Reappraisal of ERP abnormalities in ASD based on previously published studies

New data collection

Reappraisal of ERP abnormalities in ASD based on previously published studies

Published studies included in our analysis are listed in Table 1, along with relevant methodological details and effect sizes for selected ERP characteristics.

A first purported ERP abnormality, the reduction in face over object superiority of N170 timing in ASD subjects compared with controls, has been observed in five studies [16, 24, 25, 27, 29], although three studies have failed to find a significant between-group difference [26, 28, 30]. Two of these studies [26, 30] were excluded from our analysis due to the lack of data required for effect size calculation. Another study with negative findings [28] was excluded due to highly atypical ERP responses [see Additional file 1]. Analyses of the remaining five studies yielded a weighted average effect size of d = 0.68, a medium-size effect according to Cohen’s classification [65], which provided a quantitative estimate of between group differences. However, it is important to note that, because analyses were based on data from studies with positive findings only, this estimate is likely to be overestimated. Of note, parents of children with ASD also exhibited reduced face over object superiority effect of N170 timing as compared to parents of typically developing children [46] with an effect size of 0.62.

A second purported ERP abnormality, diminished effect of face inversion, as reflected in either P1 or N170 amplitudes, was not supported by the accumulated literature. P1 amplitude inversion was examined in four studies [27–29, 31] with only two reporting significant group differences [28, 29]. Weighted effect size estimation also did not support the hypothesis that the reduced P1 face inversion effect is a distinguishing characteristic of ASD (Additional file 1). The N170 amplitude inversion effect was examined in five studies [16, 27–29, 31], among which only one reported a significant group difference [27]. The weighted average effect size from those studies (two negative and one positive) is 0.35, corresponding to small effect size. However, a new composite measure of face inversion effect introduced by Webb et al. [29], the P1/N170 slope, showed a better discrimination between ASD vs. TD. This measure combines P1 and N170 components affected by face inversion and, as noted by Webb et al., “takes into consideration the peak-to-peak change in amplitude over the peak-to-peak change in latency” [29]. This slope index differentiated ASD from neurotypical controls with an effect size of 0.63.

Therefore, collective evidence from previous studies suggests that the two ERP measures related to face processing which warrant strongest consideration as potential ASD endophenotypes are (1) the face over object superiority effect on N170 timing and (2) the face inversion on P1/N170 slope. These ERP measures differentiated ASD from neurotypical subjects with medium effect sizes [0.68 and 0.63, as assessed from [16, 24, 25, 27, 29] and [29], respectively). A power analysis revealed that an effect of this size can be detected with at least 80% power with a sample size of 33 subjects per group. Each group in our sample exceeded this threshold, with the exception of UF.

New data collection

Effects of stimulus type and orientation

In our study, the largest amplitudes of P1 and the greatest N170 face inversion effect were observed with the vertex reference. Figure 1 presents grand-averaged ERPs from P8 and O2 electrodes obtained using this reference scheme (scalp topography is shown in Additional file 2). Table 3 summarizes the results of ANOVAs obtained with different reference schemes (statistics are provided in Additional file 3). Consistent with previous studies in non-clinical samples, in our children’s group face stimuli produced earlier and larger N170 component (across all reference schemes employed: main effect of stimulus type on latency was F(1, 135) > 35.66, p < 0.001, η² > 0.209; main effect of stimulus type on amplitude: F(1, 135) > 160.96, p < 0.001, η² > 0.544). Similarly, inverted images produced earlier and larger N170 component compared to that produced by upright images (main effect of orientation on latency: F(1, 135) > 15.88, p < 0.001, η² > 0.105; main effect of orientation on amplitude: F(1, 135) > 41.79, p < 0.001, η² > 0.236). Although less consistent across reference schemes, similar differences were observed in UF. In addition, in children (ASD, US, and UC) these effects on N170 amplitude showed a hemispheric asymmetry (type X hemisphere interaction: F(1, 135) > 9.04, p < 0.003, η² > 0.063 hemisphere X orientation interaction: F(1, 135) > 18.57, p < 0.001, η² > 0.121). Of note, the inversion effect on N170 amplitude was not face-specific in children (no significant interactions involving stimulus type and orientation: all p > 0.05, but was larger for faces than houses with average and vertex references in UF (F(1, 17) > 7.93, p < 0.012, η² > 0.318).

	N170 latency	N170 amplitude	P1 latency	P1 amplitude	P1/N170 slope
General effects
Type	++++	++++	++++	+++±	++++
	++=±	++++	====	++++	++++
Orientation	++++	++++	++++	++++	++++
	++++	=±++	+++±	====	====
Hemisphere	====	====	++++	====	++++
	====	====	====	====	====
Type X orientation	==±=	====	++++	++++	++++
	++=±	==++	====	====	±±==
Type X hemisphere	+++=	++++	====	==±=	±+++
	====	====	====	====	====
Orientation X hemisphere	==±=	++++	====	±=±+	±=±=
	====	====	====	====	====
Type X orientation X hemisphere	====	====	===+	===±	====
	====	====	=±=+	====	====
Group differences (ASD/US/UC)
Group	====	====	===±	====	====
Type X group	=±==	====	====	±===	====
Orientation X group	====	===+	====	====	=±==
Hemisphere X group	====	====	====	===±	====
Type X orientation X group	====	====	====	====	====
Type X hemisphere X group	====	====	====	±===	====
Orientation X hemisphere X group	±===	====	====	====	====
Type X orientation X hemisphere X group	====	====	====	====	====

Within each cell, test results are presented respectively for nose, REST, average, and vertex references, in that order. For general effects, upper array is for children; lower array for adults (UF)

Note: ‘+’ codes for significant effects surviving Bonferroni correction, and “±” corresponds to statistical significant for a given singular test, which did not survive Bonferroni correction for multiple comparison (four reference schemes), “=” codes for insignificant effects (p ≥ 0.05) for nose/REST/average/vertex references, respectively (e.g., “===+” indicated that the effect is significant only with vertex references). For each factor, the irst line indicates effects for ASD/US/UC combined and the second line is for the separate analysis of UF

Finally, in children, the latency of the earlier P1 component was shorter for upright faces than both houses and inverted faces (type X orientation interaction: F(1, 135) > 12.47, p < 0.001, η² > 0.085), and P1 amplitude was larger for inverted faces than houses and upright faces (type X orientation interaction: F(1, 135) > 36.12, p < 0.001, η² > 0.211). In UF, only the main effect of stimulus type was significant for P1 amplitude (F(1, 17) > 11.45, p < 0.004, η² > 0.402) and the main effect of orientation for P1 latency (F(1, 17) > 7.08, p < 0.016, η² > 0.294). The face over object superiority in latency and the face inversion effects for P1 and N170 were not correlated (p > 0.05), suggesting distinct underlying mechanisms involved in the modulation of these ERP components. In addition, we confirmed the sensitivity of P1/N170 slope to face inversion [29]. Extending this finding to the child population: P1/N170 slope was steeper for faces than houses (main effect of stimulus type: F(1, 135) > 172.99, p < 0.001, η² > 0.562) and for inverted rather than upright stimuli (main effect of orientation: F(1, 135) > 62.36, p < 0.001, η² > 0.316); the inversion effect was larger for faces than houses (type X orientation interaction: F(1, 135) > 18.53, p < 0.001, η² < 0.121). In UF, only the main effect of stimulus type was significant (F(1, 17) > 27.99, p < 0.001, η² > 0.622).

Is the face over object superiority effect reduced in ASD?

The N170 latency was significantly shorter for faces than houses for UC and US, but not ASD children, irrespective of reference type (Table 4). In the UF group, the difference did reach significance but only with the average reference scheme. In spite of a qualitative difference, the magnitude of the face superiority effect (difference between N170 latencies in response to faces and houses) did not consistently differentiate ASD from other groups. A significantly reduced face superiority effect was observed in ASD subjects as compared to UC only under the average reference. The reduction in the face superiority effect was due to delayed N170 latency for faces in ASD children as compared to UC (Additional file 5). The difference between US and UC did not reach significance for any of the reference schemes.

Is the face inversion effect on P1/N170 slope diminished in ASD?

The P1/N170 slope at P8 was significantly steeper for inverted than upright faces, but, contrary to our expectations, this effect showed no significant group differences and was observed in all studied groups of children under all reference schemes irrespective of the diagnosis or family type (Table 4, Fig. 2). Furthermore, the face inversion effect, computed as the difference between peak values obtained in inverted and upright conditions for P1 and N170 amplitudes, also failed to differentiate the study groups (Additional file 5, except N170 amplitude inversion with vertex reference related to general N170 amplitude inversion effect, which is discussed in detail below).

Are the results affected by the inclusion of subjects with lower IQ and ADHD symptomatology?

To examine whether the findings might be influenced by the inclusion of individuals with low IQ and/or ADHD diagnosis, the above hypothesis-testing analyses were repeated after the exclusion of four US and 10 UC subjects with ADHD diagnosis and three ASD subjects with full-scale IQ < 70. This exclusion did not significantly impact the pattern of results described above. Moreover, our ERP effects of interest were not correlated with IQ scores (all ps < 0.15). Thus, the results obtained for P1 and N170 in our original analyses are unlikely to be driven by the inclusion of either low IQ or ADHD subjects in the analysis.

The role of performance in the secondary (control) task

A re-analysis of data after the application of stricter subject exclusion criteria based on the performance in the secondary task (responding to less than 90% of the rare target stimuli) did not affect the main findings. Furthermore, no significant correlations between accuracy in the secondary task and ERP variables of interest were observed (all ps > 0.05). Taken together, these results suggest that variability in the performance on the secondary “control” task did not impact the main findings of this study.

Effects of stimulus type and orientation on P1 and N170 components

Corroborating previous findings in the general population, the N170 component was significantly larger and peaked earlier for faces than for houses, predominantly at the right posterior sites [17–19]. The face inversion effect on N170 reported in previous studies (e.g., [17–19]) was also well replicated in the present study, although our findings challenged its specificity to faces: N170 amplitude was larger for inverted compared to upright images of both faces and houses. We confirmed the sensitivity of a new measure, proposed by Webb and colleagues [29], the P1/N170 slope, to face inversion and extended this finding to the child population: the inversion effect on P1/N170 slope was larger for faces than for houses. Furthermore, our study supported face-related effects on the P1 component [20–23]: P1 latency was shorter for faces than houses, and inverted faces elicited larger P1 than upright faces and houses. The P1 effects were not specific to the right hemisphere and observed both at the left and right occipital sites. It is important to note that P1 and N170 latency facilitation effects for faces were not correlated, suggesting that the “face processing advantage” begins as early as 120 ms post-stimulus and involves distinct underlying mechanisms at different stages of information processing.

Limited support for the hypothesized ERP endophenotypes for ASD

Many studies have examined the latency of the N170 component in response to face stimuli, although most of them have not found a significant difference between ASD and control groups (17 out of 23, [32]). One potential explanation for this variability of findings could be that N170 represents more general mechanisms of the neural processing of complex visual patterns that are not fully specific to face stimuli. To address this problem we computed the difference in the latencies of N170 elicited in response to objects and faces. In the general population, N170 latency is shorter in response to faces than objects [33, 34], and this superiority effect on N170 timing differentiated ASD from UC.

Our analyses revealed a substantial impact of EEG reference scheme on the results of comparisons between ASD and UC subjects with respect to the studied ERP components. Analysis of published literature (Table 1, [16, 24–30]) showed that five out of eight studies reported a reduced “face over object superiority” on N170 latency among ASD subjects with a weighted average effect size of 0.68. Our present data supported the reduction of the face superiority effect in ASD group as confirmed by Bayesian analysis but only under the average EEG reference scheme. Noteworthy, all studies that reported this effect previously also used the average reference, while two out of three remaining studies [26, 30] utilized a nose reference. Thus, our findings suggest that N170 latency abnormalities in ASD are sensitive to the reference scheme, and the average reference appears to be optimal for detecting that effect.

A primary aim of the present study was to examine familial aggregation of previously reported face-related ERP abnormalities in male relatives of children with ASD. The difference between US and UC groups was of small effect size even with the optimal reference schemes (d = 0.19) providing little support for the difference between first-degree relatives (parents of ASD children) and low-risk controls reported previously [46]. Moreover, contrary to a prior report, our sample of unaffected fathers ascertained exclusively from multiplex families showed a significant face over object superiority effect on N170 timing.

The systematic review by Feuerriegel and colleagues [32] suggested that ERP characteristics in response to specific manipulation of face stimuli, such as face inversion, warrant thorough investigation as potential neurophysiological biomarkers of ASD. The present study addressed this issue in a comprehensive manner and found no evidence that the face inversion effect on studied ERP components reliably differentiated ASD from healthy control groups. In a previous study [29], the P1/N170 slope differentiated ASD and controls with a medium effect size (d > 0.5), however the present data collection did not replicate this effect (Table 4); moreover neither P1 nor N170 amplitude (Additional file 5) differentiated ASD and controls in this study.

Thus, despite a clear-cut replication of previously reported, general, within-subject effects of face superiority and inversion, the differences between ASD and controls were entirely limited to N170 latency, exclusively derived from the average reference scheme. None of the proposed ERP markers of ASD met the criteria for an endophenotype; notably US and UC groups did not differ significantly with respect to the face over object superiority effect on N170 latency or the face inversion effect on P1/N170 slope. Furthermore, none of the studied ERP components showed significant correlations with a validated dimensional measure of ASD severity (Social Responsiveness Scale score) in any of the studied groups.

The N170 amplitude inversion effect is not specific to faces

The inversion effect on N170 amplitude differentiated ASD and UC groups, but only with the vertex reference. Contrary to our initial hypotheses, the effect of inversion on N170 amplitude was not specific to face stimuli or hemisphere. Of note, most prior studies of the inversion effect have failed to include a control condition (non-face object inversion) or, when such a condition was included, the results were not reported [16, 27, 28]. The only ASD study that reported data for an object inversion effect on ERP components indeed found that the N170 amplitude inversion effect was reduced in ASD both for faces and houses ([29], see Table 3 on page 585), although this interesting finding was not featured in the discussion. Additional corroborating evidence for non-specificity of the inversion effect to faces comes from a recent behavioral study [66] which reported better performance for upright than inverted images of both faces and cars. Moreover, these non-specific inversion effects were weak and slow to develop in ASD children as compared to controls. Therefore, we conclude that there is little evidence to support the notion that the diminished face inversion effect on N170 amplitude in ASD subjects reflects deficits specific to face processing, as suggested by previous studies [27, 29].

Further support for the common mechanism underlying processing of both inverted faces and objects is derived from studies using neural adaptation paradigms. These studies have shown that inverted objects (houses and Chinese characters) induce an adaptation effect on the N170 component for inverted faces [67, 68]. Additionally, both competition and adaptation effects on the N170 amplitude for inverted faces were larger in the inverted than in upright face context [68, 69], suggesting that the processing of upright and inverted faces recruits distinct neuronal populations of orientation-sensitive neurons [67, 68]. Intracranial recordings [70] have detected activation of both the face-specific and non-specific areas in the lateral occipital cortex in response to face inversion.

It is possible that preference for a part-based over a holistic processing strategy in ASD [43] generalizes to the perception of well-known prototypical objects such as houses and cars and this is what is captured by the non-specific reductions of N170 amplitude inversion. Yet another possibility relates to hypotheses regarding face inversion effects as a function of expertise [71]. Some studies have suggested that only a particular type of expertise, e.g., second-order relational (configural) characteristics [72], or prototype perceptual learning [73, 74] contribute to the effect. Behavioral studies have identified dog image inversion effects in dog experts ([71] but see [37]) as well as hand-writing inversion effects in hand-writing experts [75]; prosopagnosics with special expertise have reported an inability to identify not only faces but birds (among experienced bird watchers) and cows (among an experienced farmer) [76]. Neurophysiological correlates of the face inversion effect have also been reported to be sensitive to expertise [71, 74, 77]. Computer-generated artificial stimuli (“greebles” [77] and prototype-defined checkerboards [74]) have elicited the N170 amplitude inversion effect after extensive laboratory training. Therefore, N170 inversion may index a perceptual learning experience contributing to face and object recognition. Noteworthy, deficits in early experience-dependent learning were recently suggested to underlie the selective impairments in orientation sensitivity along the vertical axis found in ASD children [78].

Potential moderating and confounding factors

ERP measurements can be affected by a number of factors related to the sample composition (i.e., age, gender, comorbid psychiatric conditions, intellectual variation, and medications), subjects’ understanding of and compliance with the task instruction, and data analysis such as the choice of EEG electrode reference scheme. In the present study, we conducted a series of additional analyses in order to systematically examine the role of these potentially moderating or confounding factors. Details regarding the results of these analyses are elaborated in a corresponding section of Additional files (Additional file 6).

We wish to emphasize here the significant effects of the reference scheme on contrasts between ASD and UC subjects for the studied ERP components. Although within-subject effects of stimulus type and orientation were significant across multiple reference schemes, group differences in P1 and N170 were small and highly dependent on the choice of reference (Tables 3 and 4, Fig. 2). This suggests that to the extent that true differences exist, they may be highly specific to the brain regions uniquely represented by selection of electrodes in which the differences are detected.

Limitations

Although one of the largest ERP studies of ASD subjects to date, our sample size limited statistical power to detect group effects smaller in magnitude than those reported as positive findings in previous studies. Our study did not include age-matched controls for the fathers of ASD probands (UF), rendering the evaluation of potential ERP abnormalities in this group unfeasible. A direct statistical comparison of UF with other study groups would be inappropriate due to significant age-related ERP differences. However, this group represents a very unique sample of fathers of ASD probands from multiplex families and these data are included in the manuscript for the sake of reporting the entire data set collected in this project. Another limitation is the relatively sparse electrode montage used in the present study (30 EEG electrodes). Although the ERP components of interest (P1 and N170) show a relatively smooth distribution over the respective scalp areas and can be easily identified at several electrodes, a high-density montage would facilitate the detection of peaks in individuals with unusually low ERP amplitude and increase the overall accuracy of amplitude and latency measurements. An additional limitation is the lack of IQ assessments for the unaffected groups, which precluded precise matching of subjects with respect to this variable; inclusion of IQ measurement in future family-based studies will allow for a more rigorous control of potential confounders. We note, however that there is little evidence for a relationship between face ERPs and IQ, and no correlations between the studied ERPs components explored here and IQ measures obtained among the individuals affected by ASD were observed in this study affected by ASD. Finally, although clinician diagnosis with ADI–R confirmation exhibits very strong agreement with categorical designation on the Autism Diagnostic Observation Schedule [58], it was a limitation of the study that data from the latter were not available. The ADOS represents an additional diagnostic standard in ASD research that affords opportunity to test quantitative associations of biomarkers with autistic severity among ASD-affected individuals, as measured not only by caregiver report—as was done in this study using the Social Responsiveness Scale—but also by clinician rating.