Early Trajectories of Resting-State EEG Power in Autistic Children: A Longitudinal Study Across Language Profiles

Early Trajectories of Resting-State EEG Power in Autistic Children: A Longitudinal Study Across Language Profiles | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Early Trajectories of Resting-State EEG Power in Autistic Children: A Longitudinal Study Across Language Profiles Kenza Latrèche, Michel Godel, Ana Flò, Fiona Journal, Valentina Borghesani, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7204586/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Language development in autism spectrum disorder (ASD) is heterogeneous, ranging from subtle differences to significant delays. In previous work, we identified three autistic language profiles in early childhood: Language Unimpaired (LU), Language Impaired (LI), and Minimally Verbal (MV). While these profiles show distinct vocabulary, grammar, and pragmatic development, understanding their underlying neural correlates is essential to predict outcomes and develop targeted interventions. Here, we examined whole-brain resting-state EEG power across five canonical frequency bands in a longitudinal sample comprising 66 typically developing (TD) children and 122 autistic children (ages 1.6-6.0 years), yielding 358 time points. Within the ASD group, 61 children belonged to the LU profile, 44 children to LI, and 17 children to MV. Compared to TD peers, autistic children showed increased power in low-frequency (delta, theta) and high-frequency bands (beta, gamma). Gamma power varied by autistic language profile, with the highest levels in MV children. Moreover, gamma power within ASD followed a quadratic trajectory in relation to word combination acquisition, peaking around the time of acquisition and decreasing afterward. This pattern suggests a dynamic, compensatory mechanism supporting the transition to phrase speech, which is a critical milestone toward functional speech that may predict language outcomes in ASD. Developmental Neuroscience Psychiatry Psychology Figures Figure 1 Figure 2 Figure 3 Introduction Autism Spectrum Disorder (ASD) is a prevalent neurodevelopmental condition characterized by difficulties in social communication and interactions, as well as repetitive and restrictive behaviors or interests 1 . While outcomes across the spectrum vary widely, early language milestones (e.g., age of first spoken words), are one of the strongest predictors of long-term development 2 – 4 . However, language development in autism ranges from significant difficulties to subtle differences 5 , 6 . In previous work, we aimed to capture language heterogeneity among 215 autistic children (aged 1.5–5.7 years old) 7 . Using a data-driven approach, we identified three distinct clusters that corresponded to the autistic language profiles described in the 11th International Classification of Diseases, namely Minimally-Verbal (MV), Language Impaired (LI), and Language Unimpaired (LU) 8 , 9 . In our previous study, MV children exhibited poor vocabulary and limited use of word combinations 7 . The LI group showed delayed vocabulary development and reduced use of word combinations, while the LU group demonstrated vocabulary size and word combinations comparable to TD peers 7 . Understanding the neural correlates underlying these distinct language trajectories is crucial for predicting outcomes and developing targeted interventions, particularly for children with greater language difficulties 10 – 12 . Electroencephalography (EEG) has emerged as a powerful tool in research on autistic children, given its high temporal resolution and non-invasiveness 13 . Resting-state EEG (RS-EEG) paradigms are particularly valuable as they require no engagement in a cognitive task 13 . RS-EEG activity is frequently investigated with spectral analysis, where brain oscillations are decomposed and power is quantified over five canonical frequency bands: delta (2–4 Hz), theta (4–6 Hz), alpha (6–13 Hz), beta (13–30 Hz), and gamma (30–50 Hz) 13 – 15 . Prior evidence suggested a U-shaped profile in ASD, with increased power for low frequencies (delta, theta) and high frequencies (beta, gamma) and decreased power for middle frequencies (alpha) compared to TD 13 , 15 , 16 . Nevertheless, reliable RS-EEG biomarkers of ASD have yet to be identified, due to limitations in previous studies including small sample sizes (on average 30 autistic and 35 TD participants 15 ), cross-sectional designs that prevent examination of developmental processes, and analyses limited to specific electrodes and/or frequency bands 14 . Some previous studies examined the relationship between RS-EEG and language development in infants at high- and low-likelihood (HL; LL) for autism 17 , 18 . In a cross-sectional study on 70 HL infants aged 12–23 months, Cohenour and colleagues reported positive correlations between alpha power and receptive and expressive language skills, suggesting that alpha oscillations support language production and comprehension 17 . In a longitudinal study in 72 HL and 58 LL infants, increased 6-month frontal gamma power (FGP) was significantly associated with higher 24-month language functioning in HL infants without ASD (HL-NoASD), but with lower 24-month language functioning in HL infants later diagnosed with ASD (HL-ASD) 18 . As gamma oscillations reflect the brain’s balance between excitatory/inhibitory (E/I) neurotransmitters, reduced FGP in HL-ASD could represent successful compensation for underlying neurotransmitter imbalances. In contrast, reduced FGP in HL-NoASD could indicate delayed neural maturation and language abilities 18 . To more accurately assess the brain’s E/I balance, Wilkinson and colleagues analyzed aperiodic power in a subsequent longitudinal study of 3- and 12-month-old HL and LL infants 19 . They found greater increases in aperiodic power in HL-ASD infants, along with reduced language development at 18 months, suggesting that early E/I imbalance may underlie later language difficulties in HL-ASD infants. Furthermore, a cross-sectional study on preschoolers (48 ASD, 58 TD) showed decreased FGP associated with greater expressive language skills in autistic children 20 , supporting that reduced gamma power may reflect a compensation mechanism for language development 18 , 20 . While previous research has reported associations between RS-EEG power and language development, no study to our knowledge has examined RS-EEG power longitudinally in autistic children stratified by language profile. Longitudinal investigation is particularly important as RS-EEG power trajectories might change across the early years of life, following non-linear trajectories alongside language development 14 , 18 . To address these limitations, the present study conducts a longitudinal investigation of whole-brain EEG power in the five canonical frequency bands in three previously validated autistic language profiles (LU, LI, and MV) 7 . The longitudinal sample yields 358 time points and includes 122 autistic (61 LU, 44 LI, 17 MV) and 66 typically developing children, aged 1.56 to 6.01 years old. Our study has three aims, (1) to explore longitudinal trajectories of each frequency band across the TD and ASD groups, and within LU, LI, and MV; (2) to examine the trajectory of gamma power in relation to word combination acquisition within the ASD sample; (3) to classify TD and the three autistic profiles based on fine-grained brain activity features, using a supervised machine learning algorithm. We hypothesized that ASD children would demonstrate increased power in low (delta, theta) and high frequency bands (beta, gamma) and decreased power in middle frequencies (alpha), in line with the proposed U-shaped profile 15 . We expected the LU, LI, and MV subgroups to differ mainly in the gamma band, reflecting previously reported associations between language skills and higher frequencies in autistic children and HL infants 18 , 20 . As decreased gamma power may reflect a compensation mechanism in ASD, MV would exhibit the greatest gamma power, followed by LI, and finally LU. Considering that production of word combinations greatly differed across each language profile 7 , we evaluated gamma power trajectory jointly with this milestone. Material and methods Participants Our sample comes from the Geneva Autism Cohort, which is an ongoing open longitudinal cohort that follows autistic and typically developing (TD) preschoolers 7 . In this cohort, we include children who received a diagnosis of ASD before age 5, and their typically developing peers, and we follow them every 6 months for two years. Each visit comprises several longitudinal assessments, from behavioral measures to neuroimaging. Autistic participants are recruited through local clinical centers and parent associations. A licensed child psychiatrist (MS) confirmed the diagnosis of all autistic participants based on the Diagnostic and Statistical Manual of Mental Disorders − 5th ed. 1 and gold-standard tools (Autism Diagnostic Observation Schedule 21 ; Autism Diagnostic Interview-Revised 22 ). Inclusion criteria for TD participants included full-term birth, no ASD diagnosis in first-degree relatives, and no neurological or somatic concerns. TD participants completed the ADOS 21 to exclude ASD symptoms and a developmental assessment (either the Mullen Scales of Early Learning, MSEL 23 or the Psychoeducational Profile – third edition, PEP-3 24 ) to rule out significant developmental delays (i.e., a developmental quotient below 70 in any domain). Written informed consent forms were signed and provided by the participants’ caregivers. The Ethics Committee of the University of Geneva approved the research protocol. The present study builds upon our previous work 7 , where we subgrouped early expressive language skills of autistic preschoolers into three distinct profiles (LU, LI, MV). To understand the profiles’ underlying neural underpinnings, we included TD and autistic children from our previous work with at least one RS-EEG recording. One to three recordings per child were available for this study. Our final sample comprised 188 participants (358 time points), aged from 1.56 to 6.01 years old. The TD group was composed of 66 participants (130 time points, age range 1.56–5.84 years, 43.9% of females). The ASD group included 122 participants (228 time points, age range 1.74–6.01 years old, 12.3% of females), and was divided in the three language profiles subgroups: LU (61 participants, 127 time points, 1.74–5.99 years, 6.6% females), LI (44 participants, 71 time points, 1.93–5.82 years old, 11.4% females), and MV (17 participants, 30 time points, 1.80–6.01 years old, 35.3% females). Sample characteristics are detailed in Table 1 . Table 1 Sample characteristics of the TD and LU, LI, and MV samples Measure (mean ( SD )) TD ASD ASD-LU ASD-LI ASD-MV Number of participants 66 122 61 44 17 Number of time points 130 228 127 71 30 Time points per participants 2.0 ( 0.8 ) ‡ 1.9 ( 0.8 ) 2.1 ( 0.8 ) 1.6 ( 0.7 ) 1.8 ( 0.8 ) Mean age at first available TP (years) 3.3 ( 1.1 ) ‡ 3.5 ( 1.0 ) 3.4 ( 0.9 ) 3.8 ( 1.1 ) 3.2 ( 1.3 ) Age range (years) 1.6–5.8 1.7-6.0 1.7-6.0 1.9–5.8 1.8-6.0 Female biological sex 29 (43.9%) ‡ 15 (12.3%) † 4 (6.6%) 5 (11.4%) 6 (35.3%) Multilingual environment [N = 188] 24 (36.4%) 57 (46.7%) 27 (44.3%) 18 (40.9%) 12 (70.6%) Parental education [N = 188] (% college degree completed) 62 (93.9%) ‡ 65 (53.7%) † 28 (45.9%) 24 (54.5%) 9 (52.7%) Mean ADOS CSS Total at first available TP [N = 188] 1.1 ( 0.3 ) ‡ 7.3 ( 1.8 ) † 6.7 ( 1.6 ) 7.5 ( 2.0 ) 8.7 ( 1.4 ) Mean ADOS CSS SA at first available TP[N = 188] 1.1 ( 0.3 ) ‡ 6.3 ( 2.1 ) † 5.8 ( 1.9 ) 6.4 (2 .4 ) 7.6 ( 1.7 ) Mean ADOS CSS RRB at first available TP [N = 188] 2.6 ( 2.1 ) ‡ 8.6 ( 1.6 ) † 8.4 ( 1.8 ) 8.7 ( 1.5 ) 9.3 ( 1.1 ) Mean Total DQ [TP = 356] 114.8 ( 12.5 ) ‡ 79.14 ( 25.6 ) † 94.5 ( 15.7 ) 65.7 ( 13.4 ) 36.6 ( 11.5 ) Mean EL DQ [TP = 351] 107.8 ( 18.2 ) ‡ 66.6 ( 27.3 ) † 84.3 ( 18.4 ) 53.8 ( 15.0 ) 23.8 ( 9.9 ) Mean RL DQ [TP = 351] 118.3 ( 14.3 ) ‡ 72.9 ( 32.1 ) † 93.6 ( 20.3 ) 58.5 ( 21.0 ) 21.6 ( 9.3 ) Mean VR DQ [TP = 351] 123.3 ( 20.5 ) ‡ 88.3 ( 27.5 ) † 104.4 ( 20.3 ) 77.1 ( 17.0 ) 48.6 ( 17.8 ) Mean FM DQ [TP = 351] 107.9 ( 13.8 ) ‡ 81.1 ( 22.2 ) † 94.0 ( 17.0 ) 73.4 ( 15.4 ) 52.2 ( 16.8 ) DLPF Total Vocabulary Score [TP = 322] 1104.0 ( 412.3 ) ‡ 641.9 ( 503.2 ) † 852.6 ( 478.3 ) 454.7 ( 355.7 ) 24.4 ( 69.8 ) DLPF Total Grammar Score [TP = 323] 79.6 ( 29.9 ) ‡ 42.1 ( 36.9 ) † 57.7 ( 36.6 ) 26.7 ( 22.8 ) 0.79 ( 2.6 ) DLPF Total Pragmatics Score [TP = 314] 45.2 ( 6.3 ) ‡ 30.5 ( 16.8 ) † 38.4 ( 12.2 ) 26.8 ( 13.4 ) 2.4 ( 4.3 ) Probability values refer to Kruskal–Wallis, Mann-Whitney or Pearson chi-square test, as appropriate. † p < .05 vs. TD; ‡ p < .05 across TD, LU, LI, MV. ADOS = Autism Diagnostic Observation Schedule; ASD = Autism spectrum disorder; ASD-LU = Autism spectrum disorder-language unimpaired; ASD-LI = Autism spectrum disorder-language impaired; ASD-MV = Autism spectrum disorder-minimally verbal; CSS = Calibrated Severity Score; DLPF = Développement du Langage Productif en Français ; DQ = Developmental Quotient; EL = Expressive Language; FM = Fine Motor; RL = Receptive Language; RRB = Restricted and Repetitive Behaviors; SA = Social Affect; TD = Typical Development; TP = time point; VR = Visual Reception. We used the G*Power software to conduct a power analysis of our sample. With our TD group (N = 66) and ASD group (N = 122), the study is well-powered to detect both large (Cohen’s d = 0.8) and medium (Cohen’s d = 0.5) effect sizes. Power was nearly 100% for a large effect and 90.6% for a medium effect, indicating sufficient sample sizes to detect meaningful differences across groups. Detecting a small effect (Cohen’s d = 0.2) would require 394 participants per group to achieve a power of 80%. Of note, this analysis does not account for the repeated measures within participants. As no clear consensus exists on power analysis for unbalanced, nested longitudinal designs, we report cross-sectional power estimates for informational purposes, acknowledging the limitations of this approach. Measures To characterize our longitudinal cohort, behavioral and developmental measures were administered to complement the neurophysiological data. Behavioral measures We administered the ADOS-2 21 to confirm ASD in autistic participants and to exclude ASD symptoms in TD participants. The ADOS-2 is a semi-structured assessment quantifying symptoms in social affect and restricted and repetitive behaviors or interests. The ADOS-2 includes five modules, administered depending on the child’s age and language level. Modules were compared using calibrated total severity score 25 , 26 . The ADOS-2 was administered and coded by trained examiners. Developmental level was measured using the MSEL 23 , assessing fine motor (FM), visual reception (VR), receptive language (RL), and expressive language (EL) in children aged 0 to 68 months. We computed developmental quotients (DQs) by dividing the chronological age by the developmental age and multiplying by 100 7,27,28 . DQs were selected over standard scores to reduce floor-effects in lower-performing participants. As the MSEL was added later in our protocol, 23 participants (32 time points) were assessed with the PEP-3 24 or with the WPPSI-IV 29 . For the PEP-3, designed for children aged 2 to 7 years, we computed DQs for domains. The PEP-3 and MSEL showed excellent consistency for both expressive (Cronbach’s alpha = 0.899) and receptive language (Cronbach’s alpha = 0.913) 7 . Five children (5 time points) completed the WPPSI-IV, where only the full-scale IQ was used. Two children (1 ASD, 1 TD) had no developmental data at the EEG timepoint: one was too young, the other had behavioral difficulties invalidating the PEP-3. Parent-reported expressive language Expressive language skills were evaluated with the Développement du langage de production en français (DLPF) 30 parent-reported questionnaire. The DLPF assesses vocabulary, grammar, and pragmatic skills. Three total domain scores were computed 31 . The DLPF assesses children aged 18 to 42 months, as scores plateaued when approaching 42 months in TD 30 . Here, we administered the DLPF until 6 years old given the frequent expressive language delay in ASD 32 . We also examined word combination acquisition, which is a significant milestone toward functional speech 33 . We defined word combination as the use of at least two words to communicate (e.g., Open door , More milk ) 30 , 33 . We measured the age of word combination acquisition using several tools, depending on what assessment was available, i.e., either the DLPF (item B, Sentences and Grammar Section), the ADI-R (item 10), the MSEL (item 17), and the ADOS-2 (item A1). RS-EEG data collection, preprocessing, analyses RS-EEG recordings were acquired with a sampling rate of 1000 Hz using 128-channel Hydrocel Geodesic Sensor Nets and Net Station acquisition software (v 5.4.1.2) (Magstim EGI, Inc. Eugene, OR). To minimize movement artifacts and facilitate compliance, we adapted the RS-EEG paradigm: children watched an age-appropriate cartoon of their choice with sound, sitting ~ 60 cm from the screen, independently or on a caregiver’s lap. They were encouraged to sit still and quiet. EEGs were recorded for 5 minutes and extended by 1–2 minutes if children were initially agitated or talking. We excluded 16 recordings under 300 seconds (5 from TD, 2 from LU, 8 from LI, 1 from MV). For longer recordings, the last 300 seconds were selected to retain the most stable segments. Preprocessing was conducted using the Automated Pipeline for Infants Continuous EEG (APICE) 35 , which is based on EEGLAB 36 and is specialized for young children. The data was first low-pass filtered at 100 Hz and high-pass filtered at 1 Hz, and a notch filter was applied to remove line noise (50 Hz). The pop_eegfiltnew function from EEGLAB was used for filtering with the default parameters. Then, APICE was applied. APICE detects non-functional electrodes and motion artifacts based on signal properties, primarily fast signal changes, and repairs the data, when possible, to maximize data recovery. Brief artifacts (shorter than 100 ms) were corrected using PCA, while longer artifacts affecting only a subset of the channels (less than 30%) were corrected using spatial spline interpolation. Channels identified as non-functional during the whole recording were also interpolated using spatial splines. Finally, data was visually inspected to check for undetected corrupted segments, and those segments were excluded from the analysis. Two RS-EEG recordings (1 LU, 1 MV) with less than 100 seconds of time free of artifacts were excluded from the power analysis. We performed a multitaper spectral analysis to decompose EEG signal into power within the 2–50 Hz range over all electrodes. A 5-second time window, spaced at 1-second intervals, was used to balance temporal resolution and spectral stability. Spectral estimates were computed using a time-bandwidth of 2.5 and four orthogonal tapers 37 . We applied equal weighting to all tapers. Mean log(10) whole-brain power was calculated across the five canonical frequency bands: delta (2–4 Hz), theta (4–6 Hz), alpha (6–13 Hz), beta (13–30 Hz), and gamma (30–50 Hz) 18 . Statistical analyses Longitudinal analyses We conducted mixed model analyses to examine developmental, linguistic, and RS-EEG power trajectories across age. Mixed modeling is a successful approach for nested data with varying numbers of time points per participant and is well suited to longitudinal behavioral and neurophysiological research 38 , 39 . Age and group (TD, LU, LI, MV) were included as fixed effects, sex as a covariate, and behavioral and EEG measures as random effects. Analyses were conducted using the myMixedModelsTrajectories toolbox in MATLAB® R2019b (MathWorks, Natick, MA), fitting linear and quadratic random-slope models to capture different age-related patterns. Behavioral outcomes included Total DQ, EL DQ, RL DQ, FM DQ, VR DQ, and DLPF scores (Vocabulary, Grammar, Pragmatics). RS-EEG power trajectories were estimated across the five frequency bands, and model fit was evaluated using the Bayesian Information Criterion (BIC). As the word combination acquisition was a binary categorical variable (i.e., acquired or not), we performed a chi-square test with a 12-month sliding window comparing the proportion of children acquiring combinations at each age bin and applying FDR correction. Moreover, to examine word combination acquisition in relation to whole-brain gamma power, we applied a mixed model analysis fitting linear, quadratic, and cubic random-slope models. The BIC selected the quadratic fit to model gamma power relative to the estimated timing of word combination acquisition. This analysis was limited to the ASD group, as most TD children (63/66, 95.5%) had acquired combinations by their first time point. To compare gamma power before and after acquisition, we excluded ASD children with only one time point (N = 48) or two non-consecutive time points (N = 6). The final ASD sample included 68 children (41 LU, 18 LI, 9 MV) with 168 RS-EEG recordings (104 LU, 42 LI, 22 MV) between 1.74 and 5.99 years. Cross-sectional classification analysis We conducted two exploratory classification analyses on the cross-sectional sample (N = 188; 122 ASD, 66 TD) to determine if resting-state EEG power features could distinguish between diagnostic groups. The first analysis was a binary classification of ASD versus TD participants. The second was a multi-class classification of three autistic language subgroups and the TD group (a total of four groups). For these analyses, we used 6 400 features derived from the RS-EEG data, comprising power values from 128 electrodes across 5 canonical frequency bands (delta, theta, alpha, beta, gamma) and the first ten 10-second artifact-free time bins. We opted to retain discrete time bins per participant instead of using a single average across time as fluctuations in RS neural activity may contain diagnostically relevant information. While there is no synchronized 'time course' across participants, a classifier can leverage these multiple temporal snapshots to learn patterns of intra-individual variability. For instance, the stability or range of power fluctuations across the ten bins may itself be a distinguishing characteristic between groups, information that would be obscured by averaging. To perform the classifications, we implemented a supervised machine learning pipeline using scikit-learn in Python 40 . This pipeline integrated several steps to ensure a robust evaluation: data scaling, feature selection, and classification with a linear Support Vector Machine (SVM). The entire pipeline was evaluated using a nested cross-validation approach to prevent information leakage and provide an unbiased estimate of model performance. The outer loop of the nested cross-validation used a 5-fold stratified split to divide the data, preserving the percentage of samples for each class. The inner loop employed a 3-fold stratified cross-validation on the training data from the outer loop to perform hyperparameter tuning using a GridSearchCV. Specifically, we optimized the C parameter of the linear SVM (SVC(kernel='linear')) over a range of values ([0.01, 0.1, 1, 10, 100]). Within this pipeline, we also incorporated a feature selection step using SelectKBest to retain the 500 most informative features based on their ANOVA F-value relative to the diagnostic labels. To assess the statistical significance of our classification results, we performed a permutation test. For each classification task (2-group and 4-group), we ran the entire nested cross-validation procedure 10 000 times, each time with the diagnostic labels randomly shuffled. This process generated a null distribution of accuracy scores that would be expected by chance. The p -value was then calculated as the proportion of permutation-based accuracies that were equal to or greater than the accuracy achieved with the true labels. This rigorous procedure allowed us to determine if the model's performance was statistically significant. For further model interpretation, we also generated a confusion matrix for each classification to visualize the specific patterns of correct and incorrect predictions across the groups. The matrix is generated from the aggregated out-of-sample predictions from each fold of the outer cross-validation loop, which gives a realistic estimate of performance on unseen data. The code of these analyses can be found at https://github.com/NoCe-Lab/rsEEG_ASD . Results Description of TD and language profiles Descriptive analyses of the 66 TD (130 time points) and the 122 ASD (228 time points) participants show no significant age difference at the first available time point (mean age 3.3 ± 1.1 for TD and 3.5 ± 1.0. for ASD) (Table 1 ). As there were significantly more TD females (43.9%) than autistic (12.3%), sex was included as a covariate in all analyses. As expected, autistic participants exhibited higher autism symptom severity, and lower non-verbal and verbal skills. Additionally, there were no significant differences in age at the first timepoint across the LI, LU, MV, and TD groups ( p > 0.05). As female proportions differed across the four groups, sex was included as a covariate. The monolingual vs. multilingual environment did not significantly differ. Parental education was higher in TD than in the ASD subgroups (LU, LI, MV). ASD symptom severity and DQs differed across the four groups. To examine developmental trajectories across ASD, TD, and the three language profiles (LU, LI, MV), we analyzed longitudinal changes in non-verbal and verbal skills. Additionally, we examined expressive language development with specific measures of their vocabulary, grammar, and pragmatics. Our findings align with our previous work 7 , with distinct verbal and non-verbal trajectories within the three language profiles compared to TD children (Supplementary Analysis 1). Feasibility of RS-EEG acquisition in TD and ASD samples Due to anxiety about unfamiliar experiences and/or tactile defensiveness, it is challenging to collect EEG data in young autistic children 41 , 42 . Habituation strategies have been shown to improve compliance, including in MV children 41 , 43 . In our study, we provided materials to help children, and their parents familiarize themselves with the EEG, including pictograms and a short story illustrating the EEG room and net, a video demonstrating the setup and procedure, and a mock EEG net. Here, we defined the feasibility of RS-EEG acquisition as having at least one ≥ 300-second recording per child across the three yearly time points. Feasibility was 77.6% in TD (66/85) and 56.7% in ASD (122/215), consistent with previous reports of EEG compliance difficulties in ASD (χ²(1) = 11.38, p < 0.001) 41,42 . Feasibility varied by language profile: 70.9% (LU), 51.8% (LI), and 38.6% (MV) (χ²(2) = 13.79, p = 0.001), indicating greater challenges with limited language 43 . RS-EEG power trajectories: TD vs. ASD We compared whole-brain RS-EEG power trajectories of TD and ASD participants over the five frequency bands (Fig. 1 , Table S3). In delta, theta, and beta frequency bands, children with ASD showed significantly increased whole-brain power ( p = 0.013, p = 0.034, p < 0.001 respectively), with non-significant interactions between band and age. No significant difference was found for whole-brain alpha power. Finally, whole-brain gamma power was significantly higher in the ASD group than in the TD group ( p < 0.001) and the interaction was also significant ( p = 0.025), revealing different trajectories over development. While whole-brain gamma power slowly decreased with age in TD children, it remained stable with age in the ASD group. RS-EEG power trajectories : TD, LU, LI, and MV Building upon the observed differences between the broader ASD and TD groups, we further analyzed whole-brain RS-EEG power trajectories across the three autistic language profiles (LU, LI, MV) (Fig. 2 , Table S4). Whole-brain delta trajectories showed significant group effect ( p = 0.009), but non-significant age-interaction ( p > 0.05). In both theta and alpha bands, we found no significant group effect or age-interaction ( p > 0.05). Significant group effect ( p < 0.001) and age-interaction ( p = 0.018 and p = 0.012, respectively) were observed in beta and gamma bands. While beta and gamma power decreased with age for the TD and the LU groups, it increased over development for LI and MV. Given our hypothesis that whole-brain gamma power differs across the autistic language profiles, we conducted additional analyses comparing the four groups two-by-two. We found significant differences between TD vs . LU, TD vs . LI, TD vs . MV, and LU vs . MV (Figure S5, Table S4-5). RS-EEG gamma power and word combination acquisition Considering the heterogeneity in word combination acquisition and the association between gamma power and language in ASD 18 , 20 , we examined the trajectory of gamma power in our ASD sample. The quadratic model showed a progressive increase in gamma power, before plateauing near phrase speech acquisition (Fig. 3 ), with the peak occurring 0.384 years (approximately 4 months) post-acquisition. Gamma power then decreased after acquisition, suggesting its crucial role in the transition from single words to phrases. RS-EEG power features-based classifications Binary Classification: ASD vs. TD The supervised machine learning pipeline distinguished between individuals with ASD and Typically Developing (TD) controls with a mean accuracy of 69.2%. A permutation test, which involved running the entire cross-validation procedure 10 000 times with shuffled labels, confirmed that this accuracy was highly significant ( p = 0.0004). The observed true accuracy was well beyond the distribution of accuracies obtained from the permuted data, as illustrated in the permutation plot (Figure S6). The confusion matrix from the cross-validation reveals the specific performance of the classifier: Sensitivity (ASD identification): The model correctly identified 94 of the 122 individuals with ASD, yielding a sensitivity of 77.0%. Specificity (TD identification): The model correctly identified 36 of the 66 TD individuals, for a specificity of 54.5%. Multi-Class Classification: 4 Language Profiles For the four-group classification task (distinguishing between the three ASD language subgroups and the TD group), the model achieved a mean accuracy of 44.6%. This performance is substantially higher than the chance level of 25%. The permutation test demonstrated that this result was highly statistically significant ( p < 0.0001), with the true accuracy falling far outside the null distribution. An analysis of the confusion matrix shows that classification performance varied considerably across the four groups: The model was most successful in identifying TD participants, classifying correctly 66.7% of cases (44 out of 66). The model performed poorly in identifying the ASD-MV group, with only 1 correct classification out of 17 cases (5.9% accuracy). There was considerable confusion between the ASD-LI and ASD-LU subgroups, which the model often misclassified as each other (respectively 40.90% and 34.4% of accuracy). Discussion In this study, we investigated whole-brain RS-EEG power trajectories in a substantial longitudinal sample of 188 children (N = 122 ASD, N = 66 TD), yielding 358 time points. Given the heterogeneous language abilities in autism, we explored RS-EEG power trajectories among three autistic language profiles: Language Unimpaired (LU, N = 61), Language Impaired (LI, N = 44), and Minimally Verbal (MV, N = 17). We provide valuable insights into the neural correlates of language development in autism. In line with our hypothesis, our results on the RS-EEG power trajectories provide support to the U-shaped profile in autism (Fig. 1 ) 13 . The ASD group exhibited increased power in low-frequency bands (delta, theta) and high-frequency bands (beta, gamma), and similar alpha power compared to the TD group. While the U-shaped profile has been corroborated 13 , 15 , 16 , findings have been inconsistent. Some evidence associated enhanced delta power to lower functioning children 13 , 44 , 45 , which aligns with our LI group showing significantly higher delta power than TD peers. However, other studies report enhanced delta power in high functioning children 13 , 46 . Theta power has been shown to increase in autistic individuals 46 , 47 , though a meta-analysis found no difference compared to TD participants 15 , consistent with our findings. Additionally, our results align with the established decline of delta and theta power with age 14 , reflecting brain maturation, gray matter tissue loss, and increased processing efficiency 48 . While reduced alpha power in ASD has been proposed as a biomarker 15 , 48 , our findings and those from large cohort studies 14 do not support this. Findings on beta power in ASD are inconsistent with some meta-analyses reporting both non-significant differences and increased power 13 – 15 . We suggested that fine-grained RS-EEG features can distinguish TD from autistic participants and, to a degree, between language-based subgroups within the autism spectrum. Our binary classification of ASD versus TD achieved a statistically significant accuracy of 69.2% ( p < 0.0014), establishing that EEG signatures can differentiate these groups. The more complex multi-class model also performed significantly above chance, achieving 44.6% accuracy in differentiating the four language profiles ( p < 0.0001). Despite this statistical significance, the practical utility of these models must be carefully considered. The confusion matrices reveal challenges that preclude immediate clinical application; for instance, the binary classifier was more sensitive in identifying ASD participants than it was specific in identifying TD controls. Furthermore, the multi-class model showed inconsistent performance, struggling to accurately classify certain ASD subgroups, particularly the MV profile. Consequently, while these findings provide a robust proof-of-concept for the utility of RS-EEG in characterizing neurodevelopmental heterogeneity, the current models are not yet suitable for individual-level diagnostics. Future studies are needed to explore the full RS-EEG power spectrum in ASD, by using larger samples that span early childhood to adulthood and include varying levels of functioning. As expected, autistic children showed increased gamma compared to TD peers. Autistic children with LU exhibited increased gamma power than TD children (Figure S5). However, LU trajectories were more similar to TD than those of LI and MV children, who showed more divergent patterns (Figure S5). These results align with previous frontal EEG studies on autistic children 20 , and on HL-infants later diagnosed with ASD 18 , and our whole-brain approach extends their findings. Furthermore, since gamma oscillations might reflect the balance between E/I (or glutamatergic/GABAergic) systems, globally reduced gamma power may indicate compensatory mechanisms to early aberrant neurocircuitry, potentially supporting language development in young autistic children and HL-infant siblings later diagnosed 18 – 20 , 49 , 50 . Future research could test this hypothesis by investigating aperiodic power, as it may reflect the E/I balance in the brain 19 . Moreover, it could be insightful to combine EEG with neurochemical imaging techniques (e.g., Magnetic Resonance Spectroscopy) to directly assess the role of glutamatergic and GABAergic systems as compensatory processes in language development. Taken together, our results highlight the relevance of gamma oscillations to language abilities in ASD. Our analysis of word combination acquisition indeed revealed a complex developmental pattern. Gamma power in ASD increased prior to this milestone, peaked near acquisition, then declined. This trajectory suggests that increased gamma power, reflecting heightened neural activity 49 – 51 , may support the emergence of phrase speech in ASD. The decline of gamma power, reflecting diminished excitation, could suggest greater language and cognitive processing, as proposed previously 18 , 20 , 50 , 51 . To our knowledge, our study is the first to link RS-EEG gamma power to word combination. This finding needs replication in autistic children who reach this milestone after age 5 52,53 , as well as in children with developmental language disorder and TD children. In our sample, early acquisition of this milestone by TD children limited opportunities for comparison. Nonetheless, this finding may reflect neural mechanisms underlying both autistic language trajectories and the transition from single words to phrase speech, a critical step toward functional language 33 . Collecting and analyzing neural signals in typical and atypical early development presents significant challenges due to the dynamic interplay of age, divergent developmental trajectories and clinical heterogeneity. Our study reflects a considerable effort to carefully disentangle these factors through a robust longitudinal design, enabling RS-EEG analysis across distinct autistic language profiles. However, several limitations should be acknowledged. While our adapted RS-EEG paradigm (recorded during a non-silent cartoon) was necessary to ensure engagement and obtain good-quality data from young participants, it differs from typical resting-state protocols involving silent videos. While necessary for participant engagement and data quality, our approach deviates from standard RS-EEG protocols and should be considered when interpreting the results. Second, the MV group was relatively small, which highlights the need for specific habituation procedures for RS-EEG acquisition 43 . Considering the variability within MV children, especially in non-verbal and receptive language skills 12 , a larger MV group could provide a more detailed understanding of gamma power differences. It will also be important to follow MV children into school-age, as some may develop phrase speech later 12 . As our longitudinal cohort was recently extended to school-age 54 , we plan to address this question in future research. Finally, although RS-EEG features-based classification showed some promise for distinguishing clinical groups, our findings remain preliminary and are not yet clinically applicable. In conclusion, our longitudinal investigation of 122 autistic and 66 TD young children, provides evidence of distinct RS-EEG power trajectories, partially supporting the U-shaped spectral profile in autism 13 . Examining three validated autistic language profiles (LU, LI, MV) represents a significant strength of our study, revealing how gamma power trajectories vary within ASD and closely align with language abilities, positioning gamma power as a potential marker of language heterogeneity. Gamma power followed a quadratic trajectory around word combination acquisition, indicating a compensatory neural mechanism 18 , 20 facilitating the transition to phrase speech. As phrase speech is a critical milestone toward functional language and positive outcomes for autistic individuals (e.g., improved quality of life and independence) 3 , 33 , 55 , early language development should remain a priority in intervention efforts. Further understanding of the underlying neural mechanisms may further inform the development of effective early intervention approaches. Declarations Conflicts of interest The authors declare no competing interests. Author contributions Conceptualization, K.L., M.G., V.B., and M.S.; methodology, K.L., M.G., A.F., V.B., M.S; formal analysis, K.L., M.G., A.F., F.J.; writing—original draft preparation, K.L.; writing— review and editing, K.L., M.G., A.F., F.J., V.B., and M.S.; supervision, V.B., M.S.; funding acquisition, M.S. All authors have read and agreed to the published version of the manuscript. Acknowledgments The authors would like to thank all the families who participated in the study, as well as the many collaborators who contributed to data collection over the years, namely Alexandra Bastos, Stéphanie Baudoux, Lylia Ben Hadid, Aurélie Bochet, Gaia Brenner, Lucia Cantonas, Léa Chambaz, Flore Couty, Despoina Demenega, Sophie Diakonoff, Lisa Esposito, Constance Ferrat, Margot Giraud, Marie-Agnès Graf, Oriane Grosvernier, Pamela Iraci, Reem Jan, Nada Kojovic, Sara Maglio, Matthieu Mansion, Eva Micol, Priska Müller, Irène Pittet, Sonia Richetin, Tonia Rihs, François Robain, Laura Sallin, Stefania Solazzo, Myriam Speller, Holger Sperdin, Niveettha Thillainathan, Chiara Usuelli, and Ornella Vico Begara. References American Psychiatric Association (2013) Diagnostic and Statistical Manual of Mental Disorders. (American Psychiatric Association. 10.1176/appi.books.9780890425596 Gillespie-Lynch K et al (2012) Early Childhood Predictors of the Social Competence of Adults with Autism. J Autism Dev Disord 42:161–174 Howlin P, Goode S, Hutton J, Rutter M (2004) Adult outcome for children with autism. J Child Psychol Psychiatry 45:212–229 De Vries M, Geurts H (2015) Influence of Autism Traits and Executive Functioning on Quality of Life in Children with an Autism Spectrum Disorder. J Autism Dev Disord 45:2734–2743 Schaeffer J et al (2023) Language in autism: domains, profiles and co-occurring conditions. J Neural Transm 130:433–457 Tager-Flusberg H (2006) Defining language phenotypes in autism. Clin Neurosci Res 6:219–224 Latrèche K et al (2024) Early trajectories and moderators of autistic language profiles: A longitudinal study in preschoolers. Autism 13623613241253015. 10.1177/13623613241253015 Harrison JE, Weber S, Jakob R, Chute CG (2021) ICD-11: an international classification of diseases for the twenty-first century. BMC Med Inf Decis Mak 21:206 World Health Organization. International Classification of Diseases (ICD)-11 (2022) Lombardo MV et al (2015) Different Functional Neural Substrates for Good and Poor Language Outcome in Autism. Neuron 86:567–577 Tager-Flusberg H (2016) Risk Factors Associated With Language in Autism Spectrum Disorder: Clues to Underlying Mechanisms. J Speech Lang Hear Res 59:143–154 Tager-Flusberg H, Kasari C (2013) Minimally Verbal School‐Aged Children with Autism Spectrum Disorder: The Neglected End of the Spectrum. Autism Res 6:468–478 Wang J et al (2013) Resting state EEG abnormalities in autism spectrum disorders. J Neurodev Disord 5 Dede AJO, Xiao W, Vaci N, Cohen MX, Milne E (2025) Exploring EEG resting state differences in autism: sparse findings from a large cohort. Mol Autism 16:13 Neo WS, Foti D, Keehn B, Kelleher B (2023) Resting-state EEG power differences in autism spectrum disorder: a systematic review and meta-analysis. Transl Psychiatry 13:389 Vetri L et al (2022) A Preliminary Study on Photic Driving in the Electroencephalogram of Children with Autism across a Wide Cognitive and Behavioral Range. J Clin Med 11:3568 Cohenour T, Dickinson A, Jeste S, Gulsrud A, Kasari C (2024) Patterns of spontaneous neural activity associated with social communication abilities among infants and toddlers showing signs of autism. Eur J Neurosci 60:3597–3613 Wilkinson CL et al (2020) Use of Longitudinal EEG Measures in Estimating Language Development in Infants With and Without Familial Risk for Autism Spectrum Disorder. Neurobiol Lang 1:33–53 Wilkinson CL, Chung H, Dave A, Tager-Flusberg H, Nelson CA (2025) Changes in Early Aperiodic EEG Activity Are Linked to Autism Diagnosis and Language Development in Infants With Family History of Autism. Autism Res 18:1356–1368 Mukerji CE et al (2024) Resting Frontal Gamma Power is Associated with Both Expressive Language and Non-verbal Cognitive Abilities in Young Autistic Children. J Autism Dev Disord. 10.1007/s10803-024-06308-3 Lord C Autism Diagnostic Observation Schedule– 2nd Edition (ADOS-2). Lord , C., Rutter, M., DiLavore, P., Risi, S., Gotham, K., & Bishop, S. ((2012)). (2012)). (2012) Lord C, Rutter M, Le Couteur A (1994) Autism Diagnostic Interview-Revised: A revised version of a diagnostic interview for caregivers of individuals with possible pervasive developmental disorders. J Autism Dev Disord 24:659–685 Mullen EM (1995) Mullen Scales of Early Learning. American Guidance Service, Circle Pines, MN Schopler E, Lansing MD, Reichler RJ, Marcus LM (2005) PEP-3, Psychoeducational Profile Gotham K, Pickles A, Lord C (2009) Standardizing ADOS Scores for a Measure of Severity in Autism Spectrum Disorders. J Autism Dev Disord 39:693–705 Hus V, Gotham K, Lord C, Standardizing ADOS (2014) Domain Scores: Separating Severity of Social Affect and Restricted and Repetitive Behaviors. J Autism Dev Disord 44:2400–2412 Lord C et al (2006) Autism From 2 to 9 Years of Age. Arch Gen Psychiatry 63:694 Shen MD et al (2013) Early brain enlargement and elevated extra-axial fluid in infants who develop autism spectrum disorder. Brain 136:2825–2835 Wechsler D (2014) Wechsler Preschool and Primary Scale of Intelligence (4th Ed.). Svensk Version. NCS Pearson, Incorporated Bassano D, Labrell F, Champaud C, Lemétayer F, Bonnet P (2005) Le DLPF: un nouvel outil pour l’évaluation du développement du langage de production en français. Enfance 57:171 Bassano D, Labrell F, Bonnet P (2020) DLPF- Guide de cotation et scores. Questionnaire sur le ‘Développement du Langage de Production en Français’: Vocabulaire, Grammaire, Communication Kwok EYL, Brown HM, Smyth RE, Oram Cardy J (2015) Meta-analysis of receptive and expressive language skills in autism spectrum disorder. Res Autism Spectr Disord 9:202–222 Byrne K, Sterrett K, Lord C (2024) Examining the Transition from Single Words to Phrase Speech in Children with ASD: A Systematic Review. Clin Child Fam Psychol Rev 27:1031–1053 Bell MA, Cuevas K (2012) Using EEG to Study Cognitive Development: Issues and Practices. J Cogn Dev 13:281–294 Fló A, Gennari G, Benjamin L, Dehaene-Lambertz G (2022) Automated Pipeline for Infants Continuous EEG (APICE): A flexible pipeline for developmental cognitive studies. Dev Cogn Neurosci 54:101077 Delorme A, Makeig S (2004) EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. J Neurosci Methods 134:9–21 Prerau MJ, Brown RE, Bianchi MT, Ellenbogen JM, Purdon PL (2017) Sleep Neurophysiological Dynamics Through the Lens of Multitaper Spectral Analysis. Physiology 32:60–92 Mutlu AK et al (2013) Sex differences in thickness, and folding developments throughout the cortex. NeuroImage 82:200–207 Pittet I Trajectories of imitation skills in preschoolers with Autism Spectrum Disorders Pedregosa F et al Scikit-learn: Machine Learning in Python. Mach Learn PYTHON Slifer KJ, Avis KT, Frutchey RA (2008) Behavioral intervention to increase compliance with electroencephalographic procedures in children with developmental disabilities. Epilepsy Behav 13:189–195 Webb SJ et al (2015) Guidelines and Best Practices for Electrophysiological Data Collection, Analysis and Reporting in Autism. J Autism Dev Disord 45:425–443 Tager-Flusberg H et al (2017) Conducting research with minimally verbal participants with autism spectrum disorder. Autism 21:852–861 Chan AS, Sze SL, Cheung M (2007) Quantitative electroencephalographic profiles for children with autistic spectrum disorder. Neuropsychology 21:74–81 Stroganova TA et al (2007) Abnormal EEG lateralization in boys with autism. Clin Neurophysiol 118:1842–1854 Coben R, Clarke AR, Hudspeth W, Barry R (2008) J. EEG power and coherence in autistic spectrum disorder. Clin Neurophysiol 119:1002–1009 Murias M, Webb SJ, Greenson J, Dawson G (2007) Resting State Cortical Connectivity Reflected in EEG Coherence in Individuals With Autism. Biol Psychiatry 62:270–273 Neuhaus E et al (2021) Resting state EEG in youth with ASD: age, sex, and relation to phenotype. J Neurodev Disord 13:33 Dickinson A, Jones M, Milne E (2016) Measuring neural excitation and inhibition in autism: Different approaches, different findings and different interpretations. Brain Res 1648:277–289 Sohal VS, Rubenstein JL (2019) R. Excitation-inhibition balance as a framework for investigating mechanisms in neuropsychiatric disorders. Mol Psychiatry 24:1248–1257 Coghlan S et al (2012) GABA system dysfunction in autism and related disorders: From synapse to symptoms. Neurosci Biobehav Rev 36:2044–2055 Broome K, McCabe P, Docking K, Doble M, Carrigg B (2023) Speech Development Across Subgroups of Autistic Children: A Longitudinal Study. J Autism Dev Disord 53:2570–2586 Wodka EL, Mathy P, Kalb L (2013) Predictors of Phrase and Fluent Speech in Children With Autism and Severe Language Delay. Pediatrics 131:e1128–e1134 Bochet A (2022) Capturing part of the heterogeneity along the Autism Spectrum Disorders by focusing on the co-occurrence of Attention Deficit/Hyperactivity Disorder. 10.13097/ARCHIVE-OUVERTE/UNIGE:158714 Moss P, Mandy W, Howlin P (2017) Child and Adult Factors Related to Quality of Life in Adults with Autism. J Autism Dev Disord 47:1830–1837 Additional Declarations The authors declare no competing interests. Supplementary Files FinalSupplementaryMaterial.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7204586","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":490259956,"identity":"e4d57120-e276-4110-8b04-7ead3f1c152d","order_by":0,"name":"Kenza Latrèche","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABA0lEQVRIiWNgGAWjYBACCQhlASaZGRhswCQQHCCkRYKBB6IljXQth2ESuLVItvce/vCDQULeXvrwwc8FFecTt7OzX3zAUHMHpxZpnnMJhj0MEoY9fGnJ0jPO3E7c2cxTbMBw7BlOLXISOQYJPAwSjD08PGbMvG23Ezcc5kmTYGw4jFuL/BuDg38YJOx7ePi/MfP+OwfSkv4DnxZpCR7DZqAtiUBb2Jh5Gw4AtbAfY8CnRbInx5hZxkAiuecMm7E0z7FkY6AtzBIJx3BrkTh+xvjjmwob2/Ye5oefeWrsZDecP/7ww4ca3FogwACFx2PAkEBAAzpgf0CihlEwCkbBKBjmAACz/k5uu2A2fwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0003-0264-4048","institution":"University of Geneva","correspondingAuthor":true,"prefix":"","firstName":"Kenza","middleName":"","lastName":"Latrèche","suffix":""},{"id":490259957,"identity":"6497bae5-0351-46bb-8532-cdbff99c3612","order_by":1,"name":"Michel Godel","email":"","orcid":"https://orcid.org/0000-0002-9789-3540","institution":"University of Geneva","correspondingAuthor":false,"prefix":"","firstName":"Michel","middleName":"","lastName":"Godel","suffix":""},{"id":490259958,"identity":"68eef73a-1509-4caf-a169-6d5ca1a4d718","order_by":2,"name":"Ana Flò","email":"","orcid":"https://orcid.org/0000-0002-3260-0559","institution":"University of Padua; NeuroSpin center","correspondingAuthor":false,"prefix":"","firstName":"Ana","middleName":"","lastName":"Flò","suffix":""},{"id":490259959,"identity":"343efbe5-6d0e-4e2b-869f-7f97e24804ce","order_by":3,"name":"Fiona Journal","email":"","orcid":"https://orcid.org/0009-0001-8269-3765","institution":"University of Geneva","correspondingAuthor":false,"prefix":"","firstName":"Fiona","middleName":"","lastName":"Journal","suffix":""},{"id":490259960,"identity":"b3ac0bdc-ce23-4194-b860-88222544591a","order_by":4,"name":"Valentina Borghesani","email":"","orcid":"https://orcid.org/0000-0002-7909-8631","institution":"University of Geneva","correspondingAuthor":false,"prefix":"","firstName":"Valentina","middleName":"","lastName":"Borghesani","suffix":""},{"id":490259961,"identity":"454ca0bb-6b08-4025-9c9d-c973fdce3563","order_by":5,"name":"Marie Schaer","email":"","orcid":"https://orcid.org/0000-0001-8479-6365","institution":"University of Geneva","correspondingAuthor":false,"prefix":"","firstName":"Marie","middleName":"","lastName":"Schaer","suffix":""}],"badges":[],"createdAt":"2025-07-24 10:32:39","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-7204586/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7204586/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":87711161,"identity":"e4269d24-1942-490a-acd1-acd90b7c8112","added_by":"auto","created_at":"2025-07-28 08:41:15","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1027152,"visible":true,"origin":"","legend":"\u003cp\u003eWhole-brain resting-state EEG power trajectories of TD and ASD young children over the five canonical frequency bands. \u003cstrong\u003eA.\u003c/strong\u003e Delta (2-4Hz) \u003cstrong\u003eB.\u003c/strong\u003e Theta (4-6Hz) \u003cstrong\u003eC.\u003c/strong\u003eAlpha (6-13Hz) \u003cstrong\u003eD.\u003c/strong\u003e Beta (13-30Hz). \u003cstrong\u003eE.\u003c/strong\u003e Gamma (30-50Hz).\u003c/p\u003e\n\u003cp\u003eThe colored bands around the estimated group-level trajectory indicate the 95% confidence interval. TD: typical development ; ASD: autism spectrum disorder.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-7204586/v1/faf87bbeee326d9deea41141.png"},{"id":87710269,"identity":"ebd9fa5e-565e-4313-8e2d-fd548c668ddb","added_by":"auto","created_at":"2025-07-28 08:33:15","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1021044,"visible":true,"origin":"","legend":"\u003cp\u003eWhole-brain resting-state EEG power trajectories of TD and of the three ASD language profiles over the five canonical frequency bands. \u003cstrong\u003eA.\u003c/strong\u003e Delta (2-4Hz) \u003cstrong\u003eB.\u003c/strong\u003e Theta (4-6Hz) \u003cstrong\u003eC.\u003c/strong\u003eAlpha (6-13Hz) \u003cstrong\u003eD.\u003c/strong\u003e Beta (13-30Hz). \u003cstrong\u003eE.\u003c/strong\u003e Gamma (30-50Hz).\u003c/p\u003e\n\u003cp\u003eThe colored bands around the estimated group-level trajectory indicate the 95% confidence interval. TD: typical development; LU: language unimpaired; LI: language impaired; MV: minimally-verbal.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-7204586/v1/2193278d730d516d6279a5b5.png"},{"id":87711163,"identity":"1aa5d322-1f1f-4e48-ac6a-fe96a92e9ffe","added_by":"auto","created_at":"2025-07-28 08:41:15","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":271945,"visible":true,"origin":"","legend":"\u003cp\u003eTrajectory of whole-brain resting-state EEG gamma power trajectory of the ASD group (N=68, 168 time points), aligned with the age of word combination acquisition. The x-axis represents time relative to individual word combination acquisition (in years). Age of word combination acquisition, or time 0, is marked by the dashed vertical line. The solid black line shows the estimated group-level trajectory, and the shaded area represents the 95% confidence interval. The yellow arrow indicates the peak of the curve, which was calculated at 0.384 years, i.e., approximately 4 months post-word combination acquisition. Whole-brain gamma power then decreases after acquisition of this milestone.\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-7204586/v1/8ad7405706681db71e6b872a.png"},{"id":87713938,"identity":"1d8f89ef-1bfa-4867-9f4e-97f02d8916a4","added_by":"auto","created_at":"2025-07-28 08:57:17","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2472981,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7204586/v1/cca68147-1397-44b6-9241-621e85043f27.pdf"},{"id":87710268,"identity":"53921a43-9663-458b-8402-f9c54f89a710","added_by":"auto","created_at":"2025-07-28 08:33:15","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":5662468,"visible":true,"origin":"","legend":"","description":"","filename":"FinalSupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-7204586/v1/0de9f313177d6b3e05b03a1e.docx"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eEarly Trajectories of Resting-State EEG Power in Autistic Children: A Longitudinal Study Across Language Profiles\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAutism Spectrum Disorder (ASD) is a prevalent neurodevelopmental condition characterized by difficulties in social communication and interactions, as well as repetitive and restrictive behaviors or interests\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. While outcomes across the spectrum vary widely, early language milestones (e.g., age of first spoken words), are one of the strongest predictors of long-term development\u003csup\u003e\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. However, language development in autism ranges from significant difficulties to subtle differences\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e,\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. In previous work, we aimed to capture language heterogeneity among 215 autistic children (aged 1.5\u0026ndash;5.7 years old)\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Using a data-driven approach, we identified three distinct clusters that corresponded to the autistic language profiles described in the 11th International Classification of Diseases, namely Minimally-Verbal (MV), Language Impaired (LI), and Language Unimpaired (LU)\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. In our previous study, MV children exhibited poor vocabulary and limited use of word combinations\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. The LI group showed delayed vocabulary development and reduced use of word combinations, while the LU group demonstrated vocabulary size and word combinations comparable to TD peers\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eUnderstanding the neural correlates underlying these distinct language trajectories is crucial for predicting outcomes and developing targeted interventions, particularly for children with greater language difficulties\u003csup\u003e\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Electroencephalography (EEG) has emerged as a powerful tool in research on autistic children, given its high temporal resolution and non-invasiveness\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Resting-state EEG (RS-EEG) paradigms are particularly valuable as they require no engagement in a cognitive task\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. RS-EEG activity is frequently investigated with spectral analysis, where brain oscillations are decomposed and power is quantified over five canonical frequency bands: delta (2\u0026ndash;4 Hz), theta (4\u0026ndash;6 Hz), alpha (6\u0026ndash;13 Hz), beta (13\u0026ndash;30 Hz), and gamma (30\u0026ndash;50 Hz)\u003csup\u003e\u003cspan additionalcitationids=\"CR14\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. Prior evidence suggested a U-shaped profile in ASD, with increased power for low frequencies (delta, theta) and high frequencies (beta, gamma) and decreased power for middle frequencies (alpha) compared to TD \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. Nevertheless, reliable RS-EEG biomarkers of ASD have yet to be identified, due to limitations in previous studies including small sample sizes (on average 30 autistic and 35 TD participants\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e), cross-sectional designs that prevent examination of developmental processes, and analyses limited to specific electrodes and/or frequency bands\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eSome previous studies examined the relationship between RS-EEG and language development in infants at high- and low-likelihood (HL; LL) for autism\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e,\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. In a cross-sectional study on 70 HL infants aged 12\u0026ndash;23 months, Cohenour and colleagues reported positive correlations between alpha power and receptive and expressive language skills, suggesting that alpha oscillations support language production and comprehension\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. In a longitudinal study in 72 HL and 58 LL infants, increased 6-month frontal gamma power (FGP) was significantly associated with higher 24-month language functioning in HL infants without ASD (HL-NoASD), but with lower 24-month language functioning in HL infants later diagnosed with ASD (HL-ASD)\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. As gamma oscillations reflect the brain\u0026rsquo;s balance between excitatory/inhibitory (E/I) neurotransmitters, reduced FGP in HL-ASD could represent successful compensation for underlying neurotransmitter imbalances. In contrast, reduced FGP in HL-NoASD could indicate delayed neural maturation and language abilities\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. To more accurately assess the brain\u0026rsquo;s E/I balance, Wilkinson and colleagues analyzed aperiodic power in a subsequent longitudinal study of 3- and 12-month-old HL and LL infants\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. They found greater increases in aperiodic power in HL-ASD infants, along with reduced language development at 18 months, suggesting that early E/I imbalance may underlie later language difficulties in HL-ASD infants. Furthermore, a cross-sectional study on preschoolers (48 ASD, 58 TD) showed decreased FGP associated with greater expressive language skills in autistic children\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e, supporting that reduced gamma power may reflect a compensation mechanism for language development\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e,\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eWhile previous research has reported associations between RS-EEG power and language development, no study to our knowledge has examined RS-EEG power longitudinally in autistic children stratified by language profile. Longitudinal investigation is particularly important as RS-EEG power trajectories might change across the early years of life, following non-linear trajectories alongside language development\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. To address these limitations, the present study conducts a longitudinal investigation of whole-brain EEG power in the five canonical frequency bands in three previously validated autistic language profiles (LU, LI, and MV)\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. The longitudinal sample yields 358 time points and includes 122 autistic (61 LU, 44 LI, 17 MV) and 66 typically developing children, aged 1.56 to 6.01 years old. Our study has three aims, (1) to explore longitudinal trajectories of each frequency band across the TD and ASD groups, and within LU, LI, and MV; (2) to examine the trajectory of gamma power in relation to word combination acquisition within the ASD sample; (3) to classify TD and the three autistic profiles based on fine-grained brain activity features, using a supervised machine learning algorithm. We hypothesized that ASD children would demonstrate increased power in low (delta, theta) and high frequency bands (beta, gamma) and decreased power in middle frequencies (alpha), in line with the proposed U-shaped profile\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. We expected the LU, LI, and MV subgroups to differ mainly in the gamma band, reflecting previously reported associations between language skills and higher frequencies in autistic children and HL infants\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e,\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. As decreased gamma power may reflect a compensation mechanism in ASD, MV would exhibit the greatest gamma power, followed by LI, and finally LU. Considering that production of word combinations greatly differed across each language profile\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e, we evaluated gamma power trajectory jointly with this milestone.\u003c/p\u003e"},{"header":"Material and methods","content":"\u003cp\u003eParticipants\u003c/p\u003e\u003cp\u003eOur sample comes from the Geneva Autism Cohort, which is an ongoing open longitudinal cohort that follows autistic and typically developing (TD) preschoolers\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. In this cohort, we include children who received a diagnosis of ASD before age 5, and their typically developing peers, and we follow them every 6 months for two years. Each visit comprises several longitudinal assessments, from behavioral measures to neuroimaging. Autistic participants are recruited through local clinical centers and parent associations. A licensed child psychiatrist (MS) confirmed the diagnosis of all autistic participants based on the \u003cem\u003eDiagnostic and Statistical Manual of Mental Disorders \u0026minus;\u003c/em\u003e\u0026thinsp;5th ed.\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e and gold-standard tools (Autism Diagnostic Observation Schedule\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e; Autism Diagnostic Interview-Revised\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e). Inclusion criteria for TD participants included full-term birth, no ASD diagnosis in first-degree relatives, and no neurological or somatic concerns. TD participants completed the ADOS\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e to exclude ASD symptoms and a developmental assessment (either the Mullen Scales of Early Learning, MSEL\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e or the Psychoeducational Profile \u0026ndash; third edition, PEP-3\u003csup\u003e24\u003c/sup\u003e) to rule out significant developmental delays (i.e., a developmental quotient below 70 in any domain). Written informed consent forms were signed and provided by the participants\u0026rsquo; caregivers. The Ethics Committee of the University of Geneva approved the research protocol.\u003c/p\u003e\u003cp\u003eThe present study builds upon our previous work\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e, where we subgrouped early expressive language skills of autistic preschoolers into three distinct profiles (LU, LI, MV). To understand the profiles\u0026rsquo; underlying neural underpinnings, we included TD and autistic children from our previous work with at least one RS-EEG recording. One to three recordings per child were available for this study. Our final sample comprised 188 participants (358 time points), aged from 1.56 to 6.01 years old. The TD group was composed of 66 participants (130 time points, age range 1.56\u0026ndash;5.84 years, 43.9% of females). The ASD group included 122 participants (228 time points, age range 1.74\u0026ndash;6.01 years old, 12.3% of females), and was divided in the three language profiles subgroups: LU (61 participants, 127 time points, 1.74\u0026ndash;5.99 years, 6.6% females), LI (44 participants, 71 time points, 1.93\u0026ndash;5.82 years old, 11.4% females), and MV (17 participants, 30 time points, 1.80\u0026ndash;6.01 years old, 35.3% females). Sample characteristics are detailed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eSample characteristics of the TD and LU, LI, and MV samples\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMeasure (mean (\u003cem\u003eSD\u003c/em\u003e))\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTD\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eASD\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eASD-LU\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eASD-LI\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eASD-MV\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNumber of participants\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e66\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e122\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e61\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e44\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e17\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNumber of time points\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e130\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e228\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e127\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e71\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e30\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTime points per participants\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2.0 (\u003cem\u003e0.8\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.9 (\u003cem\u003e0.8\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2.1 (\u003cem\u003e0.8\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.6 (\u003cem\u003e0.7\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e1.8 (\u003cem\u003e0.8\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMean age at first available TP (years)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.3 (\u003cem\u003e1.1\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.5 (\u003cem\u003e1.0\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3.4 (\u003cem\u003e0.9\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e3.8 (\u003cem\u003e1.1\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e3.2 (\u003cem\u003e1.3\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge range (years)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.6\u0026ndash;5.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.7-6.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1.7-6.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.9\u0026ndash;5.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e1.8-6.0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale biological sex\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e29 (43.9%)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e15 (12.3%)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4 (6.6%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e5 (11.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e6 (35.3%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMultilingual environment [N\u0026thinsp;=\u0026thinsp;188]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e24 (36.4%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e57 (46.7%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e27 (44.3%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e18 (40.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e12 (70.6%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eParental education [N\u0026thinsp;=\u0026thinsp;188]\u003c/p\u003e\u003cp\u003e(% college degree completed)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e62 (93.9%)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e65 (53.7%)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e28 (45.9%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e24 (54.5%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e9 (52.7%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMean ADOS CSS Total at first available TP [N\u0026thinsp;=\u0026thinsp;188]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.1 (\u003cem\u003e0.3\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e7.3 (\u003cem\u003e1.8\u003c/em\u003e)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e6.7 (\u003cem\u003e1.6\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e7.5 (\u003cem\u003e2.0\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e8.7 (\u003cem\u003e1.4\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMean ADOS CSS SA at first available TP[N\u0026thinsp;=\u0026thinsp;188]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1.1 (\u003cem\u003e0.3\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e6.3 (\u003cem\u003e2.1\u003c/em\u003e)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e5.8 (\u003cem\u003e1.9\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e6.4 (2\u003cem\u003e.4\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e7.6 (\u003cem\u003e1.7\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMean ADOS CSS RRB at first available TP [N\u0026thinsp;=\u0026thinsp;188]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2.6 (\u003cem\u003e2.1\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e8.6 (\u003cem\u003e1.6\u003c/em\u003e)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e8.4 (\u003cem\u003e1.8\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e8.7 (\u003cem\u003e1.5\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e9.3 (\u003cem\u003e1.1\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMean Total DQ [TP\u0026thinsp;=\u0026thinsp;356]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e114.8 (\u003cem\u003e12.5\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e79.14 (\u003cem\u003e25.6\u003c/em\u003e)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e94.5 (\u003cem\u003e15.7\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e65.7 (\u003cem\u003e13.4\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e36.6 (\u003cem\u003e11.5\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMean EL DQ [TP\u0026thinsp;=\u0026thinsp;351]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e107.8 (\u003cem\u003e18.2\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e66.6 (\u003cem\u003e27.3\u003c/em\u003e)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e84.3 (\u003cem\u003e18.4\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e53.8 (\u003cem\u003e15.0\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e23.8 (\u003cem\u003e9.9\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMean RL DQ [TP\u0026thinsp;=\u0026thinsp;351]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e118.3 (\u003cem\u003e14.3\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e72.9 (\u003cem\u003e32.1\u003c/em\u003e)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e93.6 (\u003cem\u003e20.3\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e58.5 (\u003cem\u003e21.0\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e21.6 (\u003cem\u003e9.3\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMean VR DQ [TP\u0026thinsp;=\u0026thinsp;351]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e123.3 (\u003cem\u003e20.5\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e88.3 (\u003cem\u003e27.5\u003c/em\u003e)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e104.4 (\u003cem\u003e20.3\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e77.1 (\u003cem\u003e17.0\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e48.6 (\u003cem\u003e17.8\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMean FM DQ [TP\u0026thinsp;=\u0026thinsp;351]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e107.9 (\u003cem\u003e13.8\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e81.1 (\u003cem\u003e22.2\u003c/em\u003e)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e94.0 (\u003cem\u003e17.0\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e73.4 (\u003cem\u003e15.4\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e52.2 (\u003cem\u003e16.8\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDLPF Total Vocabulary Score [TP\u0026thinsp;=\u0026thinsp;322]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1104.0 (\u003cem\u003e412.3\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e641.9 (\u003cem\u003e503.2\u003c/em\u003e)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e852.6 (\u003cem\u003e478.3\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e454.7 (\u003cem\u003e355.7\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e24.4 (\u003cem\u003e69.8\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDLPF Total Grammar Score [TP\u0026thinsp;=\u0026thinsp;323]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e79.6 (\u003cem\u003e29.9\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e42.1 (\u003cem\u003e36.9\u003c/em\u003e)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e57.7 (\u003cem\u003e36.6\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e26.7 (\u003cem\u003e22.8\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.79 (\u003cem\u003e2.6\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDLPF Total Pragmatics Score [TP\u0026thinsp;=\u0026thinsp;314]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e45.2 (\u003cem\u003e6.3\u003c/em\u003e)\u003csup\u003e\u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e30.5 (\u003cem\u003e16.8\u003c/em\u003e)\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e38.4 (\u003cem\u003e12.2\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e26.8 (\u003cem\u003e13.4\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e2.4 (\u003cem\u003e4.3\u003c/em\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eProbability values refer to Kruskal\u0026ndash;Wallis, Mann-Whitney or Pearson chi-square test, as appropriate.\u003csup\u003e\u0026dagger;\u003c/sup\u003e \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.05 vs. TD; \u003csup\u003e\u0026Dagger;\u003c/sup\u003e \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.05 across TD, LU, LI, MV.\u003c/p\u003e\u003cp\u003eADOS\u0026thinsp;=\u0026thinsp;Autism Diagnostic Observation Schedule; ASD\u0026thinsp;=\u0026thinsp;Autism spectrum disorder; ASD-LU\u0026thinsp;=\u0026thinsp;Autism spectrum disorder-language unimpaired; ASD-LI\u0026thinsp;=\u0026thinsp;Autism spectrum disorder-language impaired; ASD-MV\u0026thinsp;=\u0026thinsp;Autism spectrum disorder-minimally verbal; CSS\u0026thinsp;=\u0026thinsp;Calibrated Severity Score; DLPF\u0026thinsp;=\u0026thinsp;\u003cem\u003eD\u0026eacute;veloppement du Langage Productif en Fran\u0026ccedil;ais\u003c/em\u003e; DQ\u0026thinsp;=\u0026thinsp;Developmental Quotient; EL\u0026thinsp;=\u0026thinsp;Expressive Language; FM\u0026thinsp;=\u0026thinsp;Fine Motor; RL\u0026thinsp;=\u0026thinsp;Receptive Language; RRB\u0026thinsp;=\u0026thinsp;Restricted and Repetitive Behaviors; SA\u0026thinsp;=\u0026thinsp;Social Affect; TD\u0026thinsp;=\u0026thinsp;Typical Development; TP\u0026thinsp;=\u0026thinsp;time point; VR\u0026thinsp;=\u0026thinsp;Visual Reception.\u003c/p\u003e\u003cp\u003eWe used the G*Power software to conduct a power analysis of our sample. With our TD group (N\u0026thinsp;=\u0026thinsp;66) and ASD group (N\u0026thinsp;=\u0026thinsp;122), the study is well-powered to detect both large (Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;0.8) and medium (Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;0.5) effect sizes. Power was nearly 100% for a large effect and 90.6% for a medium effect, indicating sufficient sample sizes to detect meaningful differences across groups. Detecting a small effect (Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;0.2) would require 394 participants per group to achieve a power of 80%. Of note, this analysis does not account for the repeated measures within participants. As no clear consensus exists on power analysis for unbalanced, nested longitudinal designs, we report cross-sectional power estimates for informational purposes, acknowledging the limitations of this approach.\u003c/p\u003e\u003cp\u003eMeasures\u003c/p\u003e\u003cp\u003eTo characterize our longitudinal cohort, behavioral and developmental measures were administered to complement the neurophysiological data.\u003c/p\u003e\u003cp\u003eBehavioral measures\u003c/p\u003e\u003cp\u003eWe administered the ADOS-2\u003csup\u003e21\u003c/sup\u003e to confirm ASD in autistic participants and to exclude ASD symptoms in TD participants. The ADOS-2 is a semi-structured assessment quantifying symptoms in social affect and restricted and repetitive behaviors or interests. The ADOS-2 includes five modules, administered depending on the child\u0026rsquo;s age and language level. Modules were compared using calibrated total severity score\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e. The ADOS-2 was administered and coded by trained examiners.\u003c/p\u003e\u003cp\u003eDevelopmental level was measured using the MSEL\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e, assessing fine motor (FM), visual reception (VR), receptive language (RL), and expressive language (EL) in children aged 0 to 68 months. We computed developmental quotients (DQs) by dividing the chronological age by the developmental age and multiplying by 100\u003csup\u003e7,27,28\u003c/sup\u003e. DQs were selected over standard scores to reduce floor-effects in lower-performing participants. As the MSEL was added later in our protocol, 23 participants (32 time points) were assessed with the PEP-3\u003csup\u003e24\u003c/sup\u003e or with the WPPSI-IV\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. For the PEP-3, designed for children aged 2 to 7 years, we computed DQs for domains. The PEP-3 and MSEL showed excellent consistency for both expressive (Cronbach\u0026rsquo;s alpha\u0026thinsp;=\u0026thinsp;0.899) and receptive language (Cronbach\u0026rsquo;s alpha\u0026thinsp;=\u0026thinsp;0.913)\u003csup\u003e7\u003c/sup\u003e. Five children (5 time points) completed the WPPSI-IV, where only the full-scale IQ was used. Two children (1 ASD, 1 TD) had no developmental data at the EEG timepoint: one was too young, the other had behavioral difficulties invalidating the PEP-3.\u003c/p\u003e\u003cp\u003eParent-reported expressive language\u003c/p\u003e\u003cp\u003eExpressive language skills were evaluated with the \u003cem\u003eD\u0026eacute;veloppement du langage de production en fran\u0026ccedil;ais\u003c/em\u003e (DLPF)\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e parent-reported questionnaire. The DLPF assesses vocabulary, grammar, and pragmatic skills. Three total domain scores were computed\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. The DLPF assesses children aged 18 to 42 months, as scores plateaued when approaching 42 months in TD\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. Here, we administered the DLPF until 6 years old given the frequent expressive language delay in ASD\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eWe also examined word combination acquisition, which is a significant milestone toward functional speech\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. We defined word combination as the use of at least two words to communicate (e.g., \u003cem\u003eOpen door\u003c/em\u003e, \u003cem\u003eMore milk\u003c/em\u003e)\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e,\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. We measured the age of word combination acquisition using several tools, depending on what assessment was available, i.e., either the DLPF (item B, Sentences and Grammar Section), the ADI-R (item 10), the MSEL (item 17), and the ADOS-2 (item A1).\u003c/p\u003e\u003cp\u003eRS-EEG data collection, preprocessing, analyses\u003c/p\u003e\u003cp\u003eRS-EEG recordings were acquired with a sampling rate of 1000 Hz using 128-channel Hydrocel Geodesic Sensor Nets and Net Station acquisition software (v 5.4.1.2) (Magstim EGI, Inc. Eugene, OR). To minimize movement artifacts and facilitate compliance, we adapted the RS-EEG paradigm: children watched an age-appropriate cartoon of their choice with sound, sitting\u0026thinsp;~\u0026thinsp;60 cm from the screen, independently or on a caregiver\u0026rsquo;s lap. They were encouraged to sit still and quiet. EEGs were recorded for 5 minutes and extended by 1\u0026ndash;2 minutes if children were initially agitated or talking. We excluded 16 recordings under 300 seconds (5 from TD, 2 from LU, 8 from LI, 1 from MV). For longer recordings, the last 300 seconds were selected to retain the most stable segments.\u003c/p\u003e\u003cp\u003ePreprocessing was conducted using the Automated Pipeline for Infants Continuous EEG (APICE)\u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e, which is based on EEGLAB\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e and is specialized for young children. The data was first low-pass filtered at 100 Hz and high-pass filtered at 1 Hz, and a notch filter was applied to remove line noise (50 Hz). The pop_eegfiltnew function from EEGLAB was used for filtering with the default parameters. Then, APICE was applied. APICE detects non-functional electrodes and motion artifacts based on signal properties, primarily fast signal changes, and repairs the data, when possible, to maximize data recovery. Brief artifacts (shorter than 100 ms) were corrected using PCA, while longer artifacts affecting only a subset of the channels (less than 30%) were corrected using spatial spline interpolation. Channels identified as non-functional during the whole recording were also interpolated using spatial splines. Finally, data was visually inspected to check for undetected corrupted segments, and those segments were excluded from the analysis.\u003c/p\u003e\u003cp\u003eTwo RS-EEG recordings (1 LU, 1 MV) with less than 100 seconds of time free of artifacts were excluded from the power analysis. We performed a multitaper spectral analysis to decompose EEG signal into power within the 2\u0026ndash;50 Hz range over all electrodes. A 5-second time window, spaced at 1-second intervals, was used to balance temporal resolution and spectral stability. Spectral estimates were computed using a time-bandwidth of 2.5 and four orthogonal tapers\u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. We applied equal weighting to all tapers. Mean log(10) whole-brain power was calculated across the five canonical frequency bands: delta (2\u0026ndash;4 Hz), theta (4\u0026ndash;6 Hz), alpha (6\u0026ndash;13 Hz), beta (13\u0026ndash;30 Hz), and gamma (30\u0026ndash;50 Hz)\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eStatistical analyses\u003c/p\u003e\u003cp\u003eLongitudinal analyses\u003c/p\u003e\u003cp\u003eWe conducted mixed model analyses to examine developmental, linguistic, and RS-EEG power trajectories across age. Mixed modeling is a successful approach for nested data with varying numbers of time points per participant and is well suited to longitudinal behavioral and neurophysiological research\u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e,\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. Age and group (TD, LU, LI, MV) were included as fixed effects, sex as a covariate, and behavioral and EEG measures as random effects. Analyses were conducted using the myMixedModelsTrajectories toolbox in MATLAB\u0026reg; R2019b (MathWorks, Natick, MA), fitting linear and quadratic random-slope models to capture different age-related patterns. Behavioral outcomes included Total DQ, EL DQ, RL DQ, FM DQ, VR DQ, and DLPF scores (Vocabulary, Grammar, Pragmatics). RS-EEG power trajectories were estimated across the five frequency bands, and model fit was evaluated using the Bayesian Information Criterion (BIC).\u003c/p\u003e\u003cp\u003eAs the word combination acquisition was a binary categorical variable (i.e., acquired or not), we performed a chi-square test with a 12-month sliding window comparing the proportion of children acquiring combinations at each age bin and applying FDR correction. Moreover, to examine word combination acquisition in relation to whole-brain gamma power, we applied a mixed model analysis fitting linear, quadratic, and cubic random-slope models. The BIC selected the quadratic fit to model gamma power relative to the estimated timing of word combination acquisition. This analysis was limited to the ASD group, as most TD children (63/66, 95.5%) had acquired combinations by their first time point. To compare gamma power before and after acquisition, we excluded ASD children with only one time point (N\u0026thinsp;=\u0026thinsp;48) or two non-consecutive time points (N\u0026thinsp;=\u0026thinsp;6). The final ASD sample included 68 children (41 LU, 18 LI, 9 MV) with 168 RS-EEG recordings (104 LU, 42 LI, 22 MV) between 1.74 and 5.99 years.\u003c/p\u003e\u003cp\u003eCross-sectional classification analysis\u003c/p\u003e\u003cp\u003eWe conducted two exploratory classification analyses on the cross-sectional sample (N\u0026thinsp;=\u0026thinsp;188; 122 ASD, 66 TD) to determine if resting-state EEG power features could distinguish between diagnostic groups. The first analysis was a binary classification of ASD versus TD participants. The second was a multi-class classification of three autistic language subgroups and the TD group (a total of four groups). For these analyses, we used 6 400 features derived from the RS-EEG data, comprising power values from 128 electrodes across 5 canonical frequency bands (delta, theta, alpha, beta, gamma) and the first ten 10-second artifact-free time bins. We opted to retain discrete time bins per participant instead of using a single average across time as fluctuations in RS neural activity may contain diagnostically relevant information. While there is no synchronized 'time course' across participants, a classifier can leverage these multiple temporal snapshots to learn patterns of intra-individual variability. For instance, the stability or range of power fluctuations across the ten bins may itself be a distinguishing characteristic between groups, information that would be obscured by averaging.\u003c/p\u003e\u003cp\u003eTo perform the classifications, we implemented a supervised machine learning pipeline using scikit-learn in Python\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e. This pipeline integrated several steps to ensure a robust evaluation: data scaling, feature selection, and classification with a linear Support Vector Machine (SVM). The entire pipeline was evaluated using a nested cross-validation approach to prevent information leakage and provide an unbiased estimate of model performance. The outer loop of the nested cross-validation used a 5-fold stratified split to divide the data, preserving the percentage of samples for each class. The inner loop employed a 3-fold stratified cross-validation on the training data from the outer loop to perform hyperparameter tuning using a GridSearchCV. Specifically, we optimized the C parameter of the linear SVM (SVC(kernel='linear')) over a range of values ([0.01, 0.1, 1, 10, 100]). Within this pipeline, we also incorporated a feature selection step using SelectKBest to retain the 500 most informative features based on their ANOVA F-value relative to the diagnostic labels.\u003c/p\u003e\u003cp\u003eTo assess the statistical significance of our classification results, we performed a permutation test. For each classification task (2-group and 4-group), we ran the entire nested cross-validation procedure 10 000 times, each time with the diagnostic labels randomly shuffled. This process generated a null distribution of accuracy scores that would be expected by chance. The \u003cem\u003ep\u003c/em\u003e-value was then calculated as the proportion of permutation-based accuracies that were equal to or greater than the accuracy achieved with the true labels. This rigorous procedure allowed us to determine if the model's performance was statistically significant. For further model interpretation, we also generated a confusion matrix for each classification to visualize the specific patterns of correct and incorrect predictions across the groups. The matrix is generated from the aggregated out-of-sample predictions from each fold of the outer cross-validation loop, which gives a realistic estimate of performance on unseen data. The code of these analyses can be found at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/NoCe-Lab/rsEEG_ASD\u003c/span\u003e\u003cspan address=\"https://github.com/NoCe-Lab/rsEEG_ASD\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eDescription of TD and language profiles\u003c/p\u003e\u003cp\u003eDescriptive analyses of the 66 TD (130 time points) and the 122 ASD (228 time points) participants show no significant age difference at the first available time point (mean age 3.3\u0026thinsp;\u0026plusmn;\u0026thinsp;1.1 for TD and 3.5\u0026thinsp;\u0026plusmn;\u0026thinsp;1.0. for ASD) (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). As there were significantly more TD females (43.9%) than autistic (12.3%), sex was included as a covariate in all analyses. As expected, autistic participants exhibited higher autism symptom severity, and lower non-verbal and verbal skills. Additionally, there were no significant differences in age at the first timepoint across the LI, LU, MV, and TD groups (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05). As female proportions differed across the four groups, sex was included as a covariate. The monolingual \u003cem\u003evs.\u003c/em\u003e multilingual environment did not significantly differ. Parental education was higher in TD than in the ASD subgroups (LU, LI, MV). ASD symptom severity and DQs differed across the four groups.\u003c/p\u003e\u003cp\u003e To examine developmental trajectories across ASD, TD, and the three language profiles (LU, LI, MV), we analyzed longitudinal changes in non-verbal and verbal skills. Additionally, we examined expressive language development with specific measures of their vocabulary, grammar, and pragmatics. Our findings align with our previous work\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e, with distinct verbal and non-verbal trajectories within the three language profiles compared to TD children (Supplementary Analysis 1).\u003c/p\u003e\u003cp\u003eFeasibility of RS-EEG acquisition in TD and ASD samples\u003c/p\u003e\u003cp\u003eDue to anxiety about unfamiliar experiences and/or tactile defensiveness, it is challenging to collect EEG data in young autistic children\u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e,\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e. Habituation strategies have been shown to improve compliance, including in MV children\u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e,\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e. In our study, we provided materials to help children, and their parents familiarize themselves with the EEG, including pictograms and a short story illustrating the EEG room and net, a video demonstrating the setup and procedure, and a mock EEG net.\u003c/p\u003e\u003cp\u003eHere, we defined the feasibility of RS-EEG acquisition as having at least one\u0026thinsp;\u0026ge;\u0026thinsp;300-second recording per child across the three yearly time points. Feasibility was 77.6% in TD (66/85) and 56.7% in ASD (122/215), consistent with previous reports of EEG compliance difficulties in ASD (χ\u0026sup2;(1)\u0026thinsp;=\u0026thinsp;11.38, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001)\u003csup\u003e41,42\u003c/sup\u003e. Feasibility varied by language profile: 70.9% (LU), 51.8% (LI), and 38.6% (MV) (χ\u0026sup2;(2)\u0026thinsp;=\u0026thinsp;13.79, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.001), indicating greater challenges with limited language\u003csup\u003e\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eRS-EEG power trajectories: TD vs. ASD\u003c/p\u003e\u003cp\u003eWe compared whole-brain RS-EEG power trajectories of TD and ASD participants over the five frequency bands (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, Table S3). In delta, theta, and beta frequency bands, children with ASD showed significantly increased whole-brain power (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.013, \u003cem\u003ep\u0026thinsp;=\u003c/em\u003e\u0026thinsp;0.034, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001 respectively), with non-significant interactions between band and age. No significant difference was found for whole-brain alpha power. Finally, whole-brain gamma power was significantly higher in the ASD group than in the TD group (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and the interaction was also significant (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.025), revealing different trajectories over development. While whole-brain gamma power slowly decreased with age in TD children, it remained stable with age in the ASD group.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eRS-EEG power trajectories : TD, LU, LI, and MV\u003c/p\u003e\u003cp\u003eBuilding upon the observed differences between the broader ASD and TD groups, we further analyzed whole-brain RS-EEG power trajectories across the three autistic language profiles (LU, LI, MV) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, Table S4). Whole-brain delta trajectories showed significant group effect (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.009), but non-significant age-interaction (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05). In both theta and alpha bands, we found no significant group effect or age-interaction (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05). Significant group effect (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and age-interaction (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.018 and \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.012, respectively) were observed in beta and gamma bands. While beta and gamma power decreased with age for the TD and the LU groups, it increased over development for LI and MV. Given our hypothesis that whole-brain gamma power differs across the autistic language profiles, we conducted additional analyses comparing the four groups two-by-two. We found significant differences between TD \u003cem\u003evs\u003c/em\u003e. LU, TD \u003cem\u003evs\u003c/em\u003e. LI, TD \u003cem\u003evs\u003c/em\u003e. MV, and LU \u003cem\u003evs\u003c/em\u003e. MV (Figure S5, Table S4-5).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eRS-EEG gamma power and word combination acquisition\u003c/p\u003e\u003cp\u003eConsidering the heterogeneity in word combination acquisition and the association between gamma power and language in ASD\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e,\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e, we examined the trajectory of gamma power in our ASD sample. The quadratic model showed a progressive increase in gamma power, before plateauing near phrase speech acquisition (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e), with the peak occurring 0.384 years (approximately 4 months) post-acquisition. Gamma power then decreased after acquisition, suggesting its crucial role in the transition from single words to phrases.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eRS-EEG power features-based classifications\u003c/p\u003e\u003cp\u003e\u003cem\u003eBinary Classification: ASD vs. TD\u003c/em\u003e\u003c/p\u003e\u003cp\u003eThe supervised machine learning pipeline distinguished between individuals with ASD and Typically Developing (TD) controls with a mean accuracy of 69.2%. A permutation test, which involved running the entire cross-validation procedure 10 000 times with shuffled labels, confirmed that this accuracy was highly significant (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.0004). The observed true accuracy was well beyond the distribution of accuracies obtained from the permuted data, as illustrated in the permutation plot (Figure S6). The confusion matrix from the cross-validation reveals the specific performance of the classifier:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eSensitivity (ASD identification): The model correctly identified 94 of the 122 individuals with ASD, yielding a sensitivity of 77.0%.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eSpecificity (TD identification): The model correctly identified 36 of the 66 TD individuals, for a specificity of 54.5%.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003eMulti-Class Classification: 4 Language Profiles\u003c/em\u003e\u003c/p\u003e\u003cp\u003eFor the four-group classification task (distinguishing between the three ASD language subgroups and the TD group), the model achieved a mean accuracy of 44.6%. This performance is substantially higher than the chance level of 25%. The permutation test demonstrated that this result was highly statistically significant (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.0001), with the true accuracy falling far outside the null distribution. An analysis of the confusion matrix shows that classification performance varied considerably across the four groups:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eThe model was most successful in identifying TD participants, classifying correctly 66.7% of cases (44 out of 66).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eThe model performed poorly in identifying the ASD-MV group, with only 1 correct classification out of 17 cases (5.9% accuracy).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eThere was considerable confusion between the ASD-LI and ASD-LU subgroups, which the model often misclassified as each other (respectively 40.90% and 34.4% of accuracy).\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, we investigated whole-brain RS-EEG power trajectories in a substantial longitudinal sample of 188 children (N\u0026thinsp;=\u0026thinsp;122 ASD, N\u0026thinsp;=\u0026thinsp;66 TD), yielding 358 time points. Given the heterogeneous language abilities in autism, we explored RS-EEG power trajectories among three autistic language profiles: Language Unimpaired (LU, N\u0026thinsp;=\u0026thinsp;61), Language Impaired (LI, N\u0026thinsp;=\u0026thinsp;44), and Minimally Verbal (MV, N\u0026thinsp;=\u0026thinsp;17). We provide valuable insights into the neural correlates of language development in autism.\u003c/p\u003e\u003cp\u003eIn line with our hypothesis, our results on the RS-EEG power trajectories provide support to the U-shaped profile in autism (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e)\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. The ASD group exhibited increased power in low-frequency bands (delta, theta) and high-frequency bands (beta, gamma), and similar alpha power compared to the TD group. While the U-shaped profile has been corroborated\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e, findings have been inconsistent. Some evidence associated enhanced delta power to lower functioning children\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e,\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e, which aligns with our LI group showing significantly higher delta power than TD peers. However, other studies report enhanced delta power in high functioning children \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e. Theta power has been shown to increase in autistic individuals\u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e,\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e, though a meta-analysis found no difference compared to TD participants\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e, consistent with our findings. Additionally, our results align with the established decline of delta and theta power with age\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e, reflecting brain maturation, gray matter tissue loss, and increased processing efficiency\u003csup\u003e\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e. While reduced alpha power in ASD has been proposed as a biomarker\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e, our findings and those from large cohort studies\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e do not support this. Findings on beta power in ASD are inconsistent with some meta-analyses reporting both non-significant differences and increased power\u003csup\u003e\u003cspan additionalcitationids=\"CR14\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. We suggested that fine-grained RS-EEG features can distinguish TD from autistic participants and, to a degree, between language-based subgroups within the autism spectrum. Our binary classification of ASD versus TD achieved a statistically significant accuracy of 69.2% (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.0014), establishing that EEG signatures can differentiate these groups. The more complex multi-class model also performed significantly above chance, achieving 44.6% accuracy in differentiating the four language profiles (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.0001). Despite this statistical significance, the practical utility of these models must be carefully considered. The confusion matrices reveal challenges that preclude immediate clinical application; for instance, the binary classifier was more sensitive in identifying ASD participants than it was specific in identifying TD controls. Furthermore, the multi-class model showed inconsistent performance, struggling to accurately classify certain ASD subgroups, particularly the MV profile. Consequently, while these findings provide a robust proof-of-concept for the utility of RS-EEG in characterizing neurodevelopmental heterogeneity, the current models are not yet suitable for individual-level diagnostics. Future studies are needed to explore the full RS-EEG power spectrum in ASD, by using larger samples that span early childhood to adulthood and include varying levels of functioning.\u003c/p\u003e\u003cp\u003eAs expected, autistic children showed increased gamma compared to TD peers. Autistic children with LU exhibited increased gamma power than TD children (Figure S5). However, LU trajectories were more similar to TD than those of LI and MV children, who showed more divergent patterns (Figure S5). These results align with previous frontal EEG studies on autistic children\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e, and on HL-infants later diagnosed with ASD\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e, and our whole-brain approach extends their findings. Furthermore, since gamma oscillations might reflect the balance between E/I (or glutamatergic/GABAergic) systems, globally reduced gamma power may indicate compensatory mechanisms to early aberrant neurocircuitry, potentially supporting language development in young autistic children and HL-infant siblings later diagnosed\u003csup\u003e\u003cspan additionalcitationids=\"CR19\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e,\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e,\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e. Future research could test this hypothesis by investigating aperiodic power, as it may reflect the E/I balance in the brain\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. Moreover, it could be insightful to combine EEG with neurochemical imaging techniques (e.g., Magnetic Resonance Spectroscopy) to directly assess the role of glutamatergic and GABAergic systems as compensatory processes in language development.\u003c/p\u003e\u003cp\u003eTaken together, our results highlight the relevance of gamma oscillations to language abilities in ASD. Our analysis of word combination acquisition indeed revealed a complex developmental pattern. Gamma power in ASD increased prior to this milestone, peaked near acquisition, then declined. This trajectory suggests that increased gamma power, reflecting heightened neural activity\u003csup\u003e\u003cspan additionalcitationids=\"CR50\" citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e, may support the emergence of phrase speech in ASD. The decline of gamma power, reflecting diminished excitation, could suggest greater language and cognitive processing, as proposed previously\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e,\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e,\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e,\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e. To our knowledge, our study is the first to link RS-EEG gamma power to word combination. This finding needs replication in autistic children who reach this milestone after age 5\u003csup\u003e52,53\u003c/sup\u003e, as well as in children with developmental language disorder and TD children. In our sample, early acquisition of this milestone by TD children limited opportunities for comparison. Nonetheless, this finding may reflect neural mechanisms underlying both autistic language trajectories and the transition from single words to phrase speech, a critical step toward functional language\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eCollecting and analyzing neural signals in typical and atypical early development presents significant challenges due to the dynamic interplay of age, divergent developmental trajectories and clinical heterogeneity. Our study reflects a considerable effort to carefully disentangle these factors through a robust longitudinal design, enabling RS-EEG analysis across distinct autistic language profiles. However, several limitations should be acknowledged. While our adapted RS-EEG paradigm (recorded during a non-silent cartoon) was necessary to ensure engagement and obtain good-quality data from young participants, it differs from typical resting-state protocols involving silent videos. While necessary for participant engagement and data quality, our approach deviates from standard RS-EEG protocols and should be considered when interpreting the results. Second, the MV group was relatively small, which highlights the need for specific habituation procedures for RS-EEG acquisition\u003csup\u003e\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e. Considering the variability within MV children, especially in non-verbal and receptive language skills\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e, a larger MV group could provide a more detailed understanding of gamma power differences. It will also be important to follow MV children into school-age, as some may develop phrase speech later\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. As our longitudinal cohort was recently extended to school-age\u003csup\u003e\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e\u003c/sup\u003e, we plan to address this question in future research. Finally, although RS-EEG features-based classification showed some promise for distinguishing clinical groups, our findings remain preliminary and are not yet clinically applicable.\u003c/p\u003e\u003cp\u003eIn conclusion, our longitudinal investigation of 122 autistic and 66 TD young children, provides evidence of distinct RS-EEG power trajectories, partially supporting the U-shaped spectral profile in autism\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Examining three validated autistic language profiles (LU, LI, MV) represents a significant strength of our study, revealing how gamma power trajectories vary within ASD and closely align with language abilities, positioning gamma power as a potential marker of language heterogeneity. Gamma power followed a quadratic trajectory around word combination acquisition, indicating a compensatory neural mechanism\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e,\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e facilitating the transition to phrase speech. As phrase speech is a critical milestone toward functional language and positive outcomes for autistic individuals (e.g., improved quality of life and independence)\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e,\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e,\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e\u003c/sup\u003e, early language development should remain a priority in intervention efforts. Further understanding of the underlying neural mechanisms may further inform the development of effective early intervention approaches.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eConflicts of interest\u003c/h2\u003e\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\u003ch2\u003eAuthor contributions\u003c/h2\u003e\u003cp\u003eConceptualization, K.L., M.G., V.B., and M.S.; methodology, K.L., M.G., A.F., V.B., M.S; formal analysis, K.L., M.G., A.F., F.J.; writing\u0026mdash;original draft preparation, K.L.; writing\u0026mdash; review and editing, K.L., M.G., A.F., F.J., V.B., and M.S.; supervision, V.B., M.S.; funding acquisition, M.S. All authors have read and agreed to the published version of the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgments\u003c/h2\u003e\u003cp\u003e The authors would like to thank all the families who participated in the study, as well as the many collaborators who contributed to data collection over the years, namely Alexandra Bastos, St\u0026eacute;phanie Baudoux, Lylia Ben Hadid, Aur\u0026eacute;lie Bochet, Gaia Brenner, Lucia Cantonas, L\u0026eacute;a Chambaz, Flore Couty, Despoina Demenega, Sophie Diakonoff, Lisa Esposito, Constance Ferrat, Margot Giraud, Marie-Agn\u0026egrave;s Graf, Oriane Grosvernier, Pamela Iraci, Reem Jan, Nada Kojovic, Sara Maglio, Matthieu Mansion, Eva Micol, Priska M\u0026uuml;ller, Ir\u0026egrave;ne Pittet, Sonia Richetin, Tonia Rihs, Fran\u0026ccedil;ois Robain, Laura Sallin, Stefania Solazzo, Myriam Speller, Holger Sperdin, Niveettha Thillainathan, Chiara Usuelli, and Ornella Vico Begara.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAmerican Psychiatric Association (2013) Diagnostic and Statistical Manual of Mental Disorders. (American Psychiatric Association. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1176/appi.books.9780890425596\u003c/span\u003e\u003cspan address=\"10.1176/appi.books.9780890425596\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGillespie-Lynch K et al (2012) Early Childhood Predictors of the Social Competence of Adults with Autism. J Autism Dev Disord 42:161\u0026ndash;174\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHowlin P, Goode S, Hutton J, Rutter M (2004) Adult outcome for children with autism. J Child Psychol Psychiatry 45:212\u0026ndash;229\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDe Vries M, Geurts H (2015) Influence of Autism Traits and Executive Functioning on Quality of Life in Children with an Autism Spectrum Disorder. J Autism Dev Disord 45:2734\u0026ndash;2743\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchaeffer J et al (2023) Language in autism: domains, profiles and co-occurring conditions. J Neural Transm 130:433\u0026ndash;457\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTager-Flusberg H (2006) Defining language phenotypes in autism. Clin Neurosci Res 6:219\u0026ndash;224\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLatr\u0026egrave;che K et al (2024) Early trajectories and moderators of autistic language profiles: A longitudinal study in preschoolers. Autism 13623613241253015. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/13623613241253015\u003c/span\u003e\u003cspan address=\"10.1177/13623613241253015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHarrison JE, Weber S, Jakob R, Chute CG (2021) ICD-11: an international classification of diseases for the twenty-first century. BMC Med Inf Decis Mak 21:206\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWorld Health Organization. International Classification of Diseases (ICD)-11 (2022)\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLombardo MV et al (2015) Different Functional Neural Substrates for Good and Poor Language Outcome in Autism. Neuron 86:567\u0026ndash;577\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTager-Flusberg H (2016) Risk Factors Associated With Language in Autism Spectrum Disorder: Clues to Underlying Mechanisms. J Speech Lang Hear Res 59:143\u0026ndash;154\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTager-Flusberg H, Kasari C (2013) Minimally Verbal School‐Aged Children with Autism Spectrum Disorder: The Neglected End of the Spectrum. Autism Res 6:468\u0026ndash;478\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang J et al (2013) Resting state EEG abnormalities in autism spectrum disorders. J Neurodev Disord 5\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDede AJO, Xiao W, Vaci N, Cohen MX, Milne E (2025) Exploring EEG resting state differences in autism: sparse findings from a large cohort. Mol Autism 16:13\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNeo WS, Foti D, Keehn B, Kelleher B (2023) Resting-state EEG power differences in autism spectrum disorder: a systematic review and meta-analysis. Transl Psychiatry 13:389\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVetri L et al (2022) A Preliminary Study on Photic Driving in the Electroencephalogram of Children with Autism across a Wide Cognitive and Behavioral Range. J Clin Med 11:3568\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCohenour T, Dickinson A, Jeste S, Gulsrud A, Kasari C (2024) Patterns of spontaneous neural activity associated with social communication abilities among infants and toddlers showing signs of autism. Eur J Neurosci 60:3597\u0026ndash;3613\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWilkinson CL et al (2020) Use of Longitudinal EEG Measures in Estimating Language Development in Infants With and Without Familial Risk for Autism Spectrum Disorder. Neurobiol Lang 1:33\u0026ndash;53\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWilkinson CL, Chung H, Dave A, Tager-Flusberg H, Nelson CA (2025) Changes in Early Aperiodic EEG Activity Are Linked to Autism Diagnosis and Language Development in Infants With Family History of Autism. Autism Res 18:1356\u0026ndash;1368\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMukerji CE et al (2024) Resting Frontal Gamma Power is Associated with Both Expressive Language and Non-verbal Cognitive Abilities in Young Autistic Children. J Autism Dev Disord. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10803-024-06308-3\u003c/span\u003e\u003cspan address=\"10.1007/s10803-024-06308-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLord C \u003cem\u003eAutism Diagnostic Observation Schedule\u0026ndash; 2nd Edition (ADOS-2). Lord\u003c/em\u003e, C., Rutter, M., DiLavore, P., Risi, S., Gotham, \u003cem\u003eK., \u0026amp;\u003c/em\u003e Bishop, S. ((2012)). (2012)). (2012)\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLord C, Rutter M, Le Couteur A (1994) Autism Diagnostic Interview-Revised: A revised version of a diagnostic interview for caregivers of individuals with possible pervasive developmental disorders. J Autism Dev Disord 24:659\u0026ndash;685\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMullen EM (1995) Mullen Scales of Early Learning. American Guidance Service, Circle Pines, MN\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchopler E, Lansing MD, Reichler RJ, Marcus LM (2005) \u003cem\u003ePEP-3, Psychoeducational Profile\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGotham K, Pickles A, Lord C (2009) Standardizing ADOS Scores for a Measure of Severity in Autism Spectrum Disorders. J Autism Dev Disord 39:693\u0026ndash;705\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHus V, Gotham K, Lord C, Standardizing ADOS (2014) Domain Scores: Separating Severity of Social Affect and Restricted and Repetitive Behaviors. J Autism Dev Disord 44:2400\u0026ndash;2412\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLord C et al (2006) Autism From 2 to 9 Years of Age. Arch Gen Psychiatry 63:694\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShen MD et al (2013) Early brain enlargement and elevated extra-axial fluid in infants who develop autism spectrum disorder. Brain 136:2825\u0026ndash;2835\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWechsler D (2014) \u003cem\u003eWechsler Preschool and Primary Scale of Intelligence (4th Ed.). Svensk Version. NCS Pearson, Incorporated\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBassano D, Labrell F, Champaud C, Lem\u0026eacute;tayer F, Bonnet P (2005) Le DLPF: un nouvel outil pour l\u0026rsquo;\u0026eacute;valuation du d\u0026eacute;veloppement du langage de production en fran\u0026ccedil;ais. Enfance 57:171\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBassano D, Labrell F, Bonnet P (2020) DLPF- Guide de cotation et scores. Questionnaire sur le \u0026lsquo;D\u0026eacute;veloppement du Langage de Production en Fran\u0026ccedil;ais\u0026rsquo;: Vocabulaire, Grammaire, Communication\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKwok EYL, Brown HM, Smyth RE, Oram Cardy J (2015) Meta-analysis of receptive and expressive language skills in autism spectrum disorder. Res Autism Spectr Disord 9:202\u0026ndash;222\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eByrne K, Sterrett K, Lord C (2024) Examining the Transition from Single Words to Phrase Speech in Children with ASD: A Systematic Review. Clin Child Fam Psychol Rev 27:1031\u0026ndash;1053\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBell MA, Cuevas K (2012) Using EEG to Study Cognitive Development: Issues and Practices. J Cogn Dev 13:281\u0026ndash;294\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFl\u0026oacute; A, Gennari G, Benjamin L, Dehaene-Lambertz G (2022) Automated Pipeline for Infants Continuous EEG (APICE): A flexible pipeline for developmental cognitive studies. Dev Cogn Neurosci 54:101077\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDelorme A, Makeig S (2004) EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. J Neurosci Methods 134:9\u0026ndash;21\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePrerau MJ, Brown RE, Bianchi MT, Ellenbogen JM, Purdon PL (2017) Sleep Neurophysiological Dynamics Through the Lens of Multitaper Spectral Analysis. Physiology 32:60\u0026ndash;92\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMutlu AK et al (2013) Sex differences in thickness, and folding developments throughout the cortex. NeuroImage 82:200\u0026ndash;207\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePittet I Trajectories of imitation skills in preschoolers with Autism Spectrum Disorders\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePedregosa F et al Scikit-learn: Machine Learning in Python. Mach Learn PYTHON\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSlifer KJ, Avis KT, Frutchey RA (2008) Behavioral intervention to increase compliance with electroencephalographic procedures in children with developmental disabilities. Epilepsy Behav 13:189\u0026ndash;195\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWebb SJ et al (2015) Guidelines and Best Practices for Electrophysiological Data Collection, Analysis and Reporting in Autism. J Autism Dev Disord 45:425\u0026ndash;443\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTager-Flusberg H et al (2017) Conducting research with minimally verbal participants with autism spectrum disorder. Autism 21:852\u0026ndash;861\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChan AS, Sze SL, Cheung M (2007) Quantitative electroencephalographic profiles for children with autistic spectrum disorder. Neuropsychology 21:74\u0026ndash;81\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStroganova TA et al (2007) Abnormal EEG lateralization in boys with autism. Clin Neurophysiol 118:1842\u0026ndash;1854\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCoben R, Clarke AR, Hudspeth W, Barry R (2008) J. EEG power and coherence in autistic spectrum disorder. Clin Neurophysiol 119:1002\u0026ndash;1009\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMurias M, Webb SJ, Greenson J, Dawson G (2007) Resting State Cortical Connectivity Reflected in EEG Coherence in Individuals With Autism. Biol Psychiatry 62:270\u0026ndash;273\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNeuhaus E et al (2021) Resting state EEG in youth with ASD: age, sex, and relation to phenotype. J Neurodev Disord 13:33\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDickinson A, Jones M, Milne E (2016) Measuring neural excitation and inhibition in autism: Different approaches, different findings and different interpretations. Brain Res 1648:277\u0026ndash;289\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSohal VS, Rubenstein JL (2019) R. Excitation-inhibition balance as a framework for investigating mechanisms in neuropsychiatric disorders. Mol Psychiatry 24:1248\u0026ndash;1257\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCoghlan S et al (2012) GABA system dysfunction in autism and related disorders: From synapse to symptoms. Neurosci Biobehav Rev 36:2044\u0026ndash;2055\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBroome K, McCabe P, Docking K, Doble M, Carrigg B (2023) Speech Development Across Subgroups of Autistic Children: A Longitudinal Study. J Autism Dev Disord 53:2570\u0026ndash;2586\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWodka EL, Mathy P, Kalb L (2013) Predictors of Phrase and Fluent Speech in Children With Autism and Severe Language Delay. Pediatrics 131:e1128\u0026ndash;e1134\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBochet A (2022) Capturing part of the heterogeneity along the Autism Spectrum Disorders by focusing on the co-occurrence of Attention Deficit/Hyperactivity Disorder. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.13097/ARCHIVE-OUVERTE/UNIGE:158714\u003c/span\u003e\u003cspan address=\"10.13097/ARCHIVE-OUVERTE/UNIGE:158714\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMoss P, Mandy W, Howlin P (2017) Child and Adult Factors Related to Quality of Life in Adults with Autism. J Autism Dev Disord 47:1830\u0026ndash;1837\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7204586/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7204586/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eLanguage development in autism spectrum disorder (ASD) is heterogeneous, ranging from subtle differences to significant delays. In previous work, we identified three autistic language profiles in early childhood: Language Unimpaired (LU), Language Impaired (LI), and Minimally Verbal (MV). While these profiles show distinct vocabulary, grammar, and pragmatic development, understanding their underlying neural correlates is essential to predict outcomes and develop targeted interventions. Here, we examined whole-brain resting-state EEG power across five canonical frequency bands in a longitudinal sample comprising 66 typically developing (TD) children and 122 autistic children (ages 1.6-6.0 years), yielding 358 time points. Within the ASD group, 61 children belonged to the LU profile, 44 children to LI, and 17 children to MV. Compared to TD peers, autistic children showed increased power in low-frequency (delta, theta) and high-frequency bands (beta, gamma). Gamma power varied by autistic language profile, with the highest levels in MV children. Moreover, gamma power within ASD followed a quadratic trajectory in relation to word combination acquisition, peaking around the time of acquisition and decreasing afterward. This pattern suggests a dynamic, compensatory mechanism supporting the transition to phrase speech, which is a critical milestone toward functional speech that may predict language outcomes in ASD.\u003c/p\u003e","manuscriptTitle":"Early Trajectories of Resting-State EEG Power in Autistic Children: A Longitudinal Study Across Language Profiles","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-28 08:33:11","doi":"10.21203/rs.3.rs-7204586/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"3f4fbad7-8507-402a-88e5-b2476880955d","owner":[],"postedDate":"July 28th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":52059404,"name":"Developmental Neuroscience"},{"id":52059405,"name":"Psychiatry"},{"id":52059406,"name":"Psychology"}],"tags":[],"updatedAt":"2025-08-07T23:45:10+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-28 08:33:11","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7204586","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7204586","identity":"rs-7204586","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}