Skip to main content
NCBI home page
Human Brain Mapping logoLink to Human Brain Mapping
. 2026 Aug 30;47(13):e70635. doi: 10.1002/hbm.70635

Cerebral Cortical Structural Variation and General Cognitive Ability: Evidence From Mendelian Randomization

Chun‐Ju Chou 1,2, Mark Fiecas 3, Elisabetta C del Re 4,5, Eero Vuoksimaa 6,✉, Chi‐Hua Chen 1,✉
PMCID: PMC13526639  PMID: 42669588

ABSTRACT

Understanding the cortical architecture underlying individual differences in general cognitive ability (GCA) remains a central question in cognitive neuroscience. Prior work has established associations between global brain size and GCA, yet the regional effects and directionality of these relationships remain debated. Using a genetically informed cortical parcellation in 11,289 UK Biobank participants, we examined associations between cortical surface area (SA), cortical thickness (CT), and GCA measured via verbal–numerical reasoning. Total SA showed a robust positive association with GCA. At the regional level, dorsolateral prefrontal and superior temporal SA exhibited the strongest positive associations, which persisted after adjustment for global SA. In contrast, CT showed comparatively modest associations. Using Mendelian randomization (MR) with genome‐wide significant genetic instruments, we observed evidence consistent with a bidirectional relationship between total SA and GCA. At the regional level, dorsolateral prefrontal and temporal SA demonstrated evidence of MR‐inferred directional effects on GCA, while GCA showed evidence of MR‐inferred directional effects on total SA and perisylvian thickness. These findings support a polyregional SA architecture underlying GCA, with prominent contributions from prefrontal and temporal association cortices. Our results refine global brain–GCA models and highlight the value of genetically informed parcellation for identifying regional cortical contributions.

Keywords: cortical morphology, general cognitive ability, genetically‐informed parcellation, genome‐wide association study, Mendelian randomization

Using a genetically‐informed cortical parcellation of 11,289 UKB participants, this study identified positive associations between fluid intelligence and cortical surface area, particularly in the dorsolateral‐prefrontal and superior‐temporal regions. Furthermore, Mendelian randomization revealed a bidirectional causal relationship between total surface area and intelligence, highlighting a complex cortical architecture underlying cognitive ability.

graphic file with name HBM-47-e70635-g002.webp

1. Introduction

Genetic influences account for as high as 80% of the variance in individual differences in intelligence, also known as general cognitive ability (GCA) or the g‐factor (Panizzon et al. 2014), yet its neural underpinnings remain unclear. In this study, we use the terms intelligence and GCA interchangeably. Over three decades of magnetic resonance imaging (MRI) studies have established a positive correlation of about 0.24 between brain volume and GCA (Pietschnig et al. 2022). In the past decade, studies have shifted focus from brain volume to cortical surface area (SA) and cortical thickness (CT)—highly heritable, developmentally distinct, and anatomically orthogonal traits (Geschwind and Rakic 2013). Evidence suggests the brain volume‐GCA link is largely due to genetic effects and driven by total SA rather than mean CT (Vuoksimaa et al. 2015).

Prior work indicates that GCA correlates with greater SA and thinner cortex in high‐expanded regions (Walhovd et al. 2016; Vuoksimaa et al. 2016). More recent large‐scale studies, such as those leveraging the UK Biobank (UKB) (Miller et al. 2016) and the Adolescent Brain Cognitive Development (ABCD) (Jernigan 2017) cohorts, provide valuable MRI data and have reported regional associations with GCA (Marek et al. 2022; Williams et al. 2023; Cox et al. 2019). Other investigations have utilized genome‐wide association study (GWAS) summary statistics rather than individual‐level data to examine brain–cognition relationships (Grasby et al. 2020; Grotzinger et al. 2023; Lin et al. 2025). Together, these studies showed global brain measures contribute to cognitive ability. However, the choice of cortical parcellation (e.g., number and boundaries of regions) and imaging metrics (e.g., volume or surface‐based measures) impacts detection power and can influence the reproducibility and localization of associations (Arslan et al. 2018). Consequently, although global brain measures show the strongest associations with GCA, further research is needed to establish consistent evidence for regional contributions. Given the strong genetic influences on cortical structure and GCA (Vuoksimaa et al. 2015, 2016), it is beneficial to consider the genetic nature of cortical regionalization (Makowski et al. 2022; Chen et al. 2013).

To address these gaps, we utilized a genetically defined cortical parcellation, which aligns with key functional specializations of the human cortex, to measure both SA and CT within each subdivision. Unlike traditional anatomical atlases, genetically‐based parcellation delineates boundaries that are intended to better capture biological and genetic (Chen et al. 2013). They further offer greater sensitivity for detecting regional associations (Makowski et al. 2022). We aimed to identify the cortical correlates underlying GCA in the UKB sample (N = 11,289). In this study, we operationalize GCA using the UKB “fluid intelligence” score (verbal–numerical reasoning; Data‐Field ID 20016) and refer to this measure as a proxy for GCA throughout. We hypothesized that genetically derived regions would reveal a polyregional neural substrate of GCA, with SA and CT associations following distinct spatial patterns. Additionally, we expected to identify MR‐inferred directional effects between cortical structure and GCA within this polyregional network through MR analysis.

2. Materials and Methods

2.1. Sample

The genomic, neuroimaging, cognitive, and demographic data were extracted from the UKB population cohort (Elliott et al. 2018; Sudlow et al. 2015; Bycroft et al. 2018; Miller et al. 2016). Quality control (QC) procedures for imaging and demographic variables followed our previous work (Makowski et al. 2022, 2023). In brief, this study included individuals of European ancestry, identified based on genetic principal components provided by the UKB (Bycroft et al. 2018). In this study, we included 34,842 participants with available neuroimaging. We accounted for genetic relatedness and removed related individuals from our analyses. Utilizing Genome‐wide Complex Trait Analysis (GCTA) (Yang et al. 2011), the pairwise genetic relationship matrix (GRM) based on genome‐wide autosomal variants was calculated, and 859 individuals were removed due to relatedness from pairs with an estimated GRM greater than 0.1 (indicating relatedness closer than third cousins), resulting in 33,861 participants. Excluding participants with missing cognitive scores, we resulted in a final dataset of 11,289 participants (age range: 45.13–80.17 years, male–female ratio of 0.94) for subsequent analyses. Demographic analysis confirmed the sample reduction reflects data availability rather than a systematic selection bias.

2.2. Genotype Data

We used UKB Version 3 release of imputed genotype data and removed individuals with more than 10% missingness, as well as Single Nucleotide Polymorphisms (SNPs) with more than 5% missingness, failing the Hardy–Weinberg equilibrium test at p = 1e−6 or with minor allele frequencies (MAF) below 0.01.

2.3. MRI Data and Atlases

T1‐weighted MRI scans were collected from three scanning sites throughout the United Kingdom, all on identical Siemens Skyra 3 T scanners (Miller et al. 2016). Automated surface‐based morphometry segmentation was performed using the standard ‘recon‐all’ processing pipeline in FreeSurfer v5.3. Both SA and CT were defined at the vertex level; SA was extracted from the white matter surface, and CT was calculated as the shortest distance between the white and pial surfaces. Scan quality was quantitatively assessed using FreeSurfer's Euler number (Dale et al. 1999). Following this standard processing, the vertex‐level data were parcellated using our genetically‐informed atlases (Chen et al. 2012, 2013).

For cortical phenotypes, we adopted two genetically‐informed atlases, including 12 regions for SA and 12 for CT, and 2 global measures of total SA and mean CT. These atlases were developed using a data‐driven fuzzy clustering technique to identify cortical parcels that are maximally genetically correlated, based on MRI scans from over 400 twins (Chen et al. 2012, 2013). We combined measures of each phenotype across both hemispheres, in view of the largely bilateral symmetry of genetic patterning demonstrated previously (Chen et al. 2011; Makowski et al. 2022).

2.4. Fluid Intelligence Scores

GCA was operationalized as fluid intelligence scores that were extracted from the UKB using field ID 20016. This is a simple unweighted sum of the number of correct answers given to the 13 fluid intelligence questions. Participants who did not answer all questions within the allotted 2‐min limit received a score of zero for each unattempted question. For this study, we included fluid intelligence scores from the first visit.

GCA was assessed using the UKB “fluid intelligence” score, which is more accurately described as a verbal–numerical reasoning (VNR) measure. Although this cognitive measure is labeled as assessing fluid cognitive abilities, which typically decline with aging, we note that its correlation with age in this UKB sample was very close to zero (−0.026). This suggests that the score reflects both fluid and crystallized (stable) abilities. Indeed, the UKB fluid intelligence score has been reported to relate to crystallized abilities and is one of the UKB cognitive measures with the highest correlation to GCA (Fawns‐Ritchie and Deary 2020).

2.5. Cortex‐GCA Association Analysis

Prior to association analysis, we regressed out scanner site, a proxy of scan quality (FreeSurfer's Euler number) (Dale et al. 1999), age, sex, and diagnosis from each morphometric measure, resulting in residualized absolute measurements. In a parallel analysis of the regional measures, we also regressed out corresponding global measures (i.e., total SA and mean cortical CT for area and thickness phenotypes, respectively), referring to residualized relative measurements. This was done to distinguish regional from global effects. Regional SA, CT, and intracranial volume (ICV) represent distinct neurobiological phenotypes with different developmental and aging trajectories (Bethlehem et al. 2022; Wierenga et al. 2014; Yu 2024). Although ICV is commonly used as a proxy for head size, longitudinal studies suggest that it also undergoes modest age‐related changes (Caspi et al. 2020). For analyses focused on regional cortical morphology, we considered regional SA and CT to be most appropriately interpreted relative to their respective global measures (total cortical SA and mean CT, respectively), rather than ICV. We present both absolute and relative models to address complementary biological questions. The absolute models capture total morphological associations, while adjusting for phenotype‐matched global measures in the relative models distinguishes specific regional associations from overall cortical morphology (Wang et al. 2024; Peelle et al. 2012). Specifically, the adjusted estimates should be interpreted as relative regional deviations conditional on the global measure, rather than as evidence of region‐specific effects independent of global brain measures. In this framework, the estimated associations reflect whether a given region shows enrichment or depletion beyond what would be expected given the corresponding global measure. Subsequently, the residuals of each measure were scaled using rank‐based inverse normal transformation to ensure normally distributed input.

Because age can have nonlinear effects, we also compared models including a spline term for age (ns(Age, 3)) to linear age models. Results for associations between fluid intelligence scores and brain morphometric measures were essentially unchanged, so linear age was retained in the main analyses, with spline models noted as a sensitivity check.

Linear regression models examined associations between fluid intelligence scores (retained in raw units) and standardized brain morphometric measures, yielding semi‐standardized coefficients. Model 1: fluid intelligence scores as the dependent variable, with a cortical region's residualized absolute measurement as the independent variable. Model 2: Same as Model 1, but using the cortical region's residualized relative measurement.

For the diagnosis covariate, we controlled for whether an individual had a brain‐related diagnosis based on International Classification of Diseases, 10th Revision (ICD10) diagnostic information collected by the UKB under field ID 41202. Individuals were classified as having a brain diagnosis if they met criteria for at least one class F (mental and behavioral disorders) or class G (disorders of the nervous system) diagnosis, with the exception of G56—carpal tunnel syndrome, which is an extremely common condition and thus we did not consider it as a neurological diagnosis. We did not exclude any participants based on diagnosis; instead, diagnosis was included as a covariate.

2.6. Determining Effective Number of Independent Phenotypes

To consider the potential correlation between phenotypes, we applied matSpD to determine the effective number of independent phenotypes (te) (Li et al. 2012), using correlation matrices of cortical measures. Statistical significance was then defined by Bonferroni correction for multiple comparisons (p < 0.05/te).

In this study, we included regional measures (12 SA and 12 CT, averaged across hemispheres) and global measures (total SA and mean CT). To account for multiple comparisons, we calculated the effective number of independent traits as t e = 22. Therefore, we set the study‐wise significance threshold at p < 2.3e−3 (0.05/t e).

2.7. Mendelian Randomization

We performed generalized summary‐based MR (GSMR) (Zhu et al. 2018) to evaluate the MR‐inferred directional effects of the above 24 regional and 2 global morphometric measures on GCA, based on the summary statistics from previous genome‐wide association analyses on the exposure (Makowski et al. 2022) and outcome (Savage et al. 2018) phenotypes (detailed in Supplementary Methods 1, 2). These GWAS studies have non‐overlapping samples. Linkage disequilibrium‐independent (r 2 < 0.05, 250 kb) significant SNPs (p < 5e−8) were included as candidate instrumental variables. The final set of instrumental variables used in the analyses is shown in Supporting Information (Tables S4 and S5). Significance for GSMR analyses was defined as Bonferroni‐corrected p‐value < 2.3e−3. The methodology and reporting of our MR analyses adhere to the STROBE‐MR (Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization) guidelines to ensure transparent and complete reporting (Skrivankova, Richmond, Woolf, Yarmolinsky, et al. 2021; Skrivankova, Richmond, Woolf, Davies, et al. 2021).

MR uses genetic variants as instrumental variables to estimate the causal effect of an exposure on an outcome, provided its core assumptions are satisfied. Because alleles are randomly allocated at conception, MR provides a natural experiment that can reduce confounding and reverse causation. Although MR is designed to estimate causal effects, some of its core assumptions cannot be directly verified but can sometimes be disproven through sensitivity analyses (Sanderson et al. 2022). Therefore, throughout this study, we use the term “MR‐inferred directional effects,” or similar terminology, to refer to effect estimates inferred from MR analyses rather than “causal effects,” reflecting the inferential nature of the evidence.

2.8. Instrument Strength and Pleiotropy Control

Instrument strength was evaluated using F‐statistics; all retained variants across both forward and reverse MR analyses exceeded the threshold for strong instruments (F > 10, with many > 50) (see Table S4), minimizing weak instrument bias (Brion et al. 2013; Pierce et al. 2010). While reverse MR discovered fewer genome‐wide significant variants, prior methodological work indicates that valid inferences can still be drawn when retained instruments are highly robust (Bowden et al. 2019). To mitigate horizontal pleiotropy, GSMR applied the Heterogeneity in Dependent Instruments (HEIDI)‐outlier test (pHEIDI < 0.01) to remove heterogeneous variants prior to MR estimation (Zhu et al. 2018). Finally, the convergence of effect directions across multiple complementary MR instruments (Supplementary Methods) indicates that residual pleiotropy is unlikely to account for the observed associations.

3. Results

3.1. Sample Descriptive Demographic

In this study sample, males were slightly older than females (64.54 ± 7.56 vs. 63.24 ± 7.32 years) but did not differ significantly in years of education (Table 1). Although males exhibited significantly higher GCA scores, the effect size was small (Cohen's d = −0.12). As expected, males had significantly greater SA than females (Cohen's d = −1.27), while females had slightly thicker cortex (Cohen's d = 0.27). The global brain differences align with males' generally larger body size, as SA correlates with height. The thicker cortex in females may reflect the modest negative correlation between SA and CT observed here (r = −0.1) and in prior studies (Tadayon et al. 2020; Vuoksimaa et al. 2016; Makowski et al. 2022; Walhovd et al. 2016).

TABLE 1.

Means and standard deviations by sex for demographic characteristics, GCA measure and global brain metrics.

Parameters Group Sex difference
Female (n = 5814) Male (n = 5475) Total T‐test p‐value Cohen's d
Age (yrs) 63.24 ± 7.32 64.54 ± 7.56 63.87 ± 7.46 < 1e−10 −0.18
Education (final age) 17.10 ± 2.39 17.17 ± 2.84 17.13 ± 2.61 0.31 −0.03
GCA 6.71 ± 1.93 6.94 ± 2.06 6.82 ± 2.00 3.41e−10 −0.12
Total SA (cm2) 1605.86 ± 126.70 1776.58 ± 141.60 1688.66 ± 158.97 < 1e−10 −1.27
Mean CT (mm) 2.43 ± 0.09 2.40 ± 0.10 2.41 ± 0.10 < 1e−10 0.27
Open in a new tab

Note: Data is presented as mean ± standard deviation (SD). General cognitive ability as measured with UKB fluid intelligence score, cortical surface area. p‐values from the t‐test for sex differences. In Cohen's d, negative values indicate greater values in males.

Abbreviations: CSA, cortical surface area; CT, cortical thickness; GCA, general cognitive ability.

3.2. Absolute and Relative Cortical Morphology Associated With GCA

Two regression models were performed to examine the cortical‐GCA associations: one unadjusted and one adjusted for overall brain size (total SA or mean CT), corresponding to absolute and relative measures, respectively.

When analyzing absolute cortical morphology, we confirmed a statistically significant and modest positive association between total SA and GCA (β = 0.36, p = < 1e−10, R 2 = 2.4%). For regional SA, all cortical regions displayed significant positive associations with GCA (Figure 1, Table S1). The dorsolateral prefrontal area, subserving working memory and executive functions, exhibited the highest effect size (β = 0.35; R 2 = 1.4%) among all regions. Other top regions included the superior temporal and orbitofrontal areas, which are critical for auditory/language processing and emotional working memory, respectively. In contrast to the SA results, mean CT showed a nominally significant positive association (β = 0.06, p = 7.36e−03) and regional CT exhibited generally modest associations with GCA (Figure 1, Table S1).

FIGURE 1.

FIGURE 1

Open in a new tab

(a) Association between cortical morphology and GCA, with and without adjustment for global measures. Large or small asterisks denote statistical (p < 2.3e−3) or nominal (p < 0.05) significance. Global measures are presented twice at the end of both rows of heatmaps. (b) Brain maps with color shading represent beta coefficients for surface area and cortical thickness. Asterisks indicate cortical regions with significant associations after correction for multiple comparisons (p < 2.3e−3). The bottom row shows atlas brain maps for reference, with numbering corresponding to the regions listed above the heatmap.

After adjusting for total SA, relative cortical SA showed smaller associations. However, the superior temporal area (β = 0.09, p = 2.05e−06) and dorsolateral prefrontal area (β = 0.07, p = 4.98e−04) remained positively associated with GCA, whereas the occipital area showed a significant negative association (β = −0.08, p = 5.24e−05). For CT, negative associations for relative CT were observed after adjusting for mean CT, particularly in ventral frontal (β = −0.13, p = 1.97e−11), middle temporal (β = −0.10, p = 1.11e−07), and medial prefrontal (β = −0.11, p = 1.53e−09) regions.

3.3. Bidirectional Mendelian Randomization of Cortical Morphology and GCA

Beyond associations, we next studied the bidirectional relationships between cortical structure and GCA to better contextualize these links (Figure 2, Table S2). In forward MR analyses, total SA exhibited a positive directional effect on GCA (βxy = 0.13, CI = [0.10, 0.18], p = 2.08e−10). At the regional level, larger dorsolateral prefrontal (βxy = 0.09, CI = [0.02, 0.15], p = 1.05e−02), posterolateral temporal (βxy = 0.06, CI = [0.02, 0.10], p = 4.00e−03), and orbitofrontal (βxy = 0.06, CI = [0.00, 0.13], p = 4.19−02) areas had positive directional effects on GCA, whereas the superior parietal area had a negative directional effect (Figure 2). For CT, no global or regional CT measures demonstrated a directional influence on GCA. To evaluate the robustness of these forward MR estimates, we compared them across five complementary MR methods (Supplementary Methods and Figure S1) and results (Table S3). Effects for total, posterolateral temporal, and dorsolateral prefrontal surface areas remained directionally consistent across models, with no substantial directional pleiotropy observed, indicating that our primary findings are stable under potential MR assumption violations.

FIGURE 2.

FIGURE 2

Open in a new tab

Bidirectional effects between cortical morphology and GCA. The heat maps display both (a) forward and (b) reverse MR analyses. Large or small asterisks denote statistical (p < 2.3e−3, Bonferroni correction) or nominal (p < 0.05) significance. Gray boxes indicate missing data due to limitations such as restricted sample size or insufficient number of filtered single nucleotide polymorphisms (SNPs) meeting the inclusion threshold. (c) Brainmap reference with numbers corresponding to the regions listed above. The left two brain maps show the area atlas, and the right two show the thickness atlas.

In reverse MR analyses, GCA was found to directionally influence total SA (βxy = 0.31, CI = [0.12, 0.49], p = 1.02e−03), suggesting a reciprocal relationship. At the regional level, higher GCA directionally influenced perisylvian thickness (βxy = 0.33, CI = [0.14, 0.51], p = 5.19e−04). Sensitivity analyses demonstrated that these reverse effects also maintained directional consistency across complementary approaches (Figure S1); the attenuated statistical strength in more conservative models is consistent with the smaller number of available genetic instruments in the reverse direction.

4. Discussion

4.1. Global and Regional Cortical Contributions to GCA

Our study replicates and extends prior evidence that total cortical SA, rather than mean CT, is associated with GCA (Vuoksimaa et al. 2015; Tadayon et al. 2020). Importantly, although total SA remains the strongest global correlate, it explains a modest 2.4% of the variance in GCA. This effect size contextualizes our findings, suggesting that cortical morphometry represents only a partial factor of GCA. Using a genetically informed cortical parcellation, we further refine the regional architecture underlying this relationship. Adjusting for global SA attenuated regional associations, consistent with previous findings (Reardon et al. 2018; Vuoksimaa et al. 2016), yet dorsolateral prefrontal and superior temporal regions remained positively associated with GCA. These findings suggest that regional cortical configuration contributes beyond overall brain size. Given that global SA is highly genetically correlated with dorsolateral prefrontal SA (Makowski et al. 2023), adjustment for total SA likely removes substantial shared variance, making the persistence of these regional effects particularly noteworthy. We note that GCA was operationalized using the UKB “fluid intelligence” score, which likely captures a combination of fluid and crystallized abilities (see Methods); thus, our findings should be interpreted in the context of this proxy rather than as a comprehensive measure of intelligence.

We observed a spatial gradient in SA–GCA associations, with strongest effects in anterior prefrontal and lateral temporal association cortices, including regions near the insula and Sylvian fissure (Chen et al. 2013; Vogel et al. 2024). In contrast, CT associations were more modest and demonstrated a dorsal‐to‐ventral gradient pattern, with the strongest effects observed in dorsal regions that diminished toward the ventral surface. These results are consistent with prior work on cortical genetic topography and expansion gradients (Walhovd et al. 2016; Vogel et al. 2024; Vuoksimaa et al. 2016) and extend these spatial principles to the domain of cognitive ability.

4.2. Regional Effects and Dorsolateral Prefrontal Prominence

Among the cortical regions examined, dorsolateral prefrontal SA exhibited the largest regional effect size and the strongest evidence consistent with a directional relationship relative to other regions. Superior and posterolateral temporal areas also demonstrated notable associations. These regions overlap with previously identified “high‐expanded” cortical territories (Walhovd et al. 2016; Vuoksimaa et al. 2016) and are genetically correlated with GCA.

The prominence of dorsolateral prefrontal cortex aligns with extensive evidence implicating this region in working memory, executive control, and abstract reasoning (Cox et al. 2019). Importantly, the use of a genetically informed atlas allowed more direct delineation of dorsolateral prefrontal cortex than traditional anatomical atlases. For example, Lin et al. (Lin et al. 2025) relied on the Desikan atlas, which includes proxy middle frontal regions rather than isolating dorsolateral prefrontal cortex directly. Methodological differences, including instrument selection thresholds for MR, may contribute to variability across studies.

4.3. Bidirectional Relationships Between Cortical Structure and GCA

MR analyses provided evidence consistent with bidirectional effects between cortical morphology and GCA. Total SA demonstrated a consistent directional influence on GCA, and GCA showed evidence of influencing total SA, replicating and extending prior findings (Grasby et al. 2020; Korologou‐Linden et al. 2023). This aligns with MR evidence suggesting that educational attainment influences structural brain reserve, potentially offering a neuroprotective buffer against Alzheimer's Disease (Seyedsalehi et al. 2023). These results point toward a complex developmental relationship between cortical expansion and cognitive ability that unfolds progressively over time. While MR does not directly capture individual developmental change, the results may provide insight into the neurobiological architecture underlying cognitive variation and inform future research in neuropsychiatric and aging‐related conditions. At the regional level, dorsolateral prefrontal and posterolateral temporal SA exhibited the strongest directional signals among the cortical regions examined. Although these regional effects did not survive conservative correction for multiple testing, their convergence with association analyses supports their potential relevance. Reverse MR analyses further indicated that GCA may influence perisylvian thickness, a region implicated in language processing (Brans et al. 2010), possibly reflecting experience‐dependent cortical plasticity; however, this interpretation should be viewed as tentative, as MR does not directly capture within‐individual developmental change or environmental effects. We note that statistical power varied across regional phenotypes (Makowski et al. 2022), limiting inference for some regions.

4.4. Implications for Distributed Models of GCA

Our findings refine, rather than contradict, distributed frameworks such as the Parieto‐Frontal Integration Theory (P‐FIT) (Jung and Haier 2007; Deary et al. 2022; Prabhakaran et al. 1997; Mitchell et al. 2023). While parietal cortex associations were comparatively weaker in our study, this does not preclude their role within distributed functional networks. Differences across studies may reflect variation in imaging metrics (SA and CT versus volume), adjustment for global measures, parcellation strategies, and cognitive phenotypes (Cox et al. 2019; Lin et al. 2025; Barbey et al. 2013).

The dissociation observed between SA and CT effects is consistent with distinct developmental and genetic mechanisms underlying these traits (Vuoksimaa et al. 2015; Reardon et al. 2018; Panizzon et al. 2009). SA is thought to reflect early neurodevelopmental processes related to cortical expansion, whereas thickness may capture later maturational and pruning‐related dynamics (Schnack et al. 2015). The prominence of SA associations therefore suggests that early cortical expansion plays a particularly important role in shaping individual differences in GCA.

4.5. Sensitivity Analyses for Assumption Violation

The sensitivity of our findings is consistent with the convergence of MR estimates across multiple MR frameworks with distinct sensitivity profiles to horizontal pleiotropy. The alignment of effect directions across these complementary estimators, combined with HEIDI filtering and strong instrument diagnostics, indicates that our primary findings are unlikely to be driven by invalid or pleiotropic SNPs. Although more conservative estimators yielded weaker statistical support, particularly for phenotypes with fewer available instruments, their point estimates remained directionally consistent. Collectively, these sensitivity analyses support the directional consistency of the observed relationships, particularly for the dorsolateral prefrontal and posterolateral temporal surface areas, while highlighting the differences in instrument strengths and statistical support between bidirectional inference.

5. Limitations

Given the high inter‐hemispheric genetic correlation in cortical measures, combining hemispheres provides a stable summary phenotype, albeit with reduced sensitivity to lateralized effects. We acknowledge that this approach may obscure hemispheric specialization, and that future hemisphere‐specific modeling will be important to better resolve these patterns. We note that scan quality was modeled as a covariate affecting the mean, rather than explicitly modeling error in the variance; thus, differences in measurement precision may not be fully accounted for. Future work could address this by incorporating approaches that directly model measurement uncertainty, such as heteroskedastic or measurement error models. Additionally, we did not perform cross‐site statistical harmonization. Although all scanning sites used identical 3 T scanners, hardware matching cannot entirely remove site‐specific variance. Therefore, the lack of statistical harmonization may represent a limitation of our current pipeline.

Although we used phenotype‐matched global measures in our relative models for global adjustment (total SA for regional SA and mean CT for regional CT), other studies have used ICV for global adjustment (Brzezinski‐Rittner et al. 2026). However, the optimal strategy for global adjustment, particularly with respect to brain size, age, and sex, remains an active area of investigation.

Finally, given the sensitive nature of cognitive research, we emphasize that these findings reflect population‐level statistical associations explaining a small proportion of variance of GCA. These results cannot be used for individual prediction and do not support biological determinism. These metrics do not comprehensively capture human intelligence.

6. Conclusion

In summary, our results support a polyregional model of GCA characterized by distributed SA expansion within higher‐order association cortex. While total SA remains the strongest global correlate, regional configuration, particularly within dorsolateral prefrontal and temporal cortices, provides additional explanatory insight. By integrating genetically informed parcellation (Chen et al. 2013) with MR modeling (Zhu et al. 2018), this study advances understanding of how global and regional cortical organization jointly contribute to cognitive ability.

Funding

This research was supported by the National Institute of Mental Health under R01MH132783. E.V. was supported by the Sigrid Jusélius Foundation and the Research Council of Finland (grants 314639 and 345988).

Ethics Statement

The authors have nothing to report.

Supporting information

Figure S1: Bidirectional causal effects between cortical morphology and GCA with different MR methods comparison. The heat maps display both (a, b) forward and (c, d) reverse Mendelian Randomization analyses. Big or small asterisks denote statistical (p < 2.3e−3, Bonferroni correction) or nominal (p < 0.05) significance. Gray boxes indicate missing data due to limitations such as restricted sample size or insufficient number of filtered single nucleotide polymorphisms (SNPs) meeting the inclusion threshold.

Table S4: Instrument strength of genome‐wide significant SNPs used in forward GSMR analyses across absolute cortical regions after pleiotropy filtering.

Table S5: Instrument strength of genome‐wide significant SNPs used in reverse GSMR analyses across absolute cortical regions after pleiotropy filtering.

HBM-47-e70635-s001.docx (4.6MB, docx)

Table S1: Results of Fluid Intelligence‐Cortical morphology association analysis.

Table S2: Results of MR analysis.

Table S3: Results of bidirectional GSMR analysis.

HBM-47-e70635-s002.xlsx (52.1KB, xlsx)

Acknowledgements

This research was supported by the National Institute of Mental Health under R01MH132783. E.V. was supported by the Sigrid Jusélius Foundation and the Research Council of Finland (grants 314639 and 345988). This research has been conducted using data from the UK Biobank, a major biomedical database, under application number 27412.

Contributor Information

Eero Vuoksimaa, Email: eero.vuoksimaa@helsinki.fi.

Chi‐Hua Chen, Email: chc101@ucsd.edu.

Data Availability Statement

This study utilizes individual‐level genetic and imaging data from the UK Biobank (https://www.ukbiobank.ac.uk/). The genome‐wide association data for cortical regions were provided from our previously published studies, which can be accessed via the GWAS Catalog (https://www.ebi.ac.uk/gwas/publications/35113692 and https://www.ebi.ac.uk/gwas/publications/36893272). The genome‐wide association data for intelligence were obtained directly from the authors upon request, as a means of avoiding overlap with UKB samples, and the original source was from the Psychiatric Genomics Consortium (PGC) (https://pgc.unc.edu/) or the GWAS Catalog (https://www.ebi.ac.uk/gwas/publications/29942086). GSMR method and code are publicly available via the GCTA software repository on GitHub (https://github.com/JianYang‐Lab/gsmr/releases).

References

  1. Arslan, S. , Ktena S. I., Makropoulos A., Robinson E. C., Rueckert D., and Parisot S.. 2018. “Human Brain Mapping: A Systematic Comparison of Parcellation Methods for the Human Cerebral Cortex.” NeuroImage 170: 5–30. [DOI] [PubMed] [Google Scholar]
  2. Barbey, A. K. , Colom R., and Grafman J.. 2013. “Dorsolateral Prefrontal Contributions to Human Intelligence.” Neuropsychologia 51, no. 7: 1361–1369. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Bethlehem, R. A. I. , Seidlitz J., White S. R., et al. 2022. “Brain Charts for the Human Lifespan.” Nature 604, no. 7906: 525–533. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bowden, J. , Del Greco M F., Minelli C., et al. 2019. “Improving the Accuracy of Two‐Sample Summary‐Data Mendelian Randomization: Moving Beyond the NOME Assumption.” International Journal of Epidemiology 48, no. 3: 728–742. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Brans, R. G. H. , Kahn R. S., Schnack H. G., et al. 2010. “Brain Plasticity and Intellectual Ability Are Influenced by Shared Genes.” Journal of Neuroscience: The Official Journal of the Society for Neuroscience 30, no. 16: 5519–5524. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Brion, M.‐J. A. , Shakhbazov K., and Visscher P. M.. 2013. “Calculating Statistical Power in Mendelian Randomization Studies.” International Journal of Epidemiology 42, no. 5: 1497–1501. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Brzezinski‐Rittner, A. , Moqadam R., Zeighami Y., and Dadar M.. 2026. “Intracranial Volume: To Adjust or Not to Adjust? It Is Not a Matter of if, but How.” Imaging Neuroscience (Cambridge, Mass.) 4, no. 5: e1235. 10.1162/IMAG.a.1235. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Bycroft, C. , Freeman C., Petkova D., et al. 2018. “The UK Biobank Resource With Deep Phenotyping and Genomic Data.” Nature 562, no. 7726: 203–209. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Caspi, Y. , Brouwer R. M., Schnack H. G., et al. 2020. “Changes in the Intracranial Volume From Early Adulthood to the Sixth Decade of Life: A Longitudinal Study.” NeuroImage 220, no. 10: 116842. [DOI] [PubMed] [Google Scholar]
  10. Chen, C.‐H. , Fiecas M., Gutiérrez E. D., et al. 2013. “Genetic Topography of Brain Morphology.” Proceedings of the National Academy of Sciences of the United States of America 110, no. 42: 17089–17094. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Chen, C.‐H. , Gutierrez E. D., Thompson W., et al. 2012. “Hierarchical Genetic Organization of Human Cortical Surface Area.” Science 335, no. 6076: 1634–1636. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Chen, C.‐H. , Panizzon M. S., Eyler L. T., et al. 2011. “Genetic Influences on Cortical Regionalization in the Human Brain.” Neuron 72, no. 4: 537–544. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Cox, S. R. , Ritchie S. J., Fawns‐Ritchie C., Tucker‐Drob E. M., and Deary I. J.. 2019. “Structural Brain Imaging Correlates of General Intelligence in UK Biobank.” Intelligence 76: 101376. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Dale, A. M. , Fischl B., and Sereno M. I.. 1999. “Cortical Surface‐Based Analysis. I. Segmentation and Surface Reconstruction.” NeuroImage 9, no. 2: 179–194. [DOI] [PubMed] [Google Scholar]
  15. Deary, I. J. , Cox S. R., and Hill W. D.. 2022. “Genetic Variation, Brain, and Intelligence Differences.” Molecular Psychiatry 27, no. 1: 335–353. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Elliott, L. T. , Sharp K., Alfaro‐Almagro F., et al. 2018. “Genome‐Wide Association Studies of Brain Imaging Phenotypes in UK Biobank.” Nature 562, no. 7726: 210–216. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Fawns‐Ritchie, C. , and Deary I. J.. 2020. “Reliability and Validity of the UK Biobank Cognitive Tests.” PLoS One 15, no. 4: e0231627. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Geschwind, D. H. , and Rakic P.. 2013. “Cortical Evolution: Judge the Brain by Its Cover.” Neuron 80, no. 3: 633–647. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Grasby, K. L. , Jahanshad N., Painter J. N., et al. 2020. “The Genetic Architecture of the Human Cerebral Cortex.” Science 367, no. 6484: aay6690. 10.1126/science.aay6690. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Grotzinger, A. D. , Mallard T. T., Liu Z., Seidlitz J., Ge T., and Smoller J. W.. 2023. “Multivariate Genomic Architecture of Cortical Thickness and Surface Area at Multiple Levels of Analysis.” Nature Communications 14, no. 1: 946. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Jernigan, T. 2017. “Adolescent Brain Cognitive Development Study (ABCD).” Preprint, NIMH Data Repositories. 10.15154/CQDY-5453. [DOI]
  22. Jung, R. E. , and Haier R. J.. 2007. “The Parieto‐Frontal Integration Theory (P‐FIT) of Intelligence: Converging Neuroimaging Evidence.” Behavioral and Brain Sciences 30, no. 2: 135–154 Discussion 154–87. [DOI] [PubMed] [Google Scholar]
  23. Korologou‐Linden, R. , Schuurmans I. K., Cecil C. A. M., et al. 2023. “The Bidirectional Effects Between Cognitive Ability and Brain Morphology: A Life Course Mendelian Randomization Analysis.” Preprint, medRxiv, November 17. 10.1101/2023.11.17.23297145. [DOI]
  24. Li, M.‐X. , Yeung J. M. Y., Cherny S. S., and Sham P. C.. 2012. “Evaluating the Effective Numbers of Independent Tests and Significant P‐Value Thresholds in Commercial Genotyping Arrays and Public Imputation Reference Datasets.” Human Genetics 131, no. 5: 747–756. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Lin, B. D. , Li Y., Goula A. A., et al. 2025. “Dissecting Causal Relationships Between Cortical Morphology and Neuropsychiatric Disorders: A Bidirectional Mendelian Randomization Study.” Nature Mental Health 3, no. 6: 613–625. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Makowski, C. , van der Meer D., Dong W., et al. 2022. “Discovery of Genomic Loci of the Human Cerebral Cortex Using Genetically Informed Brain Atlases.” Science 375, no. 6580: 522–528. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Makowski, C. , Wang H., Srinivasan A., et al. 2023. “Larger Cerebral Cortex Is Genetically Correlated With Greater Frontal Area and Dorsal Thickness.” Proceedings of the National Academy of Sciences of the United States of America 120, no. 11: e2214834120. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Marek, S. , Tervo‐Clemmens B., Calabro F. J., et al. 2022. “Reproducible Brain‐Wide Association Studies Require Thousands of Individuals.” Nature 603, no. 7902: 654–660. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Miller, K. L. , Alfaro‐Almagro F., Bangerter N. K., et al. 2016. “Multimodal Population Brain Imaging in the UK Biobank Prospective Epidemiological Study.” Nature Neuroscience 19, no. 11: 1523–1536. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Mitchell, D. J. , Mousley A. L. S., Shafto M. A., Cam‐CAN , and Duncan J.. 2023. “Neural Contributions to Reduced Fluid Intelligence Across the Adult Lifespan.” Journal of Neuroscience: The Official Journal of the Society for Neuroscience 43, no. 2: 293–307. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Panizzon, M. S. , Fennema‐Notestine C., Eyler L. T., et al. 2009. “Distinct Genetic Influences on Cortical Surface Area and Cortical Thickness.” Cerebral Cortex 19, no. 11: 2728–2735. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Panizzon, M. S. , Vuoksimaa E., Spoon K. M., et al. 2014. “Genetic and Environmental Influences of General Cognitive Ability: Is a Valid Latent Construct?” Intelligence 43: 65–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Peelle, J. E. , Cusack R., and Henson R. N. A.. 2012. “Adjusting for Global Effects in Voxel‐Based Morphometry: Gray Matter Decline in Normal Aging.” NeuroImage 60, no. 2: 1503–1516. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Pierce, B. L. , Ahsan H., and VanderWeele T. J.. 2010. “Power and Instrument Strength Requirements for Mendelian Randomization Studies Using Multiple Genetic Variants.” International Journal of Epidemiology 40, no. 3: 740–752. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Pietschnig, J. , Gerdesmann D., Zeiler M., and Voracek M.. 2022. “Of Differing Methods, Disputed Estimates and Discordant Interpretations: The Meta‐Analytical Multiverse of Brain Volume and IQ Associations.” Royal Society Open Science 9, no. 5: 211621. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Prabhakaran, V. , Smith J. A., Desmond J. E., Glover G. H., and Gabrieli J. D.. 1997. “Neural Substrates of Fluid Reasoning: An fMRI Study of Neocortical Activation During Performance of the Raven's Progressive Matrices Test.” Cognitive Psychology 33, no. 1: 43–63. [DOI] [PubMed] [Google Scholar]
  37. Reardon, P. K. , Seidlitz J., Vandekar S., et al. 2018. “Normative Brain Size Variation and Brain Shape Diversity in Humans.” Science 360, no. 6394: 1222–1227. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Sanderson, E. , Glymour M. m., Holmes M. V., et al. 2022. “Mendelian Randomization.” Nature Reviews Methods Primers 2: 6. 10.1038/s43586-021-00092-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Savage, J. E. , Jansen P. R., Stringer S., et al. 2018. “Genome‐Wide Association Meta‐Analysis in 269,867 Individuals Identifies New Genetic and Functional Links to Intelligence.” Nature Genetics 50, no. 7: 912–919. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Schnack, H. G. , van Haren N. E. M., Brouwer R. M., et al. 2015. “Changes in Thickness and Surface Area of the Human Cortex and Their Relationship With Intelligence.” Cerebral Cortex 25, no. 6: 1608–1617. [DOI] [PubMed] [Google Scholar]
  41. Seyedsalehi, A. , Warrier V., Bethlehem R. A. I., Perry B. I., Burgess S., and Murray G. K.. 2023. “Educational Attainment, Structural Brain Reserve and Alzheimer's Disease: A Mendelian Randomization Analysis.” Brain: A Journal of Neurology 146, no. 5: 2059–2074. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Skrivankova, V. W. , Richmond R. C., Woolf B. A. R., et al. 2021. “Strengthening the Reporting of Observational Studies in Epidemiology Using Mendelian Randomisation (STROBE‐MR): Explanation and Elaboration.” BMJ 375, no. 10: n2233. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Skrivankova, V. W. , Richmond R. C., Woolf B. A. R., and Yarmolinsky J.. 2021. “Strengthening the Reporting of Observational Studies in Epidemiology Using Mendelian Randomization: The STROBE‐MR Statement.” JAMA 326, no. 16: 1614–1621. [DOI] [PubMed] [Google Scholar]
  44. Sudlow, C. , Gallacher J., Allen N., et al. 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, no. 3: e1001779. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Tadayon, E. , Pascual‐Leone A., and Santarnecchi E.. 2020. “Differential Contribution of Cortical Thickness, Surface Area, and Gyrification to Fluid and Crystallized Intelligence.” Cerebral Cortex (New York, N.Y. 1991) 30, no. 1: 215–225. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Vogel, J. W. , Alexander‐Bloch A. F., Wagstyl K., et al. 2024. “Deciphering the Functional Specialization of Whole‐Brain Spatiomolecular Gradients in the Adult Brain.” Proceedings of the National Academy of Sciences of the United States of America 121, no. 25: e2219137121. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Vuoksimaa, E. , Panizzon M. S., Chen C.‐H., et al. 2015. “The Genetic Association Between Neocortical Volume and General Cognitive Ability Is Driven by Global Surface Area Rather Than Thickness.” Cerebral Cortex 25, no. 8: 2127–2137. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Vuoksimaa, E. , Panizzon M. S., Chen C.‐H., et al. 2016. “Is Bigger Always Better? The Importance of Cortical Configuration With Respect to Cognitive Ability.” NeuroImage 129, no. April: 356–366. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Walhovd, K. B. , Krogsrud S. K., Amlien I. K., et al. 2016. “Neurodevelopmental Origins of Lifespan Changes in Brain and Cognition.” Proceedings of the National Academy of Sciences of the United States of America 113, no. 33: 9357–9362. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Wang, J. , Hill‐Jarrett T., Buto P., et al. 2024. “Comparison of Approaches to Control for Intracranial Volume in Research on the Association of Brain Volumes With Cognitive Outcomes.” Human Brain Mapping 45, no. 4: e26633. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Wierenga, L. M. , Langen M., Oranje B., and Durston S.. 2014. “Unique Developmental Trajectories of Cortical Thickness and Surface Area.” NeuroImage 87: 120–126. [DOI] [PubMed] [Google Scholar]
  52. Williams, C. M. , Peyre H., and Ramus F.. 2023. “Brain Volumes, Thicknesses, and Surface Areas as Mediators of Genetic Factors and Childhood Adversity on Intelligence.” Cerebral Cortex 33, no. 10: 5885–5895. [DOI] [PubMed] [Google Scholar]
  53. Yang, J. , Lee S. H., Goddard M. E., and Visscher P. M.. 2011. “GCTA: A Tool for Genome‐Wide Complex Trait Analysis.” American Journal of Human Genetics 88, no. 1: 76–82. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Yu, J. 2024. “Age‐Related Decline in Thickness and Surface Area in the Cortical Surface and Hippocampus: Lifespan Trajectories and Decade‐By‐Decade Analyses.” GeroScience 46, no. 6: 6213–6227. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Zhu, Z. , Zheng Z., Zhang F., et al. 2018. “Causal Associations Between Risk Factors and Common Diseases Inferred From GWAS Summary Data.” Nature Communications 9, no. 1: 224. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Figure S1: Bidirectional causal effects between cortical morphology and GCA with different MR methods comparison. The heat maps display both (a, b) forward and (c, d) reverse Mendelian Randomization analyses. Big or small asterisks denote statistical (p < 2.3e−3, Bonferroni correction) or nominal (p < 0.05) significance. Gray boxes indicate missing data due to limitations such as restricted sample size or insufficient number of filtered single nucleotide polymorphisms (SNPs) meeting the inclusion threshold.

Table S4: Instrument strength of genome‐wide significant SNPs used in forward GSMR analyses across absolute cortical regions after pleiotropy filtering.

Table S5: Instrument strength of genome‐wide significant SNPs used in reverse GSMR analyses across absolute cortical regions after pleiotropy filtering.

HBM-47-e70635-s001.docx (4.6MB, docx)

Table S1: Results of Fluid Intelligence‐Cortical morphology association analysis.

Table S2: Results of MR analysis.

Table S3: Results of bidirectional GSMR analysis.

HBM-47-e70635-s002.xlsx (52.1KB, xlsx)

Data Availability Statement

This study utilizes individual‐level genetic and imaging data from the UK Biobank (https://www.ukbiobank.ac.uk/). The genome‐wide association data for cortical regions were provided from our previously published studies, which can be accessed via the GWAS Catalog (https://www.ebi.ac.uk/gwas/publications/35113692 and https://www.ebi.ac.uk/gwas/publications/36893272). The genome‐wide association data for intelligence were obtained directly from the authors upon request, as a means of avoiding overlap with UKB samples, and the original source was from the Psychiatric Genomics Consortium (PGC) (https://pgc.unc.edu/) or the GWAS Catalog (https://www.ebi.ac.uk/gwas/publications/29942086). GSMR method and code are publicly available via the GCTA software repository on GitHub (https://github.com/JianYang‐Lab/gsmr/releases).

Articles from Human Brain Mapping are provided here courtesy of Wiley

RESOURCES