Our ALE meta-analysis of grey matter in people with Asperger syndrome and autism comprehensively synthesized 15 VBM studies and revealed overlap, as well striking differences, between both conditions. For both Asperger syndrome and autism studies, commonality was noted in the volume excess in the ventral temporal lobe of the left hemisphere around the middle and inferior temporal gyrus and lingual gyrus (Brodmann area [BA] 37 and BA 19). The few studies that have directly compared people with Asperger syndrome or autism and controls in the same study
30,32,41,42 are not completely consistent with the findings reported here. In particular, results from a study by Kwon and colleagues
30 and by our own group
41 found no areas of higher grey matter volume in participants on the autism spectrum compared with controls, whereas the study by Toal and colleagues
32 reported higher volumes in the high-functioning autism group in fronto-temporal regions. The study by Kwon and colleagues and that by our own group were conducted in a childhood sample, whereas that by Toal and colleagues was conducted using an adult sample. Potentially, this age difference had a substantial impact on the results, as in the current meta-analysis the mean age across all the studies was more in line with that reported in the study by Toal and colleagues.
32Greater functional activation associated with preserved performance on the embedded figures test of visual search strategy has been reported in this ventral temporal lobe region in a mixed group of individuals with autism or Asperger syndrome.
73 This finding was significant despite the small sample size (
n = 6 on the autism spectrum), suggesting that the results were likely contributed by participants with Asperger syndrome and autism. There is good proximity of the brain activation results in the study by Ring and colleagues
73 to the structural differences in the ventral temporal lobe generated in our ALE summary of autism and Asperger syndrome studies. An explanation for the superior embedded figures test performance of both groups with autism
15 and Asperger syndrome
74 and their corresponding ventral temporal activation/hyperactivation has been suggested to be reliance on visually mediated strategies rather than working-memory systems.
73 Others have not observed expansion but rather a preference in topology for certain tasks (e.g., people on the autism spectrum tend to recruit posterior brain regions during cognitive processing, and again this has been interpreted as preference for visual-based processing mechanisms in the spectrum
75,76).
In studies of Asperger syndrome, the summary ALE generated a fairly restricted pattern of differences compared with typically developing controls. Grey matter volumes were lower in the bilateral medial temporal regions, including the ventral hippocampus extending toward the amygdala, as well as the right putamen and precuneus. The latter posterior midline region is usually active at rest and may regulate introspection.
77 It is thought that a failure to deactivate this and other components of the default network during goal-directed behaviour disrupts cognitive processes in people with autism spectrum disorders.
78 The lower limbic striatal grey matter volumes reported in studies of people with Asperger syndrome may contribute to some of the socioemotional difficulties experienced by these individuals. Region of interest (ROI)–based measurement approaches have yielded much less consistent findings. Aylward and colleagues
79 reported that in people with autism spectrum disorders, amygdala and hippocampus volumes were lower than in controls. Conversely, Schumann and colleagues
80 reported that the amygdala was enlarged only in children with autism spectrum disorders, but that the hippocampus was enlarged at all ages. Thus, as noted, the age range of the sample studied may be a critical factor in the pattern of results observed.
Only for studies of Asperger syndrome did we note clusters of grey matter volume excess relative to controls to be primarily located in the left hemisphere, medial temporal lobe and inferior parietal lobule, with only 1 cluster of grey matter excess identified in the right hemisphere in the inferior parietal lobule. Thus, lower grey matter volumes tend to be found in the right hemisphere, and higher volumes are identified mainly in the left hemisphere in people with Asperger syndrome. Between the ages of 1 and 3 years, typically developing children have a pattern of right-hemisphere dominance that is superseded by left-hemisphere dominance arising with the development of language after the age of 3 years.
81 Rinehart and colleagues
82 have argued that the time at which language develops may be pivotal for determining expression of an autistic or Asperger syndrome phenotype. Certainly the language development of children with Asperger syndrome is often precocious and in marked contrast to the great difficulty experienced by children with even high-functioning autism. This may be reflected in the larger regional grey matter volumes in the language-dominant left hemisphere of individuals with Asperger syndrome compared with controls. Also of note is the volumetric excess for the Asperger syndrome group in the inferior parietal lobule and for the autism group in the fusiform gyrus, since both regions are linked to synaesthesia, the experience whereby sensory perception in one modality is cross-wired to another modality.
83 Sensory interests, and even synaesthesia, have been noted in individuals with Asperger syndrome,
84 and the latter phenomenon is understood to be associated with larger parietal and fusiform gyri.
83,85 The inferior parietal lobule and fusiform gyrus also form part of the mirror neuron system
86 and face-processing circuit, respectively.
87 Thus, anomalies in these regions in people with autism spectrum disorders may contribute to the social difficulties experienced by individuals on the spectrum.
88We observed that in the studies of autism, many more clusters of greater regional grey matter volumes were characterized in the resulting ALE map. Some support for the pattern of volume differences that we observed comes from an ROI study that directly compared autism and Asperger syndrome and found larger caudates to be a feature of autism more than Asperger syndrome.
89 Caudate enlargement has been confirmed in other autism samples,
63,90 including medication-naïve participants.
91 These higher striatal volumes have been shown to correlate positively with repetitive behaviour scores measured on the ADI-R,
92 but this is not always the case.
93 Interestingly, children who score highly on the repetitive behaviours domain of the ADI-R are more likely to have fathers with obsessive–compulsive disorder (OCD),
94 and people with OCD have bigger striatal regions than controls.
95–97 However, just as for studies of autism spectrum disorders, there are inconsistencies in the ROI literature on OCD with smaller
98,99 or no
100 size increases in the caudate nucleus but a lower globus pallidus volume in treatment-naïve patients.
100The direction and location of volume changes in the basal ganglia reported in studies of Asperger syndrome and autism were wholly distinct. Asperger syndrome was associated with a lower right putamen volume, and autism was associated with larger bilateral caudate volumes. The striatum (caudate and putamen) integrates information from the entire cerebral cortex and regulates output to motor and thalamic targets.
101 Striato-limbic circuits interconnect social brain areas strongly associated with autism, including the amygdala and fusiform gyrus,
102,103 superior temporal sulcus
104 and the medial prefrontal lobe.
76,105 Anatomic differences in striatal systems fit with disruption of social behaviours in people with autism spectrum disorders. In addition, structural differences in the striatum are also consistent with the multiple motor symptoms in people with autism spectrum disorders,
106–109 which would be predicted to arise from pathology in the basal ganglia.
109,110 However, we speculate that distinct intrastriatal differences observed here may parallel the distinct motor phenotype described in people with Asperger syndrome and autism. Clumsiness is more common in people with Asperger syndrome, whereas children with autism tend to have postural abnormalities.
111,112 A greater impairment in motor preparation in children with high-functioning autism compared with those with Asperger syndrome has been interpreted as a downstream effect of a quantitative dissociation in motor planning in people with autism and Asperger syndrome.
111 Thus, pathology within different components of the striatal system may contribute to both the shared triadic symptoms of autism and Asperger syndrome described by the ADI-R, and their divergent executive and motor function profile.
82,111,113–116Classic language-processing regions were not shown to be different between autism and Asperger syndrome groups compared with controls. Although somewhat surprising, the Broca inferior frontal and Wernicke posterior temporal regions are now regarded as part and parcel of a more distributed cortico-cerebellar network for multimodal comprehension of language encompassing the inferior parietal, occipital and cerebellar regions.
117,118 Since language impairment is relatively absent in people with Asperger syndrome but present in those with autism, our findings of fewer clusters of volumetric difference in the Asperger syndrome group compared with controls may indicate relative sparing of brain systems during the process of neurodevelopment. The more extensive clustering of grey matter differences in the right hemisphere in the autism group may be more linked with the pragmatic language difficulties experienced by these individuals. It has been previously postulated that right hemisphere dysfunction contributes to both autism and the nonverbal language impairment observed in people with semantic–pragmatic language disorder. Indeed, some have considered that semantic–pragmatic language disorder may comprise part of the autism spectrum,
119 whereas others have queried whether semantic–pragmatic language disorder is a valid discrete clinical entity.
120 Clearly, diagnostic boundaries across these complex developmental conditions are not completely clear, and further research is needed.
Determining the basis of the neuroanatomic differences in people with Asperger syndrome and autism observed in our synthesis of the voxel-based literature is still a challenge. There is some indication that the genetics of people with Asperger syndrome and autism may be dissimilar. People with Asperger syndrome are more likely than those with autism to have a family history of depression, schizophrenia and the broader autistic phenotype.
121 Although some genetic susceptibility loci in people with Asperger syndrome overlap with loci associated with autism, others overlap with schizophrenia susceptibility loci.
122 One possibility is that on the spectrum of complex neurodevelopmental disorders, including autism, schizophrenia and Asperger syndrome,
123 Asperger syndrome sits closer to schizophrenia-like conditions. Evidence in support of this idea comes from observations of increased dopamine activity in the caudate in people with Asperger syndrome,
124 which is a feature of schizophrenia;
125 higher paranoia scores in people with Asperger syndrome compared with controls;
126 and negative symptoms in people with Asperger syndrome responsive to risperidone, an antipsychotic commonly used to treat patients with schizophrenia.
127 Indeed, 2 decades ago, the similarities between childhood schizoid disorder and Asperger syndrome were described. Wolff
128 documented the transition of individuals with schizoid personality in childhood to adults with schizophrenia spectrum diagnoses, including schizotypal personality disorder, and described these children as similar to those originally described by Asperger. Thus, the specific mix of susceptibility genes and environmental factors acting early in development may determine the specific diagnostic subgroup into which people with these traits are categorized.
41 Echoing van Os and Kapur,
123 it is not schizophrenia or autism or Asperger syndrome that is inherited or acquired, but rather a neurodevelopmental vulnerability.
Limitations
Our study has a number of limitations. First, the inferences drawn from our analysis are necessarily drawn from the samples contained therein. Although the aggregate sample sizes are large, and we believe that they are reasonably representative of people with autism and Asperger syndrome, we cannot be certain that our findings apply to the entire clinical population. This is particularly important in consideration of the impact of sex and learning disability within the spectrum, and the studies we included under-represented girls and women and people with learning disabilities. Moreover, our ALE approach involved separate analyses of data on participants with autism and Asperger syndrome, and we did not carry out a direct statistical comparison of the 2 groups. Unfortunately, few studies to date have directly compared autism and Asperger syndrome using VBM, and mapping the true overlap is therefore beyond the reach of this paper. Even in more conventional use of the ALE technique, differences across conditions have generally been arrived at by subtraction rather than direct group comparison. As further studies of autism spectrum disorders continue to be published, we expect that future application of continually updated ALE techniques (listed at
http://brainmap.org/ale/readme.html) to expanded study numbers will facilitate a direct group comparison in due course. The current study also suffers the problem of all meta-analyses in that studies reporting null results are under-represented in the literature. That is, ALE tends to compound this problem, as a VBM study reporting no group differences contains no coordinates, and therefore the information cannot be included in the ALE. In addition, VBM methodology is continually being adapted, and the data from original studies incorporated into our analysis most certainly had been preprocessed and analyzed in different ways (e.g., whether the results have been modulated or not).
45,129–132 Unfortunately there were insufficient studies to control for confounds, such as modulation and smoothing, and in one study, a support vector machine learning procedure was used to extract VBM diagnostic differences.
67