Severely Attenuated Visual Feedback Processing in Children on the Autism Spectrum

IC effect

The results of the primary analysis are summarized in Table 4. As evident in Figure 5, for the IC effect time windows, there was a group × stimulus configuration × stimulus location interaction (F₍₂₁₁₄₎ = 5.746, p = 0.004, η_p² = 0.092). For the central stimulus presentation, NT individuals demonstrated a strong IC effect characterized by greater negativity in response to the IC versus NC stimulus configuration. The IC effect was absent in NT participants for laterally presented stimulus configurations. In contrast, ASD individuals demonstrated no IC effect regardless of stimulus location. As expected, there was also a stimulus location × hemisphere interaction (F_{(1.177,67.063)} = 4.865, p = 0.025, η_p² = 0.079, Greenhouse-Geisser–corrected) whereby both hemispheres showed roughly equivalent activation in response to central stimulus presentations, whereas lateral stimulus presentations evoked greater magnitude visual potentials in the contralateral hemisphere. There were no main effects of group, stimulus configuration, stimulus location, or hemisphere; and no other interactions were significant after controlling for age.

Figure 5.

A, Violin plots represent the distribution of the difference in IC-NC mean amplitude (μV) for centrally presented stimuli during the IC effect time window (167-177 ms) at bilateral lateral occipital sites (PO3 and PO4) among ASD (purple) and NT (blue) participants. B, Violin plots represent the distribution of the difference in IC-NC mean amplitude (μV) for centrally presented stimuli during the N_cl time window (305-315 ms) at bilateral lateral occipital sites (PO3 and PO4) among ASD (purple) and NT (blue) participants. Horizontal dotted lines represent the group mean.

N_cl

For the N_cl component, there was a main effect of group (F_(1,57) = 5.077, p = 0.028, η_p² = 0.082) with NTs showing greater magnitude VEPs than ASD participants during the N_cl time windows. There was also a continuation of the stimulus location × hemisphere interaction observed during IC effect time windows (F_{(1.430,81.531)} = 8.711, p = 0.001, η_p² = 0.133, Greenhouse-Geisser–corrected) whereby lateral stimulus presentations evoked greater magnitude visual potentials in the contralateral hemisphere. However, there were no group differences related to stimulus configuration (IC vs NC); neither group demonstrated significant N_cl response to the IC versus NC stimuli after controlling for age (Fig. 5). No other main effects or interactions were significant.

Post hoc analyses

To assess the likelihood that ASD-related findings observed in the IC effect time window would have been driven by differences in sex distribution between the ASD and NT groups, we assessed for any sex-related influences on contour integration in the IC effect time window via a post hoc mixed-design ANOVA comparing males and females within the NT group only (Table 5). Within the NT group, there was no effect of sex nor any sex-related interactions significantly influencing the strength of the IC effect. As a final post hoc analysis, we assessed for any group-related differences in latency of the N1 evoked by centrally presented stimuli across groups (Table 6). There was no difference in N1 latency between ASD and NT groups for either hemisphere or stimulus type (NC or IC). Younger age was associated with longer N1 latency (F_(1,57) = 12.574, p < 0.001, η_p² = 0.181).

As the initial a priori defined analysis was restricted in both space and time, we conducted an exploratory analysis to examine for group differences in contour integration of centrally presented stimuli across the whole brain over a more liberal timespan. Topographic maps depicting the differences in evoked response to IC minus NC stimuli for central and lateral presentations in each of the NT and ASD groups allow for visualization of effects across the entire array (Fig. 4). This secondary analysis replicated was consistent again with prior literature in demonstrating that IC processing is most pronounced between 150 and 180 ms over the midline and bilateral occipital regions (significant electrodes at t = 156 ms: PO7, O1, Oz, Iz, O2, PO8, PO4) (Fig. 6A). In addition, this analysis highlighted a group × stimulus configuration in the left frontal-temporal and parietal regions (significant electrodes at t = 172 ms: FT7, T7, CP5, TP7, P1, P3, P5, P7) during this same IC effect time window whereby individuals with autism demonstrated an attenuated negative response to IC versus NC stimuli in this region in addition to the lateral occipital region revealed in the primary analysis (Fig. 6B). There was no main effect of group at any of the electrodes or windows tested.

Figure 6.

Results of the exploratory factorial mass univariate analysis across all electrodes and time points between 100 and 300 ms with a between-subjects factor of group (NT, ASD) and within-subject factor of stimulus configuration (IC, NC). A, Regions and time points demonstrating a main effect of stimulus configuration. B, Regions and time points demonstrating a group × stimulus configuration interaction. There was no significant main effect of group.

Finally, in response to a reviewer suggestion that presentation of data with respect to an average reference may be of additional value to aid in comparison of study findings between research groups with different recording setups, topographic maps of the difference in IC-NC amplitude evoked by centrally presented stimuli using an average reference are depicted in Figure 7. The primary analysis results are compared between the two types of references in Table 7. For a brief discussion of the relative merits of the different reference types, see EEG acquisition and preprocessing.

Figure 7.

Topographic representation of the difference in instantaneous amplitude of evoked response between IC and NC stimuli for central stimulus locations comparing the FPz frontal reference electrode (top) with an average reference (bottom) for NT (A) and ASD (B) participants at 50 ms intervals between 100 and 350 ms after stimulus onset.

Electrophysiologic-phenotypic correlations

There were no relationships between the magnitude of the IC effect and Full-Scale IQ, Block Design, or Matrix Reasoning subscales in either diagnostic group (NT: F_(3,21) = 1.415, p = 0.271, adjusted R² = 0.056; ASD: F_(3,27) = 0.066, p = 0.978, adjusted R² = –0.116) (Fig. 8A–C). A stronger IC effect was associated with elevated parent-reported symptoms of inattention, hyperactivity, and impulsivity (as measured by the SNAP-IV) in the NT group only (NT: F_(1,21) = 9.032, p = 0.007, R² = 0.311; ASD: F_(1,26) = 0.973, p = 0.333, R² = –0.037) (Fig. 8D). There was no relationship between the magnitude of the IC effect and SRS-2 Total T score in either group (NT: F_(3,21) = 1.415, p = 0.271, adjusted R² = 0.056; ASD: F_(3,27) = 0.066, p = 0.978, adjusted R² = –0.116) (Fig. 8E). The ADOS-2 was completed only for participants in the ASD group and is not included in the analysis, but a scatter plot depicting the relationship between the magnitude of the IC effect and ADOS-2 comparison scores is included in Figure 8F in response to a reviewer suggestion.

Figure 8.

Scatter plot represents in ASD (purple) and NT (blue) participants the relationship between difference in IC-NC mean amplitude (μV) for centrally presented stimuli during the IC effect time window and (A) Full-Scale IQ, (B) Block Design scaled score, (C) Matrix Reasoning scaled score, (D) SNAP-IV measure of ADHD symptoms, (E) SRS-2 Total T Score, and (F) ADOS-2 Comparison Score (ASD participants only).

Centrally presented IC processing

Children with ASD demonstrated a marked attenuation of the early and automatic contour integration that is present in NT children viewing centrally presented IC stimuli. This finding provides clear electrophysiologic evidence of disordered contour integration in children with ASD, extending beyond the preschool age cohort previously studied by Stroganova et al. (2007). Thus, despite several behavioral studies suggesting that children with ASD can accurately perceive ICs (Milne and Scope, 2008; Hadad et al., 2019; Gowen et al., 2020), accumulating evidence points to clear differences in underlying neurophysiologic mechanisms that appear early in development and persist across childhood and adolescence.

Contour integration has been demonstrated in a variety of work to be dependent on feedback connections from higher-order sensory cortices (Murray et al., 2002; Halgren et al., 2003; Altschuler et al., 2012; Shpaner et al., 2013; Wokke et al., 2013). Typically, ambiguous sensory inputs, like those of ICs, are shaped by statistical predictions about configuration acquired through prior exposure, processes that are thought to be conveyed by feedback circuitry (Lee and Mumford, 2003). The ultimate percept in typical visual processing therefore reflects a balance of feedforward and feedback processing. Initial input to V1 is detectable at ∼55 ms after stimulus presentation (Foxe and Simpson, 2002; Kelly et al., 2008), so at this latency a measurement would reflect largely feedforward information and, given the receptive field properties of V1 neurons, low-level featural information. During the N1 latency (beginning ∼100 ms later), assays have convincingly implicated feedback circuitry (Lamme et al., 1998a; Hupe et al., 2001; Angelucci et al., 2002). The initial pass of feedforward information from V1 reaches dorsolateral frontal cortex in ∼30 ms (Foxe and Simpson, 2002). Considering this, it is not hard to see how quickly the influence of feedback may permeate every level of the visual hierarchy (Zipser et al., 1996; Lamme et al., 1998a), and indeed this has been convincingly demonstrated in the integration of IC stimuli (Wokke et al., 2013; Pak et al., 2020; Zeng et al., 2020). Therefore, diminished electrophysiologic evidence of contour integration in ASD implies a reduction in the ability to deploy this higher-order feedback information in support of early visual processing. Prior investigation has demonstrated that VEPs are robust and highly reproducible, arguing against a neural unreliability in autism and indicating that group differences generally cannot be accounted for simply by increased variability in the ASD relative to NT group (Butler et al., 2017). Thus, this study provides convincing empiric evidence of disordered feedback processing in ASD and offers a conceptual link between many prominent ASD theories, including weak central coherence, superior local and reduced global processing, and predictive coding deficits (Bertone et al., 2005; Mottron et al., 2006; Happe and Booth, 2008; Pellicano and Burr, 2012). Each of these theories postulates a diminished ability to apply global contextual features, experience, and/or expectation to facilitate efficient sensory perception in a complex and rapidly changing environment, and thus depends on timely and robust integration of higher-order feedback with incoming sensory information.

Although not explicitly accounted for by other predominant theories of weakened top-down input to lower-level sensory processing regions, an additional higher-order cognitive process that is increasingly important to consider is the possible modulating role of attention on processing in primary sensory cortices in children with developmental disabilities. Selective attention operates at multiple levels in the visual system and appears to be mediated by higher feedback to hierarchically early visual regions (Hopfinger et al., 2004). Although the relationship between attentional abnormalities and ASD phenotypes is incompletely understood, impairments in selective attention and attentional disengagement within multiple sensory modalities have been described in ASD (Ciesielski et al., 1990; Landry and Bryson, 2004; Kawakubo et al., 2007; Keehn et al., 2013; Sacrey et al., 2014). In this study, electrophysiologic measures were recorded during a task in which ICs were passively viewed while attention was explicitly directed to an alternative task of identifying color change. Therefore, one possible interpretation is that attenuated IC processing in the ASD group may reflect, in part, a tendency to overfocus on the color change task with less allocation of attentional resources toward the IC stimuli. A recent investigation involving other visual tasks has indicated a modulating role of attention in contributing to electrophysiological visual processing deficits observed in ASD, whereby children with ASD may have specific deficits in automatic processing that can be partially to completely ameliorated by explicit attention to processing these stimuli (Knight et al., 2022a). Potentially supportive of this possibility are the ASD-associated differences in contour integration that extend beyond the lateral occipital region into the left frontotemporal and parietal regions that were identified here through exploratory analysis, although this finding should be interpreted with much caution given the post hoc design. Although outside the scope of this study, additional investigation to evaluate neural processing of ICs under varied attention conditions would therefore be of much interest.

It does not appear from these results that the development of contour integration is merely delayed in ASD children compared with chronologically age-matched peers. Younger participants across both groups in the study tended to demonstrate higher-amplitude VEPs, which is consistent with a wide body of previous research and thought to be related to developmental maturation of electrophysiologic mechanisms and/or age-related changes in skull thickness affecting volume conduction (Adeloye et al., 1975). However, even the youngest NT participants in our study demonstrated evidence of IC processing, echoing findings from previous studies on the developmental trajectory of contour integration that this phenomenon matures relatively early (Brandwein et al., 2011, 2013; Altschuler et al., 2014). Additionally, age did not significantly interact with diagnostic group nor with stimulus configuration suggesting that the magnitude of the group differences in IC effect were consistent across all ages studied. Finally, both groups showed a trend toward some limited contour integration in the N_cl timeframe that did not reach significance after correcting for age and was not substantially different across groups. In younger NT children, more pronounced conceptual stage processing has been described as a marker of developmental maturity that subsequently decreases with age (Altschuler et al., 2014). The ASD children in this study did not demonstrate this immature pattern of an overreliance on later effortful conceptual phase processing as compensation for attenuation in early automatic IC processing. Thus, the similarity in processing during the N_cl period regardless of diagnosis provides further evidence against simple developmental immaturity of contour integration mechanisms in children with ASD. Of note, the majority of the participants were between age 10 and 17 years old with relatively few participants (n = 8) in the youngest end of the age range between 7 and 9 years old, which may impact study sensitivity to detect more subtle group differences in developmental trajectory during this early period. However, additional evidence against intact but slower more inefficient IC processing in ASD is provided by the equivalency in latency of VEP components between groups. Irrespective of group, younger age was associated with longer N1 latency, and this was not more exaggerated in the ASD group. Thus, the results are more suggestive of deviant rather than delayed development of early automatic contour processing in children with ASD across a broad age range. Furthermore, we did not observe any evidence the IC effect was intact while the N_cl was selectively attenuated in ASD. Certainly, closure deficits with more complex object completion cannot be ruled out. Indeed, one would predict given the very early disruption in feedback processing underlying the IC effect observed here in combination with prior psychophysical study of object closure in autism (Dehaqani et al., 2016; Arun, 2022), that neural processing in more complex visual object recognition is likely also impacted. Based on the disrupted IC effect and the strong dependence on feedback systems in object closure and figure-ground separation implied by previous studies (Lamme et al., 1998a,b, 2002; Lamme and Roelfsema, 2000; Lamme and Spekreijse, 2000; Layton et al., 2014; Park et al., 2022), a logical next step would be to more closely examine perceptual closure as well as object segmentation. To specifically test the robustness of these processes, more complex stimuli would be needed (Dehaqani et al., 2016; Arun, 2022).

Notably, the IC effect in the school-age and adolescent ASD children in this study was absent as opposed to inverted as has been described previously in preschool age ASD children (Stroganova et al., 2007). This apparent discrepancy in the degree of atypicality between studies may be attributable to age-related developmental changes, but it is possible that alterations in paradigm or EEG preprocessing could contribute to some of the differences across studies as well. For instance, IC perception is thought to be largely scale invariant but does depend on other stimulus characteristics, such as the support ratio and exposure duration (Liinasuo et al., 1997). Despite this, both studies identified clear electrophysiologic evidence of disordered automatic contour integration in ASD participants. These findings may contrast with those obtained via imaging, as an fMRI study using similar Kanizsa figures found no difference between adolescents with ASD and NT controls in patterns of primary visual cortex activation evoked by IC versus NC stimuli (Utzerath et al., 2019). This difference across studies involving different methodologies could possibly be explained either by the superior temporal resolution of EEG and/or by the localization of ASD-associated deficits more toward higher-order visual cortices, such as lateral occipital cortex as opposed to primary visual cortex. Performing both imaging and electrophysiology in the same subjects would therefore be illuminating to help afford a more comprehensive picture of contour integration mechanisms in ASD.

Laterally presented IC stimuli

When stimuli were presented lateral to the fixation point, neither NT nor ASD children demonstrated early automatic IC processing in the N1 timeframe as indexed by the IC effect. However, there was some evidence of a trend toward contour integration emerging later in time (∼250 ms) in NT children only for stimuli presented to the right of fixation. Since laterally presented stimuli are more challenging to process and in adults result in greater evidence of effortful processing, the most parsimonious explanation is that the lack of difference between groups here reflects a floor effect. Without clear consistent evidence of contour integration during this time window in NT children, it would be near impossible to detect any subtle ASD-related deficits. An alternative explanation would be that mechanisms that support processing of laterally presented IC are intact in ASD while mechanisms supporting processing of centrally presented stimuli are selectively disrupted. As NT adults do show evidence of feedback-supported automatic IC processing in the N_cl period for laterally presented stimuli (Murray et al., 2002; Senkowski et al., 2005), comparing NT and ASD adults may be more illuminating to probe these mechanisms.

Hemispheric lateralization of contour integration

Cortical source localization was not performed in this study, so any estimation of cortical generators represents an extrapolation from data on NT children (Altschuler et al., 2014). However, consistent with the greater body of electrophysiologic research on contour integration, the IC effect was most pronounced at electrodes overlying the lateral occipital cortex among both groups of children in this study. For centrally presented stimuli, there was a trend toward more pronounced IC-specific activation over the left hemisphere in both the NT and ASD groups. There remains some controversy over the pattern of hemispheric lateralization in IC processing. While some studies have suggested a right-sided lateralization for IC processing (Hirsch et al., 1995; Atchley and Atchley, 1998; Larsson et al., 1999; Brighina et al., 2003; Halgren et al., 2003; Seghier and Vuilleumier, 2006), which would be consistent with right hemispheric specialization for perceptual grouping and feature integration (Atchley and Atchley, 1998; Evans et al., 2000; Han et al., 2002), left-sided lateralization of early IC processing similar to that observed here has been previously reported in NT adults (Proverbio and Zani, 2002). Likewise, studies of split-brain patients or those with right hemisphere lesions help clarify these conflicting results through provision of evidence that the right hemisphere does not have a preferential role in contour integration but instead may be more involved in other aspects of perception (i.e., amodal perception) (Corballis et al., 1999; Olk et al., 2001; Vuilleumier et al., 2001). As expected, within both groups, laterally presented stimuli preferentially activated the contralateral hemisphere. Likewise, for laterally presented stimuli, contour integration was more pronounced when stimuli were presented on the right, again suggestive of possible left-sided dominance in contour integration among participants in this study. Importantly for the questions of interest in this study, there were no hemisphere × group interactions, indicating that there was no significant difference in lateralization between groups. Thus, while several studies have implicated atypical lateralization among individuals with ASD in language, motor, or visual perception (Jeste and Nelson, 2009; Fiebelkorn et al., 2013; Kroger et al., 2014; Luckhardt et al., 2014; Keehn et al., 2015; Floris et al., 2016, 2021), this does not appear based on our findings to play a role in IC perception.

Electrophysiologic-phenotypic correlations

Given that contour integration is a process that involves visuospatial perception, we tested for any relationship the strength of the IC effect and performance on block design and matrix reasoning subtests as well as Full-Scale IQ. The magnitude of the IC effect did not correlate with any of these measures. Thus, the reduction in IC effect among autistic children does not seem to be related to cognitive delays. Of note, IQ testing is a relatively broad measure of cognition; and it remains possible that more detailed neuropsychological investigation could reveal more domain-specific brain–behavior relationships. Additionally, all participants in this study had a Full-Scale IQ > 70, which was deemed a necessary inclusion criterion given the task demands. However, this does limit the range over which cognitive phenotypes may have varied, making it more difficult to detect any relationship between IC processing and cognitive factors. Likewise, we did not observe relationships between contour integration and social communication ability as measured by the SRS-2, although again it is important to note that all participants in the study were fluent verbal communicators, which limits generalizability to more severe subtypes of ASD. There was an apparent relationship between augmented contour integration and elevated symptoms of inattention, hyperactivity, and impulsivity in NT participants only (though no NT participants had received a medical diagnosis of ADHD). No such effect was noted in the ASD group, making it unlikely that comorbid ADHD symptoms in the patient group contributed to the observed findings. However, as discussed above, it would be interesting to assay contour integration processing under varied attention conditions in children with ASD and in children with other neurodevelopmental disorders (e.g., ADHD, learning disability, intellectual disability).

Endophenotypic implications

There has been much interest in defining objective endophenotypes that may link the behavioral phenotype we recognize as ASD with discrete alterations in neurophysiology. Thus, the electrophysiologic demonstration of an attenuated IC effect as well as similar electrophysiologic deficits observed in other types of sensory perception (Jeste and Nelson, 2009; Luckhardt et al., 2014) indicates the usefulness of electrophysiology in detecting subtle visual perceptual changes even when not apparent in behavioral measures. As many of these types of visual processing mechanisms, including contour integration, mature over early childhood (Altschuler et al., 2014), monitoring trajectories across this critical intervention period may afford a useful window into perceptual and cognitive development.

Limitations

The paradigm was designed to elicit automatic processing of ICs without explicit attention directed to contour presence or absence. We did attempt to control for some aspects of overt attention through the use of infra-red eye-tracking to ensure gaze did not deviate from the central fixation dot. However, given the primacy of IC perception, it is certainly possible that some participants noticed the presence of ICs, resulting in covert attentional reorienting. This makes it difficult to completely ensure that the data reflect only unattended contour integration. Additionally, the participants in the study represent a relatively narrow range of the phenotypic variation that characterizes ASD. As noted above, given the requirement for children to understand task directions and provide behavioral responses, only children with average to above average intelligence and fluent verbal ability participated. This limits the generalizability of the findings to children with comorbid intellectual disability and/or severe language impairment. Although outside the scope of this study, future adaptations of the paradigm to eliminate requirements for a specific behavioral response would allow for inclusion of children with lower language and cognitive abilities who remain underrepresented in studies of sensory perception in ASD. Finally, we are unable to determine whether the reduced contour integration noted in this study is specific to ASD or characteristic of other developmental disabilities (e.g., ADHD), which are frequently comorbid with ASD. Additional studies to compare across developmental disability populations and to directly assess the role of attention by comparing contour integration in attended versus passive processing would be highly interesting.

In conclusion, children with ASD demonstrate attenuated early automatic neural responses to IC stimuli compared with children with NT development, suggesting reduced deployment of higher-order feedback mechanisms during visual processing of global stimulus features in ASD.