The ability to attend to relevant stimuli and to filter out irrelevant ones is fundamental to the normal development of a child. Joint social attention, in particular, rests on an ability to shift the attentive focus rapidly between a caregiver and some external object [Bakeman & Adamson 1984]. When a parent shows an infant a toy, for example, the infant must register not only the image of the toy, the texture of the toy, and the sound made by the toy, but also the parent's voice, facial expression, and gestures in response to the infant and the toy. In order to integrate all these stimuli into coherent percepts, the developing child must rapidly alter the scope and focus of attention among many sensory modalities and locations. An accumulation of behavioural evidence indicates that such task-based control over the scope of attention is lacking in autism [Burack & al. 1997]. Physiological studies suggest that this deficit in attentional control reflects a lack of specificity in perceptual gating, that is, in the process of selecting a few relevant stimuli from the large set of sensory inputs and conveying those stimuli into higher-order perceptual processing. A complete exploration of this hypothesis requires physiological measures of changes in perceptual gating in response to changing task demands.
Poor control over the scope of attention is most evident in tasks that demand rapid reorganisation of perceptual resources in response to changes in incoming stimuli. In a widely applied task developed by Posner & al. [1984], a cue informs subjects about the likely location of a subsequently appearing target. The cue may be spatial, as in a highlighting of the target area, or symbolic, as in an arrow pointing to the target area. After a variable delay from the onset of the cue, the target may appear in the cued (valid) location or in the uncued (invalid) location. The reaction times of normal subjects to these targets show a validity effect, that is, a cost of invalid cueing and a benefit of valid cueing. In a high-functioning group of adolescents and young adults with diagnoses of autism or Asperger syndrome, Wainwright-Sharp and Bryson [1993] found no validity effect when the target was presented after a short (100ms) delay, and a larger than normal validity effect at a long (800ms) delay. These results are consistent with a model of slowed re-orienting of attention: the 100ms cue-to-target delay gives persons with autism too little time to apply the information given by the cue, and as a result, there is no difference in reaction time between valid and invalid trials. Conversely, at the 800ms delay, persons with autism, having had sufficient time to shift their attention to the cued location, must implement another slow shift in order to respond to a target at the uncued location. These abnormal validity effects in autism are overlaid on a pattern of overall slowed responding due to motor apraxia. The two effects, one a general slowing and the other an interaction with cue-to-target delay, can sometimes be difficult to factor out. However, the autistic pattern of validity effects is present even when accuracy of discrimination is used as a measure instead of speed of response [Townsend & al. 1996], thus completely removing any motor confound.
Although differences in diagnostic criteria, age groups, IQ, and control groups make individual studies difficult to compare, the finding of disordered control over the scope of attention in autism is in general corroborated by other studies. High-functioning adults with diagnoses of autism or Asperger syndrome show a difficulty distributing attention in order to detect targets at central and lateral positions [Wainwright-Sharp & Bryson 1996]. In low-functioning children with autism, an attention-directing frame around the relevant region of the visual field improves performance, but presentation of distractor stimuli within the frame negates this effect [Burack et al. 1994]. In ten adult savants with diagnoses of autism or PDD-NOS, Casey et al. [1993] found a heightened validity effect not only at the 800ms cue-to-target delay but also at 100ms, illustrating the potential for variability in results when diagnostic criteria are loose. Even more important than diagnostic stringency is the question of the ecological validity of the tasks employed: any experiment in which individual stimuli are presented in discrete trials separated by long pauses is a poor reflection of real-world tasks, which demand continuous re-orienting of attention within a constant stream of stimuli.
In a paradigm more reflective of the constant shifting demanded by real-world situations, Courchesne & al. [1994a] measured the accuracy of target detection in two simultaneously presented streams of information, one auditory and the other visual. A target in the currently attended modality served as a cue to shift attention to the other modality. So, for example, a high tone in a background of low tones signalled subjects to stop attending to tones and to begin watching for a red flash in a background of green flashes. On detecting the red flash, subjects had to begin ignoring the flashes and listening to the tones again. Adolescents with autism uncomplicated by severe mental retardation (PIQ > 70) showed a deficit in responding to targets in different modalities when those targets were separated by less than 2.5 seconds. A like result was obtained for the case of shifting between separate visual attributes (form and colour) of a single stimulus [Courchesne & al. 1994b]. Reinforcing these results is the impairment of persons with autism at distributing attention across simultaneous auditory and visual Continuous Performance Tests [Casey & al. 1993]. These complex tasks are an advance over single-trial paradigms in addressing the problem of ecological validity. However, behavioural methods are limited in their ability to relate task performance to the underlying biology.
Electrophysiological results on autism associate the aforementioned behavioural pathologies with abnormal modulation of excitability. In response to an attended stimulus, the normal brain produces a series of electrical potentials (voltages) which can be recorded from electrodes on the scalp. Over frontal cortex, salient stimuli that call for responses or that differ from context during periods of sustained attention evoke negative potentials. At more posterior scalp sites, attention modulates a series of potentials evoked by sensory stimuli. One of the earliest of these sensory potentials, the P1, is a positive voltage appearing over occipital cortex about 100ms after presentation of a visual stimulus. The P1 becomes gradually smaller as stimulation occurs farther away from the spatial focus of attention [Mangun & Hillyard 1988]. A later, negative potential over parietal cortex, the N2, is augmented during attentional selection of task-relevant stimuli [Eimer 1996]. Finally, at a latency of about 400ms, a positive potential P3b appears with presentation of a task-relevant stimulus but not with irrelevant stimuli.
Autism presents a remarkable disruption of all these attention-related potentials. Despite normal behavioural performance in tasks of static rather than shifting attention, frontal negativities are entirely absent [Courchesne & al. 1989; Ciesielski & al. 1990] and the visual P3b is highly variable [Courchesne & al. 1989] with a slightly low average amplitude [Novick & al. 1979; Ciesielski & al. 1990; Verbaten & al. 1991]. The P1, instead of declining gradually with distance from the focus of attention, decreases either precipitously or not at all [Townsend & Courchesne 1994]. In addition to these failures of normal modulation, neural systems in the autistic brain often are inappropriately activated. The visual N2 to novel stimuli is larger when the 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 [Kemner & al. 1994]. 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 [Kemner & al. 1995]. Perceptual gating 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. Lacking normal mechanisms for gating sensory signals into higher-order processing, the autistic brain seems to accomplish attentional tasks by some other, substitute mechanism.
The aforementioned studies have assessed statically allocated attention, but by themselves they have little to say about what goes on in the autistic brain when a demand is made to shift attention. The application of evoked-potential methods to such shifts is not straightforward, since the most obvious electrophysiological indications of a shift in attention appear only after the shift has occurred. A cue to shift attention evokes a P3 response, but this response cannot be closely associated with the attentional shift itself since in normal subjects the shift is already fully implemented when the P3 is only just beginning. The P700 shift-difference (Sd) potential appears specifically in circumstances in which an attentional shift has occurred [Akshoomoff & Courchesne 1994], but again, it can furnish only retrospective information on the process of attentional shifting.
One way around this problem is to use perturbations of a steady-state visual evoked potential (SSVEP) as an index of attentional modulation. The SSVEP is an electrical resonance of the visual system, produced when stimuli are flashed periodically at frequencies in the alpha band [Regan 1977]. Just as the height attained by a child's swing grows large if the swing is pushed at appropriate intervals, the voltage generated by the visual system grows large if a visual stimulus is flashed at a resonant frequency, about ten times per second. The backward and forward swings are synchronised to the pushes, just as the SSVEP voltage peaks are phase-locked to the flashing stimulus.
In normal subjects, the SSVEP is augmented in the hemisphere contralateral to the attended visual hemifield [Morgan & al. 1996; Belmonte 1998]. Conversely, background oscillations occurring in the same frequency band as the SSVEP, which are not phase-locked to the stimulus and probably index the level of task-irrelevant processing, are decreased contralateral to the attended hemifield [Belmonte 1998]. When a subject is asked to shift attention across the midline, the pattern of SSVEP modulation inverts beginning at about 300ms [Belmonte 1998].
The hypothesis of nonspecific sensory gating in autism predicts an absence of hemispheric specificity in the modulation of the SSVEP and of background amplitude during rapid attentional shifts. Instead of being modulated in opposite directions, amplitude measures in the two hemispheres should be similar to each other when the stimulus onset asynchrony (SOA) between the cue in the previously attended location and the target in the newly attended location is brief. At longer SOAs, when people with autism are not so drastically impaired, a pattern of modulation similar to normal may emerge, but with a prolonged latency reflecting the prolonged time course of the attentional shift.