Big heads, small details and autism

Abstract

Autism is thought to be associated with a bias towards detail-focussed processing. While the cognitive basis remains controversial, one strong hypothesis is that there are high processing costs associated with changing from local into global processing. A possible neural mechanism underlying this processing style is abnormal neural connectivity; specifically reduced structural or functional connectivity between brain regions might lead to good exemplar-based processing but poor generalisation. Abnormal neural connectivity has also been suggested to account for the increased incidence of macrocephaly in autism (increased head/brain size). The present study therefore investigated the effect of head size on the ability to switch between global and local processing in autism. 49 high-functioning 7–12 year olds with autism (12 with macrocephaly) were compared to 25 normally developing children in their performance on a Local-Global Switching task. Those children with autism who also had macrocephaly showed a greater processing cost when switching into global processing, or ‘zooming out’, than both the remaining children with autism and the control children. A second experiment revealed that macrocephaly in the context of normal development is not associated with difficulty switching into global processing but rather occurs in children who are physically large. Macrocephaly in the context of autism may therefore be a biological marker of abnormal neural connectivity, and of a local processing bias.

Introduction

Individuals with autism are thought to be good at processing ‘local’ details of information rather than applying the context and extracting the ‘global’ meaning, an idea that has been strongly endorsed by the autism community (e.g. Gerland, 1997). This ability to perceive the elements rather than the whole was first attributed to a lack of drive for meaning, termed ‘weak central coherence’ (Frith, 1989) and more recently described as a local bias (Happé, 1999; Happé & Frith, 2006). There has, however, been little agreement on the mechanism that might support this processing style.

When reviewing the evidence to date, the majority of studies employ designs in which both local and global processing are possible simultaneously and are in direct competition with each other. To pick a few notable examples, the Embedded Figures Task (first used in autism research by Shah & Frith, 1983) requires the participant to both ignore and inhibit Gestalt principles when viewing a picture that is designed to elicit global processing, and process the local elements instead. This is also the case in traditional Block Design tasks (first used in autism research by Shah & Frith, 1993), where the global Gestalt image must be rejected in order to segment the design into component elements. Hierarchical figure tasks performed under divided attention (first used in autism research by Mottron, Belleville, & Menard, 1999) similarly require participants to attend to both local and global aspects of the stimuli.

A few studies of the local processing bias in autism have attempted to separate the local and global constraints of tasks. For example, the segmented version of the Block Design task (Shah & Frith, 1993) removes the global image and therefore requires the participant to simply match the local elements, and hierarchical figures performed under selective attention (first used in autism research by Ozonoff, Strayer, McMahon, & Filloux, 1994) direct a participant's attention towards either local or global aspects of the stimuli. Interestingly, although local and global task components can be isolated, it seems that these paradigms are unable to capture the essence of a local processing bias; they indicate that global processing is intact rather than deficient and local processing is normal rather than enhanced in autism.

Instead, those tasks that appear to be most sensitive in detecting a local bias therefore either pit global and local processing against each other or require fast online responses that are capable of picking up this bias (see Happé & Frith, 2006 for a review of many recent studies). This body of research has led researchers with quite different theoretical ideas to agree that, when both local and global processing are simultaneously possible, local processing often takes precedence over global processing, resulting in a bias away from global and towards local stimuli (Happé & Frith, 2006; Mottron, Dawson, Soulieres, Hubert, & Burack, 2006; Plaisted, Saksida, Alcantara, & Weisblatt, 2003). While those studies finding group differences have been useful in identifying and supporting the idea of a bias towards an alternative processing style in autism, the presence of both local and global stimuli simultaneously does not allow for an analysis of how this processing style arises and what the mechanism behind it might be. Furthermore, in addition to assessing processing capacity, the majority of tasks do not require fast online responses, allowing higher-order strategy use to also play a role in task performance; this confounds results and makes it extremely hard to interpret which factors contribute to both positive and negative findings. It is therefore still unknown how a bias away from global and towards local processing comes about, given that both global and local processing appear to be normal in autism when processing in only one of these domains is required.

More recently, a handful of studies have together begun to suggest a mechanism for this bias, in terms of difficulties broadening, but not narrowing, the spread of attention or an inability to shift out of local processing and into global processing (Mann & Walker, 2003; Rinehart, Bradshaw, Moss, Brereton, & Tonge, 2001). Once individuals with autism are processing a small stimulus or element, they are slower and have more difficulty in zooming out and processing a large stimulus or element presented subsequently, but not vice versa. The suggestion being made is therefore that both local and global processing can be performed normally; however, a deficit in global processing may be seen when such processing is required immediately after local processing, thus requiring a fast switch from local to global. This deficit may also be present when a task involving local-global competition is open-ended, such as the Embedded Figures Task, as an individual with a local bias may remain stuck in local processing once they have shifted into this processing mode; the cost of switching into global processing will be high, making it easy to ignore global stimuli and possibly resulting in an enhancement in local processing ability relative to controls. While this shifting or switching deficit may at first be reminiscent of the executive function deficits commonly attributed to this disorder (see Hill, 2004 for a review), the crucial difference here is that the shifting deficit is proposed to be unidirectional: from local to global but not vice versa.

A particularly elegant novel paradigm, known as rapid serial object transformation (RSOT), has recently been introduced into the autism literature (López, Torres, & Valdés-Sosa, 2002; Valdés-Sosa, Torres, Iglesias, & López, submitted for publication). This task avoids some of the caveats of more traditional tasks: higher-order strategy use plays little role as fast online processing is necessary; and local and global processing are temporally separated which allows for a detailed analysis of the information the participant is processing and therefore a clearer interpretation of the results. The task uses hierarchical figures in a novel way so that they can be transformed to uncover either global or local meaningful components, rather than both simultaneously, presented as consecutive pairs of stimuli in an attentional blink paradigm. Stimuli can be presented under conditions that do not require attentional switching (e.g. a global stimulus followed by a global stimulus) or those that do (e.g. a local stimulus followed by a global stimulus). Participants often have trouble reporting a second stimulus that appears in quick succession (300–400 ms) after a first, as few attentional resources remain available to direct towards it, hence the attentional blink (Raymond, Shapiro, & Arnell, 1992). Moreover, if the participant is required to switch between two different modes of processing for the two stimuli, an additional attentional cost is required, lengthening the attentional blink and further reducing the probability of correctly reporting the second stimulus (Ward, 1982). This paradigm has been used to show that individuals with autism are less likely than controls to correctly identify a global stimulus following a local stimulus but not vice versa (López, Torres, & Valdés-Sosa, 2002; Valdés-Sosa et al., submitted for publication), indicating that switching from local into global processing has a higher processing cost for individuals with autism than typical observers.

The literature regarding a local bias in autism is by no means unanimous in its support, however; negative findings have been reported alongside the positive ones (e.g. Brian & Bryson, 1996; Edgin & Pennington, 2005; Hoy, Hatton, & Hare, 2004; Kaland, Mortensen, & Smith, 2007; Ropar & Mitchell, 1999; Schlooz et al., 2006). Furthermore, one study found impaired rather than the expected enhanced performance on three visuospatial tasks (Burnette et al., 2005). While some of these findings appear to result from methodological issues, such as group matching, task design and instructions, one explanation for such mixed results is the idea of heterogeneity of this processing bias in the autistic population; perhaps one subset of individuals with autism may exhibit a local bias more than other subsets and the proportion of these individuals may vary in different studies. It is also possible that a local bias may not be a single construct but have a number of different processes contributing to it (e.g. a global deficit independent of a local enhancement), and that only subsets of individuals with autism display an abnormality in any one aspect of it (Booth, Charlton, Hughes, & Happé, 2003; Lopez, Leekam, & Arts, 2008). While heterogeneity has rarely been studied, two recent experiments have reported the presence of an abnormally local bias in almost all children with autism tested (Jarrold, Gilchrist, & Bender, 2005; Pellicano, Maybery, Durkin, & Maley, 2006), while others have found only a small proportion of individuals with autism to show this processing style (Edgin & Pennington, 2005; Jarrold & Russell, 1997; van Lang, Bouma, Sytema, Kraijer, & Minderaa, 2006). From these latter studies, the studies which fail to find group differences and many other studies finding group differences, it is possible to infer from the group means and standard deviations that the range of performances seen in autism and control groups tend to overlap greatly, indicating that heterogeneity in local/global processing may well occur in autism.

How is the subgroup of individuals with autism to be identified that may be particularly liable to show a local processing bias? At the biological level, it has been suggested that a local bias may result from abnormal neuronal connectivity, due to either structural or functional differences. These might involve a lack of synchronisation in activation between relevant brain areas (Brock, Brown, Boucher and Rippon, 2002) or reduced long-range and increased short-range physical connectivity (Just, Cherkassky, Keller and Minshew, 2004), both resulting in a lack of binding of parts into wholes; or numerous and inefficient feedback connections resulting in a lack of top-down modulation of early sensory processing and a lack of integration of sensory processing with cognitive monitoring (Frith, 2003). These ideas are supported by a growing number of functional imaging studies showing reduced connectivity and a lack of top-down modulation particularly between frontal cortex and other brain regions (e.g. Bird, Catmur, Silani, Frith and Frith, 2006; Horwitz, Rumsey, Grady and Rapoport, 1988; Just, Cherkassky, Keller, Kana and Minshew, 2007; Koshino et al., 2008). All of these ideas predict that the abnormal connectivity would give rise to a preserved or enhanced ability for exemplar-based information processing, in addition to a reduced ability to generalise across examples or process information in context, reminiscent of a locally biased style of processing. While these accounts were intended to explain autism as a whole, it is possible that, like a local bias, there is heterogeneity in connectivity and that this may relate to differences in the degree of local bias shown by different individuals with autism.

One of the more consistent neurobiological findings in autism, which was also noticed by Kanner (1943), is of increased head and brain size. This is seen to different degrees in different individuals and the distribution of head size in the autism population appears to be normal but broader than in the typically developing population, with a higher mean (Lainhart et al., 2006). Approximately 20% of the autistic population are thought to have macrocephaly, when defined as having a head circumference greater than the 97th percentile of the normal population (Bailey et al., 1995; Fombonne, Roge, Claverie, Courty, & Fremolle, 1999; Lainhart, 2003; Stevenson, Schroer, Skinner, Fender and Simensen, 1997) and two postmortem studies of increased brain weight have supported this finding (Bailey et al., 1993; Bauman & Kemper, 1985). This enlargement appears to be general across the whole of the cerebral cortex (Hazlett et al., 2005) and be heritable, being present in the parents and siblings of individuals with autism (Lainhart et al., 2006; Miles, Hadden, Takahashi, & Hillman, 2000). Very recently, a possible genetic mutation in the PTEN gene has been suggested as the cause in some cases of macrocephaly in autism (Butler et al., 2005, Buxbaum et al., 2007).

However, macrocephaly cannot normally be detected until approximately 2 years of age (Courchesne et al., 2001, Lainhart et al., 1997; Stevenson, Schroer, Skinner, Fender, & Simensen, 1997) although brain imaging studies indicate that the increased rate of head growth starts around 12 months of age (Courchesne, Carper, & Akshoomoff, 2003; Hazlett et al., 2005). There is evidence that feed forward connections are established very early in normal development whereas feedback connections are continually refined through neuronal elimination processes of pruning and apoptosis (Price et al., 2006), in order to eliminate faulty feedback connections and optimise co-ordinated neural functioning. It is plausible that there may be a decrease in these elimination processes in autism, leading to an excess number of synapses and coinciding with the increased rate of head growth from about 12 months of age (Frith, 2003). This suggestion is therefore consistent with the abnormal neuronal connectivity hypotheses mentioned above and therefore also with a local processing bias; this connection between abnormal connectivity and macrocephaly is however speculative. In support of this hypothesis, a computational model of autism has been constructed in which a lack of generalisation results from an increase in units (Cohen, 1994, Gustafsson, 1997).

Macrocephaly in autism has been linked to a number of behavioural features with mixed results, including increased social impairment and language delay (Lainhart et al., 2006) but also to reduced social impairment and improved adaptive functioning (Dementieva et al., 2005, Lainhart et al., 1997). At present, only one study has attempted to relate brain size with cognitive function however; Deutsch and Joseph (2003) examined the IQ profiles of individuals with autism and found that relatively high performance IQ compared to verbal IQ was associated with larger head sizes.

It is possible therefore that increased head size is related to the local bias seen in autism. Given that macrocephaly is not universal in autism, some individuals may display more of a local bias than others, which might help to account for the variable results seen in the literature on cognitive style in autism. As macrocephaly is so clear an index by which to identify individual cases, one approach is to focus entirely on such individuals and establish whether there is any possible connection with a difficulty broadening attention or ‘zooming out’. To this end, the effect of head size on the ability to shift from local into global processing was examined.