Participants: A total of 16 boys (33-78 months old) with full mutation of Fmr1 and 12 similarly aged (32-80 months old) typically developing boys were recruited for this study (IRB#P00025493) conducted at Boston Children’s Hospital/Harvard Medical School. 4 FXS participants did not complete baseline EEG acquisition (3 net refusal, 1 unable to maintain protocol), and 1 FXS participant’s EEG data was excluded due to excessive artifact. EEG and behavioral data were analyzed for a total of 11 FXS boys (mean age = 53.5 months; SD = 16.4 months, range 33-78 months old) and 12 typically developing boys (mean age = 47.7 months; SD = 13.1, range 32-80 months old).
FXS and age-controls: FXS participants all had documented full mutation of the Fmr1 gene, but could have size mosaicism (mixture of full and premutation) and methylation status was not known for all participants. Girls were excluded from this study given their variable expression of Fmr1 and the small size of this study. Across all groups, additional exclusion criteria included history of prematurity (<35 weeks gestational age), low birth weight (<2000gms), known birth trauma, known genetic disorders (other than FXS), unstable seizure disorder, current use of anticonvulsant medication, and uncorrected hearing or vision problems. Some participants were on stable doses of medications (Oxybutin (1 age-matched control); melatonin (2 FXS); Miralax (1 age-matched control). Children were from primarily English-speaking households with English spoken more than 50% of the time (2/11 FXS and 2/12 age-matched control participants were either in bilingual households or daycare).
Cognitive and sex-matched controls: EEG data from an additional set of 12 cognitive-matched boys were analyzed (mean age = 29.8 months; SD = 10.1 months, range 14-52 months old). 11 individuals from this group provided EEG data as part of a concurrent longitudinal study (IRB#P00018377) in the lab which used the same EEG resting-state paradigm. Exclusion criteria were the same as above. In addition infants in this group did not have a sibling with autism. Controls were identified by matching FXS participants’ Fine Motor and Visual Reception raw scores on the Mullen Scales of Early Learning (see below). In order to appropriately match all FXS participants, one EEG in this group overlapped with the above age/sex matched control group. Given this, age vs cognitive matched controls were not statistically compared.
![](https://myfiles.space/user_files/58653_1b1c6aeb34a62c68/58653_custom_files/img1612990510.jpg)
Institutional review board approval was obtained prior to starting the study. Written, informed consent was obtained from all parents or guardians prior to their children’s participant in the study. Table 1 describes participant characteristics.
EEG Assessment: Resting-state EEG data were collected in a dimly lit, sound-attenuated, electrically shielded room. The child either sat in their seated caregiver’s lap or sat independently in a chair, high-chair, or stroller based on behavioral preference. Caregivers were instructed by a research assistant to avoid social interactions or speaking with their child. Continuous EEG was recorded for 2-5 minutes
depending on compliance. Given known behavioral challenges in the FXS population, parents were asked about expected behavioral challenges, calming techniques, and motivators specific to each child. To improve compliance, participants were shown a silent screensaver of abstract colorful moving images and allowed to hold a fidget toy. EEG data were collected using a 128-channel Hydrocel Geodesic Sensor Net (Version 1, EGI Inc, Eugene, OR) connected to a DC-coupled amplifier (Net Amps 300, EGI Inc, Eugene, OR). Data were sampled at 1000Hz and referenced to a single vertex electrode (Cz), with impedances kept below 100kΩ in accordance with the impedance capabilities of the high-impedance amplifiers inside the electrically shielded room(44). Electrooculographic electrodes were removed to improve the child’s comfort.
EEG pre-processing: Raw NetStation (NetStation version 4.5, EGI Inc, Eugene, OR) files were exported to MATLAB (version R2017a) for pre-processing and power analysis using the Batch EEG Automated Processing Platform (BEAPP; (45)) with integrated Harvard Automated Preprocessing Pipeline for EEG (HAPPE; (46)). Preprocessing has previously been described in detail for similar data(37). Briefly, data were 1Hz high-pass and 100Hz low-pass filtered, downsampled to 250hz, and then run through the HAPPE module for 60Hz line noise removal, bad channel rejection and artifact removal using combined wavelet-enhanced independent component analysis (ICA) and Multiple Artifact Rejection Algorithm (MARA(47,48)). Given the short length of EEG recording, 39 of the 128 channels were selected for ICA/MARA (Standard 10-20 electrodes: 22, 9, 33, 24, 11, 124, 122, 45, 36, 104, 108, 58, 52, 62, 92, 96, 70, 83; Additional electrodes: 23, 28, 19, 4, 3, 117, 13, 112, 37, 55, 87, 41, 47, 46, 103, 98, 102, 75, 67, 77, 72). These electrodes were chosen based on their spatial location, covering frontal, temporal, central, and posterior regions of interest for later analysis (see Figure 1). After artifact removal, channels removed during bad channel rejection were interpolated, data were rereferenced to an average reference, detrended using the signal mean, and segmented into 2-second segments. Any segments with retained artifact were rejected using HAPPE’s amplitude and joint probability criteria.
EEG were rejected for data quality if they had fewer than 20 segments (40 seconds total), or did not meet the following HAPPE data quality output parameters: percent good channels >80%, mean and median retained artifact probability <0.3, percent of independent components rejected <84%, and percent variance after artifact removal <32%. Table 1 shows quality metrics for all groups.
EEG power analysis: Power spectral density (PSD) at each electrode, for each 2 second segment, was calculated with multitaper spectral analysis(49) embedded in BEAPP using three orthogonal tapers. For each electrode for a given EEG, PSD for each frequency bin (0.5Hz frequency resolution) was averaged across segments, and then averaged across the regions of interest shown in Figure 1. PSDs were normalized using Log10(Hz). Our analyses were limited to absolute power, as relative power measurements were artificially affected by normalization; increased power in high frequency bands in FXS participants artificially lowered the relative power of lower frequency bands. The PSD was also analyzed using the FOOOF v1.0.0 parameterization model across a 2-55Hz frequency range (https://github.com/fooof-tools/fooof; in Python v3.6.8) in order to model periodic and aperiodic components of the power spectra(50). The FOOOF model was used in the fixed mode (no spectral knee) with peak_width_limits set to [1, 18.0], max_n_peaks = 7, and peak_threshold = 2). For each subject’s power spectrum FOOOF provides two parameters to describe the aperiodic 1/f background signal: offset and slope. To determine an aperiodic-adjusted gamma power, the FOOOF estimated aperiodic signal was subtracted from the raw power spectrum, resulting in a flattened spectrum. FOOOF model fit to the original spectrum for each group is shown in Supplemental Figure 1.
Behavioral Measures: The Mullen Scales of Early Learning (MSEL(51)) is a standardized cognitive measure for children 0-69 months of age. Non-verbal subscales (fine motor, visual reception) were administered to all FXS participants regardless of age, age-matched controls under 70 months of age, and all cognitive matched controls, and the Nonverbal Developmental Quotient (NVDQ) was calculated. The Preschool Language Scale 5th Edition (PLS(52)), a comprehensive developmental language assessment standardized for children 0-83 months of age, was administered to FXS and age-matched participants. The PLS was used instead of the MSEL to assess language, as it covers the full age range of the study sample and has been recently updated to include toys and images that are more consistent with items children interact with today. Standard scores of subtests for receptive (Auditory Comprehension) and expressive (Expressive Communication) language, as well as the total standard score were calculated. Note that per research administration protocol augmentative communication devices are not used during this assessment, so scores may underestimate a child’s non-verbal expressive language skills. The following clinical questionnaires were completed by primary caregivers of FXS and age-matched participants: Aberrant Behavior Checklist-Community Edition(53) (ABC-FXS, scored using FXS specific factoring system(54)), Vineland Adaptive Behavior Scales, Third Edition(VABS-3(55), Repetitive Behavior Scale-Revised(56), and Sensory Profile, Child-2(57). The ABC-FXS scoring included 6 subscales: irritability, hyperactivity, lethargy, social avoidance, stereotypy, and inappropriate speech.
Statistical Analyses: T-test, or Mann Whitney if data was not normal in distribution, was used to compare differences in behavioral scores or EEG measures between either FXS vs. age-matched controls, or FXS vs. cognitive-matched controls. To examine group differences in the power spectra, a non-parametric clustering method, controlling for multiple comparisons using Monte Carlo estimation (1000 permutations)(58) was employed with MNE-Python(59) using a F-statistic threshold of 4.32. Regression analysis was used to characterize the relationship between gamma power and behavioral measures within the FXS group. Analysis were performed used Stata software, version 14.2 (Stata). Figures were created using Python v3.6.8 and python data visualization libraries (matplotlib(60) and Seaborn (https://seaborn.pydata.org/index.html).