Multiple bases of human intelligence revealed by cortical thickness and neural activation

Abstract

We hypothesized that individual differences in intelligence (Spearman's g) are supported by multiple brain regions, and in particular that fluid (gF) and crystallized (gC) components of intelligence are related to brain function and structure with a distinct profile of association across brain regions. In 225 healthy young adults scanned with structural and functional magnetic resonance imaging sequences, regions of interest (ROIs) were defined on the basis of a correlation between g and either brain structure or brain function. In these ROIs, gC was more strongly related to structure (cortical thickness) than function, whereas gF was more strongly related to function (blood oxygenation level-dependent signal during reasoning) than structure. We further validated this finding by generating a neurometric prediction model of intelligence quotient (IQ) that explained 50% of variance in IQ in an independent sample. The data compel a nuanced view of the neurobiology of intelligence, providing the most persuasive evidence to date for theories emphasizing multiple distributed brain regions differing in function.

Figure 1.

Structural and functional correlates of intelligence and their contrasting laterality. A, Correlations between cortical gray matter thickness and WAIS FSIQ. The color bar indicates the statistical significance of the correlations (left, negative correlation; right, positive correlation). Lines point to ROIs with high statistical significance (p < 0.001 uncorrected): ATC, OTC, ITC, MTC, and LPC. B, Correlations between cortical activation level during reasoning tasks and WAIS FSIQ. The color bar indicates the statistical significance of the correlations. Lines point to ROIs with high statistical significance (p < 0.001 uncorrected): ACC, PFC, and PPC. C, D, Hemispheric (C) and lobar (D) area sizes of the structural correlates manifesting left dominance. E, F, Hemispheric (E) and lobar (F) volume sizes of the functional correlates demonstrating bilateral symmetry. The sizes of the cortical areas (C, D) and the activation clusters (E, F) were defined using the correlations of statistical significance (p < 0.001 uncorrected). L, Left; R, right; Fro, frontal; Tem, temporal; Ins, insular; Par, parietal; Occ, occipital lobes.

Figure 2.

General, crystallized, and fluid intelligence selectively explained by structural and functional ROIs. A, C, The radar graphs show simple correlations of the cortical thickness (A) or peak t score (C) of each ROI with the major components of intelligence: g (gray line), gC (orange line), and gF (blue line). B, D, The bar graphs display the multiple correlations of all structural (B) or functional (D) ROI values with the intelligence components. Each bar or line indicates the amount of explained variance (R²) of individual performance in the intelligence component scores. g, Principal component of all WAIS subtests and RPM; gC, principal component of WAIS Verbal Comprehension subtests; gF, principal component of WAIS Perceptual Organization subtests and RPM.

Figure 3.

Intellectual domain effects on cortical thickness changes as a function of IQ level. A, Cortical thickness differences between adjoining levels of IQ as affected by intelligence criteria and brain lobes. The superior, high, and average IQ groups were evenly divided according to four intelligence criteria, FSIQ, VIQ, PIQ, and RPM scores. The cortical thickness of each lobe is represented by the averaged value of all ROIs within the lobe. Sup., Superior; Avg., average. *p < 0.05; **p < 0.01; ***p < 0.001, two-tailed t test. B, C, Cortical thickness deviations from the thickness of the average IQ group used as zero reference. VIQ groups are better described by a linear or quadratic function, whereas PIQ groups are better described by a logarithmic one. The brain maps show absolute thickness changes at each cortical point, based on VIQ and PIQ levels.

Figure 4.

Prediction of psychometric IQ from neurobiological measures. A, Specification of a neurometric IQ model. To specify potential predictors, we correlated IQ with cortical thickness (top) or activation level (bottom) at each voxel, and thresholded (p < 0.001) to define ROIs. We reduced redundancy by averaging ROIs that covaried (p < 0.001), leaving brain volume and three ROI-based candidate predictor variables. Interaction terms allowed sex to moderate the relation between the MR-based variables and IQ. To specify a model, we screened potential predictors by regressing IQ on them simultaneously, retaining five as the model. B, Test of the model. We estimated IQ using the model in an independent sample. Estimated IQ correlated with measured IQ, indicating successful prediction. For the strict model (fixed parameters), parameter values were obtained by fitting the model (regression) in the datasets for model specification.