Mechanisms of Diminished Attention to Eyes in Autism

Abstract

Objective: Two hypotheses, gaze aversion and gaze indifference, are commonly cited to explain a diagnostic hallmark of autism: reduced attention to others' eyes. The two posit different areas of atypical brain function, different pathogenic models of disability, and different possible treatments. Evidence for and against each hypothesis is mixed but has thus far focused on older children and adults. The authors evaluated both mechanistic hypotheses in two sets of experiments at the time of initial diagnosis.

Method: Eye-tracking data were collected in 86 2-year-olds: 26 with autism, tested at initial diagnosis; 38 matched typically developing children; and 22 matched developmentally delayed children. In two experiments, the authors measured response to direct and implicit cueing to look at the eyes.

Results: When directly cued to look at the eyes, 2-year-olds with autism did not look away faster than did typically developing children; their latency varied neither categorically nor dimensionally by degree of eye cueing. Moreover, direct cueing had a stronger sustained effect on their amount of eye-looking than on that of typically developing children. When presented with implicit social cues for eye-looking, 2-year-olds with autism neither shifted their gaze away nor more subtly averted their gaze to peripheral locations.

Conclusions: The results falsify the gaze aversion hypothesis; instead, at the time of initial diagnosis, diminished eye-looking in autism is consistent with passive insensitivity to the social signals in others' eyes.

Keywords: Autism; Child Psychiatry; Eye Contact; Gaze Aversion.

Figure 1. Experimental design for direct (Experiment 1) and implicit (Experiment 2) cueing for increased eye-looking; reduction in eye-looking in toddlers with ASD at time of initial diagnosis; and completion-of-protocol controls

(A) In Experiment 1, toddlers with and without ASD, matched on age and cognitive function, were directly cued to look either more or less at the eyes by means of a prestimulus cueing target presented on an otherwise blank screen. The cueing target was presented and then replaced with video of a face looking directly at the child or with video presenting non-face information. (B) Location of the cueing target varied with respect to the eyes and mouth of each face. Degree of direct cueing for eye-looking was measured relative to the center of the nose. (C) In Experiment 2, the same children were implicitly cued to look either more or less at the eyes. (D) The timeline of implicit cueing for eye-looking was defined by normative data. We measured the probability of eye-looking at each moment in time in typically-developing toddlers, and then tested the validity of these implicit cues by measuring how successfully they predicted the eye-looking of independent typically-developing toddlers. (E) Toddlers with ASD displayed a reduction in attention to others' eyes, consistent with either hypothesis being tested. (F) As a control for completion of protocol, we measured the percentage of trials successfully cued in Experiment 1 within each group. (G) As a control for completion of protocol in Experiment 2, we measured the percentages of time spent fixating, saccading, blinking, and looking offscreen while video trials were presented. (H, I) As a control for the capacity to execute saccadic eye movements, we measured the main sequence relationship between saccade velocity and amplitude in (H) typically-developing toddlers and (I) in toddlers with ASD. (J) As a control for differences in reaction time, we measured latency to first saccade across all trials. ASD = autism spectrum disorder; TD = typically developing.

Figure 2. Following direct cueing to look at the eyes, two-year-olds with ASD do not look away more rapidly than their typically-developing peers

(A, B) Expected results for (A) the gaze aversion hypothesis and (B) gaze indifference hypothesis. (C) In typically-developing toddlers, latency to first saccade did not vary categorically as a function of the content of cued target region: eyes, mouth, or non-face. (D) In toddlers with ASD, latency to first saccade did not vary categorically as a function of the content of cued target region: eyes, mouth, or non-face. (E) In typically-developing toddlers, latency to first saccade did not vary dimensionally by degree of physical cueing for eye-looking: closer to or farther from the eyes. (F) In toddlers with ASD, latency to first saccade did not vary dimensionally by degree of physical cueing for eye-looking: closer to or farther from the eyes. ASD = autism spectrum disorder; TD = typically developing. ^a r = -0.40, p = 0.28 ^b r = -0.39, p = 0.30

Figure 3. After direct cueing, despite equivalent latencies to first saccade, two-year-olds with ASD show higher levels of sustained looking than their typical peers

(A, B) Expected results for (A) the gaze aversion hypothesis and (B) gaze indifference hypothesis. (C) Sustained effects of physical cueing are measured as the time-until-decay of a significant association between physical cueing location and percentage of looking to the cued region. For discrete time periods, plotted in 133.3 ms intervals, the association between cueing target location and percentage of fixation to cued region is plotted as scatter plots with regression lines for typically-developing toddlers (top row, yellow) and for toddlers with ASD (bottom row, green). Regression lines for time periods with significant associations (p < 0.05) are plotted with more saturated shades for each group. (D) Measured continuously, the association (r value of correlation) between degree of physical cueing for eye-looking and percentage of eye-looking shows rapid decline in typically-developing toddlers (yellow) and slower decline in toddlers with ASD (green). Shaded regions show bootstrapped 95% confidence intervals. Despite equivalent latencies to first saccade (Figures 1 & 2), toddlers with ASD show a stronger, more persistent association between cueing target location and subsequent fixation location, looking more at the eyes when cued for eye-looking and more at the mouth when cued for mouth-looking—an indicator of relative insensitivity to the content at the cued target location in toddlers with ASD. ASD = autism spectrum disorder; TD = typically developing.

Figure 4. At moments with strongest implicit cueing for increased eye-looking, two-year-olds with ASD neither look fully away from the eyes nor shift gaze to the periphery

(A, B) Expected results for eye-looking in response to implicit cueing for eye-looking based on (A) the gaze aversion hypothesis and (B) gaze indifference hypothesis. (C) In typically-developing toddlers, implicit cueing for increased eye-looking, ranked by quartiles, positively predicts increased eye-looking among independent typically-developing children using leave-one-out cross-validation. (D) In toddlers with ASD, implicit cueing for eye-looking does not result in gaze aversion; instead, a significant increase in probability of eye-looking is observed. (E-H) Fixation density in two-year-olds with ASD is unrelated to levels of implicit cueing for increased eye-looking. (E, F) Expected results for fixation density in response to implicit cueing for eye-looking based on (E) gaze aversion hypothesis and (F) gaze indifference hypothesis. (G) In typically-developing toddlers, implicit cueing for increased eye-looking, ranked by quartiles, positively predicts greater fixation density; as implicit cueing increases, typically-developing toddlers are more likely to fixate on the same location at the same time. (H) In toddlers with ASD, fixation density is unrelated to levels of implicit cueing for eye-looking, indicating that they do not avert gaze to peripheral locations during periods of increased implicit cueing for eye-looking. ASD = autism spectrum disorder; TD = typically developing. ^a F_3,111 = 34.30, p < 0.001 ^b F_3,75 = 12.78, p < 0.001 ^c F_3,111 = 23.69, p < 0.001 ^d F_3,75 = 1.60, p = 0.20