Here we report on detailed three-dimensional maps revealing how brain structure is influenced by individual genetic differences. A genetic continuum was detected, in which brain structure was increasingly similar in subjects with increasing genetic affinity. Genetic factors significantly influenced cortical structure in Broca’s and Wernicke’s language areas, as well as frontal brain regions (r2 MZ > 0.8, p < 0.05). Preliminary correlations were performed suggesting that frontal gray matter differences may be linked to Spearman’s g, which measures successful test performance across multiple cognitive domains (p < 0.05). These genetic brain maps reveal how genes determine individual differences, and may shed light on the heritability of cognitive and linguistic skills, as well as genetic liability for diseases that affect the human cortex.
The degree to which genes and environment determine brain structure and function is of fundamental importance. Largescale neuroimaging and genetic studies are beginning to uncover normal and disease-specific patterns of gene and brain function in large human populations. Yet, little is known about the genetic control of human brain structure, and how much individual genotype accounts for the wide variations among individual brains. Recent reports show that many cognitive skills are surprisingly heritable, with strong genetic influences on IQ, verbal and spatial abilities, perceptual speed and even some personality qualities, including emotional reactions to stress. These genetic relationships persist even after statistical adjustments are made for shared family environments, which tend to make members of the same family more similar. Given that genetic and environmental factors, in utero and throughout lifetime, shape the physical development of the brain, we aimed to map patterns of brain structure that are under significant genetic control, and determine whether these structural features are linked with measurable differences in cognitive function. The few existing studies of brain structure in twins suggest that the overall volume of the brain itself and some brain structures, including the corpus callosum and ventricles, are somewhat genetically influenced, whereas gyral patterns, observed qualitatively or by comparing their twodimensional projections, are much less heritable. To make the transition from volumes of structures to detailed maps of genetic influences, advances in brain mapping technology have allowed the detailed mapping of structural features of the human cortex, including gray matter distribution, gyral patterning, and brain asymmetry. These features each vary with age, gender, handedness, hemispheric dominance and cognitive
performance in both health and disease. Composite maps of these features, generated for large populations, can reveal patterns not observable in an individual. Such patterns include statistical maps that show whether heredity and nongenetic factors are involved in determining specific aspects of brain structure.
Among the structural features that are genetically regulated and have implications for cortical function is the distribution of gray matter across the cortex. This varies widely across normal individuals, with developmental waves of gray matter gain and loss
subsiding by adulthood, and complex deficit patterns observed in Alzheimer’s disease, schizophrenia, and healthy subjects at genetic risk for these disorders. In this study, we began by comparing the average differences in gray matter (Fig. 1) in groups of unrelated subjects, dizygotic (DZ) and monozygotic (MZ) twins (see Methods). Although both types of twins share gestational and postgestational rearing environments, DZ twins share, on average, half their segregating genes, whereas MZ twins are normally genetically identical (with rare exceptions due to somatic mutations).
We found that brain structure is under significant genetic control, in a broad anatomical region that includes frontal and language-related cortices. The quantity of frontal gray matter, in particular, was most similar in individuals who were genetically alike; intriguingly, these individual differences in brain structure were tightly linked with individual differences in IQ (intelligence quotient). The resulting genetic brain maps reveal a strong relationship between genes, brain structure and behavior, suggesting
that highly heritable aspects of brain structure may be fundamental in determining individual differences in cognition.
Fig. 1. Genetic continuum of similarity in brain structure. Differences in the quantity of gray matter at each region of cortex were computed for identical and fraternal twins, averaged and compared with the average differences that would be found between pairs of randomly selected, unrelated individuals (blue, left). Color-coded maps show the percentage reduction in intra-pair variance for each cortical region. Fraternal twins exhibit only 30% of the normal inter-subject differences (red, middle), and these affinities are largely restricted to perisylvian language and spatial association cortices.
Genetically identical twins display only 10–30% of normal differences (red and pink) in a large anatomical band spanning frontal (F), sensorimotor (S/M) and Wernicke’s (W) language cortices, suggesting strong genetic control of brain structure in these regions, but not others (blue; the significance of these effects is shown on the same color scale).
RESULTS
MZ within-pair gray matter differences were almost zero (intraclass r ≈0.9 and higher, p < 0.0001 corrected; Fig. 1, right column) in a broad anatomical band encompassing frontal, sensorimotor and linguistic cortices, including Broca’s speech
and Wernicke’s language comprehension areas. Because MZ twins are genetically identical, any regional differences would be interpreted as being attributable to environmental effects or gene–environment interactions. Meanwhile, sensorimotor and
parietal occipital but not frontal territory was significantly more similar in DZ twins than random pairs (Figs. 1 and 2). Affinity was greatest in the MZ pairs, suggesting a genetic continuum in the determination of structure.
A genetic continuum
In population genetics, a feature is heritable if it shows a genetic cascade in which within-pair correlations (Fig. 2) are highest for MZ twins, lower for DZ twin pairs and lowest of all for unrelated subjects. As we expected specific regions of cortex to be more heritable than others, we plotted these correlations across the cortex (Fig. 2) and assessed their statistical significance (see Methods). This uncovered a successively increasing influence of common genetics. A 95–100% correlation was revealed between MZ twins in frontal, linguistic and parietooccipital association cortices, suggesting individual differences in these regions can be largely attributed to genetic factors. DZ twins, who share half their genes on average, were still nearidentical in the supramarginal component of Wernicke’s language area (r2 = 0.7–0.8; p < 0.0001) and highly similar in parieto-occipital association areas (60–70% correlation; p < 0.001). They also showed significantly less affinity (p < 0.05) in a sharply defined region that included the frontal cortices (p > 0.05). The resulting pattern of twin correlations suggests substantial genetic influences in this region.
Mapping genetic correlations
With a sample size of only 40 twins, heritability coefficients cannot be estimated precisely, and limited statistical power precludes the detection of differences in heritability between individual regions of cortex. Preliminary comparisons of MZ and DZ correlations suggested that frontal, sensorimotor and anterior temporal cortices were under significant genetic control (p < 0.05, rejecting the hypothesis that heritability (h2) = 0; one-tailed). Preliminary estimates suggested that discrete middle frontal regions, near Brodmann areas 9 and 46 (ref. 27) displayed a 90–95% genetic determination of structure (that is, h2 ≈0.90–0.95). Many regions are under tight genetic control (bilateral frontal and sensorimotor regions, p < 0.0001; Fig. 3). Due to small sample sizes, any provisional heritability estimates should be interpreted with caution, but were comparable with twinbased estimates for the most highly genetically determined human traits, including fingerprint ridge count (h2 = 0.98), height (h2 = 0.66) and systolic blood pressure (h2 = 0.57)28. Genetic influences here are far higher than for the most environmentally influenced characters (such as, social maturity, for which h2 = 0.16; ref. 29).
Language asymmetry
Given the high heritability of reading skills and performance on linguistic tasks, we were interested in whether the structure of
language cortices would also be heritable, and if so, whether heritability would be higher in the left hemisphere, which is dominant for language in most (right-handed) subjects. The heritability of brain size does not vary markedly by hemisphere, but one MZ twin study of cortical surface areas suggested the possibility of differing left/right genetic influences. Intriguingly, when a three-dimensional map was made subtracting the heritability of the structures in one hemisphere from their counterparts in the other, differences in Wernicke’s language area were highly significant, even after hemispheric differences in
gyral patterning were directly accommodated. Heritability was significantly greater on the left (p < 0.05, corrected). Although no other regions displayed this lateralized effect, we cannot infer that there are no such asymmetries elsewhere, as we were underpowered, in a sample of 40, to make general comparisons of heritability among cortical regions. Nonetheless, the asymmetry in language-related cortex was significant, and was corroborated by the genetic correlation maps as well (Figs. 2 and 3), in that Wernicke’s and Broca’s speech area displayed highly significant heritability on the left (p < 0.0001) but not on the right (p > 0.05).
Fig. 2. Correlation between twins in gray matter distribution.
Genetically identical twins are almost perfectly correlated in their gray matter distribution, with nearidentity in frontal (F), sensorimotor (S/M) and perisylvian language cortices. Fraternal twins are significantly less alike in frontal cortices, but are 90–100% correlated for gray matter in perisylvian language-related cortex, including supramarginal and angular territories and Wernicke’s language area (W). The significance of these increased similarities, visualized in color, is related to the local intra-class correlation coefficents (r).
Cognitive linkages
To make a preliminary assessment of whether gray matter differences between subjects were significantly linked with differences in cognitive function, a cognitive measure termed ‘Spearman’s g’ was assessed for all 40 MZ twins. Like IQ, this widely-used measure isolates a component of intellectual function common to multiple cognitive tests, and has been shown to be highly heritable across many studies, even more so than specific cognitive abilities (h2 = 0.62 (ref. 4, compare with ref. 24 [Author: As meant?]); h2 = 0.48 (ref. 33); h2 = 0.6–0.8 (ref. 34, compare with refs. 35–38 [Author: As meant?])). We found that differences in frontal gray matter were significantly linked with differences in intellectual function (Table 1; p < 0.0044; p < 0.0176 after correction for multiple tests) as quantified by g, which was itself also highly heritable (h2 = 0.70±0.17 in this study). Although these preliminary correlations should be evaluated in a larger sample, a recent abstract also observed that differences in regional gray matter volume were significantly correlated with differences in IQ, in a sample of 28 pediatric MZ twin pairs (mean age, 12.1 years) studied volumetrically (E. Molloy et al., 7th Annual Meeting of the Organization for Human Brain Mapping, 447, Brighton, England, 2001). In frontal brain regions, a regionally specific linkage has previously been found39 between g and metabolic activity measured by positron emission tomography (PET), suggesting that general cognitive ability may in part derive from a specific frontal system important in controlling diverse forms of behavior. Frontal regions also show task-dependent activity in tests involving working (short-term) memory, divided and sustained attention, and response selection40. Genetic factors may therefore contribute to structural differences in the brain that are statistically linked with cognitive differences. This is especially noteworthy, as cognitive performance seems to be linked with brain structure in the very regions where structure is under greatest genetic control (Figs. 2 and 3). This emphasizes the pronounced contribution of genetic factors to structural and functional differences across individuals, as detected here in frontal brain regions.
Fig. 3. Significance of genetic control of gray matter distribution.
Brain regions for which cortical gray matter distribution is under significant genetic control are shown in red. Frontal (F) and lateral temporal (T) regions show significant heritability, consistent with their near-identity in identical twins (Fig. 2) and the weaker patterns of correlations observed in fraternal twins, who have less similar genotypes. Wernicke’s area shows significantly higher heritability in the left hemisphere (Wleft), which is generally dominant for language function (p < 0.05 for asymmetry).
DISCUSSION
Genetic brain mapping
Influences of nature and nurture in the determination of individual brain structure are not independent; genes necessarily operate through the environment, particularly if they concern susceptibilities to environmental stressors or hazards. Nonetheless, twin designs can reveal the degree to which heredity is involved, and the extent to which individual differences can be attributed to genetic and environmental factors. Whereas genetic influences strongly determine aspects of intellect and its closely related traits, the extent to which genes shape brain structure is heterogeneous. The gene control of brain structure displays asymmetries that mirror asymmetries in the brain’s functional organization, and genes strongly control a broad anatomical band encompassing frontal, linguistic and sensorimotor cortex [Author: OK?]. As with any polygenic trait, multiple genes are likely to combine additively or interact at the same or different loci (dominance or epistasis) to structure the adult brain. Future studies mapping quantitative trait loci are likely to provide insight into the genes that determine brain structure, and neurocognitive skills that in some cases depend on it. The tight coupling of brain structure and genetics, particularly in frontal brain regions, may contribute to the genetic liability for diseases that affect the integrity of the cortex. Frontal gray matter deficits are found in both schizophrenia patients and their healthy first-degree relatives, and there is a strong familial risk for many neurodegenerative diseases that affect the frontal cortex, including frontotemporal dementia and primary progressive aphasia. The genetic cascades implicated in these diseases may or may not overlap with those involved in cortical determination, but the genetic coupling of brain structure we report here may result in increased familial liability to cortical degenerative disease, specifically in highly genetically determined frontal regions. By controlling for nongenetic factors, twin studies may offer unique advantages in isolating disease-specific differences in these highly heritable brain regions. Genetic brain maps, such as those introduced in this study, may reveal how genes determine individual differences in brain structure and function. Additional linkages were observed between cortical differences and intellectual function, suggesting that genetic brain mapping may shed light on the heritability of cognitive and linguistic skills, as well as familial liability for diseases that affect the human cortex.
METHODS
Subjects. 40 healthy normal subjects, consisting of 10 monozygotic (MZ) and 10 dizygotic (DZ) twin pairs were drawn from a twin cohort consisting of all the same-sex twins born in Finland between 1940 and 1957, inclusive, in which both members of each pair were alive and residing in Finland as of 1967 (n = 9,562 pairs, 2,495 MZ; 5,378 DZ; 1,689 of unknown zygosity)14. Pairs were excluded if either member or any of their first-degree relatives had a history of hospitalization, medicine prescriptions, or work disability due to a psychiatric indication from 1969 to 1991. MZ pairs were matched with the DZ pairs for age (48.2 ± 3.4 years), gender, handedness, duration of cohabitation and parental social class.
Each zygosity group included five male pairs and five female pairs. The study protocol was reviewed and approved by the institutional review
boards of the University of California (Los Angeles), and the National Public Health Institute of Finland, and all subjects signed IRB-approved
informed-consent forms.
Cognitive testing. Each twin in a pair received a neuropsychological test battery from a different examiner blind to zygosity [Author: OK?]. All subjects received the same test battery in a fixed order. Seventeen different cognitive domains were assessed, including verbal and spatial working memory, selective and divided attention, verbal knowledge, motor speed and visuospatial ability. A measure of general cognitive ability, in the form of an overall IQ was prorated from age-scaled scores on the Vocabulary, Similarities, Block Design, and Digit Symbol subtests of the Wechsler Adult Intelligence Scale—Revised (WAIS-R; D. Wechsler, WAISR Manual, Psychological Corporation, Cleveland). This measure exhibited 98% correlation with full-scale IQ based on all of the WAIS-R subtests.
Zygosity. For all pairs, zygosity was determined by DNA analysis using the following markers: DIS80 (20 alleles), DI7S30 (13 alleles), apoB (20 alleles), COL2A1 (10 alleles), vWA (9 alleles) and HUMTH01 (6 alleles). Assuming an average heterozygosity rate of 70% per marker, this procedure will falsely classify a DZ pair as MZ in approximately 1/482 cases.
Magnetic resonance imaging. Three-dimensional maps of gray matter and models of cortical surface anatomy were derived from high-resolution three-dimensional (2562 124 resolution) T1-weighted (MPRAGE) magnetic resonance images acquired from all 40 subjects on a 1.5 T
Siemens scanner (Siemens, [Author: City of manufacturer?] New York).
Image processing and analysis. A radio-frequency bias field orrection algorithm eliminated intensity drifts due to scanner field inhomogeneity.
A supervised tissue classifier generated detailed maps of gray matter, white matter and cerebrospinal fluid (CSF). Briefly, 120 samples of each tissue class were interactively tagged to compute the parameters of a aussian mixture distribution that reflects statistical variability in the intensity of each tissue type. A nearest-neighbor tissue classifier assigned each image voxel to a particular tissue class (gray, white or CSF),
or to a background class. The inter/intra-rater reliability of this protocol, and its robustness to changes in image acquisition parameters, have been described previously. The error variance, that is, the variation associated with map error and reproducibility, was further confirmed to be small by the extremely high intra-class correlations in the MZ pairs (around 1.0), which would not otherwise be obtainable (Fig. 2). Gray matter maps were retained for subsequent analysis.
Three-dimensional cortical maps. To facilitate comparison and pooling of cortical data across subjects, a high-resolution surface model of the cortex was automatically extracted for each subject. Thirty-eight gyral and sulcal boundaries, representing the primary gyral pattern of each subject, were digitized on the highly magnified three-dimensional surface models. Gyral patterns and cortical models were used to compute a three-dimensional vector deformation field, which reconfigures each subject’s anatomy to the average configuration of the entire group (n = 40), matching landmark points, surfaces and curved anatomic interfaces. Data were accordingly averaged or compared, to the maximum possible degree, across corresponding cortical regions12. Additional three-dimensional vector deformation fields reconfigured one twin’s anatomy into the shape of the other, matching landmark points, surfaces and curved anatomic interfaces the pair of three-dimensional image sets. Given that the deformation maps associate cortical locations with the same relation to the primary folding pattern across subjects, a local measurement of gray matter density was made in each subject and averaged across equivalent cortical locations.
Gray matter mapping. To quantify local gray matter, we used a measure termed ‘gray matter density,’ which has been used in previous studies to compare the spatial distribution of gray matter across subjects. This measures the proportion of tissue that segments as gray matter in a small region of fixed radius (15 mm) around each cortical point. Given the large anatomic variability in some cortical regions, high-dimensional elastic matching of cortical patterns21, constrained by all three-dimensional sulcal models, associated measures of gray matter density from homologous cortical regions across subjects. Maps of intra-pair gray matter differences, generated within each MZ and DZ
pair, were subsequently elastically realigned for averaging across the 10 pairs within each group, before inter-group comparisons.
Mapping genetic correlations and asymmetry. Intra-class correlation between pairs of each zygosity was computed at each cortical point, after testing for heteroscedastic variance across each group. First, to assess whether it was significantly non-zero, broad-sense heritability was computed using Falconer’s method22 to determine all genic influences on the phenotype (with heritability, h2, defined as twice the difference between MZ and DZ intra-class correlation coefficients). Because nongenetic familial effects contribute to the resemblance between relatives, such effects were accommodated, if not entirely eliminated, by assuming the same common environmental variance for MZ and DZ pairs (compare with ref. 4). For this study, random field models were preferred as opposed to a full structural equation model (for example, ref. 23) given the low degrees of freedom per point available to estimate dominance and epistatic variance terms and reject simpler models based on the available database of 40 scans. Interaction and gene–environment covariance terms, as well as unique and shared environment factors, may be estimable with a more general familial design, an adoption design, or by using sample sizes much larger than available in the present study23 compare with 24). The significance of genetic effects was computed pointwise by reference to an analytical null distribution (F-test) and was confirmed separately by assembling an empirical null distribution using 1,000,000 random pairings to avoid assuming bivariate normality. Allmap-based inferences were corrected for multiple comparisons by permutation. We used permutations to make statistical inferences that were not based on any assumptions about the error covariances. To correct for the multiple comparisons implicit in our brain maps we established the null distribution of the largest statistic over the voxels analyzed. By adopting the critical threshold of this largest statistic, we could then maintain both strong and weak control over false-positive rates over the voxels analyzed. Asymmetric heritability was tested by computing a set of 40 flows driving each subject’s left hemisphere model onto the right, matching gyral patterns, and computing a field of heritability differences. This field was compared with its standard error (pooled across contralateral cortical points, after testing equality of variance across hemispheres (compare with ref. 25). Regions of asymmetric heritability were detected and their significance was assessed by permuting the covariate vector coding for hemisphere. Maps of MZ intra-pair gray matter differences associated with intra-pair differences in the cognitive measure, Spearman’s g, were generated by elastically realigning three-dimensional maps for averaging across all MZ twin pairs and modeling g as a continuous covariate for linkage with local gray matter distribution. Maps identifying these linkages were computed point-wise across the cortex and assessed statistically by permutation by computing the area of the average cortex with statistics above a fixed threshold in the significance maps (P < 0.01). Null distributions were assembled from random pairings of unrelated subjects. We preferred this to an analytical null distribution to avoid assuming that the smoothness tensor of the residuals of the statistical model was stationary across the cortical surface26. In each case, the covariate vector was permuted 1,000,000 times on an SGI RealityMonster supercomputer with 32 internal R10000 processors. An algorithm was then developed to report the significance probability for each map as a whole12, so the significance of intra-pair variance reduction by zygosity, heritability, asymmetry and cognitively linked patterns of gray matter distribution could be assessed after the appropriate correction for multiple
comparisons.