A recurring theme in analyses of autistic cognition is a difficulty updating the scope and focus of attention [14]. Behavioural studies suggest that this operational rigidity may stem from an inability to reorient attention rapidly. An autistic deficit in rapid shifting of attention has been observed in cases of shifts between sensory modalities [29], between spatial locations [88,89,80,82,81,44,9], and between object features [30,71]. Related to this theme of inefficient selective attention are findings of impairment in engaging visual attention in the presence of distractors [13], and, in patients with pervasive developmental disorder, a deficit in dividing attention between auditory and visual channels and between visual attributes [19].
These behavioural observations of autism are complemented by neurophysiological results which reveal that even when people with autism produce normal behavioural output, they tend to do so by abnormal physiological means. Frontal negativities associated with sustained attention are reduced or absent in the autistic brain [28,21], the frontal late positive component to peripheral visual stimuli is delayed [83], and the visual P3b is highly variable [28] with a somewhat low average amplitude [67,21,87,83]. Abnormality in the allocation of spatial attention manifests from the very earliest attention-sensitive stages of processing: whereas normally the distance of a stimulus from the spatial focus of attention is indexed by a smooth decrease in P1 amplitude, the P1 in autism decreases with distance either precipitously or not at all [79].
In addition to these failures of normal modulation, neural systems in the autistic brain are often inappropriately activated. The visual N2 to novel stimuli is larger when a person with autism is performing a task than when (s)he is passively observing, even when these novel stimuli are not relevant to the task in question [49]. This inappropriate activation occurs across modalities, also: when a response is required to an auditory stimulus, autistic children manifest an enhanced P3 at occipital sites overlying visual processing areas [50]. During shifts of attention between hemifields, the normal, spatiotopically selective augmentation of the visual steady-state evoked potential is absent, and instead both hemispheres activate indiscriminately in response to any demand to shift attention [9]. In general, perceptual filtering in autism seems to occur in an all-or-none manner, with little specificity for the location of the stimulus, for the behavioural relevance of the stimulus, or even for the sensory modality in which the stimulus appears. Autistic attention, it seems, is founded more on the coarse control of general arousal than on selective activation of specific perceptual systems.
As people with autism are prone to excessive motor activity and anxiety, movement artefacts and performance difficulties have limited the population on which fMRI can be performed successfully. Most studies have been restricted to subjects in small numbers (often six or seven) in often heterogeneous groups combining individuals with high-functioning autism and those with Asperger syndrome. Despite these limitations, fMRI has produced consistent support for the hypothesis of inappropriately intense and inappropriately distributed sensory activation, across a variety of tasks. These results include abnormally high activity in ventral occipital visual areas during performance of the Embedded Figures Test, even while prefrontal and parietal activations are abnormally low [72]; heightened activity during face processing in peristriate cortex [32], inferior temporal gyrus [75], and other areas outside the fusiform `face area' [70] while fusiform activity is abnormally low; heightened activity in superior temporal gyrus during inference of mental state from pictures of eyes, while prefrontal and medial temporal activations are abnormally low [7]; decreased connectivity between extrastriate visual areas and prefrontal and temporal areas associated with inference of mental state, while prefrontal and temporal activations are again abnormally low [20]; and, during motor tasks, a lack of deactivation of early visual areas [63] and oddly distributed patterns of activation throughout Rolandic cortex [63] and cerebellum [2].
At least as important as this hyperactivation and abnormal patterning of primary perceptual processes is the accompanying hypoactivation in regions that normally subserve more complex cognitive processes: a wide range of tasks from Embedded Figures to face perception evokes unusually low activations in regions that normally integrate individual features or manage subordinate processes. This apparent impairment in complex cognitive processing suggests a failure of top-down attentional control, and behavioural observations of autism are consistent with such a view. An investigation of autism in terms of Mirsky's [61] factor analysis of attention demonstrates that autistic impairments load on the Focus-Execute and Shift components - elements of attention that involve flexibility in the application of cognitive and psychomotor processes [40]. This impairment in top-down control is evident even in the domain of eye movements, where people with autism show failures of inhibition in anti-saccade and delayed-saccade tasks in the context of normal performance in a pro-saccade task [60], and reduced dorsolateral prefrontal and posterior cingulate activity in a comparison of the delayed-saccade condition to the pro-saccade condition [54]. Although clear physiological and behavioural evidence exists both for abnormalities of top-down attentional control and for abnormalities of perceptual processing, the developmental relationship between these two domains of impairment remains an open question. The maturation of top-down processes may be influenced by abnormal perceptual inputs, and conversely, perturbations in perceptual systems may be induced by abnormal top-down control.
Remarkably, despite all these persistent abnormalities in the physiology of attention and perception, in many attention tasks that do not involve rapid shifting people with autism perform at normal or near-normal levels [29,9]. This divergence between behaviour and physiology indicates the operation of some compensatory process. The physiological evidence of high general arousal combined with low selectivity suggests that this compensatory process operates at a higher stage of processing, belatedly sorting out relevant stimuli from an indiscriminately amplified background. Electroencephalography [56,41,62,8] and functional imaging [69,25,45,46,86,85,84,57,48] have pinpointed an effect of early visual selection in ventral occipital cortex, and two fMRI studies suggest that intraparietal cortex mediates a complementary process in which irrelevant distractors that have passed through earlier filtering are actively inhibited [90,11].
Our examination of these phenomena involves observing the effect of unilaterally directed attention on brain activation driven by bilateral rapid serial presentation of visual stimuli. If indeed autism involves a failure of early selectivity, one could expect not only an absent effect of left-directed versus right-directed attention in ventral occipital visual areas, but also a heightened effect in intraparietal cortex, as this higher-level process becomes overloaded with demands to suppress stimuli from the irrelevant hemifield. In order to test this hypothesis, we applied fMRI (1) to map the specific loci of attention-related activations within ventral occipital and intraparietal cortices in individual subjects and (2) within these individually identified loci, to quantify the effect of left-hemifield versus right-hemifield foci of attention on the level of brain activation. Superior parietal lobule, a region known to be involved in visual spatial attention, was also included for purposes of comparison.
Because of individual variability in the detailed anatomy of intraparietal sulcus [64,11] and because of the small number of high-functioning, compliant subjects available, in our attention comparison we have purposely avoided analytical techniques based on blurring the data and warping onto a standard reference brain. The within-subjects region-of-interest approach that we apply has been suggested by several authors as a way of dealing with anatomical variations between individuals [66,24,90,11] and between groups of patients and controls [75,70,63,16,1]. Our statistical methods are tailored to this approach of detailed within-subjects analysis of a small sample: our technique of permutation testing is more powerful than the analogous parametric techniques [10] and is particularly suited to the problem of detecting differences between small groups of subjects [65]. Other fMRI studies on groups as small as six subjects with autism [72,7,70] have yielded significant results.