Comparison of white matter integrity between autism spectrum disorder subjects and typically developing individuals: a meta-analysis of diffusion tensor imaging tractography studies

1Atypical brain connectivity has repeatedly been implicated in neuroimaging studies of people with autism spectrum disorders (ASDs) (reviewed in), and suggested as the source of atypical behavioral characteristics found in ASDs. Accumulated evidence from previous postmortem studies and structural and functional magnetic resonance imaging (MRI) studies provided the hypothesis that the autistic brain is characterized by long-distance underconnectivity. However, it remains unclear which of the major seven long-distance tracts that compose the cortical network in the human brain is disordered. These tracts include main association (intrahemispheric) fibers – namely the uncinate fasciculus (UF), cingulum, superior longitudinal fasciculus (SLF), inferior longitudinal fasciculus, inferior frontal occipital fasciculus (IFOF), and fornix – and one commissural (interhemispheric) fiber – namely the corpus callosum (CC).

2Currently, diffusion tensor imaging (DTI) is one of the most powerful tools to investigate brain anatomical connectivity, and it can be used to study the orientation and integrity of white matter tracts (reviewed in). This is accomplished by mainly estimating the degree of anisotropy, the property of being directionally dependent, the amount of water molecules using measures including fractional anisotropy (FA), the degree of anisotropy (ranges between 0 that means unrestricted in all directions, and 1 that means restricted in one direction), and the overall displacement of molecules using mean diffusivity (MD).

3There are two different principal approaches to evaluating anatomical connectivity from DTI data. The first approach uses a voxel-wise whole brain analysis (WBA), such as voxel-based analysis or tract-based spatial statistics (TBSS). The other approach uses a region-of-interest (ROI) analysis, including using tractography to define the ROI. WBA is a powerful strategy for detecting the location of white matter changes in any part of the brain. Although there has been debate regarding how WBA can verify that any given standard space voxel contains data from the same part of the same tract from each and every subject, it is expected that TBSS will resolve these problems. Tractography is a relatively new technology that takes into account the orientation of the diffusion profile to reconstruct the trajectories of the whole fiber bundle and allows quantification of both white matter volume and microstructural organization in specific tracts.

4Although DTI is a powerful means of testing the disconnectivity hypothesis of ASDs, a number of DTI studies of subjects with ASDs yielded inconsistent results. For example, although some studies reported FA reductions in the CC of subjects with ASDs, some reported no significant differences or even higher FA values in the CC of subjects with ASDs compared with typically developing (TD) individuals. Further, previous studies have also yielded markedly inconsistent results regarding which tract shows robust differences between people with ASDs and TD individuals. Some studies reported FA differences in the CC of subjects with ASDs, whereas other reported differences in the inferior frontal occipital fasciculus or in the UF. These inconsistencies in the direction and location of the reported changes in ASDs may at least partially result from an insufficient number of participants in each study.

5To elucidate the tract that is disturbed in people with ASDs, we performed this systematic review and meta-analysis of DTI studies involving individuals with ASDs. As a number of previous neuroimaging studies have consistently reported structural and functional abnormalities in the frontal and temporal lobes (reviewed in), we hypothesized that long-distance tracts and especially those with connections in the frontal and temporal lobes exhibit consistent abnormalities in people with ASDs. Based on this hypothesis, WBA studies were excluded from the current meta-analysis to specifically focus on the anatomical connectivity of a long-distance tract that is more amenable to evaluation by ROI analyses. The present study thus conducted a meta-analysis of the ROI-based DTI studies.

6DTI studies that examined the brains of individuals with ASDs and TD individuals were obtained from the computerized databases MEDLINE, Web of Science and EMBASE. The search terms used in the systematic screening were autism, autistic, pervasive developmental disorder, and Asperger, which were also combined with diffusion, tensor, TBSS, tract-based spatial statistics and tractography. Titles and abstracts of the studies were examined to verify whether they could be included. Reference lists of included articles were also examined to search for additional studies to be included.

7Studies were included in the initial database if they were brain DTI studies published between 1980 and August 2012 and they examined people with ASDs in comparison with a TD individuals. The studies were then included in the meta-analysis if they utilized a ROI method (for example, manually-defined ROI and tractography). The literature search was performed without language restriction. If the study did not report sufficient data, we emailed the corresponding author to obtain the data. In cases where this author did not respond, we excluded the study. Two reviewers (YA and YN) performed the study screenings independently.

8In the current review, we performed a systematic screening of DTI studies of subjects with ASDs. A priori defined localized approaches such as manually-placed ROI and tractography studies were then categorized as ROI studies, and were included in the current meta-analyses. However, as voxel-wise voxel-based analysis and TBSS studies are informative, these studies were categorized as WBA studies and are summarized in Table 1.

9To perform the meta-analyses, we defined the standardized mean differences as the effect size statistic Cohen’s d, which is calculated as the difference between the mean of the experimental group and that of the comparison group divided by the pooled standard deviation. In the current meta-analyses, the mean FA and MD in autistic individuals were subtracted from those in TD individuals in each ROI, and divided by the pooled standard deviation from both groups. Tracts that were reported in a minimum of three studies were included in the meta-analysis. We followed the guidelines for the Meta-analysis of Observational Studies in Epidemiology.

10All meta-analyses were performed using Comprehensive Meta-Analysis version 2 (Biostat, Inc., Englewood, NJ, USA). Cohen’s d was calculated and used as the effect size. In studies that used multiple ROIs in a single tract (for example, five ROIs in the CC), the weighted average effect size was calculated to reflect the DTI value from the whole tracts rather than part of those tracts, and to consider differences among values from the same tract. Based on the results of previous studies that have reported asymmetry in major association fibers in subjects with ASD as well as in healthy subjects, we conducted meta-analyses for the right and left hemispheres independently. A random effects model was adapted for the meta-analysis to minimize potential heterogeneity from factors such as variations in the location of ROIs, magnetic field strength factors, and differences between ROI and tractography approaches. Studies that located their ROIs in the arcuate fasciculus were identified and integrated into analysis of the SLF, since the arcuate fasciculus is considered one of the subcomponents of the SLF. The significance level was set at P <0.05.

11The I2 statistics were calculated by the following formula:

12where Q is the chi-squared statistic and df its degrees of freedom. The I2 statistics were employed to examine between-study heterogeneity. Based on previous studies, thresholds for the interpretation of I2, which can be misleading but a rough guide to interpretation, suggests that 0 to 50% represents mild heterogeneity, 50 to 75% moderate heterogeneity, and 75 to 100% considerable heterogeneity.

13To determine the replicability and robustness of the findings, we performed one-study-removed sensitivity analyses in the meta-analyses that revealed significant differences. This analysis was adopted only when the number of included studies equaled or exceeded four, because fewer studies would preclude the conduct of a meta-analysis once one study was removed. In this approach, the meta-analysis is conducted through multiple permutations wherein one of the studies is removed from the dataset. This allows one to determine whether a single study is responsible for a significant result.

14Publication bias was investigated qualitatively with a funnel plot for each group and brain region. Funnel plot asymmetry was assessed quantitatively by linear regression analysis. A significance level of P <0.10 was used to conclude that the studies were heterogeneous.

15The literature search described above yielded 213 studies, of which 46 studies were identified as potential candidates for further screening. Four studies were excluded because they did not compare values between subjects with ASDs and TD individuals. Two studies were excluded because they were tensor-based morphometry studies. One study was removed because of potential overlap of the DTI data with another study from the same group. Seventeen studies were assigned into WBA DTI studies and are summarized in Table 1. Three studies were assigned into both ROI and WBA studies. Thus, 25 ROI DTI studies were identified as potentially relevant (Figure 1 and Table 2).

16Seventeen WBA-DTI studies involved 287 individuals with ASDs and 258 TD individuals. Eight studies utilized a 3 T scanner, whereas seven studies used a 1.5 T scanner. Two studies did not disclose the strength of the magnetic field. Eight studies adopted TBSS. Thirteen studies identified coordinates where FA was significantly reduced in the CC or association fibers in individuals with ASDs, whereas five studies showed coordinates where FA was significantly increased in the CC or association fibers in individuals with ASDs.

17Among the 25 candidate ROI DTI studies, two studies were excluded because they did not provide mean and standard deviation metabolite levels; instead of reporting these levels, one reported only P values while the other reported only the correlation to autism quotient. These studies were excluded after we emailed the corresponding author to obtain data to calculate the effect size, and the authors did not respond or provide them. In addition, nine studies were excluded because they did not focus on association fibers; instead, they focused on cerebellar fibers, short distance fibers, projection fibers, and the amygdalo-fusiform and hippocampo-fusiform pathways. Following these exclusions, 14 studies that examined 330 subjects with ASDs and 313 TD individuals were included in the current meta-analysis.

18Nine studies that recruited 208 subjects with ASDs and 179 TD independent individuals were integrated into the meta-analysis of FA values in the CC. We found significantly low FA values in subjects with ASDs compared with TD individuals (P = 0.023). Although the meta-analysis showed considerable heterogeneity, there was no publication bias. For the MD data, five studies with six datasets that examined 129 subjects with ASDs and 125 TD individuals were included in the meta-analysis, and this analysis found significantly increased MD values in subjects with ASDs (P = 0.006). (Figure 2 and Table 3; see Additional file 1).

19Six studies with six datasets were included in the meta-analysis of 117 subjects with ASDs and 125 TD individuals. The random-effects model demonstrated a significant FA reduction in the left hemisphere in people with ASDs (P = 0.011) (Table 3). The between-study heterogeneity was low and there was no publication bias (see Additional file 1). The analysis demonstrated no significant differences in FA in the right hemisphere between subjects with ASDs and TD individuals (P = 0.216). For the MD data, our integrated analysis of three studies with four datasets with 60 subjects with ASDs and 75 TD individuals revealed no significant difference between subjects with ASDs and TD individuals in either hemisphere (P = 0.113 and 0.790, left and right, respectively).

20Datasets from five studies (involving 101 subjects with ASDs and 109 TD individuals) investigating FA values in the cingulum were included in the analysis. A meta-analysis of these studies showed no significant differences between subjects with ASDs and TD individuals in the left (P = 0.255) and right (P = 0.921) hemispheres (Table 3; see Additional file 1). Only two studies reported MD in the cingulum in the left and right hemispheres, separately. We therefore did not conduct a meta-analysis for MD in the cingulum.

21Five studies that recruited 96 subjects with ASDs and 86 TD individuals were examined in the meta-analysis for FA. The random-effects model revealed a significant decrease in FA values in the left hemisphere (P = 0.016). By contrast, a meta-analysis of findings for the right hemisphere demonstrated no significant difference (P = 0.219). Three studies with four datasets that involved 51 individuals with ASDs and 58 TD individuals were integrated into the meta-analysis for MD. The results revealed a significant MD increase in the left (P = 0.031) and right (P = 0.011) hemispheres without publication bias and minimal between-study heterogeneity (Table 3 and Figure 3; see Additional file 1).

22A meta-analysis of four studies with four datasets with 61 subjects with ASDs and 85 TD individuals did not demonstrate significant FA reductions in the ASD group in the left (P = 0.201) and right (P = 0.050) hemispheres. A meta-analysis for MD values in this tract could not be conducted because of the limited number of studies. The meta-analysis of FA values demonstrated moderate heterogeneity but no publication bias (Table 3; see Additional file 1).

23Three studies with three datasets that involved 68 subjects with ASDs and 76 TD individuals were examined in the meta-analysis. The meta-analysis did not show any significant FA reductions in people with ASDs in either hemisphere (P = 0.191 and 0.198, left and right, respectively) (Table 3; see Additional file 1). MD data were available from only one study.

24To test the robustness of the significance of findings, we conducted one-study-removed sensitivity analyses of FA values in the CC, left UF and left SLF, and MD values in the CC, left UF, left and right SLF that included more than four studies. For FA values in the CC, five sensitivity analyses out of nine preserved the significant FA reductions, and the remaining four analyses preserved reductions that approached significance (see Additional file 2). One-study-removed sensitivity analysis of FA values in the left UF demonstrated that five sensitivity analyses out of six reached significance whereas the remaining one did not (see Additional file 2). All of the one-dataset-removed sensitivity analyses of MD values in the CC preserved the significance of the higher MD values in subjects with ASDs (Additional file 2). With regard to the left SLF, two out of five one-study-removed sensitivity analyses demonstrated a significant FA reduction. Only one out of four one-study-removed sensitivity analyses preserved the significance of the MD increase in the left SLF. Two out of four one-study-removed sensitivity analyses demonstrated a significant MD increase in the right SLF (see Additional file 3).

25To our knowledge, this is the first meta-analysis of DTI studies in individuals with ASDs. A comprehensive literature search yielded 25 ROI DTI studies that involved individuals with ASDs. Overall, DTI studies have consistently identified a reduction in anisotropy in subjects with ASDs compared with TD individuals (Tables 1 and 2). Meta-analyses of 14 studies with 330 individuals with ASDs and 313 independent TD individuals demonstrated significantly lower FA values in the CC, left UF and left SLF, and higher MD in the CC and the left and right SLF using random effects models (Figure 4).

26We used a random-effects model to minimize the effects of between-study heterogeneity, and only one meta-analysis showed a severe degree of heterogeneity for FA values of the CC in the current meta-analyses (Table 3). However, previous DTI studies exhibit inherent between-study heterogeneity for other tracts and brain areas besides FA values in the CC. This heterogeneity comes from sources such as differences in data acquisition, data analysis, and subject characteristic among the studies. The current findings should thus be interpreted with caution. Indeed, the accuracy of DTI data is affected by a number of factors, such as the signal-to-noise ratio, image resolution, image distortion from magnetic susceptibility effects, and motion artifacts. Those factors are interrelated and are influenced by the acquisition parameters including the field strength and number of directions, which differ from study to study. As the signal-to-noise ratio strongly affects diffusion tensor measures, some studies conducted several times repetitions of image acquisition to increase the signal-to-noise ratio. However, the number of repetitions used varies between 1 and 15 in previous studies (Tables 1 and 2).

27The current systematic review of previous literature showed that FA is the most often investigated DTI value, which ranges from 0 representing maximal isotropic diffusion of water to 1 representing maximal anisotropic diffusion. Although FA values generally reflect cellular membrane integrity, fiber myelination, fiber diameter, localized water content, and directionality, the abnormality underlying the reductions in FA values in subjects with ASDs is yet to be specified. As FA reductions may occur as a result of different abnormalities or a combination of abnormalities, and because there is heterogeneity in the ASD population, different subpopulations of people with ASDs may display FA reductions from different abnormality. We might possibly have combined FA reductions from different causes in the same meta-analysis. We expect that future study will account for the heterogeneities in people with ASDs by synthesizing an increased number of studies.

28In the current study, the meta-analysis demonstrated significantly reduced FA values and increased MD values, which suggests decreased myelination or axonal densities, in the CC of ASDs compared with that of TD individuals. Decreased white matter integrity in the CC is in line with findings from a previous meta-analysis of MRI studies that focused on the CC size and demonstrated a significantly smaller than normal CC in people with ASDs. The CC provides an integrated interhemispheric white matter bundle and is topographically organized. Its disruption may result in many of the sensory, cognitive, and behavioral symptoms observed in children with ASDs. In line with this notion, previous functional MRI studies and an EEG study during a REM sleep have reported interhemispheric disconnectivity, suggesting an underlying CC pathophysiology, in people with ASDs compared with TD individuals. The meta-analysis also identified similar pattern of deviations in FA of the left and MD of the bilateral SLF in ASDs. Although the SLF forms the main connection between frontal and occipital regions, this tract has extensive branching in the frontal, parietal and temporal lobes. The tract has been recognized as an important structure for information exchange between Broca’s and Wernicke’s area and may constitute underpinning of abnormal language processing, which is frequently observed in people with ASDs.

29The current meta-analysis of FA values in the left UF demonstrated a significant reduction in people with ASDs. The importance of this intrahemispheric tract that involves the frontal lobe has also been recognized in the pathophysiology of ASDs. Although there was not enough statistical power and there were no results regarding the axial diffusivity and radial diffusivity to make firm conclusions, the findings of reduced FA in combination with no change in MD may represent microstructural abnormality without gross tissue volume reduction. These significant reductions in FA are concordant with the results of a previous meta-analysis of VBM studies in subjects with ASDs that reported increased white-matter volumes in the UF. The UF is a bundle that connects the medial and lateral orbitofrontal cortices with the anterior portions of the temporal lobe, which are important regions for emotional and attentional processing and reward-related decision making. The UF also includes projections into the cortical nuclei of the amygdala, a limbic region known for its involvement in human social and emotion processing and the pathophysiology of ASDs. Furthermore, functional MRI studies have repeatedly revealed aberrant functional connectivity between the frontal and temporal lobes in people with ASDs (for example).

30The current analyses demonstrated a relatively left-dominant abnormality in the association fibers in individuals with ASDs. The results of the meta-analysis may reflect leftward asymmetry in the brain, which is unrelated to the scanner, brain volume, sex and handedness in TD individuals and ASD subjects. Brain asymmetry is reported to be a result of typical maturation and essential for typical emotional processing, speech perception, cognition and sensory and motor functions. The left-dominant abnormality in the current meta-analysis may be a result of atypical brain maturation in ASD individuals and a potential pathophysiology of disturbances of these functions.

31Several limitations of our study should be considered. First, the limited number of included studies in the current meta-analysis did not allow us to perform a meta-regression analysis to test the effects of methodological and clinical differences among the included studies. In addition, there still remains the possibility that meta-analyses did not yield significant difference in other tracts, such as the inferior longitudinal fasciculus, because of the insufficient number of studies included. We therefore cannot emphasize the specificity of abnormalities of the CC, UF and SLF. Second, methodological differences across studies such as in the strength of the magnetic field or the number of excitations and especially regional boundaries of ROIs could have strong effects on DTI findings. In terms of regional boundaries of ROIs, the included studies differ from each other considerably. Although previous meta-analyses also integrated studies employing ROI and those utilizing tractography, the difference in definition of regional borders yielded considerable between-study heterogeneity. Tractography intends to cover the entire fiber, whereas the boundaries of ROIs are determined by the researchers. To reduce the heterogeneity between DTI values from the entire fiber obtained by tractography and one part of fiber gained by ROI, we calculated the mean value of multiple ROIs in the case locating several ROIs within one fiber, although the attempt does not completely diminish the inherent heterogeneity. Third, the clinical characteristics of included study participants differ between studies considerably. The mean age of the participants may be one of the most important confounds because of the abnormal brain growth trajectory in the ASD brain. Therefore, although we listed studies included in the meta-analyses in the order of the mean age of participants to enable the reader to qualitatively assess the potential effects of age on effect size, the current meta-analysis could not exclude the possibility that age had effects on FA levels in brain areas such as the CC, UF and SLF. Fourth, although previous studies have reported abnormalities in projection fibers and fibers that connect cerebrum and cerebellum, and suggested that they potentially underlie the pathophysiology of abnormal behavior in individuals with ASD, we could not conduct meta-analyses because of a lack of sufficient studies. Lastly, as we have reviewed, there are accumulative studies that involved voxel-based analysis to investigate tract abnormality in individuals with ASDs. Although we could not conduct a meta-analysis of these studies due to their heterogeneities, in terms of the differences between TBSS and ordinary voxel-wise analysis and small volume correction, there is no doubt that these studies and their meta-analysis could contribute in the future.

32DTI studies involving individuals with ASDs demonstrated significantly low FA values in the CC and left UF and high MD in the CC and SLF bilaterally compared with TD individuals. Although some confounds could not be controlled for because of the limited number of included studies, the current meta-analysis strongly demonstrates that changes in white matter integrity in ASDs are localized in the CC, UF and SLF among association and commissural fibers. The current findings further support the long-distance underconnectivity hypothesis of ASDs.

33ASD: Autism spectrum disorder; CC: Corpus callosum; DTI: Diffusion tensor imaging; FA: Fractional anisotropy; MD: Mean diffusivity; MRI: Magnetic resonance imaging; ROI: Region of interest; SLF: Superior longitudinal fasciculus; TBSS: Tract-based spatial statistics; TD: Typically developing; UF: Uncinate fasciculus; WBA: Whole brain analysis.

34The authors declare that they have no conflicts of interest.

35YA and YN performed the study screenings independently. YA performed all of the data extraction and calculated the effect sizes twice to avoid careless mistakes. YN further performed the data extraction and an independent calculation of the effect sizes. YA, OA and HY wrote the paper. All authors read and approved the final manuscript.