Most Reported Genetic Associations with General Intelligence Are Probably False Positives

By Razib Khan | December 11, 2011 4:49 pm

The title says it all, and I yanked it from a paper that is now online (and free). It’s of interest because of its relevance to the future genetic understanding of complex cognitive and behavioral traits. Here’s the abstract:

General intelligence (g) and virtually all other behavioral traits are heritable. Associations between g and specific single-nucleotide polymorphisms (SNPs) in several candidate genes involved in brain function have been reported. We sought to replicate published associations between 12 specific genetic variants and g using three independent, well-characterized, longitudinal datasets of 5571, 1759, and 2441 individuals. Of 32 independent tests across all three datasets, only one was nominally significant at the p ~ .05 level. By contrast, power analyses showed that we should have expected 10–15 significant associations, given reasonable assumptions for genotype effect sizes. As positive controls, we confirmed accepted genetic associations for Alzheimer disease and body mass index, and we used SNP-based relatedness calculations to replicate estimates that about half of the variance in g is accounted for by common genetic variation among individuals. We conclude that different approaches than candidate genes are needed in the molecular genetics of psychology and social science.


My hunch is that these results will be unsatisfying to many people. The authors confirm and reassert the heritability of general intelligence, both by reiterating classical results, and utilizing novel genomic techniques. But, they also suggest that the candidate gene literature is nearly worthless because of the lack of power of most of the earlier studies. The latter is probably due to the genetic architecture of the trait. Intelligence may be determined by numerous genes of very small effect (e.g., 0.01% of the variance effected by one particular SNP), or, “rare, perhaps structural, genetic variants with modest to large effect sizes.” The former case is pretty obvious, but what about the latter? I’m mildly skeptical of this because I’m curious why modest-to-large effect variants didn’t show up in family-based studies (presumably within the family the same variants would localize to sections of the genetic map)? But I’m not fluent enough in the literature to know if there was a lot of work in this area with families previously.

Related: Here’s the first author’s article in Commentary from the late 1990s, IQ Since “The Bell Curve”.

CATEGORIZED UNDER: Behavior Genetics, Psychology
  • Grey

    I wonder about a computer analogy – separate genes for CPU, RAM and size of hard-drive – and high IQ is simply having the full set – which makes me wonder if idiots-savant break down into categories. If so some of those component genes may be a consequence of other physical parameters e.g. size of hard-drive might be equivalent to brain size i.e. skull size i.e. height? Larger brain sizes might not be useful without a CPU or RAM upgrade.

  • deevybee

    Heritability estimates typically come from twin studies. These are insensitive to gene x environment interactions, which get subsumed under the general heritability term. So another reason for failing to find replicable effects in association studies could be if genetic variants exert different effects according to specific environmental conditions.
    http://www.guardian.co.uk/science/blog/2010/sep/09/gene-intelligence-genetic-testing

  • http://blogs.discovermagazine.com/gnxp Razib Khan

    Heritability estimates typically come from twin studies. These are insensitive to gene x environment interactions, which get subsumed under the general heritability term.

    please note that they estimated heritability using SNP-based relatedness. so the talk of twin-studies isn’t on point here. in any case, there are models which take into account GxE.

  • http://theunsilencedscience.blogspot.com/ nooffensebut

    Beaver et al just published a study on three dopamine genes, education, and IQ based on the Peabody Picture Vocabulary Test. The sample size was about as large as study 2 in this manuscript. The correlations between the three-gene index with education and with IQ were stronger than the correlations between IQ and violence, race and poverty, and gender and educational attainment. The effect of the gene index seemed mostly due to DRD2/ANKK1, which is one of the genes addressed here. I find it hard to believe that this is just a false positive as every correlation seems as expected, and this cannot be gene shopping because these researchers always use the same database and study the same group of genes. Besides, a sample size of 1700 isn’t exactly trivial. My best guess is that the effect is just a specific vocabulary ability, and these three studies do not emphasize vocabulary.

  • dave chamberlin

    What puzzles me is a contradiction between two apparent observations. 1)There is an enormous variability in individuals inherited intellegence. 2) Intellegence may be determined by a number of genes of very small effect.

    I don’t want to get caught up in the philosophical quagmire of what intellegence is or how well it can be measured, we know there is a very large standard deviation in the human population that would not be there if intellegence variability was determined solely by a number of genes of very small effect. There wouldn’t be Razib’s and the lost low normals who look upon reality in comic book terms if inherited intellegence was due to 1000 coin flips.

    It makes me hopeful that the BGI study and others of it’s kind given enough time, super computers, and budget will yield results showing us what we do not yet know, why the enormous variation in human populations in viewing the world complexly. It makes me think that there must be combinations of genes that amplify intellegence or we wouldn’t be so variable

  • observer

    One of the most interesting things mentioned in the paper was a result that essentially confirmed, independently, the recent groundbreaking study from Ian Deary and colleagues (Davies et al below):

    “Davies et al. (2011) used data from five different genome-wide association studies (GWAS) and failed to identify any individual markers robustly associated with crystalized orfluid intelligence. They then applied a recently developed method (Yang et al., 2010; Visscher et al., 2010) for testing the cumulative effects of all the genotyped SNPs. In essence, this method calculates the overall genetic similarity between each pair of individuals in a sample and then correlates this genetic similarity with phenotypic similarity across all pairs. Following Yang et al. (2010), we dropped one twin per pair, and then estimated all pairwise genetic relationships in the resulting sample. We then dropped individuals whose relatedness exceeded .025, just as in Davies et al. (2011). Davies et al. reported that the ~550,000 SNPs in their data could jointly explain 40% of the variation in crystalized g (N = 3,254) and 51% of the variation in fluid g (N =3,181). We applied the same procedure to the STR sample from Study 3 and estimated that the ~630,000 SNPs in our data jointly account for 47% of the variance in g (p < .02), confirming the Davies et al. (2011) findings in an independent sample. These and our other results, together with the failure of whole-genome association studies of g to date, are consistent with general intelligence being a highly polygenic trait on which common genetic variants individually have only small effects."

  • toto

    we know there is a very large standard deviation in the human population that would not be there if intellegence variability was determined solely by a number of genes of very small effect.

    I’m sorry but I don’t understand this. A large number of nearly-independent small-effect variants is the optimal recipe for generating large, smooth variation in a population!

    If “intelligence” were governed by only a few large-effect variants, it would be semi-discrete, a bit like eye colour.

    There wouldn’t be Razib’s and the lost low normals who look upon reality in comic book terms if inherited intellegence was due to 1000 coin flips.

    Well, yes there would. If you can program a computer, try to calculate the sum (or the average) result of 1000 coin flips, again and again and again. When you collate the results, you will see that their distribution follows a smooth, well-defined bell curve, with a few extremely low sums, a few extremely high sums, and a whole bunch of average ones. Such is the power of the CLT.

  • toto

    please note that they estimated heritability using SNP-based relatedness.

    Hmmm… is their method robust to possible population structure with socio-economic correlates?

  • http://blogs.discovermagazine.com/gnxp Razib Khan

    #5, as toto says, central limit theorem. are you wondering about the variation across siblings? i think that’s easily handled by segregation and recombination.

  • observer

    One thing that always puzzles me when it comes to the issue of the effect size of individual genes on various traits/dispositions is this: shouldn’t it be the case that one would EXPECT the effect sizes of individual genes to be small on traits that are almost certainly either highly selected for, or selected against? Wouldn’t any allele that had a large negative effect on reproductive success, or large positive effect, be swept respectively out of, or through, a population very quickly? Shouldn’t there be a theoretical result to that effect? The only reason I can think it might be otherwise is if the gene in question also had some significant effects in the opposite direction on reproductive success.

    So why should we expect anything other than that, say, genes for intelligence would be of very small effect, as would be genes for most diseases? (Again, if these genes also have effects in the opposite direction, such as genes for (perhaps) Tach-Sachs disease on the one hand and (certainly) sickle cell disease on the other, then the effect might be considerable).

  • http://blogs.discovermagazine.com/gnxp Razib Khan

    #10, the reality is that quantitative traits tend to be those which haven’t been subject to a lot of directional selection. sweeps tend to remove the variation that they need to exist. selection doesn’t care about genetic architecture, it only sees phenotypic variation. i think large effect traits may actually be more common in those subject to selection because those ‘traits’ are mutants which exist only because of recessive alleles segregating at low frequencies. though i haven’t given it much thought.

  • observer

    Just to add to my previous post:

    If my speculation is correct, then one way to help locate alleles of relatively large effect when it comes to traits that would seem to be highly selected for, or highly selected against, is to do the following:

    Find NEGATIVE traits that seem to be correlated with the traits that are POSITIVELY selected for, and find POSITIVE traits that seem to correlated with the traits that are NEGATIVELY selected for. Genes encoding such “countervailing” traits would be good candidates for having large effect on the trait of interest. (E.g., perhaps myopia among more intelligent people).

  • observer

    “the reality is that quantitative traits tend to be those which haven’t been subject to a lot of directional selection.”

    It certainly may be true that MOST quantitative traits haven’t been subject to a lot of selection. But I’m concerned here about traits that are heavily selected for and which also can be affected by a large number of genes. Whether or not a trait can be affected by a large number of genes is an issue of physiology and molecular biology, independently of its role in selection.

    There is a very long standing experiment artificially selecting for oil content (I believe) in some kind of seed (don’t remember offhand the details) — I believe that it has gone over 100 generations. There continues to be considerable returns on the selection process (at least in the direction of increasing the oil content — it really can’t be decreased any further). Now that’s artificial selection, which is going to exhaust any genetic potential far more quickly than any natural selection. So I have to believe it takes a LONG time to exhaust the potential of small additive genes on a positive trait, if it can be so altered by genes.

    Now I guess these things can reach a limit, where the positive effects are counterbalanced by the negative effects. But the ongoing high variability in, say, intelligence would seem to suggest we haven’t reached that limit. If we had, then the downside of high intelligence would, one would think, be pretty obvious — in point of fact, it seems to be almost pure upside.

  • http://blogs.discovermagazine.com/gnxp Razib Khan

    If we had, then the downside of high intelligence would, one would think, be pretty obvious — in point of fact, it seems to be almost pure upside.

    have you looked at the reproductive fitness of the more educated in all nations?

  • http://www.isteve.blogspot Steve Sailer

    Consider a highly complex analog machine, such as a 19th Century clipper sailing ship. If you had ten built by hand according to the same blueprints and then had them race around the world, you would likely to see considerable variation in their speed. Why? Because some suffered more mistakes while being built, and some suffered damage during their race. What if much of the variation in IQ is driven by random errors rather than different blueprints?

  • http://emilkirkegaard.com Emil

    So, for eugenic purposes, how does one find the jeens for g? My idea is that once human jenome sequencing gets very popular, one can pile together a huj amount of such jenomes along with varius g corelates and identify the jeens. The sample size wud need to be very larj for one to notice the small efect of individual jeens with stat. sig. Im not sure about this, perhaps som of the experts here noes if that wud work or not.

  • JL

    What if much of the variation in IQ is driven by random errors rather than different blueprints

    What do you mean by random errors? How do they explain the high heritability?

  • Douglas Knight

    Is this any different than for height?

  • ackbark

    As intelligence is a product of small effects of a wide number of genes, that gives a lot of room for variation and idiosyncrasy.

    As an intelligent person, when you encounter another intelligent person –is there anything more annoying?

    It’s annoying because you can see there’s a lot of detail here, but you’re not sure what direction it might come from. You’re immediately threatened with having nothing to talk about.

    I would say that could be a reproductive disadvantage.

    (Apologies if this is a disadvantaged comment.)

  • dave chamberlin

    I hypothesize but cannot prove that their is too much variation in human intellegence for it to be determined by 1000 genes of small effect. It isn’t a matter of solving a fairly simple statistics problem, the range of variation of a thousand coin flips, it is that we have no way to quantify how much smarter a roomfull of Razib’s is than a roomfull dullards.

  • M

    I hypothesize but cannot prove that their is too much variation in human intellegence for it to be determined by 1000 genes of small effect. It isn’t a matter of solving a fairly simple statistics problem, the range of variation of a thousand coin flips, it is that we have no way to quantify how much smarter a roomfull of Razib’s is than a roomfull dullards.

    Not in an absolute sense, but IQ does tell us what their difference is in terms of relative position in the total population. Because the measure is normally distributed by definition, inferences like “since IQ is normally distributed, we should expect it to be the sum of a bunch of small independent coin flips” don’t seem to be warranted, however, though it may be the result of such things.

    That said, even granting that there’s a real sense in which distribution is normal, it doesn’t seem to be the case that per #7

    If you can program a computer, try to calculate the sum (or the average) result of 1000 coin flips, again and again and again. When you collate the results, you will see that their distribution follows a smooth, well-defined bell curve, with a few extremely low sums, a few extremely high sums, and a whole bunch of average ones.

    we’d get a sufficiently flat distribution. The sd of 1000 flips is, like, 16 (conveniently making the defined distribution of IQ equal to heads less 400.) With most of the sample sizes discussed in the paper you should find significant effects at a point of IQ, easily. Of course there could easily be more than 1,000 SNPs with an impact on intelligence (within the range of environmental variation considered in our universe etc,) but it also seems likely that some “coins” count more than others – that we’re flipping $10 in mixed coins rather than $10 of pennies, say – and that SNPs already singled out are among the most likely to have been high-denomination.

NEW ON DISCOVER
OPEN
CITIZEN SCIENCE
ADVERTISEMENT

Discover's Newsletter

Sign up to get the latest science news delivered weekly right to your inbox!

Gene Expression

This blog is about evolution, genetics, genomics and their interstices. Please beware that comments are aggressively moderated. Uncivil or churlish comments will likely get you banned immediately, so make any contribution count!

About Razib Khan

I have degrees in biology and biochemistry, a passion for genetics, history, and philosophy, and shrimp is my favorite food. In relation to nationality I'm a American Northwesterner, in politics I'm a reactionary, and as for religion I have none (I'm an atheist). If you want to know more, see the links at http://www.razib.com

ADVERTISEMENT

See More

ADVERTISEMENT

RSS Razib’s Pinboard

Edifying books

Collapse bottom bar
+

Login to your Account

X
E-mail address:
Password:
Remember me
Forgot your password?
No problem. Click here to have it e-mailed to you.

Not Registered Yet?

Register now for FREE. Registration only takes a few minutes to complete. Register now »