The basic conclusion from the above review is very simple: Autism cannot be described and much less explained as a localized defect, but needs to be modeled as a network disorder. This has several implications: (1) Any local findings, such as atypical cellular organization, reduced volume, or lack of expected activation, needs to be viewed in the context of the interaction of the region in question with other regions during development, as discussed above. (2) In addition to local findings, research on autism needs to examine the integrity of interregional connectivity. This can be done in a variety of ways, including the study of (a) white matter volume and integrity, (b) the study of anatomical fiber tracts, and (c) the study of functional cooperation between brain regions (functional connectivity).
The single most replicated finding of white matter volume decrease in ASD concerns the corpus callosum. While reduced callosal size has been observed repeatedly, regional patterns within the callosum are less established. A recent study by Vidal and colleagues (2006)
suggests that in children with autism, reductions of callosal thickness are most pronounced in three of its sections, an anterior section in the genu, which connects orbitofrontal cortices of the two hemispheres, a midsection in the body, which connects perirolandic somatosensory and motor regions, and a posterior region in the splenium, which connects parahippocampal and extrastriatal visual cortices. Some of the region-specific findings are intriguing – as, for example, the anterior finding with respect to the known role of orbitofrontal cortex in socio-affective functions (Adolphs 2003
). More generally, callosal reduction suggests impaired interhemispheric connectivity. As mentioned before, one should be cautious not to view callosal findings in older children and adults as evidence explaining
atypical interhemispheric processing. To be sure, the findings might reflect early-onset abnormalities as part of gene-driven growth defects. Yet, the predominant effects of these growth defects have been described as early overgrowth (Courchesne et al. 2001
), but not volume reduction.
In an alternative and more likely scenario, therefore, callosal reduction may be primarily driven by inefficient intra
-hemispheric organization of emerging functional networks. In other words, gray matter growth abnormalities mentioned above are expected to affect the layered architecture of cortex locally and to reduce its processing efficiency. Early white matter overgrowth may provide the developing brain with abundant substrates for intrahemispheric signal transfer. However, such overgrowth probably reflects non-selective and insufficiently organized connectivity, which would negatively affect the fine-tuning of functional networks. Local (gray matter) organization and network connectivity will further be affected by the mismatched timing of maturational events and experiential effects. In the typical brain, initial growth and subsequent regression (neuronal loss, synaptic pruning) are timed in ways that allow activity and experience to support the organization of functional networks (Kandel et al. 2000
). The growth profiles in autism do not seem to support such synergy between maturation and experience. Callosal volume reduction may be largely due to the repercussions of such a mismatch between brain growth and experience.
Volumetric MRI studies are limited to determining white matter (ab)normality based solely on size. Other MR modalities, which have been applied to ASD in a few studies, can assess white matter integrity regardless of size. One study by Hendry and colleagues (2006)
found increased transverse (T2) relaxation times of white matter in parieto-occipital and some prefrontal regions, suggesting abnormally high water content. A number of studies have used MR spectroscopy (MRS), which detects various brain metabolites based on chemical resonance shift. Most of these studies examined gray matter or measured from large voxels that included both gray and white matter. Friedman and coworkers (2006)
found that N
-acetylaspartate (NAA) and Myo
-inositol – considered markers of neuronal and membrane integrity, respectively – were reduced in the white matter of children with autism around age 4 years. This difference was, however, only found in comparison to typically developing children, not to those with developmental delay, suggesting that the finding is not specific to autism. Diagnostically more specific findings, such as reduced NAA, were limited to gray matter. Another MRS study of white matter in slightly older children with autism failed to detect reduced NAA in comparison to typical controls (Fayed and Modrego 2005
). There is thus currently no clear evidence of dramatic chemical white matter compromise in ASD.
As mentioned, MRS detects only a limited set of metabolites and the above null findings are far from conclusive. Another approach to the study of white matter is provided by diffusion imaging, an MRI modality that takes advantage of the effects of the movement of water molecules on magnetic resonance. Since axonal membranes in white matter prevent water molecules from diffusing freely in all directions (isotropically), fractional anisotropy, which can be measured in diffusion-tensor imaging (DTI), reflects the integrity and organization of axons. In one study applying this technique in male adolescents with autism, Barnea-Goraly and colleagues (2004)
found widespread reductions in fractional anisotropy in a comparison with matched control participants. Some of the effects found in all four lobes of the cerebrum were consistent with previous findings from other techniques, such as reductions in anterior portions of the corpus callosum and in ventromedial prefrontal cortex, whereas others in occipital and pericentral regions were less expected. DTI can also map out long-range fiber tracts between brain regions through computation of the predominant direction of diffusion in each voxel (Ramnani et al. 2004
). While this technique offers great promise for the study of developmental disorders, little evidence relevant to ASD is currently available.
Interregional connectivity in autism has instead been inspected using an off-shoot of functional MRI, commonly called functional connectivity MRI (fcMRI). Like conventional fMRI, fcMRI is based on the blood-oxygenation level dependent (BOLD) signal, which indirectly relates to local neuronal activity (Logothetis and Pfeuffer 2004
). Whereas fMRI looks at the variance in BOLD time series that can be explained by task designs (“activation”, defined as greater signal during an experimental compared to a control task), fcMRI is based on interregional cross-correlations of the BOLD signal in the low-frequency domain, typically below 0.1 Hz (Cordes et al. 2001
). These correlations may reflect low-frequency fluctuations in local field potentials (Leopold et al. 2003
). Applications of fcMRI have quite impressively shown this technique’s potential to map out complex distributed functional networks, such as the motor network (Biswal et al. 1995
) or perisylvian language areas (Hampson et al. 2002
). Interestingly, interhemispheric BOLD correlations, typically seen between homotopic areas (e.g., left and right superior temporal cortex), were disrupted in patients with callosal agenesis (Quigley et al. 2003
), further suggesting that these correlations are indirectly linked to anatomical connectivity.
The number of fcMRI studies of autism currently remains small. A frequently cited model of neurofunctional organization in autism derived from fcMRI studies is the ‘underconnectivity theory’, as put forth by Just and colleagues (2004)
in the context of a study on sentence comprehension. In this study, BOLD signal cross-correlations associated with task performance (determining the agent or recipient in a sentence by pressing a button) was found to be slightly reduced (though generally present) between a number of cortical regions of interest (). Largely consistent findings have been reported in additional studies by Just and colleagues on verbal working memory (Koshino et al. 2005
), semantic judgments of sentences (Kana et al. 2006
), and executive processing on the Tower of London task (Just et al. 2006
). Evidence from PET studies appears consistent with the underconnnectivity theory. In an early PET study, Horwitz and colleagues (1988)
found that positive correlations in glucose metabolic rates between frontal and parietal regions, as seen in a group of typical adults, were reduced in men with autism. Castelli et al. (2002)
, in their theory-of-mind PET activation study described earlier, found regional blood flow correlations between extrastriate cortex and superior temporal sulcus reduced in adults with autism.
Fig. 1 Activation effects for sentence comprehension in autism group (A) and control group (B), showing overall reduced effects in the former. (C) Correlation of mean time series between diverse cortical regions of interest (ROIs) shows generally reduced functional (more ...)
In the above mentioned study by Just and colleagues (2006)
on executive function, fcMRI findings were strengthened by correlations with structural and diagnostic measures. Thus frontoparietal connectivity was correlated positively with the size of the anterior portion (genu) of the callosum and negatively with the total score on the Autism Diagnostic Observation Schedule (Lord et al. 2001
). The callosal finding is consistent with results from Vidal et al. (2006)
discussed earlier. Taken together, these results may indicate that inefficient functional connectivity between cortical regions is linked to reduced callosal volume as well as greater symptom load.
It should be noted, however, that the concept of functional connectivity is not well defined – or “elusive” in the words of Horwitz (2003)
– and that methodological decisions heavily impact results. Functional connectivity has been defined as “observed temporal correlations between spatially remote neurophysiological events” (Friston et al. 1993
; see also Rippon et al. 2006
). While the conventional dichotomy with effective connectivity, defined as “the influence one neural system exerts over another” (Friston et al. 1993
), is conceptually clear, the lines between the two may be blurred in actual applications and due to a large number of differences in methodological approaches (Horwitz 2003
One conceivable confound in autism fMRI studies is head motion. In activation studies, a result according to which activation effects are simply weaker or absent – with little inverse effects (activation being stronger than normal) in other parts of the brain – needs to be treated with great caution, as such differences may simply be the effect of head motion, which will result in noisier BOLD time series. An analogous argument can be made for fcMRI studies.i
If time series in one region of interest are slightly noisier in an autism group than in a control group, correlations of time series across regions can also be expected to be lower. It is therefore important to consider data on detected motion.
Even if there are no significant differences in head motion, alternative explanations that do not relate to connectivity may be possible. Many early fcMRI studies (e.g., Biswal et al. 1995
) were acquired in the resting state. Even in this state, uncontrolled cognitive processing might drive interregional correlations. This possibility is much greater, however, when data are acquired during task performance. If interregional correlations are reduced, for example, during a sentence comprehension task, the argument could be made that such finding is still related to activation, rather than functional connectivity. The reasons for this are related to our incomplete knowledge of functional differentiation in autistic cerebral cortex. Very few studies to date have examined autism fMRI data on a single-subject level. In one study briefly mentioned before (Müller et al. 2001
), we found that what appeared to be slightly reduced activation in perirolandic regions during simple finger movement on a groupwise analysis could be explained in qualitatively different ways after inspecting results from single participants. These suggested that the reasons for the group finding were twofold. First, activation maps tended to be more distributed and often scattered in autistic cortex (). For fcMRI analyses, this implies that activity in a seed volume (or region of interest) will be less coherent across different voxels (or volume elements) within the seed. Incoherence of time series within a region will likely reduce correlations of time series between regions. Secondly, there was greater variability of activation maps across individuals with autism (compared to control individuals). To date, any study that examined single-subject data has supported this latter finding of atypical individual variability, which is also reasonable on theoretical grounds given the genetic and neuroanatomical findings discussed in earlier sections.
Fig. 2 Activation patterns for repetitive index finger movement in three men with autism and three gender and age-matched control participants. All participants show activation clusters in the vicinity of the central sulcus contralateral to the side of movement (more ...)
If task-driven activity in a region of interest is slightly reduced in an autism group for the two above reasons, cross-correlations across regions are likely to be similarly reduced. It is unclear, however, whether this may truly inform us about underconnectivity. Fortunately, there are some methodological decisions that can address the issues raised above, at least in part. Scanning in the resting state may appear a solution, but it should be noted that the mind is not usually ‘blank’ during rest and participants tend to engage in cognitive activities (Raichle et al. 2001
). Since this activity is uncontrolled experimentally and may be systematically different in ASD (Kennedy et al. 2006
), the resting state appears undesirable. An alternative is the statistical removal of effects driven primarily by a specific task. Are functional connectivity results based on interregional BOLD cross-correlation affected by this procedure?
One study by Villalobos and colleagues (2005)
examined functional connectivity between primary visual cortex and other parts of the brain during two visuomotor conditions, one of which was extremely easy (pressing a button with an index finger each time a blue dot appeared on a hand outline on a screen) whereas the second one was slightly harder (pressing fingers in 6-digit sequences prompted by the dot appearing on corresponding fingers on the screen). An adapted boxcar model for this design was used as an orthogonal regressor, thus removing effects in BOLD time series that were primarily driven by activation (i.e., the alternation of blocks in the harder and easier conditions). In this study, time series of detected head motion (for each axis and rotation) were also used as orthogonal regressors to minimize effects of head motion. Finally, BOLD time series were low-pass filtered at 0.1 Hz given that functional connectivity effects of interest are known to occur below this frequency (Cordes et al. 2001
). A main finding from the study by Villalobos et al. (2005)
showed reduced functional connectivity between primary visual cortex and inferior frontal cortex (). Since inferior frontal cortex is the presumed site of mirror neurons (Rizzolatti and Craighero 2004
), the finding is of interest in the context of recent proposals of impairment of the mirror neuron system in autism (Williams et al. 2001
). Although the autism group in the study by Villalobos et al. showed many regions of significant fcMRI effects, especially in parietal and subcortical regions, direct group comparisons yielded only effects of greater interregional correlations in the control group, but no inverse effects. These results would thus appear consistent with the underconnectivity theory.
Fig. 3 Effects of functional connectivity (correlation of BOLD time series) with primary visual cortex (A, B) and thalamus (C). Overlays are color-coded showing results of direct statistical group comparisons. Connectivity with V1 is reduced for a group of 8 (more ...)
Not all of the few autism fcMRI studies have been able to replicate generalized underconnectivity in terms of reduced interregional BOLD correlations, however. Welchew and colleagues (2005)
examined Pearson correlation matrices for 90 cortical and subcortical regions of interest in 13 adults ASD participants. Although they did find regionally specific disconnectivity for the amygdala and parahippocampal gyrus (in their comparison with matched controls), no evidence of generalized underconnectivity was seen. It should be noted that participants were scanned during viewing of faces at different levels of fearful expressions. The findings in the medial temporal lobe may have therefore been driven by group differences in activation effects, rather than by functional connectivity effects in the strict sense (as explained above). The overall picture of group comparisons for all 90 regions was a mixture of increased and reduced correlations, which is inconsistent with general underconnectivity.
Two studies from our group have recently explored subcortico-cortical functional connectivity in autism. Participants performed the simple visuomotor tasks described previously. One study examined functional connectivity of the thalamus (Mizuno et al. 2006
). Our expectation of reduced thalamocortical fcMRI effects – based on some previous findings implicating the thalamus (Tsatsanis et al. 2003
) and thalamo-cortical connections (Chugani et al. 1997
) – was not confirmed. On the contrary, functional connectivity between thalamus and several fronto-parietal regions (including perirolandic and insular cortices) was greater than in a matched control group (). A second study investigating functional connectivity between caudate nuclei and cerebral cortex (Turner et al. 2006
) was also inconsistent with general underconnectivity, yielding instead a picture of widespread and distributed clusters of greater caudato-cortical connectivity in frontal and parietal lobes of participants with autism.
One tempting interpretation of the current findings would be generally reduced cortical connectivity in autism, both within and between hemispheres, but partly increased functional connectivity between subcortical structures, such as basal ganglia and thalamus, and cerebral cortex. However, as much as there are methodological caveats related to the cortical underconnectivity findings, such caveats also apply to the subcortical fcMRI data, which were acquired in a small sample of eight high-functioning male participants with autism, for a single visuomotor task paradigm, and using a specific set of data processing techniques. For example, it cannot be ruled out that task performance was associated with generally greater arousal in participants with autism, which could result in overall greater interregional correlations of the BOLD signal.