How do genes affect intelligence?

Genetics of general cognitive ability


Intelligence is one of the best-studied constructs in the empirical behavioral sciences and represents a general mental capacity that includes the ability to reasonably think, to solve novel problems, to think abstractly and to learn quickly. These cognitive skills play a major role in explaining and predicting individual differences in central areas of social life, such as school and educational success, professional success, socio-economic status and health behavior. Behavioral genetic studies consistently show that genetic influences make a substantial contribution to explaining individual differences, which explain over 60% of intelligence differences in adulthood. In recent years, in large genome-wide association studies with frequent genetic variants, hundreds of loci associated with intelligence have been identified, as well as over 1300 associated genes with differential expression, predominantly in the brain. Several signaling pathways were enriched, especially for neurogenesis, regulation of the development of the nervous system and synaptic structure and activity. The majority of the associated loci were regulatory regions and interestingly half were intronic. Of the more than 1,300 genes, only 9.2% overlap with those associated with monogenic neurocognitive disorders. Overall, the findings confirm a polygenic model of thousands of additive factors, with the individual loci showing a very low effect size. Overall, the current findings explain about 10% of the total variance of the characteristic. These results are an important starting point for future research in both genetics and behavioral sciences.


Intelligence is one of the best studied constructs of empirical behavioral sciences and represents a general cognitive capacity, which includes - among others - the ability for conceptual thinking, solving challenging problems, abstract thinking, and rapid learning. These cognitive functions play an enormous role in the explanation and prediction of individual differences in central areas of societal life, e. g., schooling and educational success, professional success, socioeconomic status, and health-related behavior. Behavioral genetic studies have consistently shown that genetic influences make a substantial contribution to defining individual differences, that explain more than 60% of variations in intelligence in adults. Over the last few years, in large genome-wide association studies using frequent genetic variants, hundreds of loci associated with intelligence were identified, in addition to more than 1300 associated genes, which were differentially expressed in the brain. Several pathways were overrepresented, mainly those for neurogenesis, the regulation of nervous system development, and the regulation of synapse structure and activity. Most associated loci were located in regulatory regions and, interestingly, half of them in introns. Of the more than 1300 associated genes, only 9.2% overlapped with those associated with monogenic cognitive defects. Overall, the findings confirm a polygenic model of thousands of additive factors, in which individual loci have a very small effect. Collectively, the current results explain up to 10% of the overall variance in cognitive function. These results are an important starting point for future research, not only in genetics but also in the behavioral sciences.

The high heritability of cognitive abilities, also known as intelligence, and the great influence of this trait on many aspects of life explain a broad scientific and public interest. For a long time, studies of the genetic architecture of intelligence were limited to the establishment of general models without any individual causal factors being known. Significant progress has recently been made thanks to extremely large genome-wide association studies. This overview aims to discuss these new findings on the genetics of intelligence in the context of the findings of behavioral science and to shed light on the connection with cognitive disorders.

What is intelligence

Intelligence is one of the best-studied constructs in the empirical behavioral sciences [46]. In the tradition of psychometric intelligence research, there is also broad consensus on what the core components of intelligence are [44] and with what kind of test procedure these can be measured. As a result, intelligence is a very general mental capacity that includes - among other things - the ability to reasonably think, to solve novel problems, to think abstractly and to learn quickly [15]. In relation to the structure of intelligence, a hierarchical model concept dominates today with a global general factor (general cognitive ability, "g") at the top and specifically arranged factors of different widths below (Fig. 1; cf. [29]; for a compact For an overview of the most relevant structural approaches see [21]).

In addition, intelligence plays a major role in explaining and predicting individual differences in key areas of social life and advancement, including school and educational success [39], professional success [41], socio-economic status [47] and health behavior [5]. The effect sizes associated with intelligence are sometimes considerable. For example, Roth et al. in their metanalysis based on a total sample of 105,185 individuals from 240 independent studies, found a relationship between intelligence and school grades of p = 0.54 (corrected population correlation), which makes intelligence by far the strongest predictor of school success [39]. A comprehensive overview of other findings on intelligence as a success factor can be found in Deary (2012; [10]).

How is intelligence measured?

There are numerous psychometric tests for measuring intelligence, which can differ from one another in terms of task format and recorded partial performance. For individual diagnostics, test batteries are used (such as the intelligence structure test, I ‑ S-T 2000 R; [1]), which contain tasks to record various primary factors of intelligence, including, for example, inductive thinking, verbal understanding, figural relationships and dealing with numbers. Empirically it is shown with high consistency that the performances in these different sub-areas show positive intercorrelations, which is called positive manifold and represents the prerequisite for the aggregation of such sub-performances to a total value [7]. Relevant intelligence test methods enable such a total value to be determined for each person tested. Using standard values, these can then be converted into directly interpretable standard values, for example the intelligence quotient (IQ; mean value M = 100, standard deviation s = 15). Inter-individual differences in intelligence values ​​show a high degree of stability over time, which from young adulthood takes on values ​​of r = 0.60 and above [16].

Heredity of intelligence

Behavioral genetic studies have consistently shown in recent decades that genetic influences make a substantial contribution to explaining individual differences in intelligence [35], which is in line with early reviews based on twin data and other relatives [2]. More recent overview articles [3] also show an increase in the importance of genetic influences over the life span. While in early childhood effects of the shared environment (c2; environmental influences that contribute to the similarity of individuals growing up together) are responsible for explaining the differences in intelligence values, these no longer play a significant role from early adulthood. The decrease in the importance of shared environmental influences contrasts with the increase in the importance of genetic influences, which explain a little more than 20% of the differences in intelligence in early childhood, about 40–50% at the beginning of school, and up to 60% and more in adulthood. As a possible cause for this growing importance of genetic influences over the life span, it is assumed that people whose genotype has a beneficial effect in the context of learning and performance behavior and contributes to a sense of achievement also tend to focus more on environments that promote learning and performance (cf. . [45]). This is viewed as an active variant of the genotype-environment correlation (rGU; systematic relationship between genotypes and certain environmental conditions). In addition to the active form of the rGU, Plomin, DeFries, & Loehlin (1977) also describe the passive and evocative variants [33]. Passive rGU exists, for example, when parents provide environments that meet the genotype of the child / are associated with it, while an evocative rGU reflects the fact that the environment reacts to genetically (partly) conditioned behavior.

In contrast, the genotype-environment interaction (GxU) describes the genetically determined “susceptibility” to environmental influences. For example, what became known as the Scarr-Rowe hypothesis was the assumption that the heritable nature of intelligence is higher at the more privileged end of the socio-economic distribution, while shared environmental effects are more significant at the lower end of the socio-economic distribution. This is in line with the idea that genetic potential can develop to a greater extent under more favorable environmental conditions [51]. Evidence for this assumption currently comes primarily from US samples [49, 50].

First genome-wide association studies

Around 2005, with the establishment of genome-wide association analyzes (“genome-wide association study”, GWAS), an unprecedented wave of elucidation of genetic factors in genetically complex diseases and traits began, which continues to this day. By the end of August 2018, the international catalog of all published GWAS, which is kept at the European Bioinformatics Institute (EBI), contained a total of 3541 publications and 69,969 independent SNP characteristic associations [26]. It was hoped that this technology would also provide a tool for elucidating the genetic basis of intelligence. A quantitative trait with high heritability appeared to be an ideal field of application.

The disappointment was all the greater when, even by GWAS standards, large and thus apparently well-powered studies remained without resounding success. A large meta-analysis by the Cognitive Genomics Consortium (COGENT) on 35,298 healthy subjects of European descent from a total of 24 individual studies may serve as an example [48]. This generated only two genome-wide significantly associated SNP loci. Even the combination of different individual SNPs in a gene-based analysis resulted in only 7 gene loci that were still significant after Bonferroni correction for multiple testing. Another important parameter was also not met: the successful replication of association findings. The respective association findings from a total of 7 GWAS as well as those from numerous candidate gene studies were largely not replicated, which ultimately could not convincingly refute the possibility of a false-positive finding, despite the Bonferroni correction. In addition, these findings combined only explained about 1% of the total variance in the phenotype. Accordingly, the effect size of the individual loci was extremely small, which made it clear that intelligence is extremely heterogeneous and is influenced by thousands of genetic factors, each with a very low effect size. Therefore, extremely large samples would be required to detect them.

“Educational attainment” as a proxy phenotype

An interesting and important innovation came from studies that, instead of examining the complex cognitive function that had to be measured, only examined the completed years of school / academic education as a proxy. This parameter is known as educational attainment (EA). Various studies have shown that it correlates strongly with cognitive abilities (see above) [6]. Since this parameter can be ascertained relatively easily from anamnestic information, it was possible to identify three genome-wide significant SNPs by merging numerous individual studies in a meta-analysis on 125,000 individuals [37]. For the first time, these could be replicated in independent cohorts [36]. The effect sizes were also very small here; the strongest effect of an SNP (rs9320913) corresponded to only 0.02% of the total variance in the replication cohort.

In view of these minimal effect sizes of individual SNPs, the analysis was shifted to the totality of the associated SNPs, regardless of whether they reached the strict static thresholds of significance that are otherwise used in GWAS. This total value, known as the “polygenic score” (PGS), “polygenic risk score”, “genetic risk score” or “genome-wide polygenic score” represents the weighted sum of all (associated) SNPs and is intended to serve as the best predictor for the characteristic . (For an up-to-date overview of risk stratification using PGS in various complex diseases, see [20]) This PGS could explain 2% of the total variance in a replication study, ie 100 times more variance than the best individual SNP [37].

Encouraged by this success, a second meta-analysis on EA was carried out on 294,000 individuals in 2016 [31]. At this sample size, 74 loci were genome-wide significant, and the total variance for EA that PGS could explain reached 3%. Interestingly, the PGS even explained 4% of the variance for intelligence, a higher value than the actual target phenotype of the study.


However, the breakthrough came with large meta-analyzes for the intelligence phenotype. In the individual studies, the phenotype was measured using different, but ultimately comparable, methods so that they could be combined. The work by Sniekers et al. appeared in 2017 and found 336 associated SNPs and 52 associated genes among 78,308 individuals [43]. An even larger study by Savage et al. from June 2018 [40] from the same working group of Danielle Posthuma from Amsterdam practically quadrupled the sample size to almost 280,000 individuals from 16 independent cohorts including the large studies of the UK Biobank and the above. COGENT study, but also partially that already in Sniekers et al. used cohorts [43]. A total of 531 independent SNPs were genome-wide significant, which in turn could be assigned to 246 independent loci. As with other complex traits, the number of significantly associated genes did not increase linearly, but actually increased twentyfold to 1041. This was possible by merging all associated SNPs of individual genes, even if the respective SNPs did not exceed the significance threshold for the genome-wide association. The findings were replicated in part of the UK Biobank (n = 188,000) whose individuals had not undergone formal intelligence testing, but for whom the highly correlated phenotype EA was available. Overall, 94% of the SNPs showed the same direction of association, and 51 loci showed evidence of replication. Overall, the now identified SNPs explained 5.4% of the total variance of the attribute intelligence [40].

Davies et al. (2018) chose a similar approach, who even analyzed 300,486 individuals from 57 population-based studies [8]. However, this study uses the same UK biobank samples as Posthuma's group, so that there is considerable overlap between the cohorts. However, the studies differ in technical terms, particularly with regard to the quality assurance of the genotypes from the various original studies. Perhaps this explains that Davies et al. found only 148 independent loci and 709 genes associated with intelligence. Even if the results of both studies overlap considerably (Fig. 2), it is noteworthy that the addition of other divergent cohorts makes different loci and genes statistically significant. This suggests that the trait is extremely heterogeneous and that the results for numerous loci are likely to be just below the statistical significance threshold.

The largest GWAS so far was finally published again on the (proxy) phenotype EA in July 2018. A meta-analysis of 71 cohort studies increases the sample size compared to previous studies to a total of 1.1 million people. In this study, also known as EDU3, Lee and colleagues (2018; [25]) found a total of 1271 independent SNPs for EA that were genome-wide significant. A total of over 1,800 genes were associated, about 10 times as many as in the previous study for EA. The larger sample, but also innovations in data analysis, led to the PGS generated in replication cohorts explaining up to 9.7% of the variance for intelligence, the highest value achieved so far. The variance for EA explained by the PGS is even 12%.

Overall, the results of the various studies converge, although differences in the size and composition of the individual cohorts as well as methodological aspects play a role.

Genetic and Molecular Architecture

With these studies, the trait intelligence has finally arrived in the field of genetically complex traits. Increasing the sample size and improving the methods made the breakthrough possible and revealed over a thousand significant association findings. Despite the extremely low effect sizes of individual SNPS, some of them have already been successfully replicated. A highly polygenic characteristic is emerging, with thousands of individual additive factors, each with an extremely low effect size. The architecture of the trait is similarly heterogeneous as the trait body height, for which 24.6% of the phenotypic variance could be explained in similarly sized GWAS with 3290 independent SNPs [52].

With the results of the most recent studies from 2018 by Davies et al. and Savage et al. [8, 40] it is possible for the first time to gain insights not only into the genetic, but also into the molecular architecture of cognitive performance. The authors of both studies examined the functional categories of the associated SNPs and the expression of the associated genes as well as their signaling pathways. Noteworthy are some differences in genetic architecture compared to other complex traits. 51% of the SNPs for intelligence were intronic, 27% intergenic and 1.5% exonic [40]. Savage et al. also found 89 non-synonymous exonic variants. In contrast, in a joint analysis of various chronic inflammatory diseases, for example, 86% of the associated SNPs were within 10 kb of a neighboring gene, but intergenic, and 6.5% were exonic [12].

As expected, the SNP loci associated with intelligence most likely contain regulatory elements, because they lie in evolutionarily conserved regions and overlap with markers of open chromatin, known regulatory elements and eQTL loci. The expression pattern of the associated genes is also not unexpected. The genes were expressed differently in the brain and the thyroid gland in a statistically significant manner. Various brain areas already associated with cognitive abilities also showed a significant increase in expression, including various cortical regions, amygdala, hippocampus, caudate nucleus and putamen [8]. The cells with the highest expression of associated genes were medium-sized projection neurons of the striatum and cortical and hippocampal pyramidal neurons [40].

An analysis of the signaling pathways and biological functions of the associated genes showed a significant enrichment for neurogenesis, the process of the formation of mature neurons from stem cells. Other associated processes were the regulation of the structure and processes of the synapse as well as the regulation of the development of the nervous system and the regulation of cell development in general [40].

The functional annotation of the genes associated with EA in EDU3 confirmed earlier findings and extended them by two points [25]. The expression of the genes no longer extended preferentially to the prenatal phase, but also to postnatal development. In addition, gene networks in particular that are involved in communication between neurons were identified. Also noteworthy was the practical lack of gene sets that are differentially expressed in glial cells, although these cells make up about half of the central nervous system. This suggests that the speed of neurotransmission does not play a major role in the differences in cognition [25].

Comparison of the genes for cognitive abilities and for cognitive disorders

An obvious question is whether the same genes that, when mutated, cause neurocognitive diseases (the focus of this issue) are also associated with intelligence. For neurocognitive disorders, after more than three decades of intensive research, 1915 involved genes have been identified (SysID database, as of June 2018; [22]), so that a comparison now appears possible and useful. The two largest intelligence studies of 2018, by Savage et al. and Davies et al. [8, 40] found a total of 1321 associated genes, with about a third (472) overlapping (Fig. 2). The Venn diagram shows that 177 genes (9.2%) of the genes associated with neurocognitive disorders are also associated with intelligence (Fig. 2; Tab. 1). In terms of genetics, we currently seem to know more about the ways in which neurocognitive functions can be disrupted than about the processes involved in the modulation of skills.

The overlapping genes are significantly enriched for the same gene ontology (GO) terms as those for intelligence in general. More than half (100; 57%) were associated with an autosomal recessive inheritance, 75 with a dominant and 2 with both inheritance patterns. In the dominant groups, 51 (68%) showed signs of haploinsufficiency (pLi score> 0.9). It is possible that some recessive genes have a quantitative effect in the heterozygous state. It is also obvious that haploinsufficiency genes can also be quantitatively and thus modulated in their function via regulatory effects, since the dose is particularly critical here.

Relation to psychiatric illnesses and other characteristics

An association of intelligence with neuropsychiatric diseases such as schizophrenia and manic-depressive illness was already known. Overall, the genetic factors of many of these diseases show a strong overlap [13]. In addition, intelligence has previously been associated with various anthropometric and medical characteristics. These associations have now also been confirmed at the SNP level by comparison with published GWAS [8] or investigated using "Mendelian randomization". Mendel's randomization [40] describes a new type of biostatistical method for determining the influence of robustly associated gene variants on characteristics. It aims to provide undistorted evidence of causal effects and, if possible, to estimate the size of the effect. There was great agreement between the studies, with the strongest positive association, as expected, with EA. Also for former smoking, autism and intracranial volume, as well as longevity and wearing glasses or contact lenses. A negative correlation, i.e. a protective effect of the overlapping genetic predisposition to a high IQ, was found for Alzheimer's disease, depressive symptoms, schizophrenia, ADHD, high blood pressure, lung cancer as well as weight and body mass index [40]. This suggests pleiotropic factors that overlap for different traits and disorders.

The analysis of the genetics of the subphenotypes that contribute to the general cognitive ability “g” is also interesting (Fig. 1). The largest study to date analyzed a phenotypically and genotypically uniformly characterized cohort of 112,151 test subjects from the UK Biobank [9]. In particular, the subphenotypes “linguistic reasoning and understanding of numbers” (31%) and “EA” (21%) showed a high SNP-based heritability, while “reaction time” (11%) and “memory” (8%) had a significantly lower SNP -based heritability showed. However, as expected, the subphenotypes were strongly genetically correlated. B. “Linguistic reasoning and understanding of numbers” and “EA” with a correlation of g = 0.73. This study also shows that in addition to non-genetic factors, other genetic factors that are not associated with cognitive abilities but, for example, with personality traits, contribute significantly to the variability of the EA trait.

Will genetic tests be used to predict intelligence?

Currently, around 11–13% of the total variance of the EA and 7–10% of the intelligence can be predicted with the help of PGS [25]. It is foreseeable that this value will continue to rise in the next few years thanks to ever larger GWAS and improvements in statistical procedures. The theoretical upper limit of the variance that can be explained by frequent SNPs was estimated at around 25%. However, if even rare alleles are taken into account, this upper limit could even increase to 50% [19]. It can already be seen now that attempts will be made to use this PGS also prognostically, i.e. predictively.

However, caution should be exercised here. A PGS only allows a probabilistic statement. For example, the value in an examined group correlates significantly with the measured intelligence, but in individual cases a measured value can deviate significantly from the predicted value, which is why PGS are unsuitable as a prognostic marker for individuals. As an example, the study by Selzam et al. [42] which use the PGS of Okbay et al. Study [31] on the correlated (proxy) phenotype of the "educational years" on a cohort of 5800 students aged 16 years from the United Kingdom and examined their results in a national school examination (General Certificate of Secondary Education, GCSE). The exam result is normally distributed and the value is referred to as the “educational achievement score” (EAS) (Fig. 3a). It was known from twin studies that the heredity of this score is 60% [18]. As expected, the PGS was also normally distributed (Fig. 3a).

If the cohort is divided into deciles according to their PGS, then these differ accordingly in their EAS. According to the overall correlation, the mean values ​​of the bottom PGS decile were at the 28th percentile of the EAS and those of the top PGS decile were at the 68th percentile of the EAS (Fig. 3c). Within a PGS decile, however, there was a wide variance of the EAS, the overlap of the distribution of the EAS of both deciles was 61% (Fig. 3d). Accordingly, the individual values ​​of the EAS were far apart within one PGS decile: the individual with the second highest PGS only had an EAS just above the mean value of the total cohort, an individual with the eighth lowest PGS value had an EAS above the 75th percentile (Fig 3b). While the correlation is valid for the total cohort, the prediction accuracy for the individual is very low.

However, at the edges of the distribution, so z. B. in the lowest or highest percentiles, better predictions are possible, provided that the modifying effects of a (low) socio-economic status are excluded. In this context, a recently published work by the DDD study is relevant [30]. The authors show that PGS contributes to the risk of cognitive disorders and postulate that some of the NDD patients, especially those with a lighter degree of severity, do not have a monogenic cause but only represent the lower range of the polygenic normal distribution. In this study, PGS was associated with lower intelligence and educational success and with an increased risk of schizophrenia. In the long term, a PGS-based prediction value can be expected that corresponds, for example, to that of coronary heart disease, with the top 8% of the population showing an increase in risk (OR ≥ 3) comparable to that of monogenic familial hypercholesterolemia [20].

Genetics of the environment

An important aspect in the discussion about genetics and the environment ("nature and nurture") is the observation of Kong et al. [24] found that parental genotypes have an effect on children's EA even if they are not inherited. In a follow-up analysis of an Icelandic sub-cohort of the EA study by Okbay et al. [31] they were able to show that the PGS of the non-inherited (not transmitted) alleles of the parents explained about 30% of the variance calculated for the PGS of the inherited (transmitted) alleles. The non-transmitted alleles do not have a direct effect on the children, but on the parents. These then create a corresponding environment (“nurture”), which acts on the children in the sense of a passive genotype-environment correlation. Genetic and environmental mechanisms on the trait are therefore closely interwoven. At 30%, this value is significantly greater for EA than for body length, for which the analogous value is only 6%. This aspect, which has been apostrophized as the “genetic nurture effect” [23], has not yet been taken into account in GWAS. Predictive proportions of PGS are overestimated. The future investigation of non-transmitted alleles, however, promises new ways in the analysis of the factors and signaling pathways involved and underlines the importance of genetic strategies (trio vs. individual case) in researching cognitive traits.

Future research

The further development of research can partly be predicted using examples from the field of complex diseases. The cohorts examined will continue to grow, as was recently the case with Lee et al. [25] (1.1 million individuals), especially for easy-to-collect proxy phenotypes such as EA. Large population-based cohorts such as the UK Biobank, which have already made a significant contribution, will also continue to grow. As is already clear in the current studies, the power will grow disproportionately and the number of association findings and their statistical certainty will increase accordingly.

So far, however, mostly only frequent SNPs with a frequency of the rare allele> 1% have been investigated using the relatively inexpensive arrays. With the transition to the more expensive but more complete sequence-based genotyping techniques, rarer variants will also come into focus. Here, too, the comparison with the polygenic characteristic of body height is instructive. In a large-scale study on rare alleles, the authors found 83 height-associated coding variants with an allele frequency of 0.1–4.8% and effect sizes of up to 2 cm per allele, about 10 times as much as the average effect more frequently Alleles [27].

Previous GWAS are based on a simple additive genetic model without considering possible interactive effects (epistasis). In principle, however, multiplicative or completely different combinatorial effects between individual loci are also to be expected. However, due to the exponentially growing number of tests for which statistical corrections would have to be made, the latter can only be demonstrated in practice if pathophysiological considerations justify a reduced number of associations. Accordingly, few examples of such effects have been published (e.g. for psoriasis [14]). Virtually no study has the power to take a genome-wide approach.

Also, no copy number variants (CNVs) have yet been systematically investigated. However, from research on cognitive disorders and neuropsychiatric diseases, the enormous importance of CNVs and other structural variants is well known. Microdeletion syndromes in particular could have a significant effect on cognitive abilities that could be greater than current PGS. They therefore represent rare alleles with a large effect size. Rare CNVs with a large effect size also play a significant role in body height [53]. With the previous array-based techniques, CNVs can hardly be reliably determined in large cohorts. However, this should be much better possible with sequence-based techniques such as “whole genome sequencing”.

Future studies must also consider the contribution of the X chromosome. Current GWAS study designs largely omit this chromosome. The EDU3 study from 2018 [25] analyzed this chromosome separately on part of the data set for the first time. Only 10 associated loci were found and the contribution to the variance was estimated to be 0.3%, less than an autosome of comparable size. However, in view of the methodological questions that are still open and in view of the relatively small subgroup, many questions remain unanswered. In view of the importance of X-linked genes for neurocognitive diseases, it is to be expected that genes associated with intelligence will also be found here.

Previous studies have largely focused on Europeans. For this reason, studies are also necessary in other populations, even if no other genes or signaling pathways involved than in Europeans are to be expected, but sometimes other alleles. It is therefore not surprising that the predictive value of the PGS determined in the EDU3 study had a significantly lower predictive value in a subgroup of African-American people than was the case for people of European descent [25]. These studies will definitely increase the overall power and thus also contribute to the replication and refinement of the association findings.

It is possible that some of the genes now named represent false-positive results, as the gene-based GWAS methodology does not take into account associations outside of the genes. It cannot therefore be ruled out that the associated SNPs in one gene actually regulate another (neighboring) gene. Ultimately, functional studies must help to elucidate the molecular mechanisms. However, this will be a particular challenge because of the low effect strength of the individual loci. For this, however, monogenic cognitive disorders could be very helpful. As shown above, many overlapping genes follow a recessive inheritance. It is conceivable that heterozygous carriers have an IQ reduction, which has a significantly greater effect size than associated SNPs, but this is lost in the polygenic background. Here, too, a comparison with the height characteristic helps. Autosomal recessive mutations in the ACAN-Gen are associated with a severe form of short stature.As was only recently found, heterozygous carriers are found relatively frequently in cohorts with idiopathic short stature in the sense of an allele with a large effect size [17]. Studies on rare, not fully penetrating microdeletion syndromes can also serve as models. In the case of the microdeletion of chromosome 16p11.2, for example, a comparison with relatives who do not have the microdeletion showed a relative reduction of approx. 30 IQ points in the case of deletion carriers [54]. Depending on the family background, the IQ was still in the normal range or in the below-average range.

Despite all the difficulty, genetic associations are usually much easier and more reliable to determine than was previously the case for relevant environmental factors. However, as shown above, these also make a significant contribution to the overall variance. In addition, environmental factors promise to be easier to modulate, which is why reliable knowledge is also of great importance in this area. In the future, studies on environmental factors can also examine the genetic component, e.g. B. via a PGS, take into account in the analyzes what should improve their power and informative value. Research into the interaction between genes and the environment will now become easier with the increasing understanding of the genetic basis.

Changeability of intelligence

In view of the great public interest in the attribute intelligence and its relevance for the explanation and prediction of important areas of individual and social success, the question of the changeability of intelligence should be addressed at this point. First of all, it should be noted that, despite the sometimes substantial heredity findings, classical behavioral genetic studies do not generally allow any statement to be made about the potential for change at the level of the individual and with regard to the absolute expression of the characteristic of intelligence, since the designs used are solely aimed at influencing the to quantify existing differences in characteristics between individuals. Nevertheless, the findings reported in this article are relevant at this point.

Children from higher social classes have far greater chances of better educational qualifications than children from lower social classes. Against the background of the fact that intelligence has a positive effect on the educational and life success of a person and in view of the fact that genes have a substantial influence on the development of intelligence, it can be deduced that the greater educational success of children from higher social classes is not only accompanied by a better one Promotion can be explained. It can be assumed that the connection between educational success and class affiliation has partly genetic causes. This is possibly due to the plant-environment correlation (see above). In addition, there are also indications of class-specific promotional effects on intelligence, which fits with the previously reported finding that shared environmental influences are more significant in less educated families than in more educated families. Children from educationally disadvantaged backgrounds are therefore more likely to lag behind their potential unless they are encouraged by suitable offers. There are empirical indications of the changeability of intelligence through educational measures [4] or through lasting changes in living conditions [11]. A systematic, comparative consideration of the effectiveness of various complex early intervention programs is difficult due to the sometimes very different goals and the different program designs. However, a substantial and long-term increase in intelligence does not seem to be one of the positive effects of early intervention programs [32] (see also [38]).

There is a lack of convincing evidence of their efficient and sustainable effectiveness for programs that allegedly can train general mental functions in different phases of life. The transfer effects of working memory training on intelligence that have recently been publicized in the media do not stand up to critical scrutiny [28].

conclusion for practice

The current findings are primarily of scientific interest. In addition to further research into the characteristic of intelligence, they can possibly also be the basis for oligogenic models, which are discussed in particular for the mild spectrum of intelligence disorders and learning disabilities. In addition, they are of great importance for researching environmental factors that can influence the expression of the characteristic. Against the background of the probabilistic nature of the results and the comparatively low explanation of variance in the criterion, “polygenic scores” are unsuitable as a predictor of individual cognitive abilities and school success.


  1. 1.

    Amthauer R, Brocke B, Liepmann D, Beauducel A (2001) Intelligence Structure Test 2000 R. Hogrefe, Göttingen

    Google Scholar

  2. 2.

    Bouchard TJ Jr., Mcgue M (1981) Familial studies of intelligence: a review. Science 212: 1055-1059

    Article Google Scholar

  3. 3.

    Briley DA, Tucker-Drob EM (2013) Explaining the increasing heritability of cognitive ability across development: a meta-analysis of longitudinal twin and adoption studies. Psychol Sci 24: 1704-1713

    Article Google Scholar