Our finding of increased errors on the semantic category decision task in the ASD group (compared to controls) is consistent with previous studies suggesting impaired lexicosemantic accuracy in autism (Dunn et al., 1999
; Tager-Flusberg, 1981
; Toichi & Kamio, 2001
). Nonetheless, ASD participants were clearly cooperative, performing at levels far greater than chance.
On the perceptual control task, both groups were equally accurate, consistent with studies suggesting that visual search is a spared ability in ASD (O’Riordan, 2004
). Unexpectedly, reaction times were significantly longer for the ASD group, which may be related to impaired attention shifting. Townsend et al. (1996)
found slowed visual orienting in autism when participants had to shift attention from a central fixation point to stimuli presented more peripherally. In our control task, each letter string was preceded by a central fixation cross (). Target letter location was randomized within consonant strings and required spatial attention shifts, potentially explaining longer RTs in ASD participants.
Our imaging results showed that semantic category decision was associated with extensive left inferior frontal activation in the control group, consistent with numerous lexicosemantic studies in healthy adults (Damasio et al., 1996
; Martin & Chao, 2001
; Petersen et al., 1988
; Tyler et al., 2003
; Wiggs et al., 1999
). In the ASD group, inferior frontal activation clusters in areas 45 and 47 were comparatively small, possibly consistent with evidence of left-hemisphere dysfunction and rightward asymmetry of frontal language areas in autism (Bruneau et al., 2003
; Herbert et al., 2002
; Rojas et al., 2002
; Sussman & Lewandowski, 1990
). However, we did not observe right frontal activation for semantic decision in the ASD group, nor did direct group comparisons yield significant differences in left prefrontal cortex. Absence of group differences in inferior frontal cortex could be related to heterogeneity within the autism population. De Fosse and colleagues (2004)
reported atypical volumetric asymmetries in inferior frontal lobes only in autistic boys with language impairment, but not in autistic boys with normal language, whose mean VIQ (97.7) was close to mean VIQ in our ASD sample (91.6).
Large areas of inverse effects (reduced activation for semantic decision compared to the perceptual control condition) were found for the normal control group in middle occipital, temporoparietal, and bilateral frontal regions. These effects may be related to our control task. Manjaly et al. (2003)
found similar activation patterns for visual search in an embedded figures test compared to a visual match condition. As in our perceptual control condition, participants had to identify a stimulus in a more complex visual pattern. Our control task was relatively hard compared to easy baseline conditions often applied in functional imaging studies, since it had been calibrated to match the semantic decision condition on RTs and accuracy in a pilot sample of healthy adults. This explains why BOLD effects for the two conditions were overall balanced in our control group (with approximately equally extensive “activation” and “inverse” effects).
In the ASD group, a large additional area of activation was found in extrastriate visual cortex bilaterally, which was not seen in the control group (see ). This finding is significant in the context of previous studies suggesting qualitatively different lexicosemantic strategies in autism (Dunn et al., 1999
; Kamio & Toichi, 2000
; Toichi & Kamio, 2001
). As hypothesized, this strategy may involve increased visualization of target items. The areas of activation found in extrastriate visual cortex correspond to activations seen for mental imagery (Just et al., 2004a
; Mellet et al., 2000
), even when exclusively auditory stimuli were used as prompts for visual imagery (Just et al., 2004b
; Lambert et al., 2004
). In a recent study of verbal working memory, Koshino and colleagues (2005)
reported unusually high levels of extrastriate activity in autistic adults, consistent with the present findings.
Direct statistical comparison between the ASD and control groups revealed significant differences in extrastriate visual cortex bilaterally. These effects are unlikely to be explained by the inverse effects in the control group described above, as they occurred in different loci (cf. ). Our finding suggests that lexical representations in ASD may be more perceptually based, possibly because they are anchored in reduced experience (as described in the Introduction). As a result, adolescents and adults with ASD appear to process lexicosemantic stimuli in an immature fashion, continuing to rely heavily on perceptual components and visual imagery. Furthermore, reliance on such perceptual components may be associated with performance slightly below normal (see below).
Lexical organization is considered to be affected by the sensory modalities involved in the acquisition of word meanings (Martin & Chao, 2001
). However, little direct neuroimaging evidence is available to demonstrate an initial dependence of lexicosemantic organization on sensorimotor representations in children. Potentially consistent, Mills and colleagues (1994)
found that ERP components (N200 and N350) distinguishing known from unknown words were distributed across bilateral frontal, temporal and parietal lobes in 13–17 month old infants, whereas they were more localized to left temporo-parietal areas in 20 month olds. In a recent fMRI study on lexical association in children and young adults, Brown and colleagues (2005)
observed age-related activity increases in left frontal cortex, whereas age-related decreases were seen in extrastriate cortex bilaterally. These findings may reflect initial dependence of lexical representation on perceptual (especially visual) systems. This view is also supported by behavioral studies showing that perceptual information is a guide to word learning from early stages on (Smith et al., 1996
). In particular, children’s early word learning is largely based on visual information about object shape (Gershkoff-Stowe & Smith, 2004
; Samuelson & Smith, 1999
). This normal developmental profile would be consistent with our interpretation of visual cortical activation in older autistic participants during semantic decision as reflecting an ‘immature’ pattern of lexicosemantic processing.
We further tested this hypothesis in post hoc analyses examining the relation between atypical posterior activity and performance. We found that in the entire sample (both groups), this posterior activity was positively correlated with errors and RT. This suggests that indeed extrastriate activity was associated with a relatively inefficient mode of processing. However, this correlation between performance and extrastriate activation was more robust in the control group than in our ASD sample, for which it did not reach significance. This is surprising since the control group did not show significant activation in extrastriate cortex. It suggests that activity in visual cortex occurs in the typically developing brain during lexical processing in children, but then decreases with age (as discussed above, and consistent with the findings by Brown et al., 2005
). However, residual activity identified even in typically developing adolescents and adults in our study (several of whom showed ≥10 activated voxels in these regions) was still correlated with relatively low performance on semantic decision. These control participants could be characterized as displaying a subtly immature pattern of lexicosemantic activation (cf. Brown et al., 2005
). Most – but not all – individuals with ASD in our sample showed a corresponding association between extrastriate activity and performance, albeit at the lower end of the performance spectrum. Note, however, that in a block design – as in our study – performance effects cannot be analyzed on a trial-by-trial basis. Event-related fMRI studies will be necessary for a more detailed examination of the links between performance and activation profiles in posterior cortex.
With regard to general level of functioning, extrastriate activity showed a significant negative correlation with nonverbal performance IQs – but not with verbal IQs – in the ASD group, potentially suggesting association with low level of functioning in nonverbal domains. No such correlation was seen in the control group.
Taken together, the findings suggest that posterior activity during lexicosemantic processing reflects an initial perception-based strategy in young children that is gradually substituted by top-down frontal control in older typically developing children. Individuals with ASD tend to rely on a processing mode similar to the initial perception-based strategy even in adolescence and adulthood. Some typically developing adolescents and adults show residual traces of a perception-based processing mode as well. Although these are subtle (i.e., associated with minimal activity), they result in less efficient processing and therefore slightly lower performance accuracy.
A further post hoc analysis examining effects for individual categories showed that our finding in the autism group was not solely driven by a single category (), which suggests that visual imagery may play a general role in semantic processing and lexical retrieval in autism, rather than an exclusive role only in visually based representations. This is consistent with the recent finding by Kana and colleagues (2006)
of atypically strong extrastriate activity in ASD during sentence comprehension in particular for a low-imagery condition, for which only small effects of visual imagery would be expected.
Our above interpretation may appear more obvious for the categories “color” and “tool” (which can also be visualized) than for “feeling”. However, it is known that facial expressions associated with different emotions are characterized by specific visual features (Ekman, 1993
). Processing of these features is associated in neurotypical adults with activation in occipital cortex, besides medial prefrontal cortex, amygdala, cingulate gyrus, and insula (Ishai et al., 2000
; Phan et al., 2002
). Similar sites of activation including occipital cortex have been recently identified for the processing of emotional words (Kensinger & Schacter, 2006
). Finally, activation associated with visual imagery has been demonstrated not only in extrastriate, but also in primary visual cortex (Chen et al., 1998
), which appears to be involved specifically when imagery relates to high resolution detail or physical shape (Kosslyn & Thompson, 2003
). Our finding of activation in area 17 in the ASD sample could therefore suggest such local-level imagery during semantic decision.
Our results are more broadly consistent with atypical reliance on extrastriate activity in autism during a variety of tasks. In one study on face perception in autism, unusually robust occipital activity was found in medial occipital area 19 (Hubl et al., 2003
). In a study on visually prompted finger movement, significantly greater activation was seen in autistic individuals compared to controls in lateral portions of area 19 (Müller et al., 2001
). During visually prompted sequence learning autistic individuals showed atypically strong activation in visual cortices during later learning stages, despite mild behavioral improvements (Müller et al., 2003). Together with the convergent results by Koshino et al. (2005)
on verbal working memory described above, these findings suggest that individuals with ASD may rely on perception-based processing modes even after prolonged exposure to a given task, whereas control subjects tend to use such modes only initially, either in childhood or at later ages before practice becomes effective (depending on the type of task).
Just and colleagues (Just et al., 2004a
) recently reported atypical neurofunctional profiles for sentence comprehension in high functioning autism. As in our study, inferior frontal activation clusters were smaller in their autism group compared to controls, but no direct statistical group comparison was presented. Further, BOLD signal cross-correlations between a number of cortical areas, considered measures of functional connectivity, were consistently lower for the autism group, suggesting deficient integration of individual components into more complex meaning in the autistic brain.
Although our study provides neurofunctional evidence that is consistent with an ‘immature’ lexicosemantic strategy involving visual imagery in ASD, a number of questions remain. Current literature suggests areas activated in our ASD group are involved in visual imagery regardless of input modality (Just et al., 2004b
; Lambert et al., 2004
). However, the effects of different input modalities (i.e., visual, auditory) remain to be explored. To our knowledge, only one functional imaging study to date has examined semantic functions using auditory stimulation in autism. In this study (Müller et al., 1999
), a small sample of autistic adults showed atypical absence of leftward asymmetry of perisylvian activations during passive listening to meaningful speech, but normal levels of left inferior frontal activity for sentence generation based on an auditory word prompt. Neither of the conditions was associated with significant activity in extrastriate cortex.
It is likely that atypical functional organization, as demonstrated in our study, relates to recent anatomical findings of brain overgrowth during the first two years of life in autism (Courchesne et al., 2001
). Neurofunctional abnormality may thus result in part from aberrant neuronal growth in the absence of environmental influence. Specifically, it has been proposed that the reduction of long-distance, reciprocal cortical connectivity leads to defects in the processing of complex information (Courchesne & Pierce, 2005
). Although our study was not designed to address this issue directly, our findings suggest that in ASD perceptual brain regions play a relatively strong role during lexicosemantic processing, whereas in healthy controls top-down functions of supramodal frontal regions are more predominant. One may also note that the visual areas active in our task are considered relatively preserved in ASD, possibly due to their early course of maturation (Carper et al., 2002
). Atypical reliance on posterior brain regions for language tasks may result from this relative integrity of visual cortices.
In conclusion, our study is consistent with previous findings suggesting atypical organization of the lexicosemantic system in autism. Such atypical organization may relate to lack of interpersonal experience, which is the primary basis of word learning in typically developing children. Reduced interpersonal language experience is likely to result in greater reliance on nonverbal information. This is supported by the results of the current study showing that individuals with ASD exhibit atypical activity in extrastriate visual regions during semantic category decisions.