A new paper in Molecular Psychiatry has been reported on extensively in the media, and readers have mentioned it several times in the comments. I read it. It’s titled Genome-wide association studies establish that human intelligence is highly heritable and polygenic. But the fact is that I read this paper last year. Back then it was titled Common SNPs explain a large proportion of the heritability for human height. I kid, but you get the picture. The new paper establishes for intelligence what we already suspected: most of the genetic variation in this heritable trait is accounted for by numerous genes of small effect. You inherit variants of these numerous genes from your two parents, and your own trait value is to a large extent a combination of the parental values. The issue is not if intelligence is heritable, but the extent of that heritability.
The standard way to estimate human heritability was to track similarities across individuals with varying degrees of relatedness. For example, compare identical twin correlations on a trait with fraternal twin correlations. The main objection to these methods is that one could argue that environmental factors were correlated with particular genetic relationships (e.g., you treat individuals who are presumed identical twins more similarly). There are many reasons that I’m skeptical of extreme objections in this vein, but there are out there. This particular experiment design sidesteps that issue by looking at unrelated individuals. Not just notionally unrelated individuals, but actually those who were not genomically related. That’s a key difference between quantitative genetics and quantitative genomics. The former takes biological relatedness at face value, translating from ideal categories. The relatedness between full siblings is 0.50 for example. But when you look at the genomic level you can account and correct for the variation of that relatedness amongst siblings (e.g., two of my siblings exhibit a relatedness of only 0.42)! In this study they focused on numerous widely dispersed single nucelotide polymorphisms (SNPS), specific variants within genes, and used these to infer the nature of the genetic architecture of intelligence. More specifically, the genetic variation of two forms of intelligence, crystallized and fluid. The former seems to correspond to knowledge and the latter to raw problem solving abilities. Perhaps the difference between having an excellent operative system and applications vs. top of the line hardware?
In any case, here’s their abstract:
General intelligence is an important human quantitative trait that accounts for much of the variation in diverse cognitive abilities. Individual differences in intelligence are strongly associated with many important life outcomes, including educational and occupational attainments, income, health and lifespan. Data from twin and family studies are consistent with a high heritability of intelligence, but this inference has been controversial. We conducted a genome-wide analysis of 3511 unrelated adults with data on 549 692 single nucleotide polymorphisms (SNPs) and detailed phenotypes on cognitive traits. We estimate that 40% of the variation in crystallized-type intelligence and 51% of the variation in fluid-type intelligence between individuals is accounted for by linkage disequilibrium between genotyped common SNP markers and unknown causal variants. These estimates provide lower bounds for the narrow-sense heritability of the traits. We partitioned genetic variation on individual chromosomes and found that, on average, longer chromosomes explain more variation. Finally, using just SNP data we predicted ∼1% of the variance of crystallized and fluid cognitive phenotypes in an independent sample (P=0.009 and 0.028, respectively). Our results unequivocally confirm that a substantial proportion of individual differences in human intelligence is due to genetic variation, and are consistent with many genes of small effects underlying the additive genetic influences on intelligence.
The authors suggest that these values are a floor to heritability estimates, at least with the sorts of homogeneous populations they have here. That’s because their statistical genetic method is likely to miss a lot of true genetic variance due to its diminishing power when the causal genes are at too low of a frequency. They’re working within a framework where a given typed marker is correlated with a nearby I.Q. causal marker. At very low correlations they are going to miss the causal variant.
Some of the psychologists interviewed by the media contended that on one level these were banal findings. A value in the range they report is entirely within the mainstream of behavior genetic studies, which use pedigrees and what not. But many people don’t trust behavior genetics for whatever reason. One person’s banality is another person’s profundity.
But I think these sorts of findings should tilt us away from the proposition that large effect quantitative trait loci are common for I.Q. By this, I mean an “I.Q. gene” which is responsible for a huge difference between two people. There are some of these no doubt, especially those which result in mental retardation, but they don’t play that much of a role in all likelihood in ‘normal’ variation. Earlier linkage studies which reported such genes made huge media splashes and tended to fade because of lack of replication. Those genes may actually have been real QTLs, but the huge effect was likely to have been a random chance occurrence. Genome-wide association is better able to detect smaller effect genes within populations, but even it has been notably lacking in robust results.
Overall, this is good science. The results aren’t what those of us who were hoping that the intersection between psychometrics and genomics would yield low hanging fruit were pulling for. But reality is often likely to dash our hopes. No matter the banality of the final results, I do think the figure to the left is rather cool. It shows that the larger the chromosome the greater the proportion of variance is explained by that chromosome. This is entirely expected in theory (large chromosomes would carry more causal variants), but it is gratifying to still see it born out empirically.
Razib Khan’s degrees are in biochemistry and biology. He has blogged about genetics since 2002, previously worked in software development, is an Unz Foundation Junior Fellow and lives in the western US. He loves habaneros.
August 10th, 2011 at 11:49 pm
Kevin Mitchell isn’t ready to believe that it has been shown to be as polygenic as claimed.
August 11th, 2011 at 5:53 am
I can’t access the paper. Has anyone knowledgeable looked at the stats and vouched for it?
August 11th, 2011 at 8:52 am
“The authors suggest that these values are a floor to heritability estimates, at least with the sorts of homogeneous populations they have here. That’s because their statistical genetic method is likely to miss a lot of true genetic variance due to its diminishing power when the causal genes are at too low of a frequency.”
Okay, the title of your blog is “Gene Expression.” You and the study do not acknowledge a role for gene expression elements.
“But I think these sorts of findings should tilt us away from the proposition that large effect quantitative trait loci are common for I.Q.”
It may very well be that such loci do not exist, but the study only addressed SNPs. It is still possible that large effect gene expression elements are common for IQ. Is it possible to construct a GWAS that examines genetic variation besides SNPs?
August 11th, 2011 at 9:11 am
“gene expression elements” will either be tagged by the markers they genotyped or not just like any other functional class of variant.
August 11th, 2011 at 9:52 am
Thanks for the lead. I see that they used the Illumina610-Quad v1 chip that has 60,000 copy number variant markers. Illumina now makes a 660W-Quad v1 chip with 100,000 CNV markers. So, at least GWAS will better address this issue with time.
August 11th, 2011 at 11:11 am
I feel slightly dumb but that’s fine. Is anyone willing to put this in laymens terms for me?
August 11th, 2011 at 11:18 am
last i heard people had looked at cnvs (circa 2006 or so i think). nothing showed up. don’t know if it was published or not (negative results don’t often get published). chill on the attitude. you’re not the only person who knows anything dude. and wouldn’t linkage studies looking for QTLs have found regulatory elements. what am i missing?
August 11th, 2011 at 1:43 pm
I am also confused by nooffensebut’s reference to “gene expression elements”? If these are elements in the genetic sequence that regulate gene expression then they would be picked up by GWAS, as would more and more copy number variants. Are you referring to epigenetic changes that may be heritable for one or two generations? Or something else entirely?
August 11th, 2011 at 1:45 pm
#8, i think he means more larger grain genomic structures?
August 11th, 2011 at 2:17 pm
Sorry, but it is a life-and-death issue, both figuratively for most of molecular psychiatry and literally. GWAS keep bringing genetics to square one by not finding loci. I have seen one GWAS for psychopathology, albeit childhood psychopathology, with Viding et al. There has to be some way to reconcile the trend of disappointing GWAS with the evidence of exciting candidate gene studies, like this one by Beaver and Chaviano, which has an amazing graph for Hispanic arrests versus three dopamine genes. It cannot all be massive fraud.
I meant promoters mostly. That’s what most candidate gene research seems to be today. Viding et al used the Affymetrix 6.0 GeneChip. I think that is this one with 200K probes of 5,677 CNVs. That’s genetics at 20-thousand feet, and I’m trying to make sense of it.
August 11th, 2011 at 6:39 pm
“The authors suggest that these values are a floor to heritability estimates, at least with the sorts of homogeneous populations they have here.”
Razib,
Could you elaborate on this? Based on the coverage, do you see a ceiling below the .8 (broad heritability) found from kinship studies?
August 11th, 2011 at 10:52 pm
google “genetic architecture”. you really need 10k or 100k samples for GWAS results with complex traits.
http://www.nature.com/mp/journal/vaop/ncurrent/abs/mp201165a.html
In this article, we review some of the data that contribute to our understanding of the genetic architecture of psychiatric disorders. These include results from evolutionary modelling (hence no data), the observed recurrence risk to relatives and data from molecular markers. We briefly discuss the common-disease common-variant hypothesis, the success (or otherwise) of genome-wide association studies, the evidence for polygenic variance and the likely success of exome and whole-genome sequencing studies. We conclude that the perceived dichotomy between ‘common’ and ‘rare’ variants is not only false, but unhelpful in making progress towards increasing our understanding of the genetic basis of psychiatric disorders. Strong evidence has been accumulated that is consistent with the contribution of many genes to risk of disease, across a wide range of allele frequencies and with a substantial proportion of genetic variation in the population in linkage disequilibrium with single-nucleotide polymorphisms (SNPs) on commercial genotyping arrays. At the same time, most causal variants that segregate in the population are likely to be rare and in total these variants also explain a significant proportion of genetic variation. It is the combination of allele frequency, effect size and functional characteristics that will determine the success of new experimental paradigms such as whole exome/genome sequencing to detect such loci. Empirical results suggest that roughly half the genetic variance is tagged by SNPs on commercial genome-wide chips, but that individual causal variants have a small effect size, on average. We conclude that larger experimental sample sizes are essential to further our understanding of the biology underlying psychiatric disorders.
August 12th, 2011 at 4:20 am
I would interpret these findings very differently. What the authors do is analyse GWAS data in a very unusual way – they are not interested in finding specific SNPs affecting the trait, they simply use the SNPs to measure genetic relatedness between individuals. (Well, they were interested in findings specific SNPs but once they didn’t find any significant ones they turned to this other kind of analysis). Razib, you say that the people in the study are unrelated but they are not – they are all from the same population and are distantly related. The study uses SNPs across the genome to measure this relatedness and then shows it correlates with phenotypic similarity – i.e., the trait is heritable. We knew that already.
What they claim is that you can break down this effect by chromosome or by subregion. When they use the SNPs along longer chromosomes they seem to get a bigger effect – “explaining more of the phenotypic variance”. The inference is that thousands of SNPs, scattered across the whole genome, contribute to the trait or, more specifically to variance in the trait across the population (the implication is that they contribute to the value of the trait in individuals).
There is an alternative explanation for this effect, however, which is that using more SNPs simply gives a better estimate of genetic relatedness. So, the SNPs on chromosomes 1 (the longest) give a better estimate than those on chromosome 21 (the shortest) – they index relatedness with more precision. As a result, they correlate better with phenotypic similarity – this looks like you have “explained more of the variance”. In fact, getting such a signal from SNPs on chromosome 1 does not mean that any of the causal variants are actually on chromosome 1. Nor does the fact that such signals can be derived from anywhere in the genome mean that there are thousands of variants across the genome affecting the trait.
In fact, the authors can conclude very little from this study beyond a replication of the known fact that IQ is heritable. They can say nothing about how many variants are involved across the population or how many affect the trait in each individual. Note that those could be very very different from each other – you could have hundreds or thousands of genes affecting a trait across the population, but only one, two or a handful of variants affecting the phenotype in any individual. Nor can they say whether the causal variants are common or rare. One could expect different combinations of small numbers of different rare variants to be determining phenotype in different individuals. In fact, I would say that is exactly what one should expect. Not the picture they try to sell in this paper, which is that the phenotype in any individual is determined by the combination of thousands of common variants.
August 12th, 2011 at 5:27 am
It’s heartening to know that somewhere in all the science blogs I read, there’s someone who knows the difference between ‘titled’ and ‘entitled’!
August 13th, 2011 at 10:02 am
@kjmtchl
If it weren’t many genes of small effect then this and earlier GWAS would have picked up individual hits.
A few genes of large effect accounting for the 1 SD population variation would have already been detected. They’d have to be common variants (since by assumption they account for the observed variation) and of large effect, so difficult to hide.
August 13th, 2011 at 11:33 am
Steve, the point is that the variants that are having an effect may be rare. We may each carry a different spectrum of rare variants that affects the trait. These would be completely missed by GWAS. The results they show in this paper do not definitively support the conclusion that hundreds of thousands of variants are involved. For more on why I generally think the idea that common variants explain human diversity, see: http://wiringthebrain.blogspot.com/2011/08/welcome-to-your-genome.html
August 13th, 2011 at 2:03 pm
If I understand correctly, you want to claim that the observed population variation could be due to a few rare variants of large effect. But then it would be surprising for this study to have found .5 of the total variation to be associated with SNPs — compare to earlier studies using twins/adoptions/siblings that found narrow sense heritability of about .6 or so. I would not expect the rare alleles you hypothesize to be in good LD with SNPs (which are designed to tag common variants), so we would expect to lose a big chunk of the .6 additive heritability.
For example, in the Visscher paper on height they had to hand wave about imperfect LD to recover the full .8 or so of heritability. In this case the global fit comes out very close to .6, which suggests common rather than rare variants (at least, they are well tagged by SNPs). But if they are common variants their individual effect sizes must be small and there are a lot of them. Let me know if I am missing something.
August 14th, 2011 at 6:43 pm
Razib Khan said:
“wouldn’t linkage studies looking for QTLs have found regulatory elements”
Well, those studies did find linkages, but they did not agree, so Butcher et al said, “Because linkage designs are powerful for detecting genes of large effect size, one safe conclusion from these linkage studies is that there are unlikely to be any genes that have a large effect on g, for example accounting for more than 10% of the variance.” This is a more limited statement than that of SNP GWAS, in which “estimates for the effect of any one genetic variant are well below 1% of the variance…” Gene duplication and associated regulation might be an additional source of variation unexamined by GWAS. Nevertheless, if I grant you that regulatory elements do not have large-effect loci, it might still be true that the study was wrong to fail to acknowledge gene expression elements as an additional source of many common variants of small effect. Their limited view could lead to a premature resolution that underestimates true IQ heritability. Some adult IQ heritability estimates approach 80%, and Need et al determined a principal component heritability of 88% for CANTAB, which seems to emphasize crystallized intelligence more. Comparisons between chimpanzee and human protein-coding have concluded that gene expression elements, which show positive selection, might account for most differences in behavior and cognition.
It looks to me like GWAS chips have markers for a fraction of CVNs, but not VNTR. Up to 20% of genes might have VNTRs, which have accelerated mutation rates and can affect protein coding (“little is known about the underlying molecular mechanism”) and gene expression. A good (free) review of VNTR is Gemayel et al. They write of VNTR, “They represent, in addition to single nucleotide polymorphisms (SNPs) and copy number variations (CNVs), a third, mostly ignored, category of genetic variation. Many of the examples demonstrating or suggesting that variable TRs influence phenotypes have been discovered by research teams focusing on a specific phenotype or gene rather than large-scale studies specifically aimed at characterizing the global role of repeats in genome evolution.” Therefore, GWAS should acknowledge VNTR as true variation outside their current detection ability.
August 14th, 2011 at 8:21 pm
Steve Hsu said:
“But then it would be surprising for this study to have found .5 of the total variation to be associated with SNPs — compare to earlier studies using twins/adoptions/siblings that found narrow sense heritability of about .6 or so.”
Why should we be married to 60% IQ heritability? If you will pardon the cliché, you are putting the cart before the horse. This study uses 60% but uses as support a review that says, “Again we cite older studies because it is well established that heritability increases from about 30% in very young childhood to as much as 80% in adulthood. Part of the reason that this corroboration is so helpful is that the twins studies are all over the map. I remember The Bell Curve settled on a 60% estimate, but it was a comically glib compromise between 40% and 80% intended mostly to change the subject. James Lee’s response to Nisbett cites an aggregate broad-sense heritability from monozygotic twins reared apart studies of 75%. The debate over whether twins studies under- or overestimate IQ heritability is complex with arguments for both sides. As Lee puts it, “Such factors as fetal position, order of delivery, and blood transfusion may even act to differentiate MZ twins rather than to increase their similarity.”
August 15th, 2011 at 8:29 am
Steve, I am saying that all this study does is use SNPs to estimate genetic relatedness and show that it correlates with phenotypic similarity. This does not mean that there are thousands of SNPs involved. They can say nothing about the SNPs actually being causal. And I don’t think the population variation is caused by “a few” rare variants – I think it is (or could be at least) caused by a larger number of rare variants – different ones in different people.
August 15th, 2011 at 9:57 am
kjmtchl,
I think the authors realize this limitation of their method. From a commentary paper (http://homepages.ed.ac.uk/qgjc/2010_2011/CommentaryOnYang.pdf):
“If most causal variants for human height have such low frequency in the population that they are not in LD with the (common) SNPs on the commercial SNP arrays then the method we used would not detect much more additional variance than already accounted for by the published genome-wide significant loci”.
August 15th, 2011 at 1:36 pm
kjmtchl,
In reference to your point that chr 1 would explain the most variance simply because it is the largest – Figure 2 shows chr 1 and 2 actually explain very little variance in IQ, whereas 3, 4, and 5 explain a lot more. Granted, you’re right that SNP hertability is a function of ancient relatedness, but if that’s all it was measuring, then there should always be a very strong correlation between variance explained (regardless of the trait) and chromosome size. However, the correlation is only moderate (.07 to .21 r2 or a correlation of .26 and .46).
August 15th, 2011 at 10:25 pm
James Lee’s response to Nisbett cites an aggregate broad-sense heritability from monozygotic twins reared apart studies of 75%. The debate over whether twins studies under- or overestimate IQ heritability is complex with arguments for both sides. As Lee puts it, “Such factors as fetal position, order of delivery, and blood transfusion may even act to differentiate MZ twins rather than to increase their similarity.”
if you know james personally just ask if him if you should be telling steve this stuff as if it’s going to blow his mind. i think he’ll disabuse you of this notion. if you don’t know him personally, i think you’ll understand that it seems really strange that you’re quoting an expert to enlighten steve when said expert has actually been in correspondence with steve for a while now (see “james lee” on his weblog).
August 15th, 2011 at 11:10 pm
Point taken.
August 17th, 2011 at 6:21 am
Razib, you say in a comment “last i heard people had looked at cnvs (circa 2006 or so i think). nothing showed up. ”
I would say that if that’s what the intelligence people are saying, they must be wrong or at least only talking about part of the IQ range. We know that CNVs are associated with very low IQ i.e. mental retardation or intellectual disability or whatever you choose to call it.
Typically this is defined as an IQ of <70.
In the psychiatric literature it's also striking that across all kinds of disorders, autism schizophrenia ADHD and others, whenever a study finds CNVs, the patients with the CNVs have a lower IQ than the patients without.
There's a nice paper just out showing that, above a certain size of CNV (about 1.25 megabases), everyone who carries them has a low IQ. And the larger the CNV, the less likely it is to be carried by people with normal IQ.
Cooper, G.M., et al., A copy number variation morbidity map of developmental delay. Nat Genet, 2011.
Personally I suspect that CNVs account for most of the variation at the bottom of the IQ curve, while in the middle it's common genetic variation, and at the very top it's environment – I don't think you can score 140 without the benefit of the right education on top of the right genes – but that's just my intuition.
August 17th, 2011 at 9:36 pm
#25, limit the discussion to the middle 90% and i’m comfortable with remaining skeptical of CNVs as explaining much of the variance. but i agree at the bottom of the distribution there are all sorts of weird genomic variants, including CNVs. less sure about the top of the distribution though.