Participants
Infants were enrolled in a longitudinal study conducted jointly by Boston Children’s Hospital and Boston University and approved by the institutional review board (#X06-08-0374). Written consent was obtained from a parent or guardian prior to each child’s participation.
Exclusion criteria for the study included prenatal or postnatal medical or neurological problems (e.g. seizures), genetic mutations known to affect neurodevelopment, and uncorrected hearing or visual impairment. All infants had a minimum gestational age of 36 weeks and were from households speaking primarily English (English spoken more than 75% of the time). Infants were also excluded from this analysis if they did not complete the 12-month visit or complete the ADOS assessment at a later visit.
Infants were enrolled in two groups: (1) high familial risk infants for ASD who had at least one older sibling with ASD, confirmed using the ADOS or the Social Communication Questionnaire (SCQ) (Rutter Bailey & Lord, 2003), and (2) low risk infants, defined by having a typically developing older sibling, and no first- or second-degree family member with ASD. ASD outcomes groups (LRC, HR-NoASD, and HR-ASD) were determined using behavioral assessments administered at 18-36 month time points (see Behavioral Assessment).
Of the 183 eligible infants who provided EEG data for this analysis, only a subset (n = 110) met our behavioral and data quality requirements (Supplemental Figure 1). After ERP preprocessing pipelines described below, 102 ERPs were available for Nc analysis (42 LRC, 40 HR-NoASD, 20 HR-ASD) and 64 ERPs were available for N290/P400 analyses (24 LRC, 26 HR-NoASD, 14 HR-ASD).
Behavioral Assessments
Infants were evaluated using the Mullen Scales of Early Learning (MSEL; Mullen, 1995) at 6, 9, 12, 18, 24, and 36-month visits. These evaluations assessed receptive and expressive language, fine motor skills, and visual reception developmental domains. This study utilizes standardized T-scores from expressive language and receptive language subscales of the MSEL at 12 months of age.
Final ASD outcome groups were determined using the ADOS (Lord et al., 2000), administered at 18, 24, and 36 months of age. For participants receiving an ADOS score indicative of ASD or within 3 points of cutoffs, a licensed clinical psychologist reviewed video recordings of concurrent and previous assessments, and using DSM-5 criteria, provided a best estimate clinical judgment in one of three categories: typically developing, ASD, or non-spectrum disorder (e.g., ADHD, anxiety, language delay). Of the 60 HR infants contributing data for this study, 3 children (2 HR-ASD, 1 HR-No ASD) had final outcome judgements based on only the 18 month ADOS assessment. At 18 months, all participants were administered the ADOS Module 1, and the social affect score was used as one measure of social development.
Parent Questionnaires
Parents completed the MacArthur-Bates Communicative Development Inventory (MB-CDI): Words and Gestures (Fenson et al., 2007) at the 12-month time point. The study utilizes the Early Gesture and Phrases Understood raw scores from this questionnaire. At the 18-month visit, parents completed the Communication and Symbolic Behavior Scales Developmental Profile (CSBS-DP; Wetherby & Prizant, 2002). The CSBS-DP is a norm-referenced measure of early social communication and symbolic development. The Social composite standard score (comprised of questions related to emotion, eye gaze, communication, and gestures) was used in subsequent data analyses.
Mother/Stranger Stimuli and EEG Task Procedure
For this task, infants observe color pictures of their mother and a similarly-looking stranger. Images of the mother and stranger were randomly presented for 500 ms, maintaining a ratio of 1:1 for each type of picture. Pictures of the mothers were matched with strangers according to ethnicity and whether or not they wore glasses. The mothers and strangers had neutral expressions for their pictures.
EEG sessions were conducted in a sound attenuated and electrically shielded room with minimal lighting. During the sessions, caregivers held the infant on their lap, approximately 65 cm from the experimental monitor. Continuous EEG was recorded using either 64-channel Geodesic Sensor Net System or a 128-channel Hydrocel Geodesic Sensor Nets (Electrical Geodesics, Inc., Eugene, OR, USA). Signals were amplified with a Net Amps 200 or Net Amps 300 amplifier (Electrical Geodesic Inc., Eugene, OR, USA), sampled at either 250 Hz or 500 Hz. EEG data were online-referenced to a single vertex electrode (Cz), and impedances were kept below 100kΩ. Stimulus presentation was managed via ePrime software (Psychology Software Tools, Pittsburgh, PA). Each stimulus was initiated only when the child was attending to the screen, as observed by an examiner in the adjacent room. Trials during which the child’s attention was not maintained on the visual stimulus were marked and then removed from further analysis. A maximum of 100 trials (Mother and Stranger combined) were presented. Fewer trials were presented when the infant became fussy, tired, or inattentive. There was no significant difference in number of trials administered between outcome groups (p>0.1, Supplmental Table 1).
EEG Pre-Processing
The continuous EEG data collected over the mother/stranger paradigm was first downsampled to 250Hz in Netstation and then exported to MATLAB (versionR2017b) for preprocessing analysis using a modified version of the Harvard Automated Processing Pipeline for EEG (HAPPE; Gabard-Durnam et al., 2018) to allow for ERP analyses similar to the recently released HAPPE+ER software (Monachino et al., under review). Within the modified HAPPE pipeline, artifact within the continuous EEG data is first extracted using the following steps: a copy of the data is made and that copy is high pass filtered at 1 Hz, channels for subsequent ICA analysis are selected (Supplemental Figure 2), 60Hz electrical noise is removed via Cleanline’s multi-taper regression (Mullen, 2012), bad channels are rejected, and then remaining artifact is extracted first using wavelet-enhanced independent component analysis (W-ICA), and then subsequently using ICA with MARA automated independent component rejection. Next, the original unfiltered EEG file is subjected to the same channel selection and electrical noise removal steps above and the bad channels detected from analysis on the data copy are removed. The artifact signals identified after the W-ICA step on the data copy are then subtracted from the original unfiltered EEG file, and the identified artifact ICA components rejected from the data copy are back-projected to sensor space as timeseries that are then rejected from the original unfiltered signal. This now ‘clean’ unfiltered file is filtered using standard ERP filter settings (0.3Hz-30Hz), and segmented (-100ms to 700ms) around the visual stimulus, and baseline corrected via baseline subtraction. Segments with retained artifact in the subset of electrodes used for ERP analyses (Figure 1A and B) are rejected using HAPPE’s amplitude (amplitude threshold of +80 µV) and joint probability criteria, bad channels are interpolated, and data is referenced to the average reference.
EEG Rejection Criteria
Children were excluded from the final sample if they had fewer than 10 trials for either the mother or stranger stimuli, or did not meet the following HAPPE data quality output parameters previously determined in this dataset (Wilkinson et al. 2020): percent good channels > 82%, percent of independent components rejected < 84%, percent variance of data retained after artifact removal > 32%, mean retained artifact probability < 0.3. There were no significant differences in data quality between outcome groups. Supplemental Table 1 shows quality metrics for all outcome groups for both ERP analyses.
ERP Analysis
Average waveforms for each individual participant for each stimulus condition (mother and stranger) were calculated across electrodes in corresponding regions of interest (Nc: Figure 1A, P400: Figure 1B), which were chosen based on previous literature (Guy et al., 2016, 2018; Luyster et al., 2011, 2014). To control for the effect of preceding peak/trough amplitude on the Nc, N290, and P400 amplitudes, all peak amplitudes were calculated by measuring the peak-to-peak amplitude (Richards 2003) - the magnitude of the component value subtracted from the maximum value of the previous opposite polarity peak (Supplemental Figure 3).
For the Nc waveform, the peak negative Nc component was identified as the most negative point between 300 and 600ms after the stimulus. The peak negative N290 components were identified as the most negative point between 200 and 350 ms after the stimulus. The peak positive P400 components was identified as the most positive point between 300 and 500ms.
To evaluate the difference in response for mother against stranger (Mother-Stranger), for each component, the peak amplitude response to stranger was subtracted from the peak amplitude response to mother.
Statistics
Demographics were analyzed across groups using Fischer’s Exact Test to determine any differences between groups. Continuous variables (eg. EEG HAPPE metrics, ERP component amplitudes) were analyzed for normality, and Kruskal-Wallis H tests (one-way nonparametric ANOVA) were used to compare groups when Shapiro-Wilk test was p<0.05, followed by post hoc Dunn’s tests to examine pairwise comparisons. Bonferroni’s correction was used to account for multiple comparisons such that family-wise error rate was set to a< 0.05. Two-way mixed ANOVA were used to determine the effects of group, picture, and group x picture interaction on ERP peak amplitudes.
Simple and multiple linear regressions were used to determine whether ERP peak amplitudes (Mother – Stranger) were associated with social and communication measures. Maternal education and ASD outcome group were included in models as covariates to account for their effects on social and communication measures.