Language ability and handedness are likely to be associated with asymmetry of the cerebral cortex (grey matter) and connectivity (white matter). Grey matter asymmetry, most likely linked to language has been identified with voxel-based morphometry (VBM) using T1-weighted images. Differences in white matter obtained with this technique are less consistent, probably due to the relative insensitivity of the T1 contrast to the ultrastructure of white matter. Furthermore, previous VBM studies failed to find differences related to handedness in either grey or white matter. We revisited these issues and investigated two independent groups of subjects with diffusion-tensor imaging (DTI) for asymmetries in white matter composition. Using voxel-based statistical analyses an asymmetry of the arcuate fascicle was observed, with higher fractional anisotropy in the left hemisphere. In addition, we show differences related to handedness in the white matter underneath the precentral gyrus contralateral to the dominant hand. Remarkably, these findings were very robust, even when investigating small groups of subjects. This highlights the sensitivity of DTI for white matter tissue differences, making it an ideal tool to study small patient populations.
Voxel-based statistical analysis is usually performed on T1-weighted magnetic resonance (MR) images and has revealed details of anatomical differences in vivo (Ashburner and Friston, 2000; Good et al., 2001). Studies have shown changes in grey matter density related to navigational skills (Maguire et al., 2000), specific forms of headache (May et al., 1999) and degenerative diseases (Baron et al., 2001).
A problem concerning white matter morphometry on the basis of T1-weighted images is that T1 signal intensities are not very well correlated to white matter integrity. This can be seen in disorders such as multiple sclerosis, in which white matter T1 signal intensities remain normal even in severely damaged tissue (Filippi et al., 2001).
In contrast to T1-weighted imaging, diffusion-weighted imaging provides more subtle information about white matter tissue composition (Basser, 1995). Diffusion tensor (magnetic resonance) imaging (DTI) can be used to measure the diffusion characteristics of water in vivo. White matter fiber orientation can be determined by using diffusion-tensor MR imaging because water diffuses faster parallel to the longitudinal axis of axons than perpendicular to it (Basser et al., 1994). The fractional anisotropy (FA) of diffusion is sensitive the coherence of the orientation of fibers within each voxel. Lower FA values can indicate decreased fiber coherence or myelination defects as observed in multiple sclerosis (Filippi et al., 2001). Diffusion-tensor MR imaging has also indicated subtle white matter abnormalities in developmental speech (Sommer et al., 2002) and language disorders (Klingberg et al., 2000).
Voxel-based morphometry (VBM) studies investigating brain asymmetry have focused on grey matter (Watkins et al., 2001). One study also investigated differences in white matter using T1-weighted images (Good et al., 2001). In the right hemisphere, higher grey matter density was found in the frontal lobe, whereas the opposite was found in the occipital and the anterior temporal lobe in accord with previously described larger indentations of the skull (i.e. petalia). Interestingly, this does not seem to reflect the marked difference in lateralization of language to the left hemisphere, according to which one would expect higher grey matter density in left fronto-temporal areas. Similarly, based on the profound differences in motor skills between the dominant and non-dominant hands, one would expect marked differences in the organization of the motor system depending on the handedness of the subject. However, the two studies using VBM found no such difference when investigating either grey matter (Watkins et al., 2001) or grey and white matter (Good et al., 2001). So far, only studies investigating the depth of the central sulcus revealed a left–right asymmetry related to handedness in males but not in females (Amunts et al., 2000). The negative result of these VBM studies is unlikely to be related to insufficient statistical power, as both studies investigated >100 subjects. This is in marked contrast to traditional morphometry studies that have consistently revealed a correlation between the asymmetry in the planum temporale and planum parietale and hand dominance (Steinmetz et al., 1991; Jäncke et al., 1994). This leaves the possibility that signal intensity in T1-weighted images is an insensitive marker for anatomical differences related to handedness.
We revisited this issue and investigated two independent groups of healthy subjects with diffusion-tensor imaging (DTI) and voxel-based statistical analysis to further investigate the following questions: (i) are there differences between the ultrastructure of white matter in the left versus the right hemisphere? and (ii) is handedness reflected by a different white matter composition in the hemisphere contralateral to the dominant hand?
Materials and Methods
We studied two groups of subjects with DTI. The first group consisted of 15 healthy volunteers (mean 30.0, range 23–43 years, four females) and was previously used as a healthy control group for a sample of stutterers (Sommer et al., 2002). Most of the volunteers were employees of the university hospital. None of the participants suffered from any neurological or unstable medical disease or took any CNS-active medication. All volunteers in this group were consistent right handers with a score of at least 15/22 points (Oldfield, 1971).
The second group consisted of 28 healthy volunteers (mean 29.9, range 21–40 years, 14 females). Nine subjects were left handed, 19 right handed. The mean handedness score on the 10-item version of the Edinburgh Handedness Inventory (Oldfield, 1971), calculated as 100*((R – L)/(R + L), was 86.8 ± 15.9 (range 53.8–100) for the right handers and –79.7 ± 17.2 (range –60 to –100) for the left handed volunteers. Right handers were on average 29.7 ± 5.5 years old and left handers were 30.3 ± 6.5 years old. There was no significant age difference between groups [F(1,26) = 0.1, P = 0.8]. Both studies were approved by the local ethics committee and subjects gave written informed consent prior to the experiment.
Diffusion-weighted images were acquired on a Magnetom Vision 1.5 T MR system (Siemens Erlangen, Germany) with a circularly polarized head coil and maximum gradient amplitude of 25 mT/m. Cushions were used to restrict the subject’s head movements. The participants wore earplugs for noise protection. We acquired diffusion-weighted images with a ‘stimulated echo acquisition mode’ (STEAM; Nolte et al., 2000) sequence (flip angle = 15°, TR = 8872 ms, TE = 65 ms, 56 × 64 matrix, field of view = 168 × 192 mm, voxel size = 3 × 3 × 5 mm3) of 20 slices covering the whole brain and parts of the cerebellum. The full protocol consisted of a T2-weighted image and six diffusion-weighted images sensitized for diffusion along six different directions (b-value = 750 s/mm2). These measurements were repeated 40 times to improve the signal-to-noise ratio in the tensor maps. A T1-weighted image was acquired using a 3-D ‘fast low angle shot’ sequence (flip angle = 30°, TR = 15 ms, TE = 5 ms, 256 × 256 × 196 matrix, voxel size = 1 × 1 × 1 mm3).
Image Processing and Statistical Analysis
The first diffusion-weighted image was discarded to exclude the transition to steady state. Image processing was performed with SPM 99 and SPM2 (http://www.fil.ion.ucl.ac.uk/spm/). The diffusion-weighted images were realigned to the second image without diffusion weighting, and co-registered with the high resolution T1 image, which we then spatially normalized to a standard template (Paus et al., 1999), reorienting the diffusion gradient directions accordingly. The DTI images were sinc interpolated to 2 × 2 × 2 mm3 resolution. The diffusion-tensor and fractional anisotropy (FA) were determined for every voxel according to standard methods (Basser et al., 1994). Corrections for eddy current distortions were not necessary for this STEAM DTI acquisition.
To account for different geometrical configuration of both hemispheres, we performed two additional normalization steps: In a first step, all FA volumes were averaged to create a mean FA image. This image was then averaged with a mirrored version of itself, generating a symmetrical FA template. We then spatially normalized all FA volumes to this symmetrical template.
It is conceivable that both hemispheres are not perfectly parallel (e.g. smaller interhemispheric gap between the occipital lobes). If the relative position between left and right hemisphere in an individual brain is different from that in the template, spatial normalization can shift the brain slightly to one side, which renders spatially normalized FA maps erroneously asymmetric. To avoid these artifacts, in a second step the hemispheres were separated, all right hemispheres flipped to the left and individual hemispheres spatially normalized to the left hemisphere of the symmetrical FA template.
We used a 12 parameter affine registration together with non-linear warps parameterized through a basis set of discrete cosines (7 × 8 × 7 cycles in x, y and z directions) in all normalization procedures. The non-linear step was considered important, since we wanted to register different brains and hemispheres as accurately as possible. This is in contrast to standard VBM approaches, in which a residual difference in structure is desired (Ashburner and Friston, 2000). These rather complicated procedures were necessary to avoid possible confounds of differential registration of the two hemispheres.
Finally, FA maps were smoothed with a Gaussian kernel of 4 or 12 mm full width at half maximum. FA values were compared within subjects between hemispheres using SPM2 (paired t-test) with a threshold set to P < 0.05, corrected for multiple comparisons. For the handedness analysis, we hypothesized increased FA contralateral to the dominant hand in motor/premotor areas. In this case the correction of the P-value was based on a volume of interest (sphere with radius 12 mm, i. e. a volume of 7400 mm3) centered around the hand knob (Yousry et al., 1997).
To visualize differences between hemispheres in an anatomical context of fiber bundles, we used a common scheme to represent the orientation of the first eigenvector (i.e. the eigenvector corresponding to the largest eigenvalue) with colors. The magnitude of the Cartesian components (x, y, z) of this vector are color-coded: |x| (left–right) in red, |y| (posterior–anterior) in green and |z| (inferior–superior) in blue. Note that this simple representation is ambiguous: in the worst case two perpendicular vector orientations are displayed in the same color (see, for example, Pajevic and Pierpaoli, 1999). All color plots were derived from a high resolution (2 × 2 × 3 mm voxel-size) template DTI scan of a healthy control subject, as previously reported (Koch et al., 2002).
To study the reliability of our approach, hemispheric asymmetry was assessed for both groups of volunteers individually. In group 1, testing for greater FA in the left than the right hemisphere revealed a C-shaped structure connecting the temporal and frontal cortex. This structure comprised the posterior part of the superior longitudinal fascicule, a part of the arcuate fascicle. The observed FA difference extended far into the temporal lobe (Fig. 1, top left). Performing the same analysis in group 2 revealed an identical pattern of differences with significantly greater FA the arcuate fascicle of the left hemisphere (Fig. 1, top right). Overlaying this activation onto a color-coded direction map revealed that the difference was confined to areas in which fibers are predominantly oriented anterior–posterior (green in Fig. 1, top left). The same was true for the second group (Fig. 1, top right).
Testing for greater FA in the right than in the left hemisphere revealed a smaller area of increased FA posterior to the arcuate fascicle in the white matter underneath the inferior parietal cortex, just above the Sylvian fissure (Fig. 1, bottom left). An identical difference was observed in the second group (Fig. 1, bottom right). The overlay on color coded direction maps showed this difference to be located in an area in which the predominant direction of diffusion is in the lateral direction (i.e. right–left).
Differences in fractional anisotropy in relation to handedness were only assessed for the second group of subjects in which 9 of 28 subjects were left handed (Fig. 2). Here we tested for greater FA in the right as compared to the left hemisphere for left handers and greater FA in the left as compared to the right hemisphere for right handers. This was done using a conjunction analysis, reporting the minimum t-value of both comparisons (Friston et al., 1999). In this analysis only one significant difference (Z = 4.0, P < 0.05 corrected for ) was observed in the frontal lobe (x = ±32, y = –14, z = 50 mm; Fig. 2c). Closer inspection revealed that this signal difference is located exactly within the white matter of the precentral gyrus. Both pre- and postcentral gyri are easily identified on FA images by their high fractional anisotropy value (arrows in Fig. 2a). In this area FA was consistently higher in the left compared to the right hemisphere in right handers and vice versa in left handers (Fig. 2b).
Using voxel-based statistics on fractional anisotropy maps derived from DTI we were able to show greater FA values in the arcuate fascicle in the left hemisphere and in the inferior parietal white matter in the right hemisphere. The robustness of this result was demonstrated by an excellent test–retest reliability showing an identical pattern in two independent samples. Furthermore, a detailed analysis of handedness revealed a significant greater FA in the precentral gyrus contralateral to the dominant hand.
Recent voxel-based morphometric studies investigating brain asymmetry focused on the grey matter partition. The most prominent finding was a higher grey matter density in the left planum temporale (Good et al., 2001; Watkins et al., 2001). Interestingly, this area of increased grey matter density is close to the region of FA difference between right and left hemispheres which was observed in our study. Only one study directly investigated asymmetry of ‘white matter density’ as estimated from T1-weighted images (Good et al., 2001). The extensive asymmetries reported in this study partially included the arcuate fascicle.
One important difference between our method and the one employed in previous studies (Good et al., 2001; Watkins et al., 2001) concerns the spatial preprocessing. In these studies, the data were normalized to a symmetric template that was derived from the group studied. It is conceivable that although both hemispheres are congruent, they may not be perfectly parallel, e.g. due to a slightly smaller interhemispheric gap between the occipital lobes than between the frontal lobes. If the relative position between left and right hemispheres in an individual brain is different from that in the template, this can lead to an inaccurate match, in which the brain gets slightly shifted to one side, which can in turn cause asymmetrical spatially normalized FA maps. To avoid these artifacts, we added an additional processing step in which we renormalized individual hemispheres to a single average hemisphere, thus allowing for individual normalization parameters for each hemisphere. Pilot analyses without this additional step revealed asymmetry close to the midline in the left anterior and right posterior part of the cingulate, that vanished after introducing the additional single hemisphere normalization. Such an asymmetry close to the midline is to be treated with caution because this implies abrupt changes in density of the corpus callosum, which seems biologically implausible (Watkins et al., 2001).
By smoothing segmented grey matter partitions, volume or thickness differences can be transformed into image intensity differences through the partial volume effect. These intensity differences are then compared with a voxel-based statistic. Obviously, this works best if the smoothing kernel is large relative to the structures of interest. In case of the thin gray matter sheet, smoothing kernels of 12 mm have been used (Watkins et al., 2001). Subcortical white matter structures are much thicker and therefore larger smoothing kernels are necessary. This will, however, decrease the spatial resolution of the analysis.
The left arcuate fascicule has been linked to language function in patients with conduction aphasia, which is characterized by a predominant deficit in repetition (Geschwind, 1965) and is often caused by a disruption of the arcuate fascicle, which connects temporal and frontal language regions.
Our observation of greater fractional anisotropy in the left arcuate fascicle is also interesting with regard to studies of normal and abnormal white matter development and is in accord with a recent DTI region-of-interest-study showing higher anisotropy in the white matter underneath the left insula compared to the right insula (Cao et al., 2003). Using VBM on white matter segments from T1-weighted images, an increase of white matter density was observed in the left arcuate fascicle only (Paus et al., 1999). This finding was further supported by a recent developmental DTI study showing a correlation between FA and age in the left arcuate fascicle (Schmithorst et al., 2002).
In addition, another study demonstrated that language impaired patients with dyslexia show decreased fractional anisotropy in this region (Klingberg et al., 2000). This decrease in FA was tightly correlated with reading performance. The area of higher FA in the left hemisphere, as demonstrated in our study, almost perfectly coincides with the observed decrease in FA in dyslexics. Assuming that the asymmetry in the arcuate fascicle develops during language acquisition, one could speculate that reduced FA in the left arcuate fascicle in dyslexics is a sign of a deficient lateralization.
An asymmetry with greater FA values in the right hemisphere was found in the white matter of the inferior parietal lobe. Observations of neglect after right inferior parietal lesions that include the angular gyrus indicate a specific role of the right hemisphere in spatial processing (Mort et al., 2003). Hence, it is possible that the observed asymmetry reflects an asymmetry in the neural structures of spatial processing. Although this finding is robust as indicated by a replication in a second group, we refrain from interpreting the difference any further, because even in the analysis with a small smoothing kernel it is difficult to assess where precisely the difference is located and how it relates to cortical areas. Thus further high resolution studies focusing on this area are necessary.
Asymmetry Related to Handedness
Using MR morphometry the only morphological difference related to handedness in the motor system is a deeper central sulcus in the hemisphere contralateral to the dominant hand (Amunts et al., 2000). Interestingly this finding was obtained in male subjects only. In contrast the difference in FA related to handedness in our study was not significantly influenced by gender (data not shown).
Previous VBM studies failed to demonstrate any handedness-related asymmetry in the human brain. Good and colleagues reported asymmetry, but no effect of handedness, using VBM on grey and white matter segmented T1-weighted images in 465 volunteers (Good et al., 2001). Another study used multimodal imaging in 142 volunteers to better delineate the grey matter segment and performed voxel-based analyses on these segments (Watkins et al., 2001). Again, no consistent effect of handedness was found. Both studies were based on sample sizes much larger than ours. Hence, deficient statistical power is an unlikely reason for the null results obtained in these studies.
Recent DTI studies investigated the connectivity originating from primary motor cortex in both hemispheres (Ciccarelli et al., 2003; Guye et al., 2003). In these studies DTI data were used in combination with fiber tracking algorithms to define macroscopic fiber tracts. One of these studies (Guye et al., 2003) showed a more extensive connectivity in the dominant (left) hemisphere compared to the right hemisphere. The opposite comparison was not performed, because only right handers were investigated (Guye et al., 2003). In the other study, no handedness-related asymmetry was found (Ciccarelli et al., 2003).
Our study is the first voxel-based analysis to reveal handedness-related differences. The important difference between our study and previous studies is the imaging modality, i.e. diffusion-weighted imaging versus T1-weighted imaging. Our data strongly suggest that DTI is the imaging modality of choice for the investigation of white matter differences between groups and hemispheres.
Fractional anisotropy does not reflect a unique specific tissue property. Rather, FA is influenced by tissue hydration, myelination, cell-packing density and fiber diameter, and directional coherence (Shimony et al., 1999; Virta et al., 1999). We are thus not in a position to uncover the exact nature of ultrastructural changes in the precentral gyrus that seem to be associated with handedness.
Comparing the location of the observed handedness-related FA with a probability map of Brodmann area (BA) 4 and the pyramidal tract (Rademacher et al., 2001), shows that the coordinate falls precisely within the pyramidal tract. This suggests that the difference observed is indeed located in the white matter underneath BA 4.
Microscopic studies of Brodmann area 4 revealed that the hemisphere contralateral to the dominant hand shows increased contents of neuropil (Amunts et al., 1996). The term ‘neuropil’ denotes a tissue compartment containing dendrites and axons. Although this relates mainly to tissue within the boundary of grey matter, it is conceivable that remote effects of increased neuropil can also be observed underneath the grey matter sheet. This would suggest that increased FA in the precentral gyrus is the result of a higher cell packing density and more coherent fiber orientation.
Using DTI and voxel-based statistical analyses we were able to demonstrate a difference between the left and right arcuate fascicle. Together with developmental studies and studies of language impairment, this strongly suggests a link to the language abilities of this hemisphere. In addition, our study is the first to demonstrate a handedness-related difference in the precentral gyrus using voxel-based statistics. These findings were remarkably robust, even when investigating small groups of subjects, which highlights the sensitivity of DTI for white matter tissue composition. This high sensitivity renders the voxel-based analysis of DTI data an ideal tool to study patient populations (i) in which an involvement of white matter tracts is expected and (ii) in which a statement needs to be made with a small number of subjects. Recent examples of focal differences in FA in stuttering (Sommer et al., 2002) and in developmental dyslexia (Klingberg et al., 2000) underline this claim.
Supplementary material can be found at: http://www.cercor.oupjournals.org/
We thank the Physics and Methods group of NeuroImage Nord in Hamburg for their support with scanning and A. Castro-Caldas, H. Steinmetz and L. Jäncke for stimulating discussions. This work was supported by the Volkswagenstiftung (C.B.), a grant from the University of Göttingen (M.S.), the DFG and BMBF.