Our analysis of DTI asymmetries, in this large population (N=374 adults), had 3 main findings. First, frontal and temporal regions had significant asymmetries in FA. Frontal lobe FA is greater in the right hemisphere, but left temporal lobe FA is greater than on the right. The mean difference in this large sample reached 0.15 (which is very high, considering that FA runs on a scale of 0 to 1). Second, in a regional analysis of FA asymmetry, genetic factors accounted for 33% of the variance in asymmetry in the inferior fronto-occipital fasciculus, 37% of the variance in the anterior thalamic radiation, and 20% of the variance in the forceps major and the uncinate fasciculus. Shared environmental factors accounted for ~15% of the variance in the cortico-spinal tract and ~10% of the variance in the forceps minor. Asymmetries are therefore influenced by both genetic and environmental factors. Results were similar regardless of the anisotropy measure used (FA versus tGA). Finally, the frontal lobe FA asymmetry had higher variance in men than in women, but only a small proportion of voxels showed sex differences in the average level of asymmetry.
As expected from twin studies of brain structure (Brun et al., 2009
; Thompson et al., 2001
), monozygotic twins showed higher similarities in the intra-class correlation maps than dizygotic twins suggesting that fiber asymmetries are genetically influenced. This is in line with a large body of work by Annett, who proposed that there might be a single “right-shift” gene influencing the degree of cerebral dominance and lateralized behavior such as handedness (Annett, 1998
A preliminary map of heritability, based on Falconer’s heritability formula (), crudely estimates the genetic proportion of variance as twice the difference between the monozygotic twin correlations and the dizygotic twin correlations. As this initial analysis detected genetic involvement in fiber asymmetry, we fitted structural equation models to data from lobar regions of interest. A large proportion of the variance in fiber asymmetry in the sample is due to genetic factors in the frontal lobes, but not elsewhere in the brain. Voxelwise genetic models confirmed the genetic effects on the frontal lobe asymmetries, and also suggested high genetic contributions in some temporal lobe regions. The temporal lobe effects did not reach significance in the lobar analyses, and require replication.
Regions with high genetic contributions to fiber asymmetry (frontal lobes) were also the regions where asymmetry is greatest. The finding that frontal lobe FA is greater in the right hemisphere, but greater on the left for the temporal lobes, confirm prior DTI reports of frontal and temporal white matter asymmetries. In general, prior studies focused on specific tracts, e.g., the corticospinal tract (Westerhausen et al., 2007
) and the arcuate fasciculus, which is involved in language processing (de Jong et al., 2009
; Rodrigo et al., 2007
). A voxelwise analysis (Buchel et al., 2004
) suggested left greater than right FA white matter asymmetries in the arcuate fasciculus and found asymmetries contralateral to the dominant hand (i.e., higher on the left in right-handers) in tracts innervating the precentral gyrus (as expected, given the crossing of the cortical motor circuitry). Frontal and temporal white matter already show left greater than right FA in early infancy (Dubois et al., 2008
), suggesting greater myelination in the left hemisphere. Some developmental studies found that frontal FA differences between the two hemispheres diminish as the brain develops, but temporal lobe asymmetries persist (Barnea-Goraly et al., 2005
). These asymmetries may relate to the functional lateralization of higher-level cognitive processes such as spatial association and language, but our regressions with global cognitive measures did not reveal any associations.
Early studies of anatomical asymmetry noted a natural petalia
(torquing) of the brain, that shifts right hemisphere structures anterior to their left hemisphere counterparts (Kimura, 1973
; Toga and Thompson, 2003
). Post mortem studies found volumetric asymmetries in the planum temporale
, part of a temporal lobe auditory and language processing area (Geschwind and Levitsky, 1968
). More recently, a large MRI study of 142 young adults confirmed leftward volume asymmetries in posterior language areas, and rightward asymmetries in the cingulate gyrus and caudate nucleus (Watkins et al., 2001
). Surface-based analysis methods that adjust for the effects of structural translocation in space (e.g., torquing) also found leftward asymmetries in the Heschl’s gyrus and planum temporale
(Lyttelton et al., 2009
). Deformation-based morphometry studies have used the theory of random Gaussian vector fields to detect brain asymmetries, and have been used to detect statistical departures from the normal level of brain asymmetry (also termed “dissymmetry”; Thirion et al., 2000
; Lancaster et al., 2003
). In addition, some “apparently” lateralized effects in brain mapping may arise due to hemispheric differences in the statistical power to detect effects. This is inevitable, as the structures in the two hemispheres have different patterns of anatomical variability (Thompson et al., 1998
; Fillard et al., 2006
FA asymmetry had a higher variance in men than women, but there was only limited evidence for differences in the overall level of asymmetry (i.e., the sex difference formally passed the FDR criterion for significance but showed differences in less than 1% of the brain). Many studies report greater anatomical
asymmetries in men than women (reviewed in Toga and Thompson, 2003
). A voxel-based MRI study of 465 normal adults found sex differences in gray matter volumes and concentrations but no effects of handedness on the level of asymmetries (Good et al., 2001
). In a small sample (N=20), Szeszko et al. (2003)
reported that women had higher FA in the left frontal lobes compared to men, and a general leftward asymmetry of FA. No hemispheric asymmetry was detected in men. The level of leftward asymmetry in women was associated with better verbal comprehension and memory functioning. This result is surprising, given our finding of strong R>L asymmetry for FA; the magnitude of this effect is quite large in our much larger sample of subjects. Our finding of significant but limited sex differences could be due to our efforts to create a template that minimizes the structural differences between the hemispheres. This reduces the influence of “brain shape” on the asymmetries, which may have diminished any sex differences.
With fMRI, Shaywitz et al. (1995)
found significant sex differences in the phonological processing of language. Brain activation in men was lateralized to the left inferior frontal gyrus, while women engaged more diffuse neural systems involving both the left and right inferior frontal gyri. Men and women also have brain regions in which regional volumes correlate with intelligence. Haier et al. (2005)
found that women showed more white matter and fewer gray matter areas with volumes correlated with intelligence, when compared to men. IQ correlates with FA in normal subjects, and both IQ and FA are genetically influenced (Chiang et al., 2009a
; Kochunov et al., 2009). Because of this, we also regressed FA asymmetry against IQ, but no correlations survived FDR correction. IQ may relate closely to FA in specific brain regions, but not so much to its asymmetry.
A premise of any voxel-wise analysis of brain asymmetry is that there is a structural homology between white matter structures in the left and right hemispheres. For the major white matter tracts, such as the corpus callosum, fornix, and optic radiations, this assumption is tenable. The registration methods proposed here are likely to adjust for any macroscopic shape differences that get in the way of pairing homologous anatomy, where it exists, on both sides of the brain. Even so, as in all studies mapping brain asymmetry, there will always be a set of structures – cortical U-fibers for example – with no obvious homologs in the other hemisphere. As such, differences in hemispheric anatomy near the cortex may reflect not just a signal difference from the same structure occurring in both hemispheres but a lack of homology. These cases may ultimately be distinguishable with tools that model and cluster each hemisphere’s tracts as a graph, with known connectivity and topological relations; in that case, differences in tract composition between hemispheres would be easier to identify.
Methodological sources of variance in DTI data may also contribute to the level of asymmetries seen here. Magnetic susceptibility gradients occur at interfaces where tissue and air are in close proximity, and these are known to cause geometric distortions in the frontal and temporal poles. If severe, these distortions can cause a complete loss of signal, but in most cases they only lead to a geometrical warping of the data that can be corrected; here, we adjusted for it by using a mutual-information based 3D elastic warping approach. In individual cases, this distortion could contribute to the level of asymmetry in the DTI signal, and to its variance, but it is unlikely to produce a systematic pattern of asymmetry that favors the left or the right side of the brain. Maps such as (the mean asymmetry of FA) are unlikely to be affected much by susceptibility effects. These artifacts may contribute somewhat to the variance in asymmetry, slightly depleting the true biological correlations across members of a twin pair. Such artifacts would be lumped into the E-term of our structural equation model (which contains variance due to methodological error).
In addition, while the maps of DTI-derived measures (such as FA) were spatially normalized across subjects, no global intensity normalization was performed. We used each subject’s raw FA measures and did not adjust them for overall differences in mean FA between subjects. In group analyses of PET scans, individual data are commonly adjusted for overall (global) levels of activation or ligand binding, but this is not typically done in DTI studies. It is assumed that the FA is an absolute measure of fiber coherence, that is associated with physiological parameters such as axonal conduction speed, and with cognitive measures such as IQ. As a result, global normalization is not usually applied as the raw values are thought to provide a fundamental measure of fiber coherence. Even so, it is possible that local differences may, in part, reflect global differences in FA, or its asymmetry, across subjects.
Our study had 3 main limitations. First, we registered the FA images to a population averaged template created from the subjects’ FA images. The same registrations, based on the FA images, were applied to all the DTI-derived maps, allowing us to structurally align all the images in the same way. Even so, there may be a slight bias in using a single anisotropy measure to drive the nonlinear registration, and other measures could be used, individually or in combination (Park et al., 2003
). Several nonlinear registration algorithms have been proposed for DTI. Studholme (2008)
used DTI-derived measures as constraints when aligning standard anatomical images. Chiang et al. (2008)
and Li et al. (2009)
used orientation and multivariate information in the full diffusion tensor to find correspondences between DTI images.
Second, in this paper, we used voxel-based statistical maps, focusing on highly anisotropic white matter regions. Other approaches may also be helpful for selecting regions with high anisotropy, such as the tract-based spatial statistics method (TBSS: Smith et al., 2006
). In TBSS, a skeletonized (one pixel thick) map of the FA is created, and correspondences across subjects are based on distance, rather than by computing a correspondence field for the entire image. A third limitation of our study is that we do not fully exploit the angular information in the 105-direction diffusion-weighted images. Asymmetries in local diffusion geometry could also be examined by analyzing the local 3D diffusion profile, reconstructed using angular space deconvolution methods such as the tensor distribution function (Leow et al., 2008
). This could further probe the sources of asymmetry after adjusting for confounds due to the partial voluming and fiber crossings -- an inherent limitation of scalar DTI-derived measures. One could then distinguish whether the right greater than left asymmetries are due to higher fiber integrity in the right hemisphere or whether the anisotropy levels on the left are reduced due to a higher level of fiber crossings.