As hypothesized, we found that genetic factors contributed greatly to variation in surface area for almost all cortical parcellations, with heritabilities as high as 0.70 estimated from models with genetic and unique environmental variance components. Thus, genetic variation is an important determinant of individual differences in cortical surface area. These are likely to be genes related specifically to overall brain size, as opposed to body size more generally, because, although there were substantial reductions in regional heritabilities after adjusting for total surface area, we found a low correlation between height and total surface area in this sample (r = 0.24). Common environmental influences contributed to surface area measures in a small number of regions but generally accounted for less than 20% of the variance. Before any global adjustment, the heritabilities of regional surface areas (0.38 on average based on ACE models, 0.44 on average based on AE models) were significantly greater than zero but substantially smaller than that from an AE model of total surface area (0.95). Some of the difference may be due to greater measurement error for regional versus global measures, which would serve to increase unique environmental variance and decrease genetic variance estimates. This is supported by the fact that lobar heritabilities were intermediate in size between regional and global measures.
We found relatively little evidence for strong differences between individual parcellations in the degree of genetic and environmental contributions. Although heritabilities ranged from 0 to 0.70, there was generally considerable overlap in the confidence intervals. At the lobar level, we were able to test more directly for the reliability of differences between lobes in heritability estimates. One region in which unadjusted heritability was significantly lower for surface area was in the medial aspect of the temporal lobe. We did not observe such a large discrepancy for cortical thickness heritability in this region compared with other lobes, so to the extent that lower medial temporal lobe heritabilities for surface area are replicable and not due to greater measurement error in this region, this phenomenon may be related to environmental factors acting to expand or contract the numbers of neurons rather than their length and connections. In a previous family pedigree study using the same parcellation scheme to examine regional heritability of surface area (Winkler et al. 2010
), genetic influences on medial temporal lobe surface areas were also slightly lower than those of other regions as determined by averaging their reported regional heritabilities (0.38 for medial temporal lobe compared with an average of other regions of 0.58).
It remains to be seen what sorts of environmental factors might be more strongly related to variation among middle-aged men in surface area of this medial temporal region, but those associated with normal aging processes are likely candidates since a recent study found reduction in mean surface area across age groups in these same regions (Dickerson et al. 2009
) One might also speculate that environmental influences are more important in determining individual differences in cortical regions adjacent to subcortical structures such as the hippocampus which are particularly susceptible to toxic insults (e.g., hypoxia; Zola-Morgan et al. 1992
). It should be noted that before adjustment for total surface area, there was still a meaningful contribution of genetic factors to the surface area of all regions within the medial aspect of the temporal lobe. This is consistent with the finding of an association between specific genetic polymorphisms (of the MECP2 gene known to be affected in Rett syndrome) and surface area of a region within the fusiform gyrus in a recent map-based study (Joyner et al. 2009
). After adjustment, however, only 2 of the medial temporal regions (right entorhinal and left parahippocampal) had significant heritability estimates.
Extending our examination of variability among regions in genetic and environmental influences, we also examined whether regionally specific influences remained after controlling for genetic and environmental contributions to overall (i.e., total) surface area. We chose total surface area as our global covariate rather than intracranial or total brain volume since prior work suggests independent genetic influences on surface area and cortical thickness, both of which would contribute to intracranial volume and total brain volume (Panizzon et al. 2009
). Consistent with our hypothesis, regional heritabilities were greatly reduced after controlling for genetic contributions to total surface area, although most remained significantly larger than zero under a model with only A
effects. When region-specific C
effects were also in the model, however, significant genetic influences could not be demonstrated in most regions. Winkler et al. (2010)
also reported reductions in heritability estimates following adjustment for global variables, although their decreases were smaller (17–23% reductions). It is unclear what might account for the differences between these 2 studies, given that both used the same parcellation and image processing methods. The studies differed in being twin versus extended pedigree designs and in having a narrow versus a wide age range; however, it is not clear that these factors would account for the difference in the effect of adjusting for total surface area.
The effect of adjustment for total surface area on regional surface area heritability estimates was slightly greater than the effect of adjustment for mean cortical thickness on regional cortical thickness heritability estimates. After adjustment, there were no notable differences among lobar thickness heritabilities, whereas there were among the lobar surface area heritabilities. Specifically, adjusted left frontal surface area heritability was significantly greater than several other lobes. This suggests that there is a large amount of remaining genetic variance in left frontal surface area after accounting for genetic influences on total surface area. Since aging is known to have particular impact on the size of the frontal lobe (Raz et al. 2004
; Fjell et al. 2009
), it may be that aging-relevant genes are having strong regional influence.
Still, overall, we found that genetic effects were reduced after controlling for associations with global variables. Furthermore, almost all phenotypic correlations between total surface area and regional ROIs were strong, ranging from 0.12 to 0.96 (P
< 0.0001). These phenotypic correlations were found to be driven by significant genetic covariances ranging from 0.22 (95% CI = 0.06; 0.33) to 0.84 (95% CI = 0.06; 0.92), meaning that many of the genes responsible for variation in the area of each individual region are expected to be the same as those responsible for variation in surface area as a whole. Our finding of greater environmental influences on regional surface area after controlling for genetic contributions to determination of total surface area could be meaningful, but it also could point to limitations of using traditional cortical parcellation schemes. To the extent that genes may act developmentally on areas that cross the regional boundaries we enforced with the Desikan–Killiany atlas (Rubenstein and Rakic 1999
), we may have underestimated regional heritabilities and the extent to which there are important variations in this regional heritability across the cortex even after accounting for global effects. It also may be the case that regional variability will be more evident in the degree to which genetic or environmental influences change during aging. Our sample is characterized by a narrow age range and follow-up MRIs are being performed to detect particular regions in which genetic influences may increase or decrease with age.
Our results suggest high heritability of total cortical surface area and an apparent role of both genes and environment in the determination of individual differences in regional surface area measures. Total genetic influences on individual variation in surface area for any given region were generally high since there were both influences shared with total surface area as well as smaller, but generally significant, unique genetic influences on the area of the particular region. Although heritability is a population-based statistic having to do with variations among individuals, our findings in middle-aged men are broadly consistent with neurodevelopmental evidence of a protomap that establishes, very early on, relative position and numbers of cortical columns in human-specific cytoarchitectonic regions (Rubenstein and Rakic 1999
; Rakic 2009
). Genes that impact cell cycling in the first phase of symmetric divisions of neural stem cells could have a large effect on total surface area, and evolutionary effects on cortical surface area are thought to have acted during this phase (Rakic 1995
). Similarly, genes that affect the organization of the protomap and regulate gradients of transcription factors and signaling molecules clearly have effects on the relative size of cortical regions (O'Leary et al. 2007
). Regional cortical surface area in the mature adult human is likely a product of both these early determinants of numbers of neurons and subsequent effects (both growth and shrinkage) on synaptogenesis, dendritic arborization, intracortical myelination, and connectivity. Our data suggest that these subsequent effects are both genetic and environmental, perhaps related to genes involved in synaptogenesis and programmed cell death and to life experiences that may serve to increase connections within functional regions. Stochastic processes may also contribute to nongenetic variation in neural structure between individuals (Macagno et al. 1973
). Relative expansion of surface area from macaque to human and from human infants to human adults is not uniform across cerebral cortex (Hill et al. 2010
), with particularly large expansion in left dorsal frontal cortex and relatively less expansion in other regions, such as medial temporal cortex. Perhaps consistent with these differences in rates of expansion during development, which are thought to reflect differential maturity at birth, we found differences between these same regions in the relative contributions of genetic versus environmental influences.
There are several limitations to our study, which guide future work. First, our sample only included male twins, so the generalizability of our findings to women is unknown. Second, despite our very large sample size, we were underpowered to make inferences about shared environmental effects (Visscher et al. 2008
). Most of the estimates of shared environmental effects were quite low. However, in a small number of cases the estimates were high enough (even if nonsignificant) to suggest that heritability estimates based on AE models might be biased for those regions. These regions were specifically noted in Supplementary Table 1.
. This highlights the necessity of beginning with full models that include C
effects so that one can most accurately model the full range of genetic and environmental sources of variation and then make valid inferences about the likely contributions of purely genetic effects (Kendler and Neale 2009
). Third, although we chose a widely used cortical parcellation scheme, these boundaries may not be optimal for examining genetic contributions. Future studies will address this limitation using continuous maps of the heritability of area expansion or contraction relative to a standard template at each point on the cortex. Fourth, although we identified some regions (e.g., within the medial temporal lobe) that had relatively low heritability of surface area compared with other regions, we did not undertake an examination of whether the genes that influence surface area in one region (or set of regions) are distinct from those that influence the surface area of other regions. This is an important future direction that will help to identify independent structural phenotypes for gene association studies. Fifth, although our results are consistent with neurodevelopmental models, we can only definitively conclude from our cross-sectional data that these are the patterns of genetic influences on cortical surface area in middle age.
In this large-scale twin study, we found high heritability for global cortical surface area and moderate genetic contributions to variations in regional surface area. We found some evidence for stronger environmental contributions to medial temporal lobe surface area and stronger genetic contributions to frontal lobe surface area compared with several other lobes. Due to the substantial genetic covariance between total surface area and the area of specific regions, the influence of genetic factors on individual differences was reduced after controlling for global measures, although most regions had some unique genetic contributions and substantial unique genetic effects on surface area were still observed in a few regions, such as left frontal lobe. Even if unique genetic effects were not found for some specific regions, that does not mean that there are no genetic influences on the surface area of those regions; rather, it indicates that the genetic variance is shared with that of global surface area. The results highlight the importance of examining genes that have widespread effects in order to understand individual variation in surface area but also suggest that future work examining environmental influences on medial temporal lobe surface area and the effect of particular genes on the relative area of the left frontal cortex could be fruitful.