Jump the gun and you will be shot down

Stuart J. Ritchie (University of Edinburgh) responds to Oliver James's letter from the December edition.

“What”, asks clinical psychologist Oliver James, “would lead to acceptance of the null hypothesis of the Human Genome Project as regards psychological traits?”. The answer to this question is “a scientific literature that looks very different indeed to the existing one”. In his letter, James exploits the complex nature of behaviour genetics to sow confusion about what geneticists have found. In an attempt to clear up some of that confusion, below is a point-by-point rebuttal to each one of his arguments.

James: “The hypothesis of the Human Genome Project (HGP) was that differences in DNA sequences would significantly explain why one individual is more likely to possess a trait than another. This has been proven to be true for some physical differences, such as anatomical ones, like height. But, so far, there are no psychological traits – none at all – which have been shown to be significantly influenced by specific variations in DNA (James, 2016).”

It is very odd to concede that genes influence height, but then deny that the same is true for psychological traits, because the evidence for both inferences comes from exactly the same scientific methodology. All kinds of behaviour-genetic investigations (twin studies, family pedigree studies, adoption studies, and newer DNA-based studies) show consistent results when it comes to height and to psychological traits such as IQ, personality, and many other psychological measures. Physical and psychological traits are both under substantial genetic influence. I will get to James’s statement about “specific variations in DNA” below.

James: “Robert Plomin, the most widely cited authority in this field in the UK, told the Guardian newspaper in 2014 that, as regards psychological traits, ‘I have been looking for these genes for fifteen years and I don’t have any’ (Wilby, 2014).”

This is an out-of-context quote mine. To be clear, because this is a source of some confusion: we can know that a trait is genetically influenced (from methods such as twin studies) without knowing the specific genes involved (knowledge about which comes from different methods, such as genome-wide association studies). Thus, Robert Plomin is noting that, whereas we know that many psychological traits are strongly heritable, we do not yet know the specific genes involved (the reason for this is that very large studies are needed to have the statistical power to detect specific genetic associations; these take longer to put together than the average study). There is nothing contradictory or surprising about this.

In any case, James is out of date: genome-wide association studies in 2014 and 2015 have uncovered specific genes related to educational performance, to IQ and to the personality trait of neuroticism. Even larger studies with even more impressive genetic results are on the way.

James: “Genome Wide Association (GWA) studies have scanned up to a million sites on the individual genomes of tens of thousands of subjects. Whilst some variant DNA sequences have been established as associated with specific traits and in some cases that has been replicated, they provide negligible heritability: when the effect of the variants is summed all together, they produce heritability estimates of only 1–5 per cent (James, 2014, 2016). It seems highly probable that GWA studies will find no significant role for genes.”

Again, James elides ‘knowing a trait is heritable’ with ‘knowing the specific genes involved’. Since we are only at the very beginning of our use of genome-wide association studies, and since very large samples are needed to detect the very many genetic variants that relate to psychological traits, nobody sensible should expect that we would know a large proportion of them by now. Just take a look at Figure 1 in this short Nature article. As sample sizes have increased, so have the number of genetic variants found (in this case linked to schizophrenia).page1image32800page1image32960page1image33120page1image33280page1image33440page1image33600page1image33760page1image33920
James asserts that it is ‘highly probable’ that this progress in genetics will come to a halt. I suggest that this quotation could be added to this webpage of ‘incorrect predictions’, alongside the statement from the New York Times in 1920 that ‘a rocket will never be able to leave the Earth’s atmosphere’.

James: The only other approach which still holds out hope that DNA directly effects individual differences in psychological traits are gene–environment interaction studies of candidate genes. Studies of epigenetics, junk genes or Genome Wide Complex Trait Analysis (GCTA) do not test the HGP hypothesis because they do not link specific DNA sequences to any specific trait outcomes (James, 2014).

I have no idea what epigenetics or junk genes have to do with James’s argument, here: he appears to have thrown them in in an attempt to sound more scientific. He does, however, mention the method of Genome-Wide Complex Trait Analysis (GCTA). GCTA is a purely DNA-based method that tests whether individuals who are similar in DNA variants are also similar in their phenotypic traits. It allows estimation of the heritability of psychological traits without the assumptions of twin and family methods – assumptions that James has been attacking for years. That is, GCTA is a compelling piece of evidence against James’s worldview, but he does not appear to have realized it. For the third time, we do not need to know which specific DNA sequences relate to a trait to know it is heritable.

James: Although there have been some promising results in gene-environment studies (Belsky et al., 2009; Rutter, 2014), there have been a great many non-replications. Of 103 gene– environment studies of candidate genes conducted between 2000 and 2009, only 27 per cent of attempts to replicate proved positive (Duncan & Keller, 2013). When the direct impact of candidate genes for a variety of traits was reviewed in 100 GWA studies, very little effect was identified (Siontis et al., 2010).

Another complexity in the scientific literature; another opportunity for James to mislead. The era of ‘candidate gene’ studies, where scientists tested relations between specific, theoretically-relevant genes and phenotypic traits, did indeed have a replicability problem, as the cited Duncan and Keller (2013) study—which, incidentally, is an excellent example of behavioural geneticists carefully considering the replicability of their field—discusses. The cited Siontis et al. (2010) study backs this up, though notes that some pre-2010 candidate gene studies were worth following up. Importantly, however, researchers have learned the lessons and moved on from candidate gene studies, towards genome-wide association studies. And as noted above, these have had recent success in finding genes linked to psychological traits.

James: It is quite possible that the gene–environment hypothesis will never be supported, or only be shown to have a minor influence on psychological outcomes. At first, when the HGP had consistently found trivial or nonexistent heritabilities, the large gulf between those findings and those of twin studies was labelled Missing Heritability (Manolio et al., 2009). The evidence is increasingly suggesting the heritability is not missing; it is non-existent.

The idea of the ‘missing heritability’ refers to the fact that we know, from twin and family studies, that many traits are highly heritable, but we do not yet know the specific genes involved. That is, if one adds up the contributions of all the genes we know about that are related to a trait, they do not explain as much variation as we’d expect from the twin study estimate. But as I argued above, we should not expect to have found the vast bulk of these genes by now. This is principally because of the huge sample sizes required to find genes for complex traits: James is once again jumping the gun quite spectacularly to suggest that progress in genetics is about to dry up. page2image31960page2image32120page2image32280page2image32440

James: There have always been grave doubts about the robustness of the twin method (Joseph, 2015; James, 2016). It is possible that twin studies will turn out to have been conflating supposedly high heritability with shared environmental factors. THISE analysis – Twin Studies’ ‘Heritability’ Is Shared Environment – proposes that twin study heritability findings really indicate that for this trait, parents treat their children more similarly (James, 2014). A great deal of what has been assumed to be genetic influence could be the consequence of identical twins being treated more similarly than non-identical twins: the obvious implication of the HGP findings is that the equal environments assumption is false, a proposition which already had considerable evidence to support it before the HGP (Joseph, 2015).

And yet, the evidence for the heritability of psychological traits comes not just from twin studies, but from adoption studies, family-pedigree studies, and, as noted above, directly from DNA (the GCTA method). The estimates from all these different designs, all of which entail different assumptions, hang together remarkably well. All these issues were discussed in a recent debate in the criminology literature: see this paper for a resounding defence of twin methodology. Incidentally, a review of the cited book by Joseph (2015) by behavioural geneticist Eric Turkheimer is available here and is well worth reading.

James: At the end of the newspaper interview with Robert Plomin he was asked what he would conclude if the genes he is looking for are never found. Plomin replied: ‘I will still believe that [genetic] heritability is true’. This sounds more like faith than science. Whilst it is impossible to prove a negative, what will be regarded as sufficient evidence that the null hypothesis of the HGP should be accepted? It behoves the likes of Plomin to tell us.

In a crowning irony, James - who has spent decades flailing, Canute-like, against a tide of genetic evidence disconfirming his cherished beliefs - accuses another person of an attitude “more like faith than science”. For the fourth time, James is confusing evidence of heritability with evidence for specific genes. Bearing that in mind, Plomin’s statement is not outrageous in the slightest. Of course, as I noted above, we are beginning to find specific genes related to psychological traits, so the journalist’s question to Plomin is moot.

James: In the meantime, the practice of beginning scientific articles with the assertion that such traits as intelligence, ADHD and schizophrenia are ‘highly heritable’, based purely on studies of twins or adoptees, should be ended. This assertion is simply no longer tenable in the light of the HGP findings.

To recap: evidence from a huge range of different methodologies, using families and twins, speaks to the high heritability of traits like IQ, ADHD, and schizophrenia. The fact that we do not yet know many of the specific genetic variants that explain these traits and disorders —though the numbers are increasing all the time—says nothing about this heritability. James’s argument boils down to “show me the genes!”. It seems safe to say that he will look even sillier in a few years, when many of these genes are known.

James: Similarly, undergraduate and secondary education courses should cease teaching that any psychological traits are ‘highly’ heritable. At the very least, it should be explained that the HGP findings pose a major challenge to that claim.

In this final paragraph, James proposes that we should teach students to be as confused as he is about behaviour genetics. On the basis of all the evidence cited above, I can’t say that I agree.

Stuart J. Ritchie
The University of Edinburgh
page3image30464page3image30624

BPS Members can discuss this article

Already a member? Or Create an account

Not a member? Find out about becoming a member or subscriber

Comments

While not having quite such a negative view of genetic approaches to psychology as Oliver James (who I like and respect), I feel compelled to respond to Stuart Ritchie's comment, which I think is unfair in places, and repeats a depressingly common error in this debate: the confusion of correlation with causality.

To summarize Ritchie's argument: we shouldn't be worried by the failure of GWAS and similar methodologies to identify specific genes associated with specific traits because heritability (h2) statistics show that genes must be involved. This argument assumes that the heritability statistic is a measure of genetic causation, an assumption made by many leaders in the field of psychiatric genetics writing over many decades, especially with respect to psychosis. For example, in a 1973 paper in the British Journal of Psychiatry, the American psychologist Irving Gottesman writing with the British psychiatrist James Shields, argued that:

“It is important to understand the implications of finding that a trait such as the liability to schizophrenia has a high heritability. In the samples so far studied, it means that environmental factors were unimportant as causative agents of the schizophrenias.”

Three decades later, the conventional wisdom had not changed and the British psychiatrist Peter McGuffin (in van Os & McGuffin, 2003) asserted that:

“Although we cannot exclude the possibility that social factors contribute to the aetiology of schizophrenia in those with a genetic predisposition, the non-genetic proportion of variance in liability is small and is more likely to be explained by physical stressors or even by stochastic [random] processes” .

Indeed, I have heard the claim that 80% heritability means that 80% of the cause of schizophrenia is genetic made by an internationally-respected researcher in psychiatric genetics at a conference in Ritchie's home city within the last couple of years.

But a heritability coefficient is simply a partial correlation coefficient calculated from family, and adoption studies (I'll come on to the GCTA method shortly) so that h2 = variance in a trait associated with genetic variation divided by (variance in a trait associated with genetic variation + variance in a trait associated with environmental variation). And we all know what we were taught about correlation and causation in Statistics 101.

(As an aside, there are some serious debates to be had about both the raw data used to calculate h2 - modern twin studies tend to give much lower concordance rates for schizophrenia than older studies, for example, but researchers often persist in relying on the older studies to calculate heritability - and also about some of the assumptions involved - for example that MZ twins are not treated more similarly than DZ twins. These debates are beyond the scope of the present comment, but I refer readers to Fosse, 2015).

To see why it is a terrible mistake to mistake h2 for a causation coefficient, we should first of all remember that it is a statement about populations not individuals, and then imagine what would happen if variation in the environment was reduced to zero in a particular population. In the case of lung cancer, this would happen if there was no variation in how many cigarettes people smoked, for example, in an imaginary world in which everyone smoked exactly the same number - say 20 - cigarettes a day. In these circumstances, the figure for environmental variance in the heritability equation would fall towards zero, and the heritability estimate as a consequence would approach 100 percent. This outcome is not hard to interpret conceptually. In our imaginary world, the only differences between people who get lung cancer and people who remain disease-free would be genetic; presumably some people are more genetically susceptible to the effects of tobacco smoke than others. Nonetheless, in the same imaginary world, cigarette smoking would still be a major cause (arguably the major cause) of the disease.

This is not idle speculation as there are some real life analogues of this thought experiment. For example, studies of IQ show that intelligence is highly heritable in middle class families but has very low heritability – close to zero– in working class families (Turkheimer et al. 2003). This is obviously not because genes influence IQ differently in people from different socioeconomic classes, and is probably because middle class families are relatively uniform in terms of aspects of the environment that are critical for promoting learning (they all encourage their children to do their homework, read books etc.) whereas working class families are more variable (some encourage learning and others do not). Interestingly, a recent study found a similar effect for internalizing disorders (depression and anxiety), with much greater variance attributable to genes in wealthy families compared to poor families (South & Krueger, 2011). Again, it seems unlikely that genes differ in their influence on mood disorders depending on socioeconomic circumstances. I am guessing but perhaps MZ twins are more likely to be treated similarly when resources are plentiful. Or possibly, in more benign environments, the onset of internalizing disorders is associated with a lower threshold of genetic risk (because severe stress tends to be less common, the majority of those develop mood disorders have a high level of genetic sensitivity) with the result that a greater proportion of the phenotypic variance is associated with genes. Who knows?

Another problem with conventional thinking about h2 is that it assumes an additive model, in which the total variation in a phenotype is assumed to be the sum of the variation associated with genes and the variation associated with the environment. But we should ask ourselves whether this model is really plausible? Putting top one side recent epigenetic studies which directly undermine this assumption, correlations between the genotype and the environment are likely to be ubiquitous. This is because individuals (not only human beings, but even small animals and plants; Lewontin, 2000) influence their environments in ways that depend on their genetic endowment, or at least their genetic endowment influences what kind of environment they are exposed to. In modern social life these sort of effects probably happen all the time: tall people get selected for basketball training, people with certain intellectual and personality traits are selected for elite educational programmes, and people who are shy and nerdy are selectively rejected by their peers. When this happens, environmental influences can bring about dramatic changes in the individual but, according to conventional thinking, these influences are nonetheless attributed to genes. In a formal mathematical analysis of this kind of gene - environment interaction, Dickins and Flynn (2001) showed that the environment can play a decisive role in determining human traits even when h2 approaches 100 percent.

In his commentary on James, Ritchie goes on to claim that the GCTA method - which estimates heritability directly from molecular data, gives further proof of the importance of genetic effects in human psychological traits. However, this is disingenuous because, as he surely knows, heritability estimates from GCTA are typically well below those calculated from family, twin and adoption methods. For example, in a recent study of childhood psychiatric disorders by Trzaskowski, Dale, & Plomin (2013) GCTA suggested that these disorders had almost zero heritability, even though twin studies suggested that the heritability was substantial. (It is striking that, when the researchers made the same comparison for physical traits and scholastic ability, the difference between heritability calculated from GCTA and from twins was not large; both methods suggested a substantial genetic contribution.) In studies of psychosis, the highest heritability estimates from GCTA come in at about 20%, well below the 80% typically cited in textbooks from more conventional genetic approaches. It is this difference between h2 calculated from the two methods that leads to the concept of 'missing heritability'. Obviously, one possibility is that the missing heritability is really phantom heritability that was never there in the first place.

Finally, I am not sure whether it is fair to accuse Oliver James of being Canute-like. In any case, it is surely just as Canute-like to insist that tides are moving when they are not, as to insist that they not move when they will. It is simply muddleheaded to assume that heritability estimates are measures of genetic causation.

References

Dickins, W. T., & Flynn, J. R. (2001). Heritability estimates versus large environmental effects: The IQ paradox resolved. Psychological Review, 108, 346-369.

Fosse, R. (2015). Schizophrenia: A critical review of genetic effects. Psychosis, available online.

Gottesman, I.I., & Shields, J. (1973). Genetic theorizing and schizophrenia. British Journal of Psychiatry, 122, 15-30.

Lewontin, R. (2000). The triple helix: Gene, organism and environment. Cambridge, Mass.: Harvard University Press

Trzaskowski, M., Dale, P.S., & Plomin, R. (2013). No genetic influence for childhood behavior problems from DNA analysis. Journal of the American Academy of Child and Adolescent Psychiatry, 52, 1048-1056

Turkheimer, E., Haley, A., Waldron, M., D'Onofio, B., & Gottesman, I.I. (2003). Socioeconomic status modifies heritability of IQ in young children. Psychological Science, 14, 623-628

van Os, J., & McGuffin, P. (2003). Can the social environment cause schizophrenia? British Journal of Psychiatry, 182, 291-292

Dr Ritchie, you do not answer the central question: what would you regard as evidence that would lead you to accept the null hypothesis of the Human Genome Project as regards psychological traits? Come on, what is your answer?

Instead, you do your best to poke your finger into the dyke of the floodwater of null findings by citing three studies which do nothing to rebut my claim that GWAs continue to find little or no heritability for any psychological traits. The first of the studies merely recites yet more variants found for educational performance but without showing significant heritability - how is that evidence that i am not up to date on the evidence? The second is a GCTA study which has no relevance to the HGP hypothesis (see James, 2014, cited in my letter). The third finds a paltry 15% heritability for neurosis, still needing replication.

The idea that because physical traits like height have some (but not nearly as much as was predicted) genetic heritability, ergo so should psychological traits, is far from necessary. For instance, the potential for language appears to be a wholly inherited human trait but which specific language we speak is wholly due to nurture. It is perfectly possible that we are all born with (presumably genetic) potentials for universal psychological traits, like levels of humour, love and intelligence, but that individual differences therein could be largely or wholly nurtured (including prenatally). There is no a priori reason to suppose that because individual differences in some physical traits, like height, are significantly genetically caused, therefore all psychological traits are also genetic. On the contrary, show me the evidence which contradicts the assertion that, so far, GWAs suggest that not a single psychological trait is significantly influenced by genes.

Dismissing Joseph's book The Trouble With Twins - a seminal one, which should be required reading on all undergraduate courses - with Turkheimer's review can be countered with Joseph's retort to that review: http://www.madinamerica.com/2015/11/twin-studies-are-still-in-trouble-a-... There is no excuse any more for university teachers to ignore the major flaws that Twin Studies suffer from.

Above all, be all this as it may, my question was what will you accept as evidence that the null hypothesis for the HGP should be accepted? We have been listening to your demands for ever more money to study ever larger samples now for a long time. Some geneticists i have spoken with have even maintained that the amount of variance which each of thousands of variants have is so tiny that a large enough sample will never be possible, that the maths will require stupendously large samples, and therefore, although 'we know its there' it will never be found. Perhaps you subscribe to that view but where does that leave you as a scientist? Karl Popper would say it leaves you with a hypothesis which cannot be tested...

Richard Bentall calls me “muddleheaded” for “assuming heritability estimates are measures of genetic causation”. This is a classic ‘straw man’ argument: I assume no such thing, and I did not claim any such simplistic theory in my response to Oliver James. I am tempted to leave it at that, but since even the straw man appears to have outwitted Bentall in several respects, I’ll make a few points in response.

Bentall summarises my argument thus: “we shouldn’t be worried by the failure of GWAS... to identify specific genes associated with specific traits because heritability statistics show that genes must be involved”. Not quite. First, my main argument was that Oliver James’s exploitation of easily-confused aspects of the genetics literature (the difference between statistical and quantitative genetics; the difference between candidate gene studies and GWAS) is an underhanded and unfair debating tactic, whatever the state of the scientific literature. Second, Bentall’s statement assumes that GWAS has ‘failed’ to find specific genes. I don’t know where he gets this idea: I cited successful GWAS (that is, GWAS that have, even at this early stage, found several genome-wide significant hits) relating to education, IQ, neuroticism, and schizophrenia in my article, but Bentall does not mention them.

The argument about a hypothetical world where environments are exactly equal for everyone (and heritability therefore necessarily being 100%) is not a new one. Indeed, this exact point is made in a recent book by Asbury & Plomin (2013), co-authored by the main target of James’s original hit piece. There isn’t much argument about this, so it’s not clear why Bentall thinks this is a killer point against heritability.

Nevertheless, we must be very clear about the specific situations under which we can expect heritability estimates to vary. As in nearly all critiques of behaviour genetics, Bentall cites the small study by Turkheimer et al. (2003) that showed the heritability of IQ to be near-zero when socioeconomic status was low. In his hopelessly one-sided presentation of the evidence, Bentall does not mention that this study is an outlier – indeed, the largest gene-by-environment (GxE) interaction effect on IQ that has ever been found. Why does Bentall cite that particular study, instead of one of the dozens of other studies of GxE for IQ that find smaller (or, often null) effects? You decide.

In any case, a meta-analysis has appeared in recent weeks collecting all the GxE interaction studies for IQ (Tucker-Drob & Bates, in press). It shows that the GxE effect on IQ occurs in the US, but not in any of the other countries where it’s been studied (England, the Netherlands, Germany, Sweden, and Australia). The authors speculate (and do read their Discussion section) that this might be due to the better access to social welfare, medical assistance, and educational resources in the non-US countries. So Bentall’s cherry-picked reference is out of date and fails to convey the complexity of this important question. Can you trust him on anything else?

Finally, Bentall discusses the method of GCTA (Genomewide Complex Trait Analysis, which estimates heritability based only on DNA, and not twins or families). He claims it is ‘disingenuous’ to cite GCTA studies in defence of heritability, because their heritability estimates are lower than those from twin studies. But the heritability estimates from GCTA, such as the 20% for schizophrenia cited by Bentall (presumably from Lee et al., 2012) are *necessarily* a lower-bound estimate of heritability, since they only use a subsample of ‘common’ SNPs (single-nucleotide polymorphisms, which are genetic variations in one of the four DNA ‘letters’), and only those that are tagged by the DNA chips used in GCTA studies. Twin studies, because they examine individuals with their full complement of common and rare genetic variants, are an upper-bound estimate of heritability. A recent study showed that much higher GCTA estimates can be obtained if one adds more common SNPs and rare genetic variants into the equation (Yang et al., 2015). This study was for the easily-measured traits of height and BMI, but it’s only a matter of time before this new method (with the slightly less snappy name GREML-LDMS), which shows “negligible missing heritability”, is applied to psychological traits.

By the way, contra Bentall, the most recent GCTA study of psychotic experiences in adolescents (Sieradzka et al., 2015) finds heritability estimates well above zero, though these are still, for the reasons discussed above, a lower bound. Note that, even with around 2,000 participants, the sample size here was relatively small for a sample-hungry method like GCTA. Future, larger studies will likely show higher estimates with higher degrees of certainty.

***

Oliver James responds to my article – which answers his question in its second sentence – by imploring me to “answer the question” (the question being about the ‘Human Genome Project hypothesis’, a rather quirky phrase I have never heard anyone use except James himself). This does not fill me with confidence that he has read or understood anything I wrote. Nevertheless, here goes.

He misreads Davies et al. (2015) to say it is ‘a GCTA study’ – it does indeed contain a GCTA analysis showing 30% lower-bound heritability for IQ, but it is mainly a GWAS study that finds several genome-wide significant IQ-related SNPs. Like Bentall, James ignores this. Perhaps he won’t ignore the new genetic findings for this trait that are currently on the way.

James scoffs that the “paltry 15% heritability for neurosis [sic]” – it’s actually neuroticism – is “still needing replication”. That’s certainly true, since the study (Smith et al., 2015) appeared less than 3 weeks ago at the time of writing. But let’s remember, just 4 weeks ago, James could have made the claim that there were no studies showing genome-wide significant hits or non-zero GCTA heritabilities for personality traits. Now that the evidence has come in, James is reduced to moving the goalposts. Incidentally, it may be the case that the genetic architecture of personality traits is less ‘additive’ than for traits like height, or even intelligence, and may involve non-additive genetic processes like dominance; this might be why GCTA studies find disproportionately lower heritabilities for personality than they do for IQ. These are open questions; hypotheses to be tested, not sneered at.

The comically extreme position James puts forward, that “not a single psychological trait is significantly influenced by genes”, has already been falsified. We can no longer accept his “null hypothesis”. Allow me to summarise, one final time. 1) twin and family studies indicate that psychological traits are heritable; 2) Looking at DNA by measuring thousands of SNPs using the GCTA method gives us lower-bound estimates of that heritability, and this has been found to be significantly and substantially above zero in the latest and largest studies; 3a) We are also now identifying specific genetic variants that rise above the stringent significance level required in GWAS for traits such as education, IQ, neuroticism, schizophrenia, and more; 3b) As sample sizes increase, the power to detect even more variants will also increase. I am not making the argument that ‘we know the genes are there, we just haven’t found them yet’. I am saying that we have *already started* finding them, and we are continuing to make progress. All this is ignored or misunderstood by James, whose only argument at this point is “I see no ships!”.

***

Allow me to speculate for a moment. The reason James and Bentall are so spooked by heritability is that they think it might imply there is no social influence on psychological traits, and that people’s traits are set in stone from birth. But no informed person believes this. Just look at height: it is clearly strongly heritable (that is, its variation in Western society is largely due to genes, hundreds of which have been found; Wood et al., 2014), and yet it can be changed by a better or poorer diet; indeed, it’s average level has increased substantially over the years (that is, its mean can be influenced by non-genetic causes). One of my abiding interests is in looking for ways to improve intelligence, which is not a goal that’s in contradiction with my acceptance of the strong evidence for its heritability.

By blundering into a complex scientific field, mis-citing and misquoting, James and Bentall do nobody any favours. The entire episode is, in fact, an insult to behaviour geneticists, who have for decades carefully tested the assumptions of their constantly-developing methods, and taught their nuances, strengths, and limitations to generations of students. For example, both Bentall and James bash twin studies for their ‘equal environments assumption’, but they don’t mention the many behaviour genetics papers that have explicitly tested the assumption, and found that it generally holds (e.g. Conley et al., 2013; Kendler et al., 1993). Both Bentall and James express frustration that they have been making these same arguments for a long time. But is it any wonder the debate has stalled, when neither of them can even give a fair summary of the literature?

References

Asbury, K., & Plomin, R. (2013). G is for Genes: The Impact of Genetics on Education and Achievement. London, UK: Wiley-Blackwell.

Conley, D., Rauscher, E., Dawes, C., Magnusson, P. K., & Siegal, M. L. (2013). Heritability and the equal environments assumption: evidence from multiple samples of misclassified twins. Behavior Genetics, 43(5), 415-426.

Davies, G., Armstrong, N., Bis, J. C., Bressler, J., Chouraki, V., Giddaluru, S., ... & van der Lee, S. J. (2015). Genetic contributions to variation in general cognitive function: a meta-analysis of genome-wide association studies in the CHARGE consortium (N = 53 949). Molecular Psychiatry, 20(2), 183-192.

Lee, S. H., DeCandia, T. R., Ripke, S., Yang, J., Sullivan, P. F., Goddard, M. E., ... & International Schizophrenia Consortium. (2012). Estimating the proportion of variation in susceptibility to schizophrenia captured by common SNPs. Nature Genetics, 44(3), 247-250.

Kendler, K. S., Neale, M. C., Kessler, R. C., Heath, A. C., & Eaves, L. J. (1993). A test of the equal-environment assumption in twin studies of psychiatric illness. Behavior Genetics, 23(1), 21-27.

Sieradzka, D., Power, R. A., Freeman, D., Cardno, A. G., Dudbridge, F., & Ronald, A. (2015). Heritability of individual psychotic experiences captured by common genetic variants in a community sample of adolescents. Behavior Genetics, 45(5), 493-502.

Smith, D. J., Escott-Price, V., Davies, G., Bailey, M. E. S., Coldoro Conde, L., Ward, J., ... & O’Donovan, M. (2015). Genome-wide analysis of over 106,000 individuals identifies 9 neuroticism-associated loci. bioRxiv, available at: http://biorxiv.org/content/early/2015/11/20/032417

Tucker-Drob, E. M., & Bates, T. C. (In press). Large cross-national differences in gene x socioeconomic status interaction on intelligence. Psychological Science, available at: http://bit.ly/1Ok9UTh

Turkheimer, E., Haley, A., Waldron, M., D'Onofrio, B., & Gottesman, I. I. (2003). Socioeconomic status modifies heritability of IQ in young children. Psychological Science, 14(6), 623-628.

Wood, A. R., Esko, T., Yang, J., Vedantam, S., Pers, T. H., Gustafsson, S., ... & Amin, N. (2014). Defining the role of common variation in the genomic and biological architecture of adult human height. Nature Genetics, 46(11), 1173-1186.

Yang, J., Bakshi, A., Zhu, Z., Hemani, G., Vinkhuyzen, A. A., Lee, S. H., ... & Snieder, H. (2015). Genetic variance estimation with imputed variants finds negligible missing heritability for human height and body mass index. Nature Genetics, 47, 1114-1120.

Dear Stuart,

Some clinical psychologists might be inclined to offer interpretations for your use of martial metaphors (“Jump the gun and you will be shot down”; “Swings and misses….”), especially as you seem to think that my previous comment was, “In fact, an insult to behaviour geneticists”. (Dear me; I thought I was joining a scientific debate; I certainly had no intention of insulting anyone.) So, can I just suggest, before we go on, that you just try and chill a little. This isn’t a matter of life and death and, after all, it is the season to be merry.

Obviously I can’t speak for Oliver; as I said before, I suspect his attitude towards genetics is a bit more sceptical than mine (although we manage to nonetheless respect and like each other, which may be a lesson to you). But, in asking when you’d be willing to accept the null hypothesis that genes have a limited effect on human behaviour, he is surely asking a fair question of a field in which, as you well know (in respect to the molecular data), failed replication has been the norm.

With respect to my own position, your suggestion that I am attacking a straw man is, well, an attack on a straw man:

1. You say, “The argument about a hypothetical world where environments are exactly equal for everyone (and heritability therefore necessarily being 100%) is not a new one. Indeed, this exact point is made in a recent book by Asbury & Plomin (2013)”. Great. Perhaps they got it from my 2009 book ‘Doctoring the mind’?

Of course, just because an argument is old does not mean that it is invalid. In this context, it is worth noting that there is no need for environments to be “exactly equal” for these kinds of effects to be evident.

Regarding the specific studies, IQ is not my specialist area so thanks for the references. The few studies in psychiatry that have examined this effect, so far as I know, have found it. In my comment, I cited a study by South & Krueger, (2011) in which the heritability of internalizing disorders was found to be higher for middle class than working class families. I am guessing but I’d place a reasonably large bet that these sorts of effects are very common, especially when the numerous potential sources of environmental variation are considered. Indeed, another study by the same group (South & Krueger, 2008) found that the heritability (henceforth h2) of the same disorders depended on marital disharmony (with higher heritability in unharmonious marriages). I notice that you did not cite these studies in your own piece, but it’s all too easy to accuse someone of selective reporting so I won’t do that.

The important point is that these effects show that it is wrong to interpret h2 as a causation coefficient, and therefore that high h2s from quantitative genetics cannot be taken as proof of strong genetic effects (and cannot be used to refute Oliver’s claim, based on the molecular data, that it’s time to accept the null hypothesis). You (and apparently Asbury and Plomin, although I haven’t read their book) now appear to agree with my point about h2, which I welcome.

2. You say, “Bentall’s statement assumes that GWAS has ‘failed’ to find specific genes. I don’t know where he gets this idea: I cited successful GWAS (that is, GWAS that have, even at this early stage, found several genome-wide significant hits) relating to education, IQ, neuroticism, and schizophrenia in my article”.

You seem to have missed the key word here, which is 'specific'. Again, I do not know the research on IQ, but psychiatric GWAS studies suggest that the same genes are associated with a wide range of psychiatric diagnoses (not just schizophrenia, but also bipolar disorder, major depression, ADHD and autism; (Psychiatric Genomics Consortium, 2013). These findings are consistent with recent results from quantitative genetic studies (e.g. Lichtenstein et al., 2009; Rasic et al. 2014).

It is also worth stating that, in GWAS studies of psychiatric disorders, the risk ratios associated with those genes that have been identified are very small (ORs < 1.1). To put some flesh on this, the SNP (rs115329265) with the highest association with schizophrenia in the recent study by the Psychiatric Genomics Consortium, (2014) was found in 86.4% of patients and 85.0% of controls. If we take a 1% estimate of the lifetime risk of schizophrenia for ease of calculation, and imagine a sample of 10,000 people, of whom about 100 can be expected to receive the diagnosis, 86 or 87 of the patients might be expected to have the SNP, but, of the remaining 9,900 healthy people, 85%, or a total of 8,415 people, can also be expected to have it. So, for every schizophrenia patient affected, there will be nearly 100 (actually 8,415/87 = 96.7) healthy people who carry the critical SNP. If we take a broader definition of psychosis with a lifetime risk of about 3%, about 260/300 patients will be affected, versus 8,245 healthy people, giving a ratio of approximately 32 healthy people to every patient. I grant you that this finding might be intriguing from a purely scientific point of view, but it is difficult to see how it is going to be much help to anyone who is interested in the wellbeing of people with mental illness.

3. Regarding GCTA estimates of h2, we will probably have to disagree, although I grant you that the Sieradzka et al., (2015) study is important, not least because it calculated h2 for different types of psychotic experiences. However, it is misleading to say that the study “finds heritability estimates well above zero”; in fact only the h2 calculation for anhedonia was significant and came in at 20%. All the other estimates were non-significant and, in the case of hallucinations, the actual estimate was zero. (It looks like it is you have “jumped the gun” here and I can only hope that a less kind critic does not inflict a bullet wound on you as a consequence.) In any case, even if GCTA led to substantial estimates of heritability it would not negate my main point (which you apparently agree with), which is that heritability estimates are just fancy correlation coefficients and not measures of causation.

4. And finally, the crucial point, and the reason why it is you who is attacking a straw man and not me. At the end of your riposte, you speculate that, “The reason James and Bentall are so spooked by heritability is that they think it might imply there is no social influence on psychological traits.”

Again I cannot speak for Oliver but in my case, you really need to retake Mindreading 101. Although maybe not, because surely, if you had just read my piece carefully you would have realized that I believe nothing of the sort. My entire argument was precisely against the idea that h2 is a genetic causation coefficient, with the implication that heritability estimates close to 100% do not preclude substantial social and environmental risk factors for mental illness. There is nothing about h2 that is scary to me; I was just pointing out that it does not mean what most people seem to think it means.

In fact, you cannot estimate social and environmental effects simply by subtracting h2 from 100%. The only way of estimating them is to measure them directly. When this is done we find that these risk factors – which, in the case of severe mental illness, include poverty in childhood (Wicks et al. 2010), exposure to an unequal distribution of socioeconomic resources (Johnson et al. 2015), exposure to urban environments (Vassos et al. 2012), exposure to dysfunctional communications from parents (de Sousa et al. 2014); childhood sexual and physical abuse and other forms of maltreatment (Varese et al., 2012), migration (Cantor-Graee & Selten, 2005) and separation from parents at an early age (Varese et al. 2012) – often dwarf the effects of individual genes (see also, Bentall et al., 2015; Bentall, Wickham, Shevlin, & Varese, 2012).

Unfortunately however, unlike you or I, many leading figures in psychiatric genetics have assumed that high h2 estimates are causation coefficients and do preclude substantial environmental influences (I am not making this up; I included some relevant quotes in the first few paragraphs of my previous comment). And this has had disastrous consequences for patients with mental illness, who are some of the most vulnerable people in the world today.

It has meant that time and effort has been devoted almost exclusively towards research on causal factors which are unlikely to be turned into therapies anytime soon. (Look at the lack of progress in treating Huntington’s Disease, which is 100% heritable and transmitted by a single dominant gene with a well-understood biological function, and then ask yourself how knowledge of rs115329265 is ever going to be turned into a novel therapy.) By contrast, factors in the social environment that we can manipulate or at least ameliorate continue to be neglected (all of the social risk factors listed a few paragraphs above are addressable by public health or therapeutic interventions). The emphasis on genes is disproportionate to the good that is likely to emerge from it.

It has led to modest effects being spun to a gullible public as major scientific breakthroughs (and don’t get me going about DISC1 – “the Rossetta stone gene” for schizophrenia according to a recent UK university press release about a study carried out on mice, which can be found at http://www.cardiff.ac.uk/news/view/125809-scientists-identify-schizophre... in fact, the evidence from patient studies suggests that this gene should perhaps be renamed as NDISC1 or not-disrupted-in-schizophrenia 1; see Mathieson, Munafo, & Flint, 2012).

And it has led to the promulgation of theories of mental illness that have increased the stigma experienced by psychiatric patients and their families (contrary to widespread belief in the neuroscientific community, theories which exclusively emphasize genetic factors are perceived as highly stigmatizing by patients, carers and the general public; see Angermeyer, Holzinger, Carta, & Schomerus, 2011).

Given how this debate has unfolded, I am not sure I want to pursue it any further in this forum. So let me conclude be reiterating that I am certainly not opposed to genetic research into mental illness (or any other human traits for that matter) but I think it is about time that this approach was integrated with social and psychological approaches, something which geneticists (certainly in psychiatry) seem reluctant to do at present.

It only remains for me to hope that you have been good this year, and that Santa brings you some lovely gifts on Xmas eve.

Richard

References

Angermeyer, M. C., Holzinger, A., Carta, M. G., & Schomerus, G. (2011). Biogenetic explanations and public acceptance of mental illness: systematic review of population studies. British Journal of Psychiatry, 199, 367-372.

Bentall, R. P., de Sousa, P., Varese, F., Wickham, S., Sitko, K., Haarmans, M., & Read, J. (2015). From adversity to psychosis: Pathways and mechanisms from specific adversities to specific symptoms. Social Psychiatry and Psychiatric Epidemiology, 49, 1011-1022.

Bentall, R. P., Wickham, S., Shevlin, M., & Varese, F. (2012). Do specific early life adversities lead to specific symptoms of psychosis? A study from the 2007 The Adult Psychiatric Morbidity Survey. Schizophrenia Bulletin, 38, 734-740.

Cantor-Graee, E., & Selten, J.P. (2005). Schizophrenia and migration: A meta-analysis and review. American Journal of Psychiatry, 163, 478-487.

de Sousa, P., Varese, F., Sellwood, W., & Bentall, R.P. (2014). Parental communication deviance and psychosis: A meta-analysis. Schizophrenia Bulletin, 40, 756-768. doi:10.1093/schbul/sbt088

Johnson, S.L., Wibbels, E., & Wilkinson, R. (2015). Economic inequality is related to cross-national prevalence of psychotic symptoms. Social Psychiatry and Psychiatric Epidemiology. doi:10.1007/s00127-015-1112-4

Lichtenstein, P., Yip, B. H., Bjork, C., Pawitan, Y., Cannon, T. D., Sullivan, P. F., & Hultman, C. M. (2009). Common genetic determinants of schizophrenia and bipolar disorder in Swedish families: a population-based study. Lancet, 373, 234-239.

Mathieson, I., Munafo, M. R., & Flint, J. (2012). Meta-analysis indicates that common variants at the DISC1 locus are not associated with schizophrenia. Molecular Psychiatry, 17, 634-641.

Psychiatric Genomics Consortium. (2013). Genetic relationship between five psychiatric disorders estimated from genome-wide SNPs. Nature Genetics, 984-994.

Rasic, D., Hajek, T., Aida, M., & Uher, R. (2014). Risk of mental illness in offspring of parents with schizophrenia, bipolar disorder, and major depressive disorder: A meta-analysis of family high-risk studies Schizophrenia Bulletin, 40, 28-38. doi: 10.1093/schbul/sbt114

Psychiatric Genomics Consortium. (2014). Biological insights from 108 schizophrenia-associated genetic loci. Nature, 511, 421-427.

South, S. C., & Krueger, R. (2011). Genetic and environmental influences on internalizing psychopathology vary as a function of economic status. Psychological Medicine, 41, 107-117.

South, S. C., & Krueger, R. F. (2008). Marital quality moderates genetic and environmental influences on the internalizing spectrum. Journal of Abnormal Psychology, 117, 826-837.

Varese, F., Smeets, F., Drukker, M., Lieverse, R., Lataster, T., Viechtbauer, W., . . . Bentall, R. P. (2012). Childhood adversities increase the risk of psychosis: A meta-analysis of patient-control, prospective and cross-sectional cohort studies. Schizophrenia Bulletin, 38, 661-671. doi:10.1093/schbul/sbs050

Vassos, E., Pedersen, C.B., Murray, R.M., Collier, D.A., & Lewis, C.M. (2012). Meta-analysis of the association of urbanicity with schizophrenia. Schizophrenia Bulletin, 38, 1118-1123.

Wicks, S., Hjern, A., & Daman, C. (2010). Social risk or genetic liability for psychosis? A study of children born in Sweden and reared by adoptive parents. American Journal of Psychiatry, 167, 1240-1246.

I have little to add to Richard Bentall's scholarly and eminently reasonable commentary.

I agree with him that your tone does seem somewhat emotionally reactive.

Like him, I do not know why you characterize either of us as anti-genetic, environmental flat earthers. Its a fact, not my opinion, that the term Missing Heritability was created and is widely used by geneticists as well as other social scientists. Its a fact that GWAs are unable to find replicated genetic variants which, even when all added together, explain more than tiny amounts of variance for all psychological traits.

You conflate the fact (which I have never disputed) that a number of SNPs have been identified as associated with numerous mental illnesses, with the equally indisputable fact that none of them, even when all added together, provide heritability estimates which are regarded as significant (often accepted as being 20% or more variance explained if it is not to be characterized as 'minor'). I repeat that the GWAs do not explain more than tiny amounts of variance.

Since GCTAs do not identify specific genetic variants linked to specific trait outcomes, they are not tests of that hypothesis, which was indisputably the one with which the HGP began.

There is nothing to stop you offering me a proposition which would lead to acceptance of the null hypothesis. For example, when whole-genome studies have been done for samples of x or y size and still produce the same negligible heritability estimates, you might think that would be sufficient?

If genes play little or no role in explaining individual difference, that does not mean physical factors have no effect. Anyone with even a passing acquaintance with the history and philosophy of science will know that it is highly probable that there are all manner of factors which are completely unknown. Some could be physical ones.

I can see that it is not very reassuring for you to hear the possibility that your geneticism is being disproved, but as Richard Bentall recommends, you must not let that upset you. Ultimately, all that matters is that we find the truth as best we can, and that we can use it to benefit humanity.

Happy Christmas.