Intelligence

Intelligence, in this genetic literature, means general cognitive function — performance summarized across several different cognitive tests into a single score, sometimes called g. Studying its genetics means asking why that score varies between people within a population, not judging anyone's worth or capability. Two large 2018 studies asked this question with different methods, and both are on this page.

A method that borrows statistical power

Hill et al. 2018 used a technique called MTAG (multi-trait analysis of GWAS) to combine a genome-wide association study of intelligence with a separate, genetically correlated GWAS of educational attainment — the two traits share a genetic correlation of 0.70, high enough that combining them raises statistical power without claiming they are the same thing. Doing this raised the effective sample size from 199,242 to 248,482 and found 187 independent genome-wide-significant loci, implicating 538 genes.

This method has a real limitation worth stating plainly: because MTAG borrows power from a correlated proxy trait, some of these 187 loci reflect genetic architecture shared between intelligence and educational attainment specifically, not necessarily intelligence considered on its own. The paper's authors built this analysis carefully and checked that it reproduced the pattern of genetic correlations seen in earlier intelligence-only studies — but the method's own logic means this list is not a pure isolate of "intelligence genes."

The same year, tested directly instead

Davies et al. 2018 asked the same question without borrowing power from a proxy trait. Combining the CHARGE and COGENT consortia with UK Biobank — 300,486 people in total, ages 16 to 102 — they tested general cognitive function directly and found 148 independent genome-wide-significant loci, implicating 709 genes in gene-based analysis. This is a larger combined sample than the MTAG analysis's effective size, on the actual trait rather than a power-boosted proxy for it — the direct complement to Hill et al.'s own stated caveat, not a repeat of the same finding.

Built from these 148 loci, a polygenic score predicted up to 4.3% of the variance in general cognitive function in independent samples — the paper's own number. The study also found significant genetic overlap between cognitive function, reaction time, and several unrelated health variables, including eyesight, hypertension and longevity — variants that affect cognition tend to also nudge these other measures, a real but modest overlap, not a claim that any one of them causes another.

What the genes point to

Hill et al.'s enrichment analyses found the MTAG-associated genes concentrated in neurogenesis, myelination, genes expressed in the synapse, and genes involved in regulating the nervous system. Two variants on this page sit in genes matching those themes directly: rs56116382 sits in BSN (Bassoon), a structural scaffolding protein at the presynaptic terminal, and rs132570 sits in CACNA1I, a neuronal calcium channel subunit — concrete instances of the "synapse" and "nervous system regulation" categories the paper's own analysis names, not variants singled out for an unusually large effect.

Across both studies' loci combined, no single variant here does more than shift the population average by a very small amount. This is a highly polygenic trait in the fullest sense: the biology is real and distributed, and no one locus is a meaningful predictor on its own.

Who this was studied in

Both are European-ancestry-dominated findings. Hill et al.'s quality-control steps explicitly removed UK Biobank participants without British ancestry, and used the European-ancestry reference panel from the 1000 Genomes Project for its statistical calculations. Davies et al.'s combined CHARGE/COGENT/UK Biobank sample was likewise drawn overwhelmingly from European-ancestry cohorts. Like most large cognitive-genetics GWAS of this era, neither result has been validated in other ancestries and neither should be assumed to generalize to them.