Participating families were recruited within the Early Autism Sweden (EASE) project, a longitudinal study of infants at risk for autism (using a prospective sibling design, Nyström et al. 2018, 2019; Falck-Ytter et al. 2018). Participants were infants with one or more siblings with an ASD diagnosis to a high risk group (HR, n = 63, final sample n = 50) and comparable low risk control infants (LR, n = 28, final sample n = 23) that had at least one typically developing older full sibling and no first or second degree relatives with ASD. Families with infants in the HR group had been contacted through advertisements, the project’s web site and from clinical units. The older sibling diagnosis was confirmed by inspection of clinical records. Infants in the LR group were recruited from population birth records in selected municipalities in the larger Stockholm area (about 20% respond to our recruitment letters), and had at least one typically developing older full sibling and no first or second degree relatives with ASD. Infants with visual or auditory impairments or with known medical conditions (including prematurity before week 36) or genetic syndromes were excluded. The EEG and the Mullen Scales of Early Learning (MSEL) was recorded at 5 months of age. The HR and LR groups were matched according to gender, age, MSEL at 5 months and socioeconomic background (see Table 1).
Table 1 Background characteristics of the groups Written informed consent was collected from all parents. The study was approved by the Ethics Board in Stockholm and conducted in accordance with the 1964 Declaration of Helsinki.
Procedure
Families were welcomed upon arrival and given verbal instructions of the tasks during the day. Different assessments were performed at the different time points; see Nyström et al. (2017, 2018, 2019) and Falck-Ytter et al. (2018) for other experimental tasks during the day. The MSEL assessment was always performed by an experienced clinician before lunch.
The EEG was recorded using an age appropriate 128-channel Geodesic Sensor Net (Age appropriate 128-channel Geodesic Sensor Nets (HCGSN 130; EGI, Eugene, OR). The signal was sampled at 500 Hz relative to the vertex reference, amplified by EGI Net amplifier (GES 300 Amp; EGI, Eugene, OR) and stored for off-line analysis.
Stimuli were generated by a MacBook Pro using the PsychToolbox in MATLAB (2013a), running under OS X EL Capitan (version 10.11.6), and presented on a BenQ (23.5 inches) monitor with 1920 × 1080 pixel resolution operating at 60 Hz frame rate. As in a previous study (Wattam-Bell et al. 2010), for both form and motion, 2000 local arcs were always present on screen, alternating between coherent motion/form and random coherent displacement every 250 ms. Each local arc consisted of eight white dots plotted on a dark background (0.29° visual degrees). Following an 8-frame lifetime, each dot was replotted in a fresh random location on the screen. When plotted simultaneously on the screen, these dots created a short, static arc segment (the form condition). When plotted successively, they create a brief sample of motion along an arc trajectory for the motion condition (displacement between frames gave a speed of 8.6 visual degrees/sec). In the form condition the coherent interval resulted in a global concentric texture (see Fig. 1 for an example), and in the motion condition the coherent interval created a globally rotating motion about a common origin at the center of the screen. Patterns were viewed at ~ 60 cm and subtended 47.4° × 27.8°. The stimuli were presented in blocks with a duration of 12 s, containing 24 cycles. Each cycle had both a random phase (250 ms) and coherent phase (250 ms); see supplementary materials for video examples. Each 12 s block contained only form or motion stimuli, to entrain brain responses to the frequency of the specific condition, and the blocks were interleaved with unrelated experimental stimuli. We presented 10 form blocks and 10 motion blocks, giving 240 cycles in total for each condition.
Analysis
All analysis was done using MATLAB (R2018b), the EEGLAB toolbox (Delorme and Makeig 2004), and the TimeStudio scientific workflow system (Nyström et al. 2015). A subset of 121 EEG channels covering most of the scalp were used for analysis. All channels were resampled to 100 Hz to reduce computer memory load, and were high pass filtered at 0.5 Hz to filter out slow drifts in the signal. All channels were then re-referenced to average reference, and segmented into stimuli cycles as described above. To exclude artifacts, all cycles with a voltage range exceeding 100uV were excluded, as well as the first and last second of the block.
To extract brain responses related to the stimuli we calculated the T2circ statistics for all cycles and all channels separately. The T2circ statistics is based on both the real and imaginary coefficients of a Fourier transform for the frequency of the stimuli (or any other specified frequency), and requires systemic responses in both the amplitude and phase domain (Victor and Mast 1991). A statistically significant signal at the fundamental stimulus frequency (2 Hz) was taken as evidence for a neural process sensitive to global coherence, and because there are twice as many global changes every cycle (from random to coherent, and from coherent to random), a statistically significant signal at the double frequency (4 Hz) was taken as evidence for neural processes sensitive to low level contrast changes, as in Wattam-Bell et al. (2010).
All subjects without any significant channel, as tested with the T2circ statistics using all cycles in each channel, were excluded from further analysis. After exclusion, in the global motion condition the HR group (n = 50) contributed a mean of 155.2 (SD = 45.9) cycles, and the LR group (n = 23) a mean of 164.0 (SD = 38.5) cycles. In the global form condition the HR group (n = 50) contributed a mean of 155.6 (SD = 44.0) cycles, and the LR group (n = 23) mean = 162.7 (SD = 37.9) cycles.
Statistical comparison of topographical distributions was based on T2circ values by interpolating the electrode values over a uniform grid (~ 3500 vertices), bounded by the outer electrodes of the nets. These interpolated values were averaged separately within nine areas of interest (AOIs) around the back of the head (see Fig. 2, row 2–5). In the previous study by Wattam-Bell et al. (2010) only five AOIs were used for the same scalp surface, but we increased the number of AOIs to get a more detailed spatial profile, and less risk of pooling different brain processes into the same AOI. At this stage, each subject’s data consisted of separate 9-point spatial brain activity profiles. The activity profiles were vector-normalized (McCarthy and Wood 1985) to eliminate overall amplitude differences between individual infants. Our dependent measure used in our main analysis was a measure of the topographic centralization of the signal, calculated by subtracting the central AOI activation from the maximum activation in all AOIs. To visualize the general central vs lateral topographic pattern across individuals, the default AOI positions were flipped along the sagittal plane so that the maximum lateral amplitude always was presented on the left hemisphere (Nb participants flipped: LR form n = 9/23, HR form n = 26/50, LR motion n = 15/23, HR motion n = 30/50). The number of hemispherical flips did not differ between groups (χ2 test, group*condition, p > .25). This way left and right laterality between infants cannot cancel out in the average response. The resulting activity profiles for global change/coherence and form/motion profiles are shown in Fig. 2.