The genetic specificity of specific cognitive abilities after controlling for general cognitive ability

doi:10.21203/rs.3.rs-5053719/v1

Download PDF

Research Article

The genetic specificity of specific cognitive abilities after controlling for general cognitive ability

https://doi.org/10.21203/rs.3.rs-5053719/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 18 Feb, 2025

Read the published version in Behavior Genetics →

You are reading this latest preprint version

Diverse cognitive abilities correlate about 0.30 phenotypically and about 0.60 genetically. Their phenotypic overlap defines general cognitive ability (g), driven largely by genetic overlap. Consequently, much of our understanding of the genetic landscape of specific cognitive abilities (SCA) likely reflects g rather than the SCA themselves. Removing this g-associated genetic variance will sharpen research on SCA. Here, we use Genomic Structural Equation Modelling (Genomic SEM) to remove the shared genetic variance among 12 diverse SCAs that capture verbal and nonverbal cognitive domains. We applied Genomic SEM to summary statistics from the largest genome-wide association studies of verbal SCA (GenLang Consortium, 5 tests) and nonverbal SCA (UK Biobank, 7 tests) to chart the genetic landscape of the 12 SCA independent of g (SCA.g) as compared to uncorrected SCA. We found that SNP heritabilities were nearly as high for SCA.g as for SCA: the average SNP heritability was 0.16 (SE = 0.02) for SCA and 0.13 (SE = 0.02) for SCA.g. Despite this, we found that the genetic landscape of SCA is transformed after controlling for genomic g. The matrix of positive genetic correlations for SCA (average genetic correlation of 0.45 among our 12 tests) disappeared for SCA.g (average genetic correlation of -0.07) and included some strong negative correlations; for instance, Memory and Word (-0.72), Fluid and Symbol (-0.72), and Tower and Spelling (-0.79). The summary statistics of SCA.g can be used by researchers to create polygenic scores that focus on the specificity of specific cognitive abilities.

specific cognitive abilities

general cognitive ability

genetic specificity

Genomic SEM

In 1904, Charles Spearman observed a phenomenon that has since become one of the most consistently replicated and widely accepted findings in psychology: positive correlations, known as a positive manifold, among tests of specific cognitive abilities. This led him to hypothesize that a common factor, general cognitive ability (g), accounts for the shared variance (Spearman, 1904). Later, Spearman introduced the concept of specific factors, which represents the unique unshared component of group factors, such as spatial and reasoning abilities (Spearman, 1927). Here we focus not on group factors of specific cognitive abilities (SCAs) but on individual cognitive tests independent of g.

Not only do cognitive tests show substantial phenotypic correlations, about 0.30 on average (Carroll, 1993), but they also demonstrate even higher genetic and genomic correlations, generally greater than 0.60 (Plomin & Deary, 2015, Grotzinger et al., 2019; Hindley et al., 2023; Rimfeld et al., 2015; Trzaskowski et al., 2013). These correlations indicate that the positive manifold among cognitive tests is largely driven by genes. Although the magnitude of genetic correlations among diverse cognitive tests is remarkable, the fact that genetic correlations are considerably less than 1.0 provides evidence for genetic specificity.

Due to the pervasive influence of g, much of our understanding of the genetic architecture of cognitive tests reflects the genetic variance that is shared by all tests, rather than variance specific to each test. In fact, genetic studies often leverage this positive manifold and combine multiple cognitive traits to increase statistical power, and to focus on shared genetic variance rather than unique genetic variance. The hypothesis underlying the present research is that removing the effects of g will sharpen genetic research on the specificity of cognitive tests.

Despite the widespread influence of g, there have been few genetic or genomic studies of cognitive tests independent of g. The few available twin studies indicate that g-corrected cognitive tests remain substantially heritable despite controlling for the heritability of g (Procopio et al., 2022; Rimfeld et al., 2015, 2017; Tosto et al., 2013). Multivariate genetic twin analyses imply genetic specificity because genetic correlations among cognitive tests are less than 1.0, but none have directly investigated the genetic relationships among cognitive tests independent of g. However, a novel application of network analysis to multivariate twin data has recently yielded the intriguing finding that some of the typically positive genetic correlations between pairs of cognitive tests switch to become strongly negative genetic correlations when the network of relationships with all other tests is partialled out (Knyspel & Plomin, 2024). This finding implies that some DNA variants might operate in opposite directions for pairs of g-corrected cognitive tests, which would obfuscate research on specific cognitive abilities such as genome-wide association analyses.

Genome-wide genotypes offer new opportunities to investigate changes in the genetic landscape before and after controlling for g (Procopio et al., 2024). The only relevant genome-wide association (GWA) analysis conducted analyses that are comparable to GWA of g-corrected cognitive tests investigated English, mathematics and science achievement phenotypically corrected for g (Donati et al., 2021). Although these subjects are educational outcomes, they are conceptualized as measures of cognitive abilities in this context due to the high phenotypic and genetic correlations between cognitive abilities and educational performance. Despite being underpowered (N = 3,000), the study found significant SNP heritabilities for residualized mathematics and science, and one GWA-significant SNP was found across the analyses for science. Genomic investigations of g-corrected cognitive tests with larger sample sizes are needed to further investigate the genetic architecture of specific cognitive traits independent of the influence of generalist genes. In lieu of sufficiently large GWA analyses of g-corrected cognitive tests, the aim of our study was to leverage Genomic Structural Equation Modelling (Genomic SEM) (Grotzinger et al., 2019) to investigate the genomics of cognitive tests independent of g. We modeled GWA summary statistics from the largest GWA association studies of five verbal tests (Eising et al., 2022) and seven tests that are primarily nonverbal (de la Fuente et al., 2021). Combining twelve summary statistics from these two GWA studies will produce a better representation of g, which includes both verbal and nonverbal abilities.

Unlike previous studies, we used Genomic SEM to remove the genetic influence of g from each of the 12 tests and created 12 sets of GWA summary statistics. Our goal was to compare the genetic correlational landscape of cognitive tests before and after controlling for g, as well as to compare downstream analyses of GWA-significant SNPs, SNP heritability, and genetic correlations with other cognitive traits and traits outside the cognitive domain. This study was preregistered on the Open Science Framework (OSF; https://osf.io/9qwbn/); deviations from the registered protocol are described in the Supplementary Note.

Samples and Measures

In order to comprehensively investigate the genetic landscape of cognitive tests, we selected GWA summary statistics for 12 tests from two studies: de la Fuente et al. (2021), which included seven largely nonverbal performance tests, and Eising et al. (2022), which analyzed five reading and language tests. This combination allowed us to construct a broader g factor that incorporates both verbal and nonverbal components. Although reading and language tests are often referred to as educational skills, we define specific cognitive abilities more broadly to include not only traditional psychometric tests such as reasoning, memory and perceptual speed but also tests traditionally considered as educational outcomes such as spelling, phoneme awareness and reading words and nonword.

The first study (de la Fuente et al., 2021) conducted a GWA analysis of cognitive tests from the UK Biobank (UKB; Sudlow et al., 2015). The UKB is one of the world’s largest biobanks; from 2006 to 2010, it recruited half a million individuals aged 40 to 69 across the UK. de la Fuente et al. used genotypic data from UKB participants of European ancestry (N = 11,263 to 331,679 adults aged 40 to 75). Participants were included who completed at least one of seven cognitive tests: Matrix Pattern Recognition (a measure of nonverbal reasoning, which we refer to as ‘Matrix’), Pairs Matching Test (episodic memory, ‘Memory’), Reaction Time (perceptual motor speed, ‘Reaction’), Symbol Digit Substitution (information processing speed, ‘Symbol’), Trail Making Test-B (executive function, ‘Trail’), Tower Rearranging (executive function, ‘Tower’), and Verbal-Numerical Reasoning (Fluid Intelligence, ‘Fluid’). Due to length restrictions in testing such a large sample, the UKB tests are at the lower end of reliability for cognitive tests (Fawns-Ritchie & Deary, 2020).

The second study (Eising et al., 2022) conducted a meta-analytic GWA analysis using data from the international Genetics of Language consortium (GenLang), focusing on speech, language, reading and related verbal skills. We used summary statistics from a meta-GWA analysis of 22 independent cohorts of individuals of European ancestry (N = 13,633 to 33,959 children and young adults aged 5 to 26 years). GWA summary statistics were available for five tests: Word Reading (number of words read aloud correctly, ‘Word’), Nonword Reading (nonwords are phonemes that look like words but have no meaning, ‘Nonword’), Spelling, Phoneme Awareness (‘Phoneme’), and Nonword Repetition (number of nonwords repeated aloud correctly, ‘Repetition’).

Scores on all tests were coded so that higher scores represented better (faster or more accurate) performance. Further information on the measures, cohorts and data collection procedures of both studies is included in the Supplementary Note and the original publications.

GWA analysis of g-corrected cognitive tests using Genomic Structural Equation Modelling

To create summary statistics for the 12 cognitive tests residualized by their genomic covariance, genomic g, we used Genomic Structural Equation Modelling (SEM) in R studio (Grotzinger et al., 2019). Genomic SEM combines SEM with factor analysis to explore the genetic relationship among complex traits. After constructing a genetic covariance matrix (S) along with its corresponding sampling matrix (V), the SEM of choice is fit to the genetic covariance matrix, and the parameters and their standard errors (SE) are estimated.

We created g-corrected cognitive test GWA summary statistics by fitting the model shown in Fig. 1 to the S and V matrices. This model extracted a single common factor, genomic g, from the 12 GWA summary statistics, while simultaneously residualizing the effect of genomic g from each test. We ran the model 12 times, correcting each test for genomic g individually. For more details on the procedure, see the Supplementary Note.

We extracted a single common factor, the ‘g factor’ (Jensen, 1998), as a conservative approach that captures as much covariance as possible among the 12 cognitive tests, rather than removing covariance that clusters more tightly, for example on verbal and nonverbal factors. In other words, our goal is not to identify the best-fitting model of cognitive abilities, which involves a hierarchical model with verbal and nonverbal group factors, but rather to assess the specificity of the cognitive tests independent of a first principal component that represents what the tests share in common. It should also be noted that our index of g is dependent on the 12 tests that are available from GWA analyses. No battery of tests can provide a perfect index of g, although the seven UKB tests and the five GenLang tests represent a broader set of nonverbal and verbal tests than previously available. Moreover, as long as a diverse battery of tests are analyzed, Spearman (1927) hypothesized ‘indifference of the indicator’ when it comes to assessing g, an hypothesis validated in empirical analyses showing correlations of 0.99 between g factors derived from different test batteries (Johnson et al., 2004).

SNP heritabilities and genetic correlations

We used LD Score Regression (LDSC) in Genomic SEM to estimate SNP heritabilities and genetic correlations from the GWA summary statistics (Bulik-Sullivan et al., 2015; Grotzinger et al., 2019) for the 12 cognitive tests uncorrected and corrected for g. We focused on the comparison between genetic correlations among the g-corrected and uncorrected cognitive tests. Given the broad and diverse association of cognitive traits with important life outcomes, we also used LDSC in Genomic SEM to investigate genetic correlations between the GWA summary statistics for the g-corrected and uncorrected cognitive tests and 64 external GWA summary statistics across seven domains: Behavioral; BMI, height and weight; Cognitive; Mental Health; Personality; Physical Health; and Socio-economic status (SES) traits (see Supplementary Table S1 for a full list).

GWA significant hits and functional mechanisms

We used the web-based platform Functional Mapping and Annotation of Genome-wide Association Studies (FUMA) to further explore GWA summary statistics for g-corrected cognitive tests (Watanabe et al., 2017). This included identifying GWA-significant SNPs and analyses of gene sets and gene properties using Multi-marker Analysis of GenoMic Annotation (MAGMA) within FUMA. Gene set analysis was used to identify biological pathways and processes which are associated with the g-corrected cognitive test GWA summary statistics, whereas, gene-property analysis explored potential tissue-specific effects of SNPs by analyzing specific tissue types (e.g. the hippocampus). For further information on the methods and results of these analyses, see the Supplementary Note and Supplementary Tables S6 - S26.

GWA analysis of SCA and SCA.g using Genomic SEM: evidence for both commonality and specificity

To investigate the unique genomics of five verbal SCAs (GenLang) and seven non-verbal SCAs (UKB), we used Genomic SEM to fit a single common factor model to the 12 SCAs. The full model parameter estimates are in Figure 1 and fit indices are in Supplementary Table S2. All 12 SCAs significantly loaded on genomic g – the loadings range from 0.27 for Reaction Time (Reaction) (UKB) to 0.92 for Fluid Intelligence (Fluid) (UKB) (M = 0.66; SE = 0.05). The residual variances of the 12 SCA after removing the variance explained by genomic g were also significant, ranging from 0.15 for Fluid to 0.93 for Reaction (mean = 0.53; mean SE = 0.18).

On average, genomic g accounted for just under half of the genetic variance for the 12 SCA (genomic g; mean = 46.8%), and specific genetic variance, that is, genetic variance not attributed to genomic g, on average, made up slightly more than half (genomic s; mean = 53.2%) of the genetic variance for the 12 SCA. The proportion of variance accounted for by genomic g and genomic s varies substantially across the 12 SCAs, nevertheless, the average distribution is comparable for the five verbal SCAs from GenLang (genomic g = 44.3%; genomic s = 51.5%), and seven non-verbal SCA from UKB (genomic g = 48.6%; genomic s = 51.5%).

SNP heritabilities for SCA.g are almost as high as uncorrected SCA

SNP heritabilities were nearly as high for SCA.g as for SCA (Figure 2 and Supplementary Table S5): the average SNP heritability was 0.16 (SE = 0.02) for SCA and 0.13 (SE = 0.02) for SCA.g. Comparing SNP heritabilities for SCA and SCA.g revealed some interesting differences across tests. For example, the SNP heritability of Fluid was 0.22 (SE = 0.01) for SCA and 0.06 (SE = 0.01) for SCA.g, suggesting that much of the SNP heritability of Fluid is due to g, which is not surprising because Fluid loads highest on genomic g (Figure 1). In contrast, high SNP heritabilities were also found for Phoneme (0.24, SE = 0.04) and Nonword (0.25, SE = 0.03), but these SNP heritabilities were not diminished for SCA.g (0.29, SE = 0.03 and 0.25, SE = 0.03, respectively). SCAs that showed significant decreases in SNP heritability before and after g-correction were Memory, Symbol, Trail, Fluid (all UKB) and Word (GenLang).

Genetic correlations among SCA before and after g-correction

To further investigate the specificity of the 12 SCAs, we used LDSC in Genomic SEM to estimate the genetic correlations among the 12 SCAs residualized and un-residualized for genomic g (Figure 3 and Supplementary Tables S3 and S4).

The un-residualized SCAs were positively correlated genetically with each other (mean genetic correlation = 0.45; mean SE = 0.09). On average, the highest correlations were found for SCAs correlated with Fluid (mean = 0.61), Phoneme (0.57), Matrix (0.55), and Trail (0.55); the lowest correlations were found for SCAs correlated with Reaction (0.19), Memory (0.29), and Repetition (0.34). The seven nonverbal SCAs from UKB correlated highly positively with one another (mean = 0.59), as did the five verbal SCAs from GenLang (mean = 0.84), but as expected, the average correlation between SCAs from the nonverbal UKB tests and the verbal GenLang tests was more modest (M = 0.32).

After residualizing the 12 SCAa for the genetic effects of g, this positive manifold disappeared. The average correlation among the SCA.g was -0.07 (mean SE = 0.09); the average decrease in genetic correlation from SCA to SCA.g was 0.53. After correcting for genomic g, over half of the genetic correlations changed from positive to negative, with some correlations between SCAs switching from highly positive to highly negative. For example, the greatest switch from positive to negative genetic correlations was from 0.75 (SE = 0.04) uncorrected for g to -0.62 (SE = 0.09) after g correction for Fluid and Trail. Fluid and Trail are both UKB tests, but this switch from positive to negative was also observed between tests across the two batteries from 0.46 (SE = 0.20) to -0.51 (SE: 0.15) between Matrix (UKB) and Repetition (GenLang) and from 0.42 to -0.69 between Tower (UKB) and Spelling (GenLang).

Although the correlations among the SCA.g are lower than those between the uncorrected SCAs, the extent to which the correlations decrease is not uniform. Figure 4 highlights the differences in profiles of genetic correlations before and after residualizing g by showing the genetic correlations between each of the 12 tests with the other 11 tests, comparing correlations uncorrected for genomic g (orange line) and corrected for genomic g (blue line). The figure shows the general decrease in genetic correlation after g-correction, as the blue lines are consistently lower than the orange lines. However, the differences in genetic correlations before and after correction across each cognitive ability are not uniform; the blue and orange lines are not parallel. For instance, Trail shows a relatively small difference (0.16) with Symbol before and after g-correction (SCA = 0.82, SE = 0.06; SCA.g = 0.66, SE = 0.09), but Trail and Fluid yield the greatest difference (-1.37), going from highly positive (0.75) to highly negative (-0.62).

Importantly, these differences are not merely a function of the extent to which the SCAs load on genomic g. For instance, Matrix and Trail load 0.84 on genomic g, but their genetic correlations with Word are substantially different before and after correcting for genomic g. The change in correlation between Matrix and Word before and after residualizing g is modest, whereas the change in correlation between Trail and Word is substantial.

We tested the effect of g-loadings more systematically by correlating, for each of the 66 pairs of 12 tests, their difference in their loadings on genomic g (Figure 1) with their difference in genetic correlations before and after residualizing g (Figure 3). The resulting correlation was -0.03, which supports our conclusion that the SCA profile differences are not due to differential loadings of the SCA on genomic g.

Genetic correlations with external traits

To further examine changes in the genetic landscape of SCA after g-correction, we used LDSC to generate correlations for each SCA and SCA.g versus 64 ‘external’ traits for which GWA summary statistics were available. These traits span seven domains: behavior, height and weight, cognition, mental health, personality, physical health, and SES. The genetic correlations for SCA and SCA.g are listed in Supplementary Table S27 and shown in Supplementary Figures S10 – S28.

Similar to the pattern observed in Figure 4 among SCAs before and after g-correction, genetic correlations with external traits show a similar pattern of attenuation after g-correction, although the overall correlations are much lower in magnitude than the genetic correlations among the SCA. As in Figure 4, the results provide evidence for both genomic g and genomic specificity. Genetic correlations between SCAs and external traits declined for SCA.g, indicating a contribution of genomic g. For example, ADHD yielded significant negative correlations with most SCAs, but after g-correction, the genetic correlations were near zero, suggesting that genetic correlations between ADHD and external traits are largely due to genomic g. Specificity was indicated by changes in genetic correlations from SCA to SCA.g that are not uniform, suggesting a different genetic landscape for SCA.g across external traits. For example, the genetic correlation between ADHD and Trail, a test of executive function, switched from negative (-0.21) to positive (+0.20) after g-correction. Another example is that the genetic correlation between the Memory SCA and educational attainment shifted from positive (0.10) to negative (-0.23) after g-correction. A more detailed summary of these results can be found in Supplementary Note p.12.

Much of our understanding of the genetics of specific cognitive abilities (SCAs) reflects general cognitive ability (g) rather than SCAs themselves. Our results indicate that the genetic landscape of SCAs is transformed after controlling for genomic g (SCA.g).

Specific cognitive abilities independent of g (SCA.g) are heritable

If genetic influences on SCAs were entirely due to g, there would be little value ~~no point~~ in trying to investigate the genetics of SCA.g. For this reason, it is important to note that the heritability of SCA does not merely reflect g. Despite removing almost half of the genetic variance due to g from each of the 12 SCAs, the average SNP heritability only decreased from 16% for uncorrected SCAs to 13% for SCAs independent of g. The substantial SNP heritability of SCA.g should encourage more large-scale GWA studies focusing on the genetic specificity of specific cognitive abilities controlling for genomic g.

The positive manifold of genetic correlations among SCAs disappears after controlling for genomic g (SCA.g)

The positive manifold among tests of cognitive abilities, the basis for Spearman’s g, is real, genetically and phenotypically. However, our most important finding is that this positive manifold of genetic correlations (average = +0.45) disappears when genomic g is controlled (average = -0.07).

Separate consideration of the nonverbal UKB tests and the verbal GenLang tests revealed more subtle findings. The average genetic correlation among the seven UKB tests was 0.59 for SCAs and 0.00 for SCA.g, indicating that the strongly positive manifold of genetic correlations among nonverbal tests disappears entirely after controlling for genomic g. In contrast, the average genetic correlation among the five verbal tests in GenLang is 0.84 for SCAs and 0.31 for SCA.g. In other words, both the nonverbal UKB tests and the verbal GenLang tests showed similarly dramatic declines in average genetic correlations greater than 0.50 (i.e., from 0.59 to 0.00 in UKB and from 0.84 to 0.31 in GenLang), but the verbal GenLang tests declined from a much higher level of genetic correlation. This finding makes sense phenomenologically in that the nonverbal UKB tests, such as matrix reading, episodic memory and reaction time, seem to differ to a greater extent than the verbal GenLang tests, such as reading words and nonwords, repeating nonwords, and spelling.

These results underscore the importance of assessing g broadly, including verbal and nonverbal cognitive measures. They validate the rationale of the present study, which was to integrate the language measures of GenLang and the mostly nonverbal measures of UK Biobank to investigate SCA.g.

Some of the genetic correlations among SCA.g are highly negative

Not only did the positive manifold among SCA disappear in the SCA.g matrix, some of the genetic correlations were highly negative in the SCA.g matrix, such as Memory and Word (-0.72), Fluid and Symbol (-0.72), and Tower and Spelling (-0.79). These large negative correlations suggest that many SNPs associated with these pairs of tests have effects in opposite directions when g is controlled. This finding suggests that GWA analyses of SCA.g could yield very different results than those of SCAs uncorrected for g, which muddle SNPs associated in one direction with g and SNPs associated in the opposite direction for SCA.g.

The decreases in genetic correlations from SCAs to SCA.g are not uniform

If the decreases in genetic correlations from SCAs to SCA.g were just due to removing genetic variance due to g, we might expect the decrease to be similar across pairs of tests. To the contrary, the average decrease of about 0.50 from SCA to SCA.g masks pairs of tests whose genetic correlations change dramatically and those whose genetic correlations are not affected much by removing the genomic covariance due to g. Examples of pairs of tests whose genetic correlations change substantially from positive to negative include Fluid and Trail (from +0.75 to -0.62), Fluid and Reaction (+0.75 to -0.62) and Matrix and Phoneme (+0.73 to -0.42). Pairs of tests that change little include Symbol and Reaction (+0.36 to +0.22) and Repetition and Phoneme (+0.58 to +0.47).

The genetic landscape is hillier for SCAs.g

Uncorrected for g, many peaks of positive genetic correlations emerge -- 29 of the genetic correlations are 0.50 or greater and six are greater than 0.90 -- and the lowest genetic correlations are near zero. Corrected for g, the genetic landscape of cognitive abilities is hillier, with peaks of positive genetic correlations between SCA.g as high as 0.66 (between Symbol and Trail) and deep valleys of negative correlations as low as -0.72 (between Fluid and Symbol and between Memory and Word). The SCA and SCA.g genetic correlations matrices among the 12 tests in Figure 2 were represented in Figure 3 to highlight the difference in the profiles of genetic correlations between each SCA and the other 11 SCAs uncorrected and corrected for genomic g. We showed that these profile differences before and after residualizing g are not merely due to g-loadings of the SCAs. Our analyses of external traits yielded similar patterns of results. These results suggest that there is much to be learned about the genetic architecture of SCAs once the mask of g is removed.

Limitations

Our conclusions are limited in three ways. The first set of limitations is the obvious one that participants in UK Biobank were all assessed using the same tests, whereas GenLang involved a meta-analysis of 22 studies that included a wide range of cognitive tests. In addition, participants in UK Biobank were adults aged 40-75, whereas GenLang participants were aged 5-26 but mostly in late childhood. Using different tests and samples of different ages in UK Biobank and GenLang could have contributed to the lower genetic correlations between the nonverbal UKB tests and the verbal GenLang tests. However, we showed that our most striking results of swings from positive genetic correlations for SCA to negative genetic correlations for SCA.g emerged not just between UKB and GenLang but also within each dataset. In addition, analyses of genetic correlations for IQ across childhood, adulthood and older adulthood have yielded genetic correlations of 0.86 (SE = 0.11) between childhood and adulthood and 0.67 (SE = 0.24) between adulthood and older adulthood (Savage et al., 2018). Nonetheless, what is needed is an even larger UKB-type study of the same individuals tested at the same ages on the same broad battery of tests of SCA, which might be possible, for example, with the 5 million participants targeted for Our Future Health (https://ourfuturehealth.org.uk/).

A second limitation is conceptual: our analyses of SCA.g are intrinsically circular in that they are based on the effect sizes of SNPs from GWA analyses of SCAs uncorrected for g. Genomic SEM makes it possible to use these summary SNP statistics to conduct GWA analyses of SCA.g independent of g by statistically correcting for the general effects of g. However, direct GWA analyses of SCA.g seem likely to yield more associations specific to SCA.g. The heritability of SCA.g should encourage these studies. However, the challenges to conducting large GWA studies of SCA.g seem daunting, especially the requirement of psychometrically strong measures of verbal and nonverbal abilities. It would be ideal to assess g using the same measures, and for this reason, a freely available 15-minute gamified measure of g and verbal and nonverbal abilities with excellent psychometric properties has been developed in the hope that it can be incorporated in new and existing GWA studies (Malanchini et al., 2021).

A third limitation is general to all GWA analyses: missing heritability. Although SNP heritabilities and twin heritabilities are higher for cognitive abilities than for other behavioral domains, SNP heritabilities are less than half the twin heritabilities and variance predicted by PGS is half the SNP heritabilities (Plomin & von Stumm, 2018). For now, the most likely strategy for increasing PGS predictive power is ever-larger GWA studies with better measures of SCA and g, but technological advances such as whole-genome sequencing and artificial intelligence offer hope for additional strategies.

Another strategy is to conduct more large-scale multi-ancestry GWA studies that not only increase statistical power, but also allow for genomic discoveries beyond European populations. This leads to the fourth limitation of this study, which is that the GWA summary statistics used are derived exclusively from white participants of European ancestry, which reduces the representativeness of the sample and limits the generalizability of our findings to other populations. This approach was chosen because most GWA studies of SCA rely on participants with European ancestry. To address this limitation, more genetic and genomic studies of under-represented ancestries are needed. Specifically, large-scale, multi-ancestry GWA studies are needed to identify genetic variants shared among different ancestries, as well as those unique to a particular ancestry.

Implications

The main implication of these findings about the changing genetic landscape of SCA after controlling genomic g is that if research on SCAs in education, neuroscience and genetics does not control for g, the same g variance will be investigated repeatedly in the guise of verbal, spatial, memory and other cognitive abilities. Removing the pervasive influence of genomic g is necessary to investigate the genetic specificity of specific cognitive abilities. To foster this new direction for research, the summary statistics from our genome-wide association analyses of 12 SCA.g are freely available to be used by researchers to create polygenic scores that focus on the specificity of specific cognitive abilities.

Author Contribution

Author Contributions: FP, RP, MM and KR conceived and designed the research. FP and EK analyzed the data. FP, RP, EK, JK, MM and KR wrote the paper. All authors approved the submitted draft of the paper.

Acknowledgement

We gratefully acknowledge UK Biobank and GenLang, and their participants. FP is supported by a National Institutes of Health [NIH] Subaward to King’s College London via The Regents of the University of California, Riverside [AG046938]. EK is supported by a PhD studentship from the UK Economic and Social Research Council. JK is supported by a doctoral studentship from the King's College London Medical Research Council Doctoral Training Partnership in Biomedical Sciences. MM is supported by a starting grant from the School of Biological and Behavioural Sciences, Queen Mary University of London.

Bulik-Sullivan, B. K., Loh, P.-R., Finucane, H. K., Ripke, S., Yang, J., Patterson, N., Daly, M. J., Price, A. L., & Neale, B. M. (2015). LD Score regression distinguishes confounding from polygenicity in genome-wide association studies. Nature Genetics, 47(3), Article 3. https://doi.org/10.1038/ng.3211
Carroll, J. B. (1993). Human Cognitive Abilities: A Survey of Factor-Analytic Studies. Cambridge University Press.
de la Fuente, J., Davies, G., Grotzinger, A. D., Tucker-Drob, E. M., & Deary, I. J. (2021a). A general dimension of genetic sharing across diverse cognitive traits inferred from molecular data. Nature Human Behaviour, 5(1), 49–58. https://doi.org/10.1038/s41562-020-00936-2
de la Fuente, J., Davies, G., Grotzinger, A. D., Tucker-Drob, E. M., & Deary, I. J. (2021b). A general dimension of genetic sharing across diverse cognitive traits inferred from molecular data. Nature Human Behaviour, 5(1), 49–58. https://doi.org/10.1038/s41562-020-00936-2
Donati, G., Dumontheil, I., Pain, O., Asbury, K., & Meaburn, E. L. (2021). Evidence for specificity of polygenic contributions to attainment in English, maths and science during adolescence. Scientific Reports, 11(1), 3851. https://doi.org/10.1038/s41598-021-82877-y
Eising, E., Mirza-Schreiber, N., de Zeeuw, E. L., Wang, C. A., Truong, D. T., Allegrini, A. G., Shapland, C. Y., Zhu, G., Wigg, K. G., Gerritse, M. L., Molz, B., Alagöz, G., Gialluisi, A., Abbondanza, F., Rimfeld, K., van Donkelaar, M., Liao, Z., Jansen, P. R., Andlauer, T. F. M., … Fisher, S. E. (2022). Genome-wide analyses of individual differences in quantitatively assessed reading- and language-related skills in up to 34,000 people. Proceedings of the National Academy of Sciences, 119(35), e2202764119. https://doi.org/10.1073/pnas.2202764119
Grotzinger, A. D., Rhemtulla, M., de Vlaming, R., Ritchie, S. J., Mallard, T. T., Hill, W. D., Ip, H. F., Marioni, R. E., McIntosh, A. M., Deary, I. J., Koellinger, P. D., Harden, K. P., Nivard, M. G., & Tucker-Drob, E. M. (2019). Genomic structural equation modelling provides insights into the multivariate genetic architecture of complex traits. Nature Human Behaviour, 3(5), Article 5. https://doi.org/10.1038/s41562-019-0566-x
Hindley, G., Shadrin, A. A., van der Meer, D., Parker, N., Cheng, W., O’Connell, K. S., Bahrami, S., Lin, A., Karadag, N., Holen, B., Bjella, T., Deary, I. J., Davies, G., Hill, W. D., Bressler, J., Seshadri, S., Fan, C. C., Ueland, T., Djurovic, S., … Andreassen, O. A. (2023). Multivariate genetic analysis of personality and cognitive traits reveals abundant pleiotropy. Nature Human Behaviour, 7(9), Article 9. https://doi.org/10.1038/s41562-023-01630-9
Knyspel, J., & Plomin, R. (2024). Comparing factor and network models of cognitive abilities using twin data. Intelligence, 104, 101833. https://doi.org/10.1016/j.intell.2024.101833
Malanchini, M., Rimfeld, K., Gidziela, A., Cheesman, R., Allegrini, A. G., Shakeshaft, N., Schofield, K., Packer, A., Ogden, R., McMillan, A., Ritchie, S. J., Dale, P. S., Eley, T. C., von Stumm, S., & Plomin, R. (2021). Pathfinder: A gamified measure to integrate general cognitive ability into the biological, medical, and behavioural sciences. Molecular Psychiatry, 26(12), 7823–7837. https://doi.org/10.1038/s41380-021-01300-0
Plomin, R., & Deary, I. J. (2015). Genetics and intelligence differences: Five special findings. Molecular Psychiatry, 20(1), 98–108. https://doi.org/10.1038/mp.2014.105
Plomin, R., & von Stumm, S. (2018). The new genetics of intelligence. Nature Reviews Genetics, 19(3), 148–159. https://doi.org/10.1038/nrg.2017.104
Procopio, F., Liao, W., Rimfeld, K., Malanchini, M., Von Stumm, S., Allegrini, A. G., & Plomin, R. (2024). Multi-polygenic score prediction of mathematics, reading, and language abilities independent of general cognitive ability. Molecular Psychiatry. https://doi.org/10.1038/s41380-024-02671-w
Procopio, F., Zhou, Q., Wang, Z., Gidziela, A., Rimfeld, K., Malanchini, M., & Plomin, R. (2022). The genetics of specific cognitive abilities. Intelligence, 95, 101689. https://doi.org/10.1016/j.intell.2022.101689
Rimfeld, K., Kovas, Y., Dale, P. S., & Plomin, R. (2015). Pleiotropy across academic subjects at the end of compulsory education. Scientific Reports, 5(1), Article 1. https://doi.org/10.1038/srep11713
Rimfeld, K., Shakeshaft, N. G., Malanchini, M., Rodic, M., Selzam, S., Schofield, K., Dale, P. S., Kovas, Y., & Plomin, R. (2017). Phenotypic and genetic evidence for a unifactorial structure of spatial abilities. Proceedings of the National Academy of Sciences, 114(10), 2777–2782. https://doi.org/10.1073/pnas.1607883114
Savage, J. E., Jansen, P. R., Stringer, S., Watanabe, K., Bryois, J., de Leeuw, C. A., Nagel, M., Awasthi, S., Barr, P. B., Coleman, J. R. I., Grasby, K. L., Hammerschlag, A. R., Kaminski, J. A., Karlsson, R., Krapohl, E., Lam, M., Nygaard, M., Reynolds, C. A., Trampush, J. W., … Posthuma, D. (2018). Genome-wide association meta-analysis in 269,867 individuals identifies new genetic and functional links to intelligence. Nature Genetics, 50(7), 912–919. https://doi.org/10.1038/s41588-018-0152-6
Spearman, C. (1904). ‘General intelligence,’ objectively determined and measured. The American Journal of Psychology, 15, 201–293. https://doi.org/10.2307/1412107
Sudlow, C., Gallacher, J., Allen, N., Beral, V., Burton, P., Danesh, J., Downey, P., Elliott, P., Green, J., Landray, M., Liu, B., Matthews, P., Ong, G., Pell, J., Silman, A., Young, A., Sprosen, T., Peakman, T., & Collins, R. (2015). UK Biobank: An Open Access Resource for Identifying the Causes of a Wide Range of Complex Diseases of Middle and Old Age. PLOS Medicine, 12(3), e1001779. https://doi.org/10.1371/journal.pmed.1001779
The abilities of man: Their nature and measurement / by C. Spearman. (n.d.). Wellcome Collection. Retrieved 26 September 2023, from https://wellcomecollection.org/works/evdevd8m
Tosto, M. G., Malykh, S., Voronin, I., Plomin, R., & Kovas, Y. (2013). The Etiology of Individual Differences in Maths beyond IQ: Insights from 12-year Old Twins. Procedia - Social and Behavioral Sciences, 86, 429–434. https://doi.org/10.1016/j.sbspro.2013.08.592
Trzaskowski, M., Davis, O. S. P., DeFries, J. C., Yang, J., Visscher, P. M., & Plomin, R. (2013). DNA Evidence for Strong Genome-Wide Pleiotropy of Cognitive and Learning Abilities. Behavior Genetics, 43(4), 267–273. https://doi.org/10.1007/s10519-013-9594-x
Watanabe, K., Taskesen, E., van Bochoven, A., & Posthuma, D. (2017). Functional mapping and annotation of genetic associations with FUMA. Nature Communications, 8(1), Article 1. https://doi.org/10.1038/s41467-017-01261-5

No competing interests reported.

Download PDF

Journal Publication

published 18 Feb, 2025

Read the published version in Behavior Genetics →

Editorial decision: Revision requested
24 Oct, 2024
Reviews received at journal
23 Oct, 2024
Reviewers agreed at journal
26 Sep, 2024
Reviews received at journal
24 Sep, 2024
Reviewers agreed at journal
24 Sep, 2024
Reviewers invited by journal
20 Sep, 2024
Editor assigned by journal
11 Sep, 2024
Submission checks completed at journal
09 Sep, 2024
First submitted to journal
08 Sep, 2024

You are reading this latest preprint version

The genetic specificity of specific cognitive abilities after controlling for general cognitive ability

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Methods

Samples and Measures

GWA analysis of g-corrected cognitive tests using Genomic Structural Equation Modelling

SNP heritabilities and genetic correlations

GWA significant hits and functional mechanisms

Results

Discussion

Declarations

Author Contribution

Acknowledgement

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1