You are on page 1of 12

REVIEWS

The new genetics of intelligence
Robert Plomin1 and Sophie von Stumm2
Abstract | Intelligence — the ability to learn, reason and solve problems — is at the forefront of
behavioural genetic research. Intelligence is highly heritable and predicts important educational,
occupational and health outcomes better than any other trait. Recent genome-wide association
studies have successfully identified inherited genome sequence differences that account for 20%
of the 50% heritability of intelligence. These findings open new avenues for research into the
causes and consequences of intelligence using genome-wide polygenic scores that aggregate
the effects of thousands of genetic variants.

Twin studies
Life is an intelligence test. During the school years, (GWAS) focused on a very different variable — years of
Studies comparing the differences in intelligence are largely the reason why education. In this Review, we discuss early attempts to
resemblance of identical and some children master the curriculum more readily find the inherited genetic differences that account for
fraternal twins to estimate than other children. Differences in school performance the substantial heritability of intelligence, and a twist of
genetic and environmental
predominantly inform prospects for further education, fate involving GWAS on years of education, before dis­
components of variance.
which in turn lead to social and economic opportuni­ cussing the results of recent large GWAS on intelligence.
Variance ties such as those related to occupation and income. In The second half of this Review focuses on genome-wide
An index of how spread out the world of work, intelligence matters beyond educa­ polygenic scores (GPSs) for intelligence that aggregate the
scores are in a study tional attainment because it involves the ability to adapt effects of thousands of DNA variants associated with
population, which is calculated
as the average of the squared
to novel challenges and tasks that describe the different intelligence across the genome (see BOX 2 for how GPSs
deviations from the mean. levels of complexity of occupations. Intelligence also are constructed). We illustrate how GPSs for intelligence
spills over into many aspects of everyday life, such as will transform research on the causes and consequences
Genome-wide association the selection of romantic partners and choices about of individual differences in intelligence before ending
studies
health care1. This is why intelligence — often called with a discussion of societal and ethical implications.
(GWAS). Studies that aim to
identify loci throughout the
general cognitive ability 2 — predicts educational We do not discuss other important research related to
genome associated with an outcomes3, occupational outcomes4,5 and health out­ intelligence, such as evolutionary research16,17 and neuro­
observed trait or disorder. comes6 better than any other trait. It is also the most science research15, in order to focus on the role of GPSs
stable psycho­logical trait, with a Pearson correlation in the new genetics of intelligence.
Heritability
The proportion of observed
co­­efficient of 0.54 from 11 to 90 years of age7. BOX 1
differences among individuals describes what intelligence is and how it is assessed. Finding the heritability of intelligence
that can be attributed to During the past century, genetic research on intel­ Similar to results for many other complex traits, early
inherited differences in ligence was in the eye of the storm of the nature– results for intelligence were disappointing for more than
genome sequence.
nurture debate in the social sciences8,9. In the 1970s and 100 candidate gene studies18 and for seven GWAS19–25.
1980s, intelligence research and its advocates were vili­ From the 1990s until 2017, no replicable associations
fied10–12. The controversy was helpful in that it raised were found. GPSs from these early GWAS, which we
1
Institute of Psychiatry, the quality and quantity threshold for the acceptance of refer to collectively as ‘IQ1’, predicted only 1% of the var­
Psychology and
genetic research on intelligence. As a result, bigger and iance of intelligence in independent samples. It became
Neuroscience, King’s College
London, 16 De Crespigny better family studies, twin studies and adoption studies clear that the problem was power: the largest effect sizes
Park, London SE5 8AF, UK. have amassed a mountain of evidence that consistently of associations between individual single-nucleotide
2
Department of Psychological showed substantial genetic influence on individual dif­ polymorphisms (SNPs) and intelligence were extremely
and Behavioural Science,
ferences in intelligence13. Meta-analyses of this evidence small, accounting for less than 0.05% of the variance of
London School of Economics
and Political Science, Queens
indicate that inherited differences in DNA sequence intelligence. The average effect size of the tens of thou­
House, 55–56 Lincoln’s Inn account for about half of the variance in measures of sands of SNPs needed to explain the 50% heritability of
Fields, London WC2A 3LJ, UK. intelligence14. intelligence is of course much lower. If the average effect
robert.plomin@kcl.ac.uk; These studies and applications in neuroscience15 were size is 0.005%, 10,000 such SNP associations would be
s.von-stumm@lse.ac.uk already pushing intelligence research towards rehabili­ needed to explain the 50% heritability of intelligence. To
doi:10.1038/nrg.2017.104 tation when it was thrust to the forefront of the DNA achieve sufficient power for detecting such small effect
Published online 8 Jan 2018 revolution 4 years ago by genome-wide association studies sizes (that is, power of 80%, P = 0.05, one-tailed), sample

NATURE REVIEWS | GENETICS ADVANCE ONLINE PUBLICATION | 1
©
2
0
1
8
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
,
p
a
r
t
o
f
S
p
r
i
n
g
e
r
N
a
t
u
r
e
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.

with many interventions failing covariance and was discovered by Spearman in 1904 (REF. 294. it was possible to accumulate to this GPS as ‘EA1’ (where EA stands for educational sample sizes with the necessary power to detect very attainment). able of the number of years spent in full-time educa­ on all SNPs regardless of the strength of their association usually weighted by the effect tion. A breakthrough cation.000 are required. nonverbal ability. such as education81. suggesting that cognitive tests tend to peak in young adulthood and decline any differences in test scores that occur within an individual are smaller thereafter80. or simply ‘g’. RU*Y *OTH Complete the sequence Match the shape ? Which picture completes the pattern above? Which of the four options is the same object shown above? Nature Reviews | Genetics 2 | ADVANCE ONLINE PUBLICATION www. the That said. with correlations of 0. It is a latent trait that cannot be directly observed but g‑factor exemplifies the generalist nature of intelligence as a complex is inferred from a battery of diverse cognitive test scores. intelligence has been shown to be than test score differences that exist between individuals. The solve problems74.000 individuals from scores had sample sizes from 18. about the to produce reliable and long-term positive effects82. . Its relevance to intelligence is that Spurred on by this success. For example.29. These associations could be replicated in (GPSs). For example. some psychometric tests assess verbal ability and lifespan. This anal­ in contrast to genome-wide In 2013.com/nrg © 2 0 1 8 M a c m i l l a n P u b l i s h e r s L i m i t e d . both untimed (see figure for examples). through major systematic words. reason and same time that Mendel’s laws of inheritance were rediscovered. a meta-analytic GWAS analysis of years of ysis produced a GPS.REVIEWS Genome-wide polygenic sizes greater than 250. Psychometric tests of cognitive abilities differ widely in form and Individual differences in intelligence are fairly stable across the content. We refer association studies.84. often referred to as educational attainment.50) and genetically (0. especially in children. a second Studies that focus on genes for which the function suggests that years of education is highly correlated phenotypically meta-analytic GWAS analysis with a sample size of they are associated with a trait. a GPS based associated with the trait. (0.6 others. which emerges from the test scores’ key challenge for intelligence research. Missing letter Complete the sentence Which one letter completes these words? Which two words complete the sentence? CA* Fingers are to nails/fish/gloves/hands *US as toes are to tip/feet/flip-flops/running. used ‘intelligence tests’ that yield a so‑called IQ score. More importantly. identifying ways to effectively improve intelligence remains a general factor of intelligence. equivalent to about 2 months of edu­ polymorphisms (SNPs) of the Breakthrough for years of education. In other malleable. p a r t o f S p r i n g e r N a t u r e . in nearly every GWAS.78. occupational status. Because with years of education predicted 2% of the variance in size of each SNP’s association with the trait in genome-wide ‘years of education’ is obtained as a demographic marker years of education in independent samples28. scores from timed cognitive test scores are positively intercorrelated75. Early IQ GWAS associations in a sample of 125. education yielded three genome-wide significant SNP the variance in years of education on average in Box 1 | What is intelligence? Intelligence can be broadly defined as the ability to learn. especially from teenage years onwards. dietary supplementation82 or other people will also do comparatively well on other cognitive tests.nature.79. This phenomenon is known as the positive manifold. Although individual SNPs of such minuscule individual’s increasing alleles for intelligence research came from the unlikely vari­ effect size are fairly useless for prediction.65) with intelligence27.000 identified 74 significant loci 30. Genetic indices of a large at the time but were not sufficiently powered to independent samples29. The largest effect size associated trait for each individual that are detect such small effects. longevity77. as in widely trait that penetrates many behavioural and psychological outcomes. adopting children away from impoverished home environments83. intelligence is also subject to change.85. health and acronym for an outdated concept of an ‘intelligence quotient’.000 to 54. Candidate gene studies small effect sizes26.000. within and between individuals. some give strict time limits. with an individual SNP accounted for a meagre 0. However. EA2.02% the sum across the genome of thousands of single-nucleotide of the variation. and some are and above32. A l l r i g h t s r e s e r v e d . which seemed 54 cohorts28. in 2016. a person who scores high on one type of cognitive test relative to interventions. in which IQ is an including educational attainment. that predicted 3% of association studies. Notwithstanding these differences. 76).

called the ‘increasing allele’. such as intelligence. the EA2 GPS predicts 3% of the indexed by parents’ years of education.023 sequence between individuals. p a r t o f S p r i n g e r N a t u r e . another reason for using the acronym GPS is that we cannot resist the metaphor of the other ‘GPS’. the possible range of the GPS is from 0 to 20.000 Proportions of variance of traits in the study population SNP 6 T A T 1 0. ing it the largest GWAS for any trait to date. We see IQ GPSs as a system to triangulate on the genetics of intelligence from all domains of the life sciences.001 (SNPs). We prefer the term ‘genome-wide polygenic score’ because ‘risk’ does not apply to quantitative traits. D. A GPS sums the number of increasing alleles across SNPs. A third GWAS currently in dren’s intelligence is 0. a genotypic score is created by adding the number of increasing alleles. although most SNPs are given near-zero weights.000 children inherit the DNA differences that predict their significant associations and a GPS. that have positive as well as negative poles86. Summing these weighted genotypic scores gives this individual a GPS of 0. the correlation with the trait is five times greater than for SNP 10. The effect cation predicted more variance in intelligence than they size of the EA3 GPS for predicting intelligence is likely predicted for the GWAS target trait of years of educa­ to rival that of family socio-economic status. SNP 8 A A A 2 0. that predicts intelligence from their parents. Koellinger. Once a GPS for intelligence is created for each individual in a sample. For each SNP. Other ways to improve the predictive power of GPSs include taking into account expected SNP effect sizes. However. Most involve the word ‘risk’.003 0. the EA3 GPS is expected to predict intelligence for each individual. implying that it accounts for progress includes more than 1 million participants. it can be used to investigate the extent to which this genetic index of intelligence mediates or moderates effects on variables of primary interest to the researcher. mak­ 9% of the variance in children’s intelligence32. The table shows how a GPS could be constructed for one individual for ten SNPs. personal com­ age association for offspring. for SNP 1.003 Effect sizes SNP 5 G C C 0 0. The ‘genome-wide’ addition to ‘polygenic score’ distinguishes GPSs from polygenic scores that aggregate candidate genes or just the top hits from genome-wide association studies (GWAS). estimate only an aver­ independent samples31 (P.000 SNP 3 A A A 2 0. such as education.002 0. GPSs for years of edu­ more than 10% of the variance in intelligence. Box 2 | Creating genome-wide polygenic scores Thousands of single-nucleotide polymorphisms (SNPs) are needed to account for the heritability of intelligence and other complex traits because the effect sizes of SNP associations are so small.004 0. variance in years of education but it predicts 4% of the correlation value for parents’ education with chil­ the variance in intelligence30. imputed as well as measured. a GPS is a composite of SNP associations.003 0. For example. For example. parental more than 10% of the variance in years of education in phenotypes. Finally. Because there are 10 SNPs. For instance. Surprisingly. The use of only genome-wide significant hits does not predict nearly as well as the use of tens of thousands of SNPs.002 0. Aggregating thousands of these minuscule effects in a genome-wide polygenic score (GPS) is the crux of the new genetics of intelligence.001 0.002 accounted for by a particular factor such as a genome-wide SNP 7 C C G 1 0.005 0. . Preliminary this association is confounded by genetics because results from this GWAS have identified more than 1. Hence. whereas GPSs predict munication). In this example. Multiplying the genotypic score by the correlation gives a weighted genotypic score (see table.002 polygenic score.003 0.023 for intelligence. GWAS results are used to determine which of the two alleles for a SNP is positively associated with the trait.001 0. How many SNPs should be included in a GPS? The goal is to maximize predictive power in samples independent from the GWAS samples. odds ratio for qualitative traits) as gleaned from GWAS results (see table). such as polygenic risk scores. Furthermore. NATURE REVIEWS | GENETICS ADVANCE ONLINE PUBLICATION | 3 © 2 0 1 8 M a c m i l l a n P u b l i s h e r s L i m i t e d .006 SNP 4 G C G 1 0. Single base pair differences in inherited DNA Polygenic score 9 0. last column).000 polymorphisms SNP 10 C C G 1 0.004 Single-nucleotide SNP 9 A T T 0 0. the individual’s GPS is 9. In the same way. which is why this is called an additive model. Programmes including LDpred88 and PRSice89 provide pipelines for the construction of GPSs.002 0. An intelligence test score is a composite of several tests. LDpred uses all SNPs. Across studies. EA3. A l l r i g h t s r e s e r v e d .30. it can be used like any other variable in analyses. There are at least a dozen labels to denote GPSs. REVIEWS independent samples30. SNP Increasing Allele 1 Allele 2 Genotypic Correlation Weighted allele score with trait genotypic score SNP 1 T A T 1 0. global positioning system. A more predictive GPS can be constructed by weighting each genotypic score by the effect size of the SNP (β for quantitative traits. often with each test weighted by its contribution to general intelligence. which is tion27. weighted by their correlation with the trait. the genetic architecture of the trait and specifically modelling linkage disequilibrium87.005 SNP 2 C G G 0 0.

 34): transforming research on intelligence41. This gap is known as ‘missing heritability’. which can be assessed the 4% of the variance explained by the EA2 GPS. which from 2% to 3%. r2 >0. 10% is a long way from the heritability 280.000 individuals. 22): n = 54. with a single self-reported item. EA2 (REF. many 100 200 300 400 500 600 700 800 900 1. such r2 = 0. of the intelligence GWAS sample sizes from 78.03. Extrapolating from the results of EA3 with a sample size of more than 1 million. predicts about 4% of the variance and other GPSs in this multivariate way can already of intelligence in independent samples34.01. it is more difficult to obtain 3% of the variance of intelligence in independent sam­ very large sample sizes for intelligence.000 to Nonetheless. the top-down approach of GPSs that GPSs derived from a GWAS of intelligence with a sample size of 1 million. D.03. 30): n = 294. yield GPSs in addition to the EA2 GPS and IQ2 GPS will similar results35.000. First. is needed that is compatible with this extreme pleiotropy personal communication) — more than twice as much as the latest IQ3 GPS. IQ3 (REF. It is possible to use multiple GPSs the UK Biobank.000.000. than for years of education. Including the EA2 GPS. analytic GWAS using the UK Biobank data. However. The predictive power of GPSs derived from genome-wide association studies for the 50% heritability estimated by twin studies. personal communication). IQ3 4 not just the common SNPs used on current SNP chips. breaking through this ceiling of 25% 8 SNP heritability to the 50% heritability estimated from 7 twin studies — assuming that twin studies yield accurate estimates of the total variance explained by inherited DNA 6 differences — will require different technologies. IQ3. 31): n = 1. A l l r i g h t s r e s e r v e d . the variance in life sciences38 (BOX 4). r2 = 0. yielded Koellinger.000 to 294. which means that there GWAS sample size (×1.10. The current limit for the variance years of education predicted by the EA GPS grew only that can be predicted by GPSs is SNP heritability. EA1 (REF.  34).02.000 boosted the predictive power of the IQ GPS estimate of 50% obtained from twin studies of intel­ only from 3% to 4%. IQ1 aggregate thousands of these tiny effects is already (REF. IQ2 (REF. A GPS (IQ2) similar jump in the predictive power of the IQ GPS when derived from these GWAS results finally broke the 1% the sample size for GWAS meta-analyses of intelligence barrier of previous GWAS of intelligence by predicting exceeds 1 million. However.000.000) are no ‘genes’ to trace through the brain to behaviour. which were Multivariate GPS analyses that incorporate multiple released in June 2017 and are publicly available. GWAS results for years of education: after increasing which is a key genetic issue for all complex traits in the sample sizes from 125.100. and polygenicity 40. It is safe to assume 12 that GPSs for intelligence using current SNP chips can 11 approach the SNP heritability limit of 25% by amassing EA3 ever-larger GWAS samples and by using multitrait GWAS Percentage of variance explained (r2) 10 that include traits related to intelligence. the biggest hits have minuscule effects — less than Nature Reviews | Genetics a function of GWAS sample size.000 with the inclusion of cognitive data from Missing heritability. EA3 (REF.000. ligence. SNP heritability is about 25%34. Koellinger. such as 5 whole-genome sequencing data that include rare variants.com/nrg © 2 0 1 8 M a c m i l l a n P u b l i s h e r s L i m i t e d . or GWAS that require very large samples in 4 | ADVANCE ONLINE PUBLICATION www. D. For intelligence. However.nature. Figure 1 | Variance explained by IQ GPSs and by EA GPSs in their target traits as Third. r2 = 0. This GWAS analysis increased the to boost the power to predict intelligence by aggre­ number of identified genome-wide significant regions gating GPSs in a way analogous to aggregating SNPs to from 18 to 206 (REF. It might seem disappointing that the increase heritability of intelligence. preliminary analyses of the latest meta-analytic GWAS gence. Other meta-­ predict up to 7% of the variance in intelligence36. We can expect a 18 genome-wide significant regions33. p a r t o f S p r i n g e r N a t u r e .000. Genome-wide polygenic score (GPS) prediction 0. Unlike quanti­ n = 280. . which included ‘only’ 78. IQ2 GPS GWAS analyses. 28): n = 125. REVIEWS Large-scale genome-wide association studies of intelli. r2 = 0.000 hits are in intergenic regions. tative genetic studies that require special samples. we predict that more than 10% of the variance in intelligence will be predicted from IQ By contrast. such as years of 9 education. A follow‑up GWAS meta-analysis reached a sample size of 280.04. A sys­ (GWAS) of intelligence has increased in the past 2 years from 1% to 4%. IQ2 still has less predictive power than tested. In 2017. as twins. IQ2 3 EA2 GPSs in intelligence research 2 EA1 IQ A bottom‑up approach to intelligence focused on 1 EA IQ1 specific genes will be difficult for three reasons.000. 0 genetic effects are extremely pleiotropic. explain substantially more than 10% of the variance These IQ and EA GPS results are summarized in in intelligence. However. this result is parallel to ligence14. The latest EA tems biology approach to molecular studies of the brain GPS. which has to be ples. 33): n = 78. which is more than 20% of the 50% FIG.39. predicts more than 10% of the variance in intelligence (P. the largest GWAS meta-analysis of intel­ (EA3) with a sample size of more than 1 million31 (P.05% of the variance — which means that hundreds of of intelligence (IQ) and educational attainment (EA) increased linearly with sample thousands of SNP associations are needed to account size.37. A GPS derived from these produce GPSs (BOX 3). r2 = 0. Second. EA3. Note that the predictive power of the estimates the extent to which phenotypic variance for a EA GPS jumped to more than 10% of the variance in trait can be explained by SNPs across the genome with­ out identifying specific SNP associations. 1.

8% of the variance in intelligence. an atheoretical developmental findings about intelligence is that its approach can be used in analyses of multiple GPSs by incorporating as many GPSs as heritability as estimated in twin studies increases dra­ possible rather than selecting a few candidate GPSs. GPSs can be used to add a lated with their genetic propensities. Although the EA2 GPS alone gence. it is possible to the same SNP chip genotype information used in GWAS aggregate many GPSs to exploit their joint predictive power. GWAS multitrait analysis of GWAS (MTAG)91. for intelligence because genetic effects in the cognitive At 2 years of age. the EA2 GPS predicts 2% not be expected to systematically change the thousands of the variance in intelligence at age 7 years. Three examples are developmental change and accounted for most of the variance.42. these mutations would their education. multiple can be used to create GPSs for hundreds of disorders and GPSs were used to predict intelligence in a sample of 6. one of which is intelligence. The EA2 GPS30. whereas age‑to‑age genetic corre­ cognitive. multiple correlated predictors while preventing overfitting on the basis of training in GPSs for intelligence will open new avenues for one sample and testing in another sample in a repeated cross-validation design.44. especially GPSs derived from genome-wide association studies (GWAS) of environment (GE) interplay. although traits rather than the variance of each trait. This expense genome-wide polygenic scores (GPSs) has been key to predicting individual differences would not be incurred specifically to predict intelligence. This approach is a multiple regression prediction model that accommodates traits. More than 7% of the for capitalizing on these opportunities is to make the variance in intelligence was predicted37 through the use of another approach called ingredients for GPSs publicly available — that is. In an analogous manner. medical and anthropometric traits available in LD Hub92 that together lations are consistently high43. What could account predicted 4. One of the most interesting in the genome rather than selecting candidate genes. other GPSs added significantly to the prediction of continuity. p a r t o f S p r i n g e r N a t u r e . Although EA2. If GPSs for individuals do not change EA GPSs predict twice as much variance in intelli­ during the lifespan. These latter associations as time goes by 45. In other words. . genetic effects correlate highly across most NATURE REVIEWS | GENETICS ADVANCE ONLINE PUBLICATION | 5 © 2 0 1 8 M a c m i l l a n P u b l i s h e r s L i m i t e d . That is. 3% at age of inherited SNPs that contribute to a GPS. currently the that they are exceptionally stable throughout the lifespan best genetic predictor of intelligence until the EA3 GPS because they index inherited differences in our DNA becomes available. multivariate links between traits and gene– intelligence. 12 years and 4% at age 16 years in a longitudinal study 46. which performs a meta-analysis from summary summary-level statistics (BOX 5). For example. GPSs derived from GWAS of any trait at any age would be expected to have a corre­lation near Multivariate genetic research. significant age‑to‑age genetic correlations? Twin studies suggest independent contributions to the prediction of intelligence were also found for major that genetic effects are amplified through GE corre­lation depressive disorder GPSs and autism spectrum disorder GPSs. one-tailed). For example. DNA obtained in adulthood. A critical requirement high‑IQ individuals57. They are an exception to the rule that corre­ intelligence increase from childhood to adolescence to lations do not imply causation in the sense that there adulthood? Are the correlations between GPSs at these can be no backward causation when GPSs are correlated ages consistently high? with traits. For example. That is.0 when GPSs are constructed from DNA obtained at research focuses on the genetic covariance between birth and in adulthood for the same individual. statistics for a few correlated GPSs and produces new summary statistics that can be used to create a multivariate GPS. in complex traits such as intelligence. REVIEWS Box 3 | The use of multiple genome-wide polygenic scores to predict a trait GPSs are unbiased in the sense that they are not sub­ ject to training. resulting in high age‑to‑age genetic correlations. This question ligence in adulthood will predict adult intelligence just as raises interesting methodological and conceptual well from DNA obtained at conception or birth as from issues (BOX 6). leading to greater genetic dimension to any research with modest sample heritability of intelligence. 5% of the variance of intelligence in late adolescence32. was derived from a GWAS meta-­ sequence. the same large set of DNA were in the direction expected on the basis of the negative genetic correlation between intelligence and depression and the surprising positive genetic variants affects intelligence from childhood to adult­ correlation between intelligence and autism (see Multivariate genetic research). childhood IQ19 and household income90. faking or anxiety. size. A specific we are not aware of any empirical evidence relevant to multivariate question for intelligence research is why this prediction. infant intelligence tests predict less than domain have been shown in twin studies to be general. intelligence Multivariate genetic research is especially important tests at birth cannot predict intelligence at age 18 years. This research into the causes and consequences of intelli­ approach predicted 4. tested more rigorously and can be extended through the GPSs are unique predictors in the behavioural use of GPSs. included a total of 81 GPSs from well-powered GWAS of adulthood (60%).05.8% of the variance in intelligence. Although mutations accrue in the salivary and analysis of years of education in adults who had completed blood cells used to collect DNA. IQ and income GPSs for this increasing heritability despite unchanging drove most of the predictive power of this multiple-GPS analysis. By contrast. costing less than US$100 per person. The success of GWAS comes from their atheoretical approach that analyses all SNPs Developmental research. That is. but these DNA variants increasingly have an impact on intelligence as individuals select environments corre­ the hundreds of thousands. A l l r i g h t s r e s e r v e d . nothing in our brains. a GPS for intelligence that predicts Developmental hypotheses about high age‑to‑age 10% of the variance needs a sample size of only 60 to genetic correlations and increasing heritability can be detect its effect with 80% power (P = 0. A related advantage of GPSs as predictors is intelligence in childhood. mentioned above36. hood. Multivariate genetic 1. Does the variance explained by GPSs for sciences. behaviour or High age‑to‑age genetic correlations for intel­ environment changes inherited differences in our DNA ligence imply that GWAS of adults should predict sequence. In the same way. the first study of this matically from infancy (20%) to childhood (40%) to sort.710 unrelated 12‑year-old children36. Nonetheless. a GPS derived from GWAS of intel­ gence as do GPSs for intelligence itself. They are also inexpen­ Aggregating thousands of single-nucleotide polymorphism (SNP) associations in sive.

such as verbal.10) and autism (0. rare variants. what is needed and positive genetic correlations with height 51. with are large GWAS focused on well-measured specific myopia52 and. Additive genetic variance refers to the independent effects of alleles or loci that ‘add up’. such as verbal and spatial abilities. most of these studies assessed multiple measures of EA GPSs are correlated genetically with a wider specific cognitive abilities.27). regresses χ2 toms (−0. years of education (0. The viduals. personal communication).34.39. A l l r i g h t s r e s e r v e d . suggesting that most genetic influence on intelligence is additive. for each other traits. SNP heritability is about 25%22.21)33. from statistics against LD scores. These three types of heritability denote two types of ‘missing heritability’. and. and other cognitive possible using existing GWAS of intelligence because abilities.22) reading disability just as much as reading ability. with autism53. which implies that EA2 GPS predicts 5% of the variance in comprehension there are ability-specific SNP associations. mathematics and language. Single-nucleotide polymorphism (SNP) heritability is estimated directly from SNP differences between individuals. for example. genetic correlations across cognitive abilities mathematics47. In addition to investigating links between different single-nucleotide smoking cessation (−0.36). However. prevention Linkage disequilibrium (LD) score regression analysis ligence significantly correlated genetically with many and intervention. Twin heritability compares the resemblance of identical and fraternal twins to estimate genetic and environmental components of variance. Narrowing the missing heritability gap between SNP heritability (25%) and twin heritability (50%) will require different technologies that consider. longevity (0.32). slow readers to speed readers48. the EA2 GPS predicts statistics from GWAS summary (−0. of course. This index intelligence. surprisingly.70).29). such as spatial ability. which were combined to range of variables than any other GPS 50 . Instead. A recent multivariate finding is that the and educational skills are not 1.28). such as mathematics. EA GPSs are also likely to predict other edu­ intelligence. This is by far the most pow­ tant direction for research is to identify ability-­specific erful GPS predictor of reading ability because there have GPSs derived from large GWAS analyses focused on as yet been no large GWAS of reading with replicable specific cognitive abilities independent of general results49. The genetic strengths and weaknesses for individuals who same LD score regression analysis33 found that intel­ could be targets for personalized prediction. having ever smoked (−0. GPS heritability is currently about 10%31 (P. traits.27). multivariate genetic research can examine genetic polymorphism in a head circumference in infancy (0. such as reading and effects. p a r t o f S p r i n g e r N a t u r e . For intelligence. finds a similar pattern of results for intelligence pay-off from these studies will be GPSs that predict spe­ using the IQ2 GWAS: the negative genetic correlation cific abilities independent of general intelligence. depressive symp­ links between dimensional and diagnostic measures of genome-wide association study (GWAS). Because GPSs are always 6 | ADVANCE ONLINE PUBLICATION www.0. Adoption studies of first-degree relatives yield similar estimates of narrow heritability of intelligence.20) and the positive genetic ability-specific GPSs could be used to create profiles of correlations with height (0. including Alzheimer disease (−0. Despite this evidence for ability-general genetic as well as most educational skills.nature. These with schizophrenia (−0. intracranial volume (0. For example. gene–gene interactions and gene–environment interactions. as shown in the figure. D. Analysis that. spatial and memory equilibrium (LD) score regression analysis54. These data could be reanalysed in pervasive genetic influence of EA GPSs extends to meta-analytic GWAS that focus on specific abilities a negative genetic correlation with schizophrenia included in multiple studies. Broad heritability involves all additive and nonadditive sources of genetic variance. attention-­deficit–hyperactivity disorder the ‘same’ domain. The missing heritability gap between GPS heritability (10%) and SNP heritability (25%) can be narrowed by increasing GWAS sample size. Nonadditive genetic variance involves effects of alleles or loci that interact. which uses abilities and specific cognitive skills taught in schools. summary GWAS statistics rather than GPSs for indi­ for example reading. It does not specify which SNPs are associated with a trait. twin estimates of broad heritability are 50% on average14.REVIEWS Box 4 | Twin. it uses chance genomic similarities across hundreds of thousands of SNPs genotyped on an SNP chip for thousands of unrelated individuals to estimate the extent to which genomic covariance accounts for phenotypic covariance in these individuals. . single-nucleotide polymorphism and genome-wide polygenic score heritabilities Heritability is the proportion of observed (phenotypic) differences among individuals that can be attributed to genetic differences in a particular population. Genome-wide polygenic score (GPS) heritability is the proportion of variance that can be predicted by the GPS. For intelligence. Koellinger.23). whereas narrow heritability is limited to additive genetic variance. Total variance 100% Twin heritability 50% SNP heritability 25% GPS heritability 10% Missing GPS Missing SNP heritability heritability Nature Reviews | Genetics cognitive abilities. For intelligence. Linkage dis­ cognitive abilities.com/nrg © 2 0 1 8 M a c m i l l a n P u b l i s h e r s L i m i t e d . SNP heritability is the ceiling for genome-wide association study (GWAS) and for GPS heritability because all three rely on the additive effects of SNPs genotyped on SNP chips93. An impor­ and efficiency of reading 48. Preliminary analyses of this sort would be cational skills.

they will show that there are no The importance of both genetics and environ­ aetiologically distinct common disorders. the EA2 GPS were correlated with ligence because they grew up in the same family and intelligence in low-socio-economic-status just as much attended the same schools. which is used to estimate genetic correlations among traits54. Associations between environmental measures mental influence. which is responsible for five of the six GWAS for which summary statistics are publicly available in the intelligence section of LD Hub. siblings were thought to be similar in intel­ action46. or even before publication. REVIEWS Box 5 | Make genome-wide association study summary-level statistics publicly available It is essential for continued rapid scientific advances using genome-wide polygenic scores (GPSs) that summary-level statistics from genome-wide association studies (GWAS) be made publicly available for all single-nucleotide polymorphisms (SNPs) following publication. The high the environment. which is genetically dis­ tigate their interplay with aspects of the environment. only about 10% of published GWAS results are publicly available on LD Hub. That is. but these fears should be allayed as it is not possible to reconstruct individual-level data from summary-level GWAS statistics in large heterogenous samples95. but it was socio-economic status than in those of high socio-­ assumed that this family resemblance was due to nur­ economic status. a paragon is the Social Science Genetic Association Consortium27. for the first time. although it should be emphasized ioural sciences show genetic influence in twin stud­ that we are referring to the normal range of environ­ ies63. a centralized database and web interface that provides an automated pipeline for entering and using GWAS summary-level data92. GE correlation ligible. The first test of this hypo­thesis using ture. consistently support this assumption. Even extremely high intelligence intelligence will greatly facilitate this research because is only quantitatively. for NATURE REVIEWS | GENETICS ADVANCE ONLINE PUBLICATION | 7 © 2 0 1 8 M a c m i l l a n P u b l i s h e r s L i m i t e d . effects59. The exception is assessing genetic propensities of individuals to inves­ severe intellectual disability. normally distributed. Research environ­mental factors that make children growing up using GPSs is beginning to confirm these twin study in the same family different 14. Intelligence has correlate less with intelligence in environments of low always been known to run in families. GPSs for very high intelligence46. which means that family environments have lit­ refers to the correlation between genetic propensities tle effect on individual differences in the long run45. For example. the EA2 GPS found no evidence for such an inter­ That is. A l l r i g h t s r e s e r v e d . By contrast. findings about the ‘nature of nurture’ by showing. socio-economic-status family environments and higher Research using genetically sensitive designs has led in high-socio-economic-status family environments61. some twin research sug­ heritability of intelligence should not obscure the gests that the heritability of intelligence is lower in low-­ fact that heritability is considerably less than 100%. which made it difficult to use the data to investigate traits across studies. A worrying trend is that several commercial organizations do not allow summary GWAS statistics from their samples to be used in open-access summary-level statistics for all SNPs when their samples are included in meta-analytic GWAS. most notably linkage disequilibrium (LD) score regression. However. The reason why public access to summary statistics is important is that the construction of GPSs requires an effect size indicator and P value for each SNP in the GWAS. only contin­ ment for cognitive development is a recommendation uous dimensions55. After adolescence. but only until GPSs provide a particularly powerful approach to test adolescence. . different genetic­ they offer. such as the Psychiatric Genomics Consortium94. and experiences. the effect of shared for GE interaction compared with twin studies62. GE correlation is the reason why Family resemblance for intelligence is due to nature most environmental measures used in the behav­ rather than nurture. the possibility of directly ally from the normal distribution56. Some GWAS consortia are exemplars for making GWAS summary-level data available immediately upon publication. tinct from the rest of the distribution of intelligence58 GE interplay refers to two different concepts — GE and affected by rare. Twin and adoption studies as in high-socio-economic-status family environments. This problem has been solved with LD Hub. little is known about the specific mediated in part by genetic differences. This is also true for very low and for for investigating the interplay between them.57. its data should not be used in ‘publication’. not qualitatively. some authors apply conditions for the use of the summary statistics from their published GWAS paper. often de novo mutations with large interaction and GE correlation. which means ‘to make public’. GWAS summary-level data were stored in different databases using different formats. GE interaction denotes a conditional relationship in which the effects of genes on intelligence depend on Research on gene–environment interplay. If a group does not want its summary-level GWAS statistics to be freely available for a published meta-analytic GWAS. to one of the most important findings about environ­ This hypothesis predicts that GPSs for intelligence will mental influence on intelligence. Concerns about privacy have been put forth as an explanation. However. Such asymmetrical data-sharing policies between industry and academia will hold back research in the field. called ‘shared family environmental influence’. family environmental influence on intelligence is neg­ In contrast to GE interaction. GWAS summary-level statistics are also necessary for other analyses. not the extremes such as neglect and behavioural traits such as intelligence are also or abuse. p a r t o f S p r i n g e r N a t u r e . true to its Latin origin publicare. although three of the five GWAS were for years of education rather than for intelligence itself. In intelligence research.60. Until 2017. Others refuse to share these statistics altogether.

modify and create environ­ about the probabilistic nature of these predictions and ments correlated with their genetic propensities. 50% genetically different. A major impact of GPSs will be to recognize and respect these large genetic Understanding ourselves. IQ GPSs could be used The most exciting aspect of GPSs is their potential for to assess discrepancies between GPSs and educational addressing novel. A l l r i g h t s r e s e r v e d . dict individuals’ genetic propensity to learn. the EA2 GPS predicts medicine’. such as EA3. will predict at least as much variance in intelligence as does the current EA3 GPS. analogous to ‘precision EA2 GPS differences67. GPSs will also make it possible to fortune telling. not only in research but also in society. D. It is also predicted by personality traits. twice as much variance in educational attainment and A novel. a mean of 100 and a standard deviation of 15. Years of education is largely bimodal. intelligence is a more refined measure than years of education that captures the commonalities among diverse tests of cognitive abilities and is normally distributed. That is. Second. First. but it predicts 4% of the variance in intelligence46. we predict that an IQ GPS derived from a GWAS of intelligence with a sample size of 1 million. socially important direction for research occupational status in the post-Soviet era as in the Soviet using IQ GPSs is to understand differences within era in Estonia. primarily indicating whether an individual completed university. 1). but the EA3 GPS predicts more than 10% of the variance in intelligence31 (P.50 also sets a limit on the extent to which EA GPSs can predict intelligence. personal communication). First. and having fewer mental health problems. such as conscientiousness and well-being. children in public and private schools of bias of GPSs make them ideal for prediction. Koellinger. EA GPSs predict intelligence much better than IQ GPSs predict intelligence themselves. which differ in their EA2 GPSs because private schools select is essential for the prevention of problems before they pupils on the basis of genetic differences in intelligence66. try information.nature. intergenerational educational mobility reflects used to customize education. intelligence is not actually predicted to a greater extent by EA GPSs than by intelligence GPSs when the powers of the discovery GWAS are similar. The GPS effect sizes for intelligence are similar to those for EA GPSs for comparable effect sizes (that is. By contrast. years of education is a coarse measure. controlling for genetic influences. the limitations of their effect sizes (BOX 7). In other words. A ‘precision education’ based on GPSs could be Second. which we achievement (that is. The use of GPSs to predict general model for how genotypes become phenotypes individuals’ genetic propensities requires clear warnings — how children select. stability and lack research.50 in twin studies96 and linkage disequilibrium score regression studies97. why do EA GPSs predict intelligence to a greater extent than they predict educational attainment itself? That is. The reliability. The educational attainment GWAS incorporates SNPs associated with any of these traits. Finally. IQ GPSs will be used for prediction from early in life before intelligence or educational achievement can be Implications for society assessed. intelligence is more heritable (60% in adults) than years of education (40%). However. as shown in twin studies98. We predict that IQ GPSs will become The average difference between parents and offspring Box 6 | Educational attainment genome-wide polygenic scores and intelligence Educational attainment (EA) genomic-wide polygenic scores (GPSs) predict intelligence because the genetic correlation between years of education and intelligence is greater than 0. 8 | ADVANCE ONLINE PUBLICATION www. such as depression. There are two likely reasons why EA GPSs currently predict intelligence to a greater extent than EA GPSs predict years of education itself. Moreover. a finding compatible with the hypothesis families. reason and For scores on an intelligence test standardized to have solve problems. Together. GPSs will greatly advance research on GE correlation Although simple curiosity will drive consumers’ by providing an individual-specific index of the ‘G’ interests. IQ2 as compared to EA1 and IQ3 as compared to EA2. Because intelligence is one of the best assess environmental influences on intelligence while predictors of educational and occupational outcomes. which means they are on average and meritocracy 68. The genetic correlation of 0. educational achievement is not just a proxy for intelligence. The IQ3 GPS from the most recent genome-wide association study (GWAS) of intelligence predicts 4% of the variance of intelligence34. with a peak at the end of secondary school and another peak for individuals who attended university.REVIEWS example. these nonability traits account for as much of the heritability of educational achievement as intelligence99. socially important questions. p a r t o f S p r i n g e r N a t u r e .com/nrg © 2 0 1 8 M a c m i l l a n P u b l i s h e r s L i m i t e d . that EA GPSs correlate with social mobil­ routinely available from direct-to‑consumer companies ity 64 and capture covariation between environmental along with hundreds of other medical and psycho­ exposures and children’s behaviour problems and logical GPSs that can be extracted from genome-wide educational achievement 65. see FIG. occur. the EA2 GPS predicts 3% of the variance in years of education30. First-degree relatives are on average only 50% that heritability is an index of equality of opportunity genetically similar. In the school years. not just with intelligence14. IQ GPSs will be used to pre­ differences within families. the current GWAS sample sizes for educational attainment are three times larger than for intelligence. the average as direct-to‑consumer genomic services provide GPS difference between pairs of individuals who are selected information that goes beyond single-gene and ances­ randomly from the general population is 17 IQ points. GPS-based overachievement and will illustrate with three recent examples from our own underachievement). Of note. For this reason. GE correlation provides a geno­typing on SNP chips. . GPSs for intelligence are more than idle of GE interplay.

and the emotional impact of However. the General Certificate of score only slightly above the average. For the highest EA2 GPS decile.30 in this example).1% ~2% ~15% 50% ~75% ~98% 99. as GPSs always are. there is great variability between individuals. GPSs are useful for individual prediction as long as the when the correlation is modest (0. the average school intelligence for individuals. the potential for dis­ a low-IQ GPS does not mean that a child is destined crimination and stigmatization. Ethical. A l l r i g h t s r e s e r v e d .9% 0. School achievement was assessed For example. a b d 99. the Twins Early Development Study100. individuals within the lowest and highest EA2 GPS deciles differences (see the figure. Specifically. Although higher EA2 GPSs can be seen to predict higher of GPS predictions using data on EA2 GPSs and school achievement from GCSE scores on average. The scatterplot between EA2 GPSs and GCSE scores (see the increase. GPSs will become available to predict at the 28th percentile. the individual with the second-highest EA2 GPS has a GCSE by scores from a UK‑wide examination. achievement is at the 68th percentile. notion of biological determinism. powerful predictions can be made at the gender-regressed. administered at the end of compulsory eighth-lowest EA2 GPS has a GCSE score above the 75th percentile. ership of information. as is the measure of school achievement the two distributions is 61%.1% ~2% ~15% 50% ~75% ~98% 99. part c). accounting for 9% of the variance46. it does mean that the child is more likely to knowledge of one’s personal genomics and intelligence. The EA2 GPS is normally vary widely in school achievement (see the figure. the predictive power will example). the question of own­ to go no further in education than secondary school. education at age 16 years. and EA2 GPSs were constructed as described extremes. most evident at the extremes (see the figure. a strong relationship example because the effect size of this association is currently the between average EA2 GPS and average GCSE scores emerged that was strongest in the behavioural sciences. part a bottom). p a r t o f S p r i n g e r N a t u r e . The overlap in distributed. GCSE scores were age-regressed and Despite this variability. We illustrate the probabilistic nature in GCSE scores. when the sample was divided into ten equal-sized elsewhere46. For example. These issues of variability in prediction are (see the figure. . Genomic research and studies help parents understand why their children differ in of intelligence face four principal ethical concerns: the school achievement. The starting point for prediction is the distribution of individual Nonetheless. and with that.9% EA2 polygenic scores EA2 polygenic score percentiles GCSE score percentiles c ~65% GCSE score percentiles ~60% 50% ~40% ~35% ~25% –4 –2 0 2 4 1 2 3 4 5 6 7 8 9 10 GCSE scores Deciles for EA2 polygenic scores Nature Reviews | Genetics NATURE REVIEWS | GENETICS ADVANCE ONLINE PUBLICATION | 9 © 2 0 1 8 M a c m i l l a n P u b l i s h e r s L i m i t e d .1% –4 –2 0 2 4 0. REVIEWS and between siblings is 13 IQ points69. part a top). IQ GPSs might Ethical implications. an individual with the Secondary Education (GCSE). Squaring this probabilistic nature of the prediction is kept in mind. find academic learning more difficult and less rewarding These and other ethical issues are explored in detail by the than a sibling with a high-IQ GPS. We used the EA2 GPS prediction of GCSE scores as an groups (deciles) on the basis of their EA2 GPS. the It will soon be possible to explain a similar amount of variance in average school achievement of individuals in the lowest EA2 GPS decile is intelligence. part d). figure.9% GCSE score percentiles ~98% ~75% 50% ~15% ~2% 0. Because GPSs are probabilistic. part b) indicates the difficulty of predicting individual outcomes In summary. Conversely. As bigger and better GPSs emerge. GPS prediction of individual differences is the same for any predictor that accounts for 9% of the variance in the based on its covariance with the target trait (school achievement in this target trait. Legal and Social Implications (ELSI) Research Box 7 | The use of genome-wide polygenic scores to predict outcomes in individuals Genome-wide polygenic scores (GPSs) must be used with caution when correlation indicates that the EA2 GPS predicts 9% of the variance predicting outcomes in individuals.

 D. V. C. none is 100% heritable. ability: a five-stage design using DNA pooling and birth cohort of 1921. Psychiatry 22. General mental ability in 12. 10. M. F. 435–446 (2008).. 2003). M. R. attainment. R. and polygenic.. & 183–192 (2015). well established and uncontroversial among highly polygenic and associated with FNBP1L. J. Psychol. 1982).. G. Strenze. 21..W. Science 212. 385–386 (2017). Kamin. not hard. J. C. Nature 545. Butcher. J. recent books discuss eth­ the depth that they deserve. J. for example. 22. J. 497–509 (2013). Sci. Childhood intelligence in relation 2017). 253–258 (2014). genetic influence on intelligence. (Cambridge Univ. Psychiatry 20. J. 3. It is impor­ studies) and more than 40% of the 25% SNP heritability tant to acknowledge the risk of discrimination that of intelligence. et al. J. “it’s 19. We must be careful not to blame the With these advances in the past few years. Psychiatry 21. Psychol. as well as to guide eugenic of the variance in intelligence (that is. G. Hare. 1976). Annu. Mol. Learn. 162–173 13.nature. R. Benyamin. R. Norton. differences in genetics Psychiatry 19. Block. A reach­ are heritable. of quantitative genetic research showing that inherited wired and deterministic like single-gene disorders. Genetic contributions to stability and 9.. function: a report from the COGENT consortium. 23. 1. & Dworkin. At the same into the causes and consequences of intelligence. Schmidt. A l l r i g h t s r e s e r v e d . 2013). 10 | ADVANCE ONLINE PUBLICATION www. However. I. 758–767 (2016). reaching our current ability to predict 10% of was infamously used to differentiate European immi­ the variance in intelligence from DNA alone. (2011). et al. S. 1314–1323 (2012). 212–214 (2012). T. I. B. recommendations. Neiderheiser. A. The Evolution of extreme selected groups. The IQ Controversy: Genes Brain Behav. Intelligence (Psychology Press. markers for allelic associations with general cognitive intelligence from age 11 to age 90 years: the Lothian Psychol. The time. L. IQ GPSs will open new avenues for research complex in our current age of information. typically because findings from Until 2016. S. 155–186 (2017). & Plomin. Mol. GPSs could predict only 1% of the var­ both fields of study were applied to justify policies that iance in intelligence. Haier. 2016). & Starr. et al. L. P. Davis. DeFries. GWAS meta-analysis reveals novel held back by its past. S. 1–7 (2017). cognitive functions and educational attainment in UK to major causes of death in 68 year follow‑up: 15. R. Soc. Huntington disease. 23. B. S. Mol. 26. 2. IQ is not determin­ in intelligence scores explained by GPSs and the SNP istic but probabilistic. Davies. It is the need to assess intelligence. & McGue. it is equally important to magnitude are large enough to be “perceptible to the realize that research does not lead directly to any policy naked eye of a reasonably sensitive observer” (REF. By contrast. Intelligence 35. E. Survival of the friendliest: Homo sapiens 24. The Science and Politics of IQ (Routledge. Mol. T. intelligence scientists or entire disciplines when their findings are steps out of the shadows and takes the lead in genomic used wrongly 9. J. Plomin. C. N. Deary. understanding and managing the emotional impact new genetics of IQ GPSs will bring the omnipotent vari­ that stems from knowledge of our genomics and intelli­ able of intelligence to all areas of the life sciences without gence have emerged as new societal responsibilities. Trampush. [No authors listed] Intelligence research should not be 18. Knopik. Strand. et al. F. J. The Mismeasure of Man (W. G. prospective population study. Sternberg. For example. 16. et al. Behavioral Genetics. Intelligence 11. Genetic contributions to variation performance.” Genome-wide quantitative trait locus association scan 2. 500K single nucleotide polymorphism microarrays. It is DNA differences are responsible for substantial indi­ important to recall here that although all complex traits vidual differences in intelligence test scores. A genome-wide scan of 1842 DNA 7. Pattie. M. Chabris. C. 996–1005 4. J. 24. 2708 (2017). Smith. et al. (2001). This is an important milestone for the occurs on the back of scientific findings about indi­ new genetics of intelligence because effect sizes of this vidual differences. G. et al. which has Conclusions 100% penetrance. ing challenge is to break through the ceiling of 25% SNP Issues of discrimination and stigmatization have heritability to reach the 50% heritability estimated by accompanied research into genetics and intelligence twin studies. I. meta-analytic review of longitudinal research. M. 14. W. 336 (2017). 7. Behav. J. Bouchard. M. Pers. J. 13–21 (2007). Most reported genetic associations 25. Pinker. J. Craig. Davies. L. Who ‘owns’ our genetic information? And who should In addition to investigating traditional issues about decide who can access it? The question of ownership of development. Sci. with general intelligence are probably false positives. 73). 68. J. P. In short. sions of ethical issues that surround personal genomics specifically in relation to education71 and occupation72. Intelligence 24. 357. Gottfredson. consortium (N = 53 949). multivariate links among traits and GE personal data has become pivotal but also increasingly interplay. between people. Press. A more daunt­ more than the sum of their genes or their IQ scores. L. Brit. The stability of evolved via selection for prosociality. GPSs are ‘less dangerous’ Genetic association studies have confirmed a century because they are intrinsically probabilistic. 79–132 (1997). GPSs grants to the United States of America who arrived at will soon be available that can predict more than 10% Ellis Island in the early 1900s. concluding. everyday life. C. Genome-wide association studies Intelligence and educational achievement. Psychiatry 16. J. The Neuroscience of Intelligence Biobank (N = 112 151). an individual is always herit­ability of intelligence of about 25%.com/nrg © 2 0 1 8 M a c m i l l a n P u b l i s h e r s L i m i t e d . 7th edn (Worth.. Deary. 8. 1055–1059 genome-wide association studies in the CHARGE 5. Intelligence and socioeconomic success: a (1981). IQ testing then. Progress has been rapid since served sociopolitical ideologies. Gould. establish that human intelligence is highly heritable 35. but we expect that the discus­ ical as well as scientific issues about personal genomics. Familial studies of in general cognitive function: a meta-analysis of (2004). S. Deary. more than 20% of ideas about sterilization in Britain and the United States the 50% heritability of intelligence estimated from twin of America throughout the 20th century 11. Sci. Many of these ethical discussions focus on single-gene disorders. the world of work: occupational attainment and job 1974). J. Plomin. Genome-wide association study of 6. Childhood intelligence is heritable. & Hunter. Nature 482. Davies. Genetics and educational geneticists that together. loci and genetic correlates for general cognitive This editorial is a landmark in the acceptance of Psychol. & Visscher.REVIEWS Programme. Mol. . & Fernandes. 401–426 (2007). Human Nature (Penguin. Med. 86. p a r t o f S p r i n g e r N a t u r e . S. 2361–2368 17. A similar logic can able objective shared with all complex traits in the life be applied to IQ scores: although they have great predic­ sciences is to close the gap between the 10% variance tive validity for key life outcomes1–6. Why g matters: The complexity of underwrite significant variation in intelligence 20. 31.. et al. will consolidate the DNA revolution. & Kaufman. which is an integral part of the Human beyond the scope of this paper to elucidate these issues in Genome Project 70. J. O. Calvin. I. research. Critical Readings (Pantheon. Genet. The Blank Slate: The Modern Denial of of general cognitive ability using pooled DNA and change in intelligence from childhood to old age. Cesarini. et al. Rev. from the beginning. In addition. intelligence: a review.

21. 33. Multi-polygenic score approach to increase power — a case–control design in which the adoption study. 28. & Davis. McEwen. D. Plomin. 27. & Kovas. et al. years. in intelligence in an independent sample. Genet. B. Dissecting the genetics of attainment and income predict 7% of the variance 61. et al. et al. Socioeconomic status and 52. Carroll. Protzko. C. Briley. 15. I. implications program of the National Human Genome This is the GWAS origin of EA1. 334–349 (2017).. attainment. Ageing 26. Marioni. Factor-Analytic Studies (Cambridge Univ. Mol. et al. 14. identifies new genes and functional links to 55. NATURE REVIEWS | GENETICS ADVANCE ONLINE PUBLICATION | 11 © 2 0 1 8 M a c m i l l a n P u b l i s h e r s L i m i t e d . trait prediction. Genetic influence on intergenerational educational and an agenda. L. Davies. R. 54. 45977 (2017). A. genome-wide association studies. 1304–1310 (2017). Preprint at https://doi. 30–36 (2015). Psychiatry 80. Am. Modeling linkage of complex diseases. USA 113. and social (2013). This book describes genetic research on behaviour 65. Genet. Large cross. Bulik-Sullivan. Smith-Woolley. S. Press. subject privacy. Pasaniuc. Deary. which yields a GPS This study shows that EA2 predicts 9% of the Research Institute: reflections on an ongoing that predicts 1% of the variance in years of variance in tested educational achievement at age experiment. et al. J. T. 91–96 (2014). including one not Abstracts.1101/160291 (2017). Behav. This study employs multiple-trait analysis of GWAS family so different from one another? Behav. R. et al. Dev. Macfarlane.. C.17030283 (2017). C. Nature 461. Natl Acad. J. maximizes the potential of summary level GWAS data 42. estimation and interpretation (1948).doi. Hum. E. Natl Acad. Deary. Selzam. recent data. Is working memory This study employs a multiple-GPS approach and 58. K. Sci. Annu. et al. & (2016). in the press). effect on IQ. Proc. and the Future (Princeton Univ. disequilibrium increases accuracy of polygenic risk (2009). genome-wide-association studies for behavioral traits. E. Spain. MTAG: Multi-Trait Analysis of 41. J. Am. J. Genetic acf.pdf (2010). Ability (Praeger. Read. A l l r i g h t s r e s e r v e d . Common genetic variants This article highlights five genetic findings that are 68. & Tomkiewicz. Turley. et al. M. M. Zheng. Ayorech.559 individuals development: a meta-analysis of longitudinal twin and 69. Developmental Influences on Adult intelligence. Krapohl. Genet. K. Plomin. . & Tucker-Drob. & DeFries. & Galloway. Psychol. C. USA 114. Mol. 15. Psychiatric genomics: an update in six studies from four countries. Genet. Krapohl. A. R. Bacanu. The future of genomics org/10. Nat. 28. A. Exp. Davies. 539–542 (2016). Hill. Widespread covariation of early 92. 31. Plomin. Euseden. 24. T. 74. Mol. Behavior Genetics Association 47th Annual Meeting 531–541 (2016). GWAS meta-analysis (N = 279. Plomin. Intelligence 4. 425–430 (2012).. et al. 44. Cell 169. for SNP heritability and genetic correlation analysis. Bouregy. R. R. analysis of 78. Natl Acad. 29. S. I. Sci. W. Behavior Genetics Association 46th Annual Meeting associated with cognitive performance identified using special to intelligence differences. Rietveld. educational attainment and occupational status This study uses EA1 SNPs to predict intelligence. Natl Acad. 803 (2016). Common 80. 78. Stability and change in intelligence This is the GWAS origin of IQ2 GPS. Blueprint: How DNA Makes Us Who We 64. This is the GWAS origin of EA2 GPS. and 87: the Lothian Birth the prediction of intelligence from 1% to 3%. K. Biol.163 (2017). Plomin. adolescence. A. 75. 34.. J. 957–972 (2016). S. Mol. 3083–3089 (2016).1038/mp. et al. 39. Gow. Sharing extended summary data from differences: five special findings. Trzaskowski. which 49. 47.org/10. Administration for Children and Families https://www. Common genetic variants explain 1993).gov/sites/default/files/opre/hs_impact_study_ correlated traits identifies 107 loci associated with studies in intellectual disability and related disorders. LD Score regression 79. 1704–1713 (2013). 272–279 (2017). disorders. Nat. Socioeconomic status (SES) scores. Esko. & Bates.hhs. et al. Honzik. determined and measured. 25. org/10. A. E. The stability of mental test performance between 66. 18.. Educ. 86. D. Proc. Finding the missing heritability status interaction on intelligence. Plomin. which yields a GPS (IQ3) that functional and exonic variation associated with adolescence raises IQ scores. D. Cohorts of 1921 and 1936.ajp. of SNP-based heritability. http://dx. Sci. Does the raising IQ−raising g distinction influencing general cognitive function include genes 57. L. J. educational attainment (n = 1.0003 of the population (mean IQ of 170). Grigorenko. J. 1977). Discontinuity in the genetic and training effective? A meta-analytic review. Psychol. 11727–11732 (2017). Nature 533. intelligence research. 1998). to predict intelligence: the generation Scotland study. C. K. A. Gale. The ethical. Psychol. Plomin. Curr. S. Psychol. 49. Genome-wide association meta. E. Intelligence: The Seattle Longitudinal Study (Oxford org/10. 39. Preprint at https://doi. Okbay. M. Psychol. Genome-wide association study 48. & Price. PLOS ONE 12. A. Ethics and Education (Cambridge Univ. Gottfredson. Concepts. Proc. & Plomin. 35. J. Natl Acad. predicts 4% of the variance in intelligence. Plomin. 10. genetic influence on “environmental” measures. S. et al. et al. Yang. C. GWAS predict 5% of the variance in intelligence. L. et al. 17. disorders are quantitative traits. C. J. & Daniels. & Plomin. A. 1098–1103 85. more than 10% of the variance in the targeted trait. et al.000 twin pairs 67. J. M. Z.2017. Molecular genetic contributions to 40.doi. EA3. Psychiatry 20.. Psychiatry (2016). Psychol. Psychol. R. Common DNA markers can account and children’s intelligence (IQ): in a UK‑representative 89. for intelligence and finds that educational 10. which increases distinguishes confounding from polygenicity in from age 11 to ages 70. Biobank. the top 0. Reichenberg. A. Y. M. GWAS of 126. A. Psychiatry 22. (N = 280. Bioinformatics 31. This paper describes the largest GWAS of 56. Sniekers. Carrion-Castillo. Psychiatry 21. Behav. C. Dumaret. S. Manolio. E. Plomin. Genet. & Pritchard. Differences in exam Bioinformatics 33. Preprint at https://doi. Schaie. 576–592 (2015). et al. GPSs that show genetic influence on contemporary genetic studies is unlikely to threaten 98–108 (2015). W. Krapohl. Tucker-Drob. J. Sci. 13790–13794 (2014). 16. M. Conley. Explaining the during and after the Soviet era in Estonia [abstract].930) 291–295 (2015). I. final. et al. complex traits using summary association statistics. Genet. et al. 1177–1186 (2017). 32. finds that 81 GPSs derived from well-powered environmental causes of the intellectual disability Psychol. 53.. A. 201–292 (1904). 17. Rev. K. et al. & Veltman.308 individuals identifies new loci and general cognitive function and major psychiatric This article is an authoritative overview of genes influencing human intelligence. which was the strongest GPS prediction of a 481–504 (2014). W. 49. M. 38. 25. E. Selzam. Proc. & Tan. A. Mol. & Hulme. et al.121 (2017). Behav. et al.. abilities. (in the press). W. Rimfeld. Sci. Plomin. Genetic influence on USA 111. Sci. von Stumm. Rev. L. J. increases the prediction of educational attainment genome-wide studies of language and reading in a 73. S. Jensen. 60. 7. E. 267–272 (2017). A twin study of the genetics of NPJ Sci. 65–71 associated with brain health and structure for extremely high intelligence. Mainstream science on intelligence. Gilissen. T. disabilities. Generalist genes and learning Genetics. 309–324 performance between pupils attending different school 93. Rev. 44.1101/176511 (2017). Psychol. Psychiatry http://dx. et al. A. An expanded 63. et al. Brinch. et al. Sciences (Lawrence Erlbaum Associates. 87. J. Phenome-wide analysis of Wall Street Journal (13 December 1994). Evaluation of results from 2017). L. 26. Psychol. Genet. 71. Abstracts. not software. F. Vilhjálmsson. Behav. J. 79. 91. PRSice: polygenic risk score for more than half of the genetic influence on cognitive sample SES moderates the environmental. W. A. Genome-wide polygenic scores 72.100. J. A. novel independent dataset.‑C. Intelligence 48. p a r t o f S p r i n g e r N a t u r e . T.org/10. & Bergeman. 747–753 138–149 (2016). Genet. O. A. Hum. D. Press.. USA 96. GWAS. increasing heritability of cognitive ability across Behav. Am. Why are children in the same for developmentalists. (2017). extremely high intelligence. Stud. B. B. This study refers to the largest GWAS of 51. 46. Univ. M. Head Start Impact Study Final Report. M. 270–291 (2013). 1467–1471 46. P. Sci. L. Haworth. Rietveld.. 562–568 (2013). & Fletcher. J. 84. Genom. Spearman. 1–16 (1987). R. J. The g Factor: The Science of Mental the growth of intelligence from infancy through correlation between myopia and intelligence. our History. 90. 1263–1278 (2013). E. Nat. PLOS ONE 7. Genet. 699 (2017). The genetics of success.1038/mp. B. W. et al. E. This GWAS of intelligence uses a novel strategy to How can we boost IQs of “dull children”?: a late 36. Schooling in intelligence to date. N. G. P. Okbay. Rev. Mol. the proxy-phenotype method. 43. et al. types mirror the genetic differences between them. 1302–1310 (2017). 62. 2005). S. 83. K.. Age-dependent pleiotropy between 453–482 (2012). Intelligence. 1466–1468 (2015). 1975–1986 (2014). although less than 1% of the variance is predicted. Annu. et al.1176/appi. Dev. D. http://dx. Genetics and intelligence This paper describes both twin analyses and EA2 95. Brain Sci. A. R. 82. & Allen... Li. Bull. Statistical Power Analysis for the Behavioral from 1% to 3% of the variance. Preprint at https://doi. education. Ninety-nine independent genetic loci 1145–1151 (2016). M. Psychopathol. 45. 88. Sci. J. from DNA. J. 37. et al. 232–240 (2011). A combined analysis of genetically 59. J. R. Haworth. C. LD Hub: a centralized database and from twin studies to the DNA revolution and its environmental exposures and trait-associated web interface to perform LD score regression that implications for science and society. E.2017. genome-wide polygenic scores. Puma. Nat. Nat. & Deary. 94. Human Cognitive Abilities: A Survey of Genet. et al. S. attainment. GWAS of educational 50. A. P. & Simpson. McIntosh. Genet. much greater for intelligence than for other traits. 47. Zabaneh. et al. et al. J. S. subjects were individuals with extremely high IQ from 8790–8794 (1999). Genes Brain Behav. R. spectrum. polygenic variation. Predicting educational achievement 70. Rep. Rietveld. 15–24 (1980). 97. org/10. R. et al. Sci.2017. 63. which the majority of the correlation between height and 76. intergenerational EA. Latham. Sullivan. e0179504 (2017).000). S. Rev. 117–127 (2017). 47. attainment – phase 3: main results [abstract].1101/118810 Are (Allen Lane/Penguin. Biol. R. A. The Genome Factor: What the identifies 74 loci associated with educational predict reading performance throughout the school Social Genomics Revolution Reveals about Ourselves. 592–617 (2005). 266–273 (2016). C. E. & von Stumm. D. Press. Brain Sci. 872–878 (2009). Learn. Psychiatry 21. USA 109. high cognitive ability selected from 11. Genetics and intelligence: identifies genetic variants associated with adoption studies. 9–18 (2016). legal. I. Hill. two and eighteen years. Sci. R. 15. et al. 49. Williams. Psychol. ‘General Intelligence’ objectively increases the power of its GPS. Genet. M. Hill.1101/184853 (2017). S. M. Belsky. 1107–1112 (2017). C. Proc. 359–370 (2009). et al. Savage. Phenotypic and genotypic 77. e30320 (2012). intelligence. S. 131. Replicability and robustness of behavioural trait at that time. Hanscombe. Krapohl. et al. 24. Duyme. Cohen. national differences in gene × socioeconomic Nat. R. Boyle.. E.360). C. Press. 1188–1193 (2015). Sci. Psychol. A. 373–386 (1991). org/10. 30. 27. Science 340. G. B. A genome-wide analysis of putative 81. C. M. A genome-wide association study explain the fadeout effect? Intelligence 56. R. R. Y. Sci. et al.doi. D. REVIEWS 27.. Sci. et al. D. educational attainment. genetic. Melby-Lervåg. Vissers. The nature of nurture: social deprivation and household income in UK view of complex traits: from polygenic to omnigenic. 2017). E. J. mentioned in this Review — assortative mating is Metspalu.

 R.. The high heritability of the European Research Council (ERC) under the European DATABASES educational achievement reflects many Union’s Seventh Framework Programme (FP7/2007‑2013) UK Biobank: www. Calvin. Acknowledgements Competing interests statement 97. 109–140 port from the US National Institutes of Health (AG046938).com/nrg © 2 0 1 8 M a c m i l l a n P u b l i s h e r s L i m i t e d . et al. Variation grant to R. 42. 15273–15278 295366.org/ 100. M. Intelligence 44. Twins Early a J a c o b s Fo u n d a t i o n R e s e a rc h Fe l l ows h i p awa rd ALL LINKS ARE ACTIVE IN THE ONLINE PDF Development Study (TEDS): a genetically sensitive (2017–2019). R. (TEDS) and their families. Soc.ac. O. Sci. Branigan. A l l r i g h t s r e s e r v e d . not just intelligence.S. with additional sup. E.v. manuscript. Genet. Behav. socioeconomic status and intelligence. M. (2013).000 and 166. S. et al. C. Krapohl.REVIEWS 96. international meta-analysis. is also supported by a Medical Research FURTHER INFORMATION (2014). McCallum. E. school children. J. C.nature. Forces 92. 12 | ADVANCE ONLINE PUBLICATION www. R. The authors contributed equally to all aspects of the population samples of 174. Council Professorship award (G19/2).. of the participants in the Twins Early Development Study 26–32 (2014). et al.P.P. Multivariate genetic analyses of investigation of cognitive and behavioral development Author contributions cognition and academic achievement from two from childhood to young adulthood.000 Genet. grant agreement 602768 and ERC grant agreement Proc.broadinstitute. 117–125 (2013). Molecular genetic contributions to The authors gratefully acknowledge the ongoing contribution The authors declare no competing interests. USA 111. claims in published maps and institutional affiliations. Marioni. Natl Acad. & Freese. A. 699–710 (2012). K. . 16. from the UK Medical Research Council (MR/ Springer Nature remains neutral with regard to jurisdictional in the heritability of educational attainment: An M021475/1 and previously G0901245). p a r t o f S p r i n g e r N a t u r e . S. Haworth. Twin Res. The research reported here has also received funding from 99. TEDS is supported by a programme Publisher’s note 98. J. & Plomin. R.uk genetically influenced traits.ukbiobank. is supported by LD Hub: http://ldsc. Hum. Davis.