAS hit does not sit inside an actual gene. Most do not. Sometimes that’s because of the innocent-bystander phenomenon mentioned above — there is a change in a gene that’s important, and the C that you found is sitting near that change. More often, though, if we can find a specific reason why a particular variant is associated with a condition we’re interested in, it relates to how genes are controlled, rather than to changes that result in a different version of a protein being produced. The genome is full of sequences that are important for regulating activity in the cell nucleus. This often works by producing signals in the form of RNA, a chemical that is nearly-but-not-quite the same as DNA. This activates that, which suppresses the other thing, which in turn changes the activity of a gene, meaning more or less of a protein being produced … which might be relevant to the thing you were interested in to begin with. Our understanding of this network of signals is far from complete, but it’s likely that many GWAS hits involve subtle shifts of balance in a tangled web of information — not something we are likely to figure out in a hurry.

Ideally, we would like to identify all of the genetic variation that affects human92 diseases, as well as characteristics like height. If we could completely understand the genetic influences that decide whether a particular person will have a heart attack, it might be possible to find new ways of preventing that from happening.

[12 My focus in this book is on humans, but the same technology is widely used in other organisms, including for the purposes of agriculture. If you can identify genetic variation that influences milk production in cows, the dairy industry will beat a path to your door. There are plenty of similar applications in other areas, from crop production to horse racing — and it goes beyond the more obvious organisms. For some years, until he retired, I collaborated with Professor Chris Moran at the University of Sydney. Chris was part of the team that sequenced the genome of the saltwater crocodile, and he worked on finding variation that affected the speed at which young crocodiles grow, as well as characteristics of their skin that are important in crocodile-leather production. Virtually every economically important organism — from honey bees to rice to farmed salmon — has had its genome studied in an effort to figure out how to make them more productive, and lucrative.]

Long before the human genome was sequenced, there were efforts to figure out how important genes are in controlling various conditions and characteristics. Think ‘nature vs nurture’, with an effort to actually measure the ‘nature’ part. A widely used measure in this area is heritability — the degree to which variation in a particular characteristic within a population is due to genes rather than environment. The name makes it sound like this is a direct measure of the ‘nature’ part of the equation, but that’s not quite right: it really is about variability within the population. To illustrate this point, imagine you are studying hair colour in two different groups. One group is composed entirely of Nigerians; the other is composed of a random sampling of Brazilians. The Nigerians all have black hair, so there is no variation to measure: heritability will be zero. That doesn’t mean genes aren’t important in controlling the hair colour of Nigerians — quite the opposite is true. The Brazilians have everything from black hair to blonde. This variability is mostly explained by genetics, so heritability will be high.

There are several ways of calculating heritability. A common and relatively simple approach is to compare the degree to which identical and non-identical twins are similar. The assumption is that twins share their environment equally, even down to the conditions they experience while still in the womb, and that environmental effects are no different for identical and non-identical twins. Since identical twins share all of their genes,93 whereas non-identical twins share, on average, only half of their genes, you expect (and usually observe) that identical twins will be more similar to each other than non-identical twins — not just in physical appearance but also in height, blood pressure, and so on. Some fairly simple maths, by the standard of this sort of thing, lets you measure that difference across a group of twins, and use it to calculate heritability. Heritability scores, however calculated, range from zero (no contribution from genes to the variation in that population) to one (the variation is entirely due to genetics); they can also be expressed as percentages.

[13 In principle. In fact, it is possible for identical twins to have differences either in the sequence of their genes (that happened after the split, and that would usually be mosaic) or in some of the settings within the cells that control the functioning of genes. Usually, these are too subtle to cause differences that you can observe just by looking at them, but they are real nonetheless.]

Heritability estimates for different characteristics vary between studies, and between populations. One measure that is relatively consistent across a number of studies has been height: in well-fed populations, height has a heritability of 0.8. Most of the variation in height is due to genetic effects. Studies in China have found lower heritability for height, around 0.65. That’s still an important contribution from genes, but suggests that environment might be more important in China than in, say, the US. A possible explanation for this relates to the fact that, if your mother is undernourished while you are in the womb, and you are undernourished as a child, you are likely to wind up shorter than you would have done otherwise. Study people who have grown up in hard times and you will see a larger impact from the environment, pushing the contribution from genes down.

GWAS became possible in the mid-2000s, and really took off in about 2007. Pretty quickly, the GWAS-ologists became aware of an awkward problem: missing heritability. By 2010, after quite a bit of effort, GWAS studies had managed to explain only 5 per cent of the variation in height — quite a gap from the 80 per cent predicted by measures of heritability! Gradually, over the past decade and a bit, that gap has narrowed — but there is still a gap and it’s not fully explained. The biggest study of the genetics of height so far used data from nearly half a million people from the United Kingdom, volunteers who had donated DNA and extensive personal and medical information as part of the UK Biobank project. A group led by Stephen Hsu (our second Hsu — of whom more later) was able to use this treasure trove of data to explain 40 per cent of the variation in height — an impressive feat, but far from allowing us to predict someone’s height accurately from their DNA alone. The predictive score they developed allowed them to estimate the heights of most of a group of people (not part of the original data set) to within a few centimetres. That sounds impressive, but the range of possible actual heights for any given predicted height was pretty wide. Imagine a witness in a court case saying, ‘I think the man was about five foot eight, give or take an inch or two. But he could have been as short as five foot two or as tall as six foot two.’94 Hsu’s group also convincingly showed that, for height, studying more people or looking at more genetic markers would be unlikely to provide more accuracy — they have taken this approach about as far as it is ever likely to go. They also looked at educational attainment, measured on a six-point scale that tops out at ‘university degree’, and found they could explain just 9 per cent of the variation in that measure, but also that an even more enormous study would have a decent chance of explaining more of the variation.

[14 Or, if you prefer metric, ‘I think the man was about 173 cm tall, give or take a few centimetres, but he could have been as short as 157 cm or as tall as 188 cm.’]

Why is it that even a superb study, using huge amounts of data, can only explain part of the variation that should be there to find? There are two main explanations that are put forward. The first argues that the traditional measures of heritability are substantial overestimates — for example, a recent study using data from Iceland and incorporating genetic data (from whole genome sequencing) in calculating heritability found rather lower values than from the traditional approaches. For height, rather than heritability of 0.8, the new estimate was only 0.55 — suggesting that Hsu’s group found more than 70 per cent of all that there was to find. For body mass index (BMI),95 the numbers were 0.65 (old method) and 0.29 (new method); for educational attainment, it was 0.43 and 0.17 — both very substantial differences.

[15 A measure of the relationship between height and weight; it is weight in kilograms, divided by height in metres squared.]

Perhaps more interestingly than just a miscalculation, an alternate explanation for missing heritability is that there are huge numbers of undiscovered genetic contributors to the characteristics that are being studied — but most of them only have very, very tiny effects, that are too hard to measure even if very large numbers of people are studied. This idea is not new — its centenary was in 2018. In 1918, R.A. Fisher, one of the luminaries of modern statistics, proposed the infinitesimal model, in which a variable characteristic like height is under the control of an infinitely large number of genes, each of which has an infinitely small impact on the characteristic, as well as, of course, environmental influences. Fisher wasn’t suggesting there really are an infinite number of genes, it was just a way of thinking about the problem.

It’s possible you haven’t heard of Fisher, but in certain circles he remains a revered figure. This is not so much for the infinitesimal model — a relatively minor achievement by his standards — but for many other major contributions to statistical and genetic theory. Ian Martin, the scientist who taught me how to breed mice,96 retired a few years ago. In December 2016, the University of Sydney honoured Ian’s contributions to the field of mouse genetics, and to the university, by awarding him an honorary doctorate in veterinary science. At a reception held before the ceremony, Ian told me in detail about the time that he met Fisher and showed him some of his work, which Fisher commented on favourably. Ian was awarded his PhD in 1962, the year that Fisher died, so the events must have occurred well over half a century before, but it was obvious that their encounter was as clear in Ian’s mind as though it had happened the previous week. I was deeply impressed to hear that my friend and teacher had known the great man.97

[16 For scientific purposes, not as a hobby.]

[17 I wish I could say that Fisher’s reputation is as a giant of statistics and nothing else. Unfortunately, his reputation is somewhat blemished by his membership of the Eugenics Society at Cambridge, and by his firmly held views that there are real and important differences between the races of humans. A small positive of this, for me at least, is that through Fisher I am connected in a surprisingly small number of steps to Charles Darwin. I know Ian Martin, who met Fisher, who knew Horace Darwin — a fellow member of the Eugenics Society, and son of Charles Darwin. That’s just four degrees of separation. From Darwin!]

Let us drag ourselves reluctantly from scientific hero-worship to look at ‘environment’ a bit more. If the potential environmental influence on your health is a speeding truck, it’s fairly obvious. If it’s an infectious disease, you might be tempted to think not only that the environmental component (the virus, bacterium, fungus, or parasite) is obvious, but that it’s the only thing that matters. If so, you’d be wrong. Some people have immune systems that deal well (or poorly) with particular types of infection. At one extreme, there are people who are almost immune to infection by HIV, as we’ve seen (in chapter 8). At the other are people with genetic immune deficiencies, whose bodies are terribly vulnerable to some types of infection, or to any infection at all. All the rest of us lie somewhere in between those extremes, along a spectrum of susceptibility.

There’s an interaction between host and infection, and not all germs are created equal, so that both sides of this equation are variable. Someone who is naturally resistant to influenza encounters a mild strain of the flu and doesn’t even notice that they are ill. Someone else, with an average immune system, encounters a flu virus that is average in nastiness, and has a miserable few days but recovers with no lasting effects. And someone with a vulnerable immune system has the bad luck to run into a particularly gnarly virus … like the H1N1 strain that rampaged across the world in 1918 (Spanish flu) and 2009 (swine flu) … and that person dies.

In genetics, a particular focus of our thinking about the interaction between genes and environment is the early months of pregnancy. Thalidomide is an extreme and infamous example of an environmental influence on early development, but there are other medicines that are known to have the potential to harm the unborn baby. Isotretinoin, marketed as Accutane or Roaccutane, is very effective at treating acne. Unfortunately, if a woman becomes pregnant while taking this drug, there is a very high risk of a variety of problems in the baby — physical malformations and intellectual disability — although many babies are unharmed. Some drugs that are used to treat epilepsy can cause problems for the baby as well, which can lead to a difficult choice in women who have severe epilepsy and need to take a drug that is risky to use in pregnancy. Do you accept the risk to the child or switch drugs, knowing the mother might have seizures, which come with their own problems?

It’s not just medications that can potentially hurt the developing baby. Alcohol can cause significant harm, particularly to the brain. Most of the evidence for this comes from the children of mothers who have consumed large quantities of alcohol for a large part of the pregnancy, but it’s not clear whether there is any truly safe amount of alcohol to consume, so the advice that’s given to pregnant women is appropriately cautious. Smoking during pregnancy can cause miscarriage, or poor growth during the pregnancy, among other problems. There are infections that are serious problems — the main reason we vaccinate against rubella is to prevent infections during pregnancy, and there are plenty of other infections which are bad news for the baby (zika virus is a recent addition to the list). The babies of mothers who are diabetics are more likely to have a range of physical malformations; this isn’t a problem for women with the version of diabetes that is caused by pregnancy (gestational diabetes), because, by the time this problem starts, the growing fetus is fully formed. Not that gestational diabetes is harmless — far from it: it carries significant risks to mother and child, but physical malformations in the baby are not among them.

All these hazards are real, but they all share the characteristic of unpredictability. You can’t be sure — even with thalidomide — exactly what will happen in a pregnancy. There’s also the problem that, for many medications, there just isn’t much evidence about risks in pregnancy; this has led to the development of efforts to study and monitor the outcomes in babies whose mothers took any kind of medication during pregnancy.

We don’t have a full understanding of why some babies are badly affected by a particular insult, while others suffer no harm, but there’s good reason to think that genetic variation is involved. That could be variation that directly protects the baby — perhaps that baby’s genes that give the instructions to make arms and legs, and to keep the blood flowing to the growing limbs, are particularly robust versions. Or it could be maternal genetic factors that protect the baby. Perhaps the mother has a genetic variant that means that her gut absorbs thalidomide poorly, so that levels in her blood never get high enough to matter.

On the whole, public awareness that the mother’s actions can sometimes harm her baby is a good thing. It leads women to stop smoking and avoid alcohol during pregnancy, and to be careful about the medications they take. But like many things, this knowledge can be a two-edged sword. I once saw a man in his early 40s, Barry, who had intellectual disability. He could walk, and talk a little, but was dependent on his parents for most things. His younger sister, in her 30s, was pregnant and was concerned that her child might also have intellectual disability. Efforts to make a diagnosis had been unsuccessful back when Barry was a child, in the 1970s, and nobody had tried since then. After a while, the question seemed less important to those who cared for him than managing his day-to-day health problems. Finally, though, Barry’s sister asked the question again. When I saw Barry, I suspected a chromosomal problem, and so it proved when I tested him — he had a chunk missing from one of his chromosomes that was undoubtedly the cause of his problems.

The news had a profound impact on his mother, a woman in her late 60s. While pregnant with Barry, her first child, she had painted the nursery; after his problems manifested, she became convinced that she had absorbed fumes from the paint that had damaged her baby’s brain. She believed that Barry’s problems were entirely her fault. For decades, she had carried this needless burden of guilt; it had eaten away at her for all that time. The chromosome result lifted the burden from her. Like others I’ve seen in this situation, her emotions were powerful and complex; relief that she was not at fault, mixed with sadness for the years she had spent believing that she was.

Asking about events during pregnancy is an important part of a clinical genetics consultation, because occasionally there can be important clues about the cause of the problem. But even when it’s obvious that nothing that happened was relevant to the child’s problem, I try to remember always to ask the parents if there is something that happened that they are worried about, because people don’t always volunteer the information. For many years, I was part of the clinic at Sydney Children’s Hospital that looks after children with clefts of their palates, lips, or both. A cleft is a split, usually caused by a failure of separate parts to join up as they should during early development. Sometimes, I would see children who had an underlying genetic condition, be that chromosomal (like velocardiofacial syndrome) or some other syndrome that was the cause of the cleft. But most children at the clinic are perfectly healthy apart from their cleft, and, thanks to the excellence of modern plastic surgery, they generally do very well once the cleft is repaired, with surprisingly subtle scars for those with cleft lips.

Their parents still want to know why this happened, and whether it was their fault. Often, the most useful thing I did in that clinic was taking explanations away from people, rather than the opposite. No, the stress you were under at work was not the cause of the cleft, and nor were those antibiotics you took, or the toner you got all over your hands that time that the printer at work went berserk.

Although we can’t usually point to a specific cause of conditions like cleft palate or congenital heart disease, we can have a decent crack at working out how important genetic factors are in causing them. For both of these conditions, we know that there are some families in which a change in a single gene is the main cause of the problem. I’ve contributed in a small way98 to identifying single-gene causes of congenital heart disease, and my friend Tony Roscioli (who is a genetic Sherlock Holmes, with an extraordinary capacity to sift through genetic data and uncover new links between genes and diseases) has found several genes linked to cleft palate. But even for ‘single-gene’ conditions, the outcomes are unpredictable — one child in a family might be born with a heart so scrambled that it is beyond repair, whereas another, despite having the same genetic variant, might have a minor problem, or even a completely normal heart. In this setting, an important part of the environment is something unexpected: it’s luck.

[18 As a minor player in a cardiac genetics group led by two embryologists, Sally Dunwoodie and Richard Harvey, and a cardiac surgeon, David Winlaw — powerhouses all.]

It may seem like a cop-out to say, ‘This child had a genetic variant that might have been fatal, but was lucky and suffered little harm, whereas this other child was unlucky and died.’ But there’s a solid scientific basis to this idea. At the scale of a single molecule of DNA, there can be some ‘wobble’ in the way that proteins and DNA interact. Likewise, the regulation of gene activity is not just a matter or on or off. Sometimes the switch flicks to a higher than usual setting, and sometimes it gets stuck on low. Over a large enough number of cells, this wobble averages out so that it (mostly) hardly matters. But when we’re talking about the very early development of an embryo, when tiny clumps of just a few cells are deciding their fate, that wobble can be enough to make a real difference. Like the proverbial butterfly that flaps its wings and causes a hurricane, small changes in just a few cells at just the wrong moment in early development can have a big effect down the track.

So, we can see that there’s a spectrum of possibilities that encompasses all human disease. There are conditions — like being hit by a car, or suffering the effects of exposure to thalidomide — that are mostly about environment, but are also about genetics — favourable genetic variation, which might protect you from thalidomide, or unfavourable variation, which might make you vulnerable to infection. There are conditions that are mostly caused by changes in a single gene — modified by the effects of other genes and by the environment. And there’s everything else in between, with a great many conditions that are the result of thousands of small genetic nudges interacting with each other, and with the environment.99

[19 Not a genetic condition, exactly, but eye and skin colour represent something of a special case, with a relatively small number of genetic variants combining to produce most of the variation. There are handy charts available online that explain that blue eyes are a recessive trait, with pictures showing brown-eyed parents having a 1 in 4 chance of having a blue-eyed child. These have the advantages of simplicity and clarity, and the disadvantage of being completely wrong. Eye colour is in fact a complex characteristic, but, unusually, looking at variation in just a few genes — especially HERC2 and OCA2 — can give you a lot of information about eye colour. A group of Dutch investigators, studying a majority blue-eyed population, were able to predict that a person had brown eyes with 93 per cent accuracy using just six variants in six genes. Skin colour has similar genetics, including — unsurprisingly — sharing several of the same genes that influence eye colour.]

With this understanding comes the possibility that we might be able to use genetic markers to make powerful genetic predictions about health, or about other characteristics. Maybe a single GWAS hit by itself can’t tell you much about your future wellbeing, but what about combinations of hundreds of them? Could you put the information together and make meaning out of it? The answer turns out to be: sort of. So-called ‘polygenic risk scores’ have been put together for various common medical problems: stroke, heart attack, diabetes, various types of cancer, and so on; essentially this is the same as the approach used by Stephen Hsu’s group to try to predict human height. The best of these are starting to become useful medical tools already. For example, a polygenic risk score can be used in combination with other information to calculate a woman’s chance of developing breast cancer, in a way that can potentially change the type of screening she is offered.

Other such scores have been less successful, so far at least. A group from Germany and the UK used data from the UK Biobank including 306,473 people aged 40–73 to develop a risk score for stroke. They succeeded: using their model, the people in the top third for risk had a 35 per cent higher risk for stroke than those in the bottom third. That sounds pretty good — until you consider that they also found that by asking four simple questions about lifestyle factors, they could do rather better: people who had zero or one healthy lifestyle factors100 had a 66 per cent higher risk of stroke than those with three or four. In even worse news for the risk score, the lifestyle benefits applied to everyone equally, regardless of their genetic profile. Conclusion: it’s best to just advise everyone to lead a healthy lifestyle. The genetic test, based on superb underlying data and being scientifically as robust as you could hope for … doesn’t add a great deal to decision-making about how to live your life in order to reduce the chance of stroke.

[20 We’re not talking about needing to be a vegan triathlete here. The factors they considered were: no current smoking, healthy diet, body mass index < 30, and moderate physical activity two or more times weekly. If you’re a (current) non-smoker and not obese, your score is two already.]

That doesn’t mean that scores like this won’t be used by someone — if not for medical reasons, then to make money. There are a variety of companies that will test your DNA and give you detailed advice about lifestyle and diet based on the results. If you come from the same genetic background as the people who were studied to develop the scores (i.e. mostly European), the results might even be scientifically valid, more or less. They just aren’t terribly meaningful for an individual. Before you pay for such a test, consider that there is really no prospect of receiving a report that says, ‘It’s fine for you to smoke, avoid exercise, and eat a diet that’s full of sugar and saturated fats and low on vegetables.’ A test result that says your body might not tolerate alcohol very well is not nearly as definitive as having a few drinks to see what happens101 … and so on.

[21 In the name of science, of course.]

A possible exception to the ‘this stuff isn’t all that useful’ rule is testing to see how well your body handles medications. There are quite a few genes involved in the metabolism of drugs, and these vary between people — some of us make a version of an enzyme that works only sluggishly and struggles to keep up, whereas others have a highly efficient version that works particularly well. Knowing about these genes might be important. For example, if your body processes a drug fast, you’ll probably need a higher dose than other people, or you’ll never have enough of that drug in your body to do you any good. On the other hand, if you’re a slow metaboliser, the drug might build up in your body and poison you, so you need a lower dose than usual. Most of us either don’t have these variations, or won’t ever need the medications for which this matters — but without either having a test or suffering the consequences, there’s no way of knowing.

Most of the companies that offer to test your DNA for complex conditions confine themselves to telling you your granny probably came from Eastern Europe,102 and that you should eat more fruit and vegetables. At worst, some of them will also suggest you might benefit from specific dietary supplements … that, by sheer coincidence, they are in a position to sell you. Mostly, though, as long as you don’t take it too seriously, this kind of ‘recreational genomics’ is relatively harmless, and might even be helpful. If being told that you have a higher than average chance of developing type 2 diabetes leads you to eat better and do more exercise, that can only be a good thing.

[22 If she’s around, you could also find this out by asking her. But if there are gaps in the information available to you, this kind of testing is getting really quite good, although it’s better for some populations than others. It will be no surprise to you at this point to learn that being of mainly European ancestry is an advantage, in terms of the likely detail and accuracy of the test results.]

There’s always someone willing to take things further, of course. Stephen Hsu is one such person. Hsu started out as a theoretical physicist, and is evidently very good at it — he is Professor of Physics at Michigan State University. Unusually for a physicist, he has branched out into genetics. Hsu hasn’t confined himself to studying height or heart disease — he has a keen interest in the genetics of intelligence, and, as we saw earlier in this chapter, has done some work looking for genetic variation that might predict intelligence. He is co-founder of a company called Genomic Prediction, which offers pre-implantation genetic testing with a difference.

As we saw in chapters 6 and 8, PGT has so far been used for situations where the outcome is very clearly defined — mainly in selecting an embryo that does not have particular genetic conditions that are known to be in the family, or chromosomal abnormalities. Genomic Prediction offers this type of single-gene and chromosomal testing, like many other companies. Unlike anyone else, they also offer testing for polygenic risk scores. Specifically, they’ll test your embryos and provide a risk score for diabetes (insulin-dependent and non-insulin-dependent), heart attack, high blood pressure and high cholesterol, various types of cancer, and short stature. There have been questions raised about how reliably they can assess risks for most of these. The published research on which the test apparently rests reports an ‘area under the curve’, a measure of test accuracy, for these conditions that sits in a range of 0.58–0.71. On this scale, a score of 1 would be a perfectly accurate test, and 0.5 a coin toss. A score in the range of 0.6–0.7 is definitely better than a coin toss, but not so good that it would be very surprising if an individual who was scored as being likely to have a heart attack didn’t, and vice versa.

Set that technical question aside for a moment, however, and let’s assume the test is very good at making these predictions. Is this useful information? If you were choosing between two embryos, equal in all other respects, but one was likely to have a heart attack some time during her life … is that really a basis for making that choice?

To put that into some kind of context, one in three people103 over the age of 45 has high blood pressure. One in 11 over the age of 45 has diabetes. Two in five have high cholesterol. Thyroid problems are less common — perhaps one in 200 people, mainly women, is on treatment for thyroid-hormone deficiency. Heart attack and stroke cause about a quarter of all deaths. Half of us will develop cancer at some time in our lives, and three in ten will die of cancer.

[23 These are Australian numbers, but figures are similar in other developed countries. In the developing world, other causes of death, including malnutrition and infectious diseases, are more common, reducing the proportion of deaths from vascular disease and cancer, and nutritional differences make diabetes and high cholesterol proportionally less common.]

So, you’re sitting in a clinic room and your fertility specialist has presented you with the results of the PGT on the four embryos from your latest cycle of IVF. Two of the embryos have major chromosomal abnormalities that would likely mean that those embryos would be miscarried if they were implanted. If not, and they somehow made it to full term and were born alive, they would be expected to have severe disabilities and short lives. Most people in this situation would think it best not to choose those embryos.

At the other end of the spectrum, though, the report says that one of the remaining two embryos is likely (not certain) to have high blood pressure, and the other, high cholesterol. Both of these are treatable problems that affect a large proportion of the population. Is this really useful information in considering whether to implant an embryo? Think about all the things this result doesn’t tell you about this baby. A third of all great artists will develop high blood pressure.104 Forty per cent of all great scientists will have high cholesterol. Thyroid disease, although fairly common, isn’t in the same league as those others. But it’s very easy to treat. A tablet once a day, a blood test now and then to check the treatment is working as it should. Not a huge burden.

[24 Perhaps I should declare at this point that I am on treatment for high blood pressure that needed a combination of two different drugs to bring it under control. I may not be an entirely neutral observer in this instance.]

Okay, but heart attacks are bad news, right? So maybe that would be a different kind of result?

Well … maybe. Remember that, already, there are things your potential child could do about their risk of heart attack. This genetic profile represents only one set of inputs into the overall risk, which will be modified by lifestyle — choosing not to smoke, eating well, exercising — and also medical interventions. At present, the latter are mainly limited to keeping an eye on blood pressure and cholesterol (and other blood lipids) and using treatments to bring them into the healthy range if needed. It’s quite possible that there will be other options available long before today’s embryo becomes a 45-year-old with a heart attack.

Nonetheless, you might take the view that it’s best for your child not to have to worry about all this stuff, or at least not more than most people. You might feel the same way about a risk of cancer, too. Mind you, if you add cancer to heart disease, you’re talking about more than half of all deaths. If this test were a perfect predictor of possible serious health problems, you may not have that many embryos left to choose from. Is there someone in your family who has had one of these problems? Do you think it would have been better, for them and for the world, if your grandmother who had colon cancer, your father who has heart disease, or your cousin with diabetes had never been born?

Of course, there is a counter argument to this. Imagine you have two embryos and must choose one. One has a high risk of stroke, the other does not. Otherwise, the two score about the same for the various risks covered by the test. Either might turn out to be a wonderful person, gifted in some way, a boon to humanity. Or, perhaps, either might live a perfectly ordinary life but be happy, loving, and loved. Either might have a terrible health problem — say, severe psychiatric disease — not covered by the test that’s available to you. With your lack of knowledge of everything else about this potential person, why not choose the one that is unlikely to have a stroke? Perhaps, in 65 years, that person will be starting a long and happy retirement, whereas the other would have just suffered a devastating stroke, permanently losing the ability to speak and to use one side of her body; her retirement would be spent dependent, in a nursing home. You may be long dead by then, but at least you had the chance to watch your child growing into adulthood without worrying about that particular future event. Worth it? Do you want to have this test?

It’s possible to push things still further. Genomic Prediction make it very clear that they do not test for intelligence. But what if a similar company could tell you which embryo was likely to be the smartest? What if they could test for homosexuality? There is good evidence that, although there isn’t a ‘gay gene’, there is a strong genetic basis to sexual orientation, and it’s complex — like height, and blood cholesterol, and high blood pressure. There’s every reason to think that one day it might be possible to score embryos for ‘likelihood to be gay’. If you could test your embryos for this, would you? Should you?

The balance between what we can do in medicine, what it is useful to do, and what we should do is not an easy one to manage. There are lots of ways we get it wrong. It seems to me that even a perfect test for common, more-or-less treatable conditions is likely to do more harm than good, for most people. It also seems to me that, based on what we know about it, this particular test is too far from perfect to make it an attractive proposition just yet, even if you think the concept is good. But there is room for the test to improve … and we’d better decide what we think about it, because this particular genie is already well and truly out of its bottle.

Speaking of mythical creatures: there’s magic of another kind, just around the corner.