It takes a certain courage to title a paper: Genetic “General Intelligence,” Objectively Determined and Measured.
Javier de la Fuente, Gail Davies, Andrew D. Grotzinger, Elliot M. Tucker-Drob, Ian J. Deary
Objectively? Is such language permissible in contemporary science? Should we not instead be cautiously shuffling towards seven types of ambiguity, hedged in with eight layers of limitations? Who are these wild types willing to risk all in search of glory? In fact, a look at the names shows that this group have established an excellent track record, so they have in all probability chosen their words carefully, with the facts on their side.
First, a digression. Since 1969 there have been factions eager to denigrate intelligence, saying that its measures are based on arbitrary mental tasks, gathered together in the statistical artefact of a make-believe common factor, which is based on precisely nothing. How can one possible counter this all-encompassing dismissal of the psychometric project?
One approach would be to link a common factor to a genetic substrate, and anchor it in the genome.
It has been known for 115 years that, in humans, diverse cognitive traits are positively intercorrelated; this forms the basis for the general factor of intelligence (g). We directly test for a genetic basis for g using data from seven different cognitive tests (N = 11,263 to N = 331,679) and genome-wide autosomal single nucleotide polymorphisms. A genetic g factor accounts for 58.4% (SE = 4.8%) of the genetic variance in the cognitive traits, with trait-specific genetic factors accounting for the remaining 41.6%. We distill genetic loci broadly relevant for many cognitive traits (g) from loci associated with only individual cognitive traits. These results elucidate the etiological basis for a long-known yet poorly-understood phenomenon, revealing a fundamental dimension of genetic sharing across diverse cognitive traits.
The authors go on to explain that tests vary in the amount of general or specific factors required for their successful completion. Some variance in each test is shared with all other tests (g) and some is specific to each test (s). Hundreds of studies show that the g factor replicates, and accounts for 40% of test variance. Twin studies show that general intelligence is strongly heritable, suggesting an overlapping genetic architecture. However, the GWAS approach does not distinguish between “g” and “s”. The authors try to search for “g” directly, using a multivariate molecular genetics approach to the hierarchy of intelligence, g at the top, cognitive domains second, and individual tests at the bottom.
They used UK Biobank, blessed be its name, and seven cognitive tests:
Reaction Time (n = 330,024; perceptual motor speed), Matrix Pattern Recognition (n = 11,356; nonverbal reasoning), Verbal Numerical Reasoning (VNR; n = 171,304; verbal and numeric problem solving; the test is called ‘Fluid intelligence’ in UK Biobank), Symbol Digit Substitution (n = 87,741; information processing speed), Pairs Matching Test (n = 331,679; episodic memory), Tower Rearranging (n = 11,263; executive functioning), and Trail Making Test – B (Trails-B; n = 78,547; executive functioning). A positive manifold of phenotypic correlations was observed across the seven cognitive traits.
The authors then investigate the genetic contribution of g to variation in each of the cognitive tests. Genetic correlation is simply the correlation between the genetic contributors to each of the measured abilities. It is correlation at the level of genes, not test scores. If the brain is made up of modules, then one would expect such genetic correlations to be low. On the other hand, a brain largely based on general ability would have strong correlations. In fact, the genetic correlations range from .14 to .87, with a mean of .53 and the first principal component accounted for a total of 62.17% of the genetic variance. The genetics of intelligence is largely g based, it would seem.
Further work identifies the tests that are most g loaded:
Trails-B (95.30% genetic g; 4.70% genetic s), Tower (72.80% genetic g; 27.20% genetic s), Symbol Digit (69.10% genetic g; 30.90% genetic s), and Matrices (68.20% genetic g; 31.80% genetic s). Verbal Numerical Reasoning (51.40% genetic g; 48.60% genetic s) and Memory (42.40% genetic g; 57.60% genetic s) are more evenly split. Reaction Time has the majority of its genetic influence from a genetic s (9.50% genetic g; 90.50% genetic s). We emphasize one important implication of these results, i.e. that genetic analyses of some of these individual traits will largely reveal results relevant to g rather than to the specific abilities thought to be required to perform the test.
Reaction time is somewhat of an outlier from the genetic point of view, as might be expected by the very simple, knee-jerk nature of the task.
Anyway, which locations on the genome are contributors to g? Getting an answer is important, since a GWAS hit could be generalizable to a broad universe of cognitive traits, or specific to a particular task, and knowing which makes a difference. In the explanation below, Q is a measure of heterogeneity, opposite to g.
Miami plot of unique, independent hits for genetic g (top) and Q (bottom). Q is a heterogeneity statistic that indexes whether a SNP evinces patterns of associations with the cognitive traits that departs from the pattern that would be expected if it were to act on the traits via genetic g. The solid grey horizontal lines are the genome-wide significance threshold (p < 5×10−8) and the dotted grey horizontal lines are the suggestive threshold (p < 1 ×10−5). The following genome wide significant loci are highlighted: Red triangles : g loci unique of univariate loci. Blue triangles : g loci in common with univariate loci. Green circles : univariate loci not in common with g loci. Yellow triangles : g loci in common with Q loci. Yellow diamonds : Q loci unique of g loci.
Overall, we identified 30 genome-wide significant (p < 5×10−8) loci for genetic g, 23 of which were common with the univariate GWAS of the individual cognitive traits that served as the basis for our multivariate analysis. We identified, in total, 24 genome-wide significant loci for Q, 3 of which were significantly associated with genetic g (and therefore likely to be relevant to more specific cognitive traits, and false discoveries on g) and 15 of which were significantly associated with at least one individual cognitive trait in the test-specific GWASs.
Although it was not intended to be part of the study, seven new locations for memory were found, and some of those locations have been associated with Schizophrenia, anti-saccade response, linguistic errors, hand grip strength and bone mineral density.
So, have the authors “got away with” their combative title? The best way to answer would be to set the question “What else do you want?” The claim is that intelligence is real, and is a real aspect of the brain. To show that that is the case you can show a link between intelligence test scores and real life (this has been done many, many times, and some examples are shown below) and a link between intelligence test scores and implied measures of genetic heritability via twin and family studies (also done many times), and now finally a link between intelligence test scores broken up into general and specific factors and measures of heritability via actual genomic studies identifying locations for general and specific factors of intelligence.
Here are some correlations between intelligence and real life measures
In my view this is a very important advance. It shows an underlying reality, at a genetic level, between general and specific aspects of cognitive ability. It allows investigations to proceed at two levels: the test score level and the genomic code level. Further studies will drill down into yet more detail.
It is fair to say that this is an objective approach, and ought to answer any reasonable critic of the reality of cognitive ability being based on brains which are under substantial genetic control.