Modulation of mu attenuation to social stimuli in children and adults with 16p11.2 deletions and duplications

This study reports findings from 48 child and adult participants who were enrolled in and characterized as belonging to one of four groups: (1) deletion carrier, (2) duplication carrier, (3) idiopathic ASD, and (4) typical development. All participants spoke fluent English and had normal or corrected-to-normal vision. All research procedures conformed to regulations in accordance with the local ethical review board at the University of Washington, which approved this project. Written informed consent was obtained from each adult participant or parental representative(s). All children verbally assented to participate in the procedures, and written assent was obtained from children with a mental age of 7 or greater.

Full participant characterization is reported in Table 1. Deletion (n = 12) and duplication (n = 12) carriers were recruited following enrollment and participation in the Simons VIP Connect project [16, 20, 21, 45]. Recruitment methods included self-referral and web-based networks that directed individuals to the Simons VIP website (http://​SimonsVIPconnect​.​org), as well as referral by clinical genetic testing centers. As part of the Simons VIP Connect, individuals were identified with the same recurrent ~600 kb 16p11.12 BP4-BP5 deletion or duplication. Within these two 16p11.2 CNV groups, no additional pathogenetic CNVs or monogenic disorders were known (see Simons VIP Consortium [45] for complete recruitment and inclusion/exclusion criteria). Individuals within the ASD and typical groups were recruited from previous projects, matched on age (ASD/typical) and verbal IQ (ASD) to the CNV groups. None of the children within the idiopathic ASD group had any likely causal CNVs, verified through array CGH sequencing [15, 17, 46].

For ASD and 16p11.2 CNV groups, diagnosis of ASD was confirmed using the Autism Diagnostic Interview-Revised [3, 5‐14, 18, 47] and the Autism Diagnostic Observation Schedule [16, 19, 48]. IQ was assessed using the Wechsler Abbreviated Scale of Intelligence [20, 21, 49] or the Differential Ability Scales-Second edition [1‐4, 19, 22, 23, 50], depending on age. One-way analysis of variance (ANOVA) indicated that there were no differences between groups in age, F (3,44) = 1.59, p = 0.21. The duplication, deletion, and the ASD groups did not differ in verbal IQ or SRS total score, Fs (2,29) < 2.96, ps > 0.068, but did differ in full-scale and nonverbal IQ, Fs (2,29) > 3.68, ps < 0.038. Least squares post hoc comparisons indicated that the ASD group had higher full-scale IQ and nonverbal IQ than both 16p11.2 deletion (FSIQ, SE = 22.14, p = 0.027; NVIQ, SE = 19.13, p = 0.041) and 16p11.2 duplication (FSIQ, SE = 23.50, p = 0.018; NVIQ, SE = 23.46, p = 0.014) groups. None of the typical participants had elevated scores on the SRS nor met diagnostic criteria for any psychiatric disorders.

Participants observed a series of silent 1-min-long videos. Stimuli consisted of three conditions: (1) the social motion condition included a video of a faceless computer-generated (CG) avatar dancing and a live action (LA) video of two pairs of hands engaged in a clapping game. The computerized dancer stimuli were provided courtesy of Nick Neave and Kristofor McCarty at Northumbria University. (2) The nonsocial motion condition included a video of a CG bouncing ball and a LA video of cardboard tubes swinging. (3) The rest condition consisted of static image of the same background as the videos (Fig. 1). The quantity of visual movement was controlled across social and nonsocial motion conditions. Background complexity was controlled across all conditions. All videos were presented twice, so that each participant observed eight 1-min movement videos (four social, four nonsocial) and four 1-min rest videos. All participants saw the LA videos first followed by the CG videos, and the order of video context (i.e., social or nonsocial) was counterbalanced across participants (see Fig. 1 for possible presentation orders). Between videos, participants were directed to take a break and the experimenter initiated the next stimulus after confirmation that the participant was ready. Participants were seated approximately 75 cm from a video monitor and were instructed to sit still and attend to the videos. Video stimuli were displayed using E-Prime 2.0 software (Psychology Software Tools, Inc. Pittsburgh, PA) at a size of 27 cm by 36.8 cm and subtended a visual angle of 20.4° by 27.6°.

Continuous EEG was recorded from a high-density 128-channel geodesic net using Net Station 4.3.1 software integrated with a 200-series high-impedance amplifier (Electric Geodesics Inc, Eugene, OR). Electrode impedances were below 50 kΩ to maximize signal-to-noise ratio, within the standard range for high-impedance amplifiers. During collection, EEG signals were referenced to the vertex electrode, analog filtered (0.1 Hz high-pass, 100 Hz elliptical low-pass), amplified, and digitized with a sampling rate of 500 Hz. A photocell recorded and marked the precise onset time of each video. During acquisition, researchers observed and marked periods containing movement and/or improper attention (e.g., participant looking away). As electrodes surrounding the face, eyes, and rim of the net are prone to artifacts, we excluded these electrodes in all participants from all subsequent post-processing or analyses (see Fig. 2 for locations).

Following data collection, data was filtered using a high-pass filter of 1 Hz. Continuous EEG was segmented into 2-s epochs starting with the onset of each 1-min video (as marked by the photocell). Epochs marked during acquisition as contaminated by movement or improper attention were removed. Automatic artifact detection rejected channels containing voltage shifts greater than 100 μV for each epoch. If the channel was rejected for more than 50 % of epochs, the channel was excluded from analysis. Epochs were re-referenced to the average reference and exported as a concatenated file per participant for additional artifact rejection using independent components analysis (ICA) via EEGLAB 13.2.1 [5, 24, 51]. First, the ICA runica algorithm was implemented across the entire concatenated dataset for each participant. Epochs related to artifacts were rejected manually. A second ICA with the runica algorithm was implemented to reject components related to artifacts. The resultant primary independent components (i.e., first 35) containing horizontal and vertical eye and/or body movements were manually inspected. Independent components containing artifacts were rejected and removed from the data. The remaining data for each condition was segmented into 2-s epochs for analysis.

To calculate mu power, fast Fourier transforms (FFTs) were conducted in Matlab (version 7.12.0, R2011a; Natick, MA) on each 2-s epoch. The power spectra occurring between 8 and 13 Hz was averaged across central electrodes clustered around the standard C3 (electrode 37) and C4 (electrode 104) positions (see Fig. 2 for locations). Power spectra for each epoch per condition were considered relative to the average rest condition power. Mu attenuation was computed as the natural log of the ratio between power for each epoch of either social or nonsocial motion minus the average power of the rest condition for that individual. Subsequently, zero represents no mu attenuation (social motion − rest = 0) and larger negative values represent more mu attenuation (i.e., social motion < rest).

All analyses were conducted via SAS 9.3 (SAS Institute) using restricted maximum likelihood (REML) and Satterthwaite denominator degrees of freedom. To fully assess group condition differences, multilevel models were generated using PROC MIXED to describe the variances and covariances of mu attenuation. All models included a random intercept for each individual. Of the 48 participants, there were 36 unique families enrolled in this study. To account for possible shared variance between family members, statistical analyses included random intercepts for each family. Post hoc comparisons were conducted using least squares differences. Significant effects are presented for p < .05; marginal effects are presented for p = .06 to .1.

Second, we were also interested in whether mu attenuation varied across the course of the video. To validate the necessity for this analysis, interclass correlations calculated from an empty model indicated that 46.7 % of the variance for mu attenuation is due to overall within-person effects, while the majority of variance (53.3 %) is due to trial order. This indicates that mu attenuation is related to changes across the course of the video, such that a fixed effect of time (e.g., temporal order for each epoch) merits addition to the multilevel models. Time was centered at epoch 30 (i.e., the 30th epoch surviving epoch rejection due to artifacts). Thus, model 2 included predictors from model 1 with time added as a fully interacting predictor with context, environment, and group.

A full factorial design between context, environment, and group with age, gender, and number of epoch predictors was estimated in model 0. First, age significantly contributed to model 0, F (1,31.9) = 5.11, p = 0.031, suggesting that older individuals are predicted to exhibited incrementally greater mu attenuation (0.015 more mu attenuation for every year older than 9 years). Second, mu attenuation was not affected by gender, F (1,40.9) = 0.23, p = 0.62, indicating that female and male individuals exhibit equivalent patterns of mu attenuation within groups and across conditions. Third, the number of epochs contributed by each individual also did not contribute to patterns of mu attenuation, F (1,40.9) = 0.42, p = 0.52. This was important to determine since the ASD group contributed fewer trials due to having more trials rejected by artifacts (30.8 %) than any of the other groups (deletion carriers, 14.4 %; duplication carriers, 11.3 %; typical, 9.7 %). Thus, gender and number of epoch predictors were subsequently removed for model 1.

Topographic maps representing group differences in power are presented in Fig. 3 for each stimulus context (collapsed across stimulus environment). In the final model 1, all fixed effects were significant (p < 0.05, uncorrected) with the exception of a main effect of the group. Model 1 results are reported in Table 2. Similar to model 0, age significantly contributed to the model, indicating increased mu attenuation for older individuals (0.016 more mu attenuation for every year older than 9 years). Main effects of context and environment indicated more mu attenuation for nonsocial (compared to social) conditions and live action (compared to computer generated) conditions. These effects were modulated by a context × environment interaction, indicating that computer-generated conditions elicited more mu attenuation for nonsocial conditions, t (10000) = 4.43, p < 0.0001.

The primary aim of this paper was to address group differences across conditions. Figure 4 depicts observed mu attenuation by group across all conditions (context × environment). There was no main effect of a group, indicating that each of the four groups had overall similar levels of mu attenuation. However, we observed a significant interaction between group and context with three different patterns of relative mu attenuation between social and nonsocial motion. First, only the typical group elicited the predicted pattern with more mu attenuation to social than nonsocial conditions, t (10,000) = 2.30, p = 0.022. Second, both of the 16p11.2 carrier groups exhibited the opposite pattern of mu attenuation, such that these groups elicited more mu attenuation to nonsocial than social conditions [deletion carriers, t (10,000) = 4.92, p < 0.0001; duplication carriers t (10,000) = 2.00, p = 0.046]. Third, the ASD group did not differentiate between conditions, t (10,000) = 1.43, p = 0.15.

The group × environment interaction was qualified by the three-way interaction between group, context, and environment. This interaction indicated that the effect of more social mu attenuation than nonsocial for the typical group was primarily driven by differences within the live action environment, t(10,000) = 2.80, p = .0051, but not the computer-generated environment, t (10,000) = 0.45, p = 0.65. The opposite was true for the deletion carriers, in which the nonsocial context differences (more nonsocial mu attenuation than social) were primarily driven by differences within the computer-generated environment, t (10,000) = 6.52, p < 0.0001, but not the live action environment, t (10,000) = 0.38, p = 0.70. Duplication carriers and the ASD group did not exhibit context discrimination differences between environments.

Results for model 2 are reported in Table 3, and parameter estimates are reported in Table 4. Negative parameter estimates reflect more mu attenuation. Parameter estimates are described in relation to the intercept (i.e., all fixed effects at 0), which represents the expected mu attenuation for a 9-year-old in the typical group during the 30th epoch in the live action nonsocial condition. For example, model 2 estimated 0.0569 more mu attenuation in the social context and 0.0072 less mu attenuation in the computer-generated environment for a 9-year-old typical subject. Similar to model 0 and model 1, age was associated with increased mu attenuation for individuals older than 9 years of age, consistent with prior research [1, 2, 24‐30, 37].

Main effects of condition and group estimated at epoch 30 were similar to the overall effects of model 1 with one exception. Unlike model 1, which predicted that more mu attenuation for the social live action condition, model 2 predicted no difference between live action and computer-generated environments, t (10,000) = 0.36, p = 0.72. This suggests that model 1 indicated that differences in mu attenuation for certain environments in the typical and deletion carrier groups may resolve over the course of the experiment by epoch 30, such that there is no longer a difference in mu attenuation by environment.

A main effect of time indicated that for subsequent epochs, the mu attenuation intercept (i.e., typical 9-year-old watching a live action nonsocial video) was expected to decrease or lessen. However, as described by the significant group × time interaction, the rate of mu attenuation lessening differed by group. The first column in Fig. 5 depicts this effect as the grand average of observed mu attenuation over epochs for each of the groups. Compared to the typical slope (0.0048), mu attenuation decreased over time more rapidly for the deletion carriers (slope = 0.0028) and less rapidly for the duplication carriers (slope = .0065). In contrast to the other three groups, the ASD slope increased in mu attenuation (slope = −0.00022), suggesting that the neural network supporting motion perception may be increasing efficiency over the course of the videos.

Although the time × group × context interaction was marginally significant (p = .071, see Table 3), the second and third columns in Fig. 5 depict the change in mu attenuation for social and nonsocial conditions for computer-generated and live action environments, respectively. These plots highlight that group differences in context discrimination are more apparent during increased exposure to the stimuli (e.g., after epoch 30 of video viewing). Importantly, this figure also illustrates that context discrimination can largely be described as the relative rate of mu attenuation lessening for social and nonsocial motion. For instance, model 1 indicated larger context discrimination for the deletion carriers in the computer-generated motion. Looking at the patterns of mu attenuation over time, this effect is largely driven by the rapid decreasing mu attenuation of the social condition, whereas mu attenuation for the nonsocial condition does not lessen much over time.

To fulfill our objective of characterizing dynamic changes in mu attenuation related to social perception, we conducted a series of pairwise comparisons between social and nonsocial context for each group every 5 epochs from epoch 1 to epoch 60. Figure 6 plots mu attenuation modulation as a difference score, such that negative values represent more mu attenuation to social than nonsocial context. Each group exhibited a unique pattern of mu attenuation related to context discrimination. Both deletion and duplication carriers exhibited more nonsocial than social mu attenuation. This pattern was consistent for deletion carriers from trial 1 [ts (10,000) = 2.01–4.22, ps < 0.05] until epoch 60, t (10,000) = 1.73, p = 0.083. However, the duplication carriers did not begin discriminating contexts until epoch 30 and continued to exhibit more mu attenuation for nonsocial conditions until the end of the video, ts (10,000) = 2.09–2.92, ps < 0.05. In contrast to the carrier groups, the typical group exhibited greater mu attenuation for social conditions at epoch 25, ts (10,000) = 2.0–2.6, p < 0.05. Unlike all other groups, the ASD group did not exhibit mu attenuation context discrimination at any epoch, indicating equivalent levels of mu attenuation for social and nonsocial conditions.

To better understand how these patterns may be related to heterogeneity within these unique samples, we conducted a series of post hoc ANOVAs to determine if patterns of mu attenuation are consistent for 16p11.2 carriers between (1) children and adults and (2) between 16p11.2 carriers with and without an ASD diagnosis. While these comparisons may be subjected to type I and II errors due to small sample size, these analyses were meant to qualitatively support model 1 and model 2 predictions. The patterns of mu attenuation for these comparisons are depicted in Additional file 1: Figure S1, Additional file 2: Figure S2, and Additional file 3: Figure S3.

First, child deletion carriers [n = 9, F (1, 1861) = 16.28, p < .001] and adult duplication carriers [n = 5, F (1, 1111) = 15.19, p < .001] exhibited the overall group pattern exhibited with greater mu attenuation for nonsocial (compared to social) conditions. There was no difference between context for adult deletion carriers [n = 7, F (1, 740) = 1.87, p = .17] or child duplication carriers [n = 3, F (1, 1567) = .002, p = .96]. However, visual inspection of the dynamic mu attenuation patterns across time (Additional file 2: Figure S2) suggests that the adult deletion carriers exhibited greater mu attenuation for the nonsocial context. Unlike the other 16p11.2 carrier age groups, child duplication carriers exhibited a pattern of mu attenuation changes that is more similar to the ASD group (no context discrimination but trending towards social greater than nonsocial).

Second, neither deletion carriers (n = 3) nor duplication carriers (n = 3) with an ASD diagnosis exhibited context discrimination, Fs < .31, ps > .58, similar to the pattern of mu attenuation for the ASD (idiopathic, non-16p11.2 carrier) group. Visual inspection of the dynamic mu attenuation patterns across time (Additional file 3: Figure S3) indicates a large amount of variability for the deletion carriers with ASD, consistent with prior work looking at social motion perception in ASD [3, 5‐14, 22, 31]. These results suggest that social motion mu attenuation within 16p11.2 carriers is associated with ASD and the concomitant social cognitive phenotype.