Introduction

Stimulus-induced and resting-state / pre-stimulus alpha-band (8–12 Hz) neural activity in magneto- and electroencephalography (MEG and EEG) is related to a variety of perceptual and cognitive processes and abilities, including visual spatial orienting (Doesburg et al. 2016; Hong et al. 2015), working memory (Bonnefond and Jensen 2012; Zheng et al. 2021), attention (Cona et al. 2020; Händel et al. 2011; Hanslmayr et al. 2011), filtering of irrelevant information (Payne et al. 2013), etc. In the sensory systems, event-related alpha-band desynchronization (ERD) is associated with the involvement of a neural tissue in stimulus-induced information processing (Bastarrika-Iriarte and Gaudes-Caballero 2019; Bastiaansen et al. 2001; Weisz et al. 2011), whereas synchronization in the alpha-band reflects the suppression or filtering external stimuli (Klimesch 2012; Klimesch et al. 2007). At the same time, resting state / baseline or pre-stimulus alpha power indicates the extent of cortical excitability, i.e., the lower power is related to the higher level of excitation of the cortical network (Mathewson et al. 2011; Thies et al. 2018). Summarizing, alpha oscillations are associated with neural excitability and inhibition and in MEG / EEG can be a non-invasive measure of the activation level in the brain (Bazanova and Vernon 2014; Bergmann et al. 2019; Sauseng et al. 2009).

According to recent studies, the neurobiology of Autism Spectrum Disorder (ASD) is hypothetically related to the imbalance between neural excitation (E) and inhibition (I) (Edgar et al. 2016; Levin and Nelson 2015; Rubenstein and Merzenich 2003; Seymour et al. 2020; Sohal and Rubenstein 2019; Tang et al. 2021; Yizhar et al. 2011) and, thus, investigation of alpha oscillations may be of particular interest in this population. ASD is a neurodevelopmental disorder characterized by the deficits in social interaction / communication and the presence of repetitive / restricted behavior (American Psychiatric Association 2013). Along with the core symptoms, children with ASD may have co-occurring intellectual disability (Polyak et al. 2015), language impairment (Kjelgaard and Tager-Flusberg 2001), attention deficit (Matson et al. 2013), reduced working memory (Habib et al. 2019), etc. Numerous studies have revealed that dysregulated stimulus-induced and resting-state / baseline alpha-band activity is reflected in core and co-occurring conditions of ASD (Dickinson et al. 2018; Larrain-Valenzuela et al. 2017; Keehn et al. 2017).

Research in alpha-band ERD has shown an abnormal suppression of the alpha power in different sensory systems and during different tasks in children with ASD (Cañigueral et al. 2021; Ewen et al. 2016; Murphy et al. 2021). For example, Keehn et al. (2017), using an attentional capture task, demonstrated that children with ASD exhibited a significantly lower alpha desynchronization level in different cortical regions compared to age- and non-verbal IQ-matched typically developing (TD) controls. In the visual system, the altered alpha-band ERD was also observed in individuals with ASD and the amplitude of ERD correlated with the severity of autistic symptoms (Martínez et al. 2019). In the auditory cortex, Bloy et al. (2019) showed that the greater ERD was associated with better language abilities of children with ASD, however, no between-group difference in ERD has been observed.

Abnormalities in resting-state / baseline or pre-stimulus alpha-band activity were also detected in a variety of cortical regions in autistic individuals, but the findings are controversial and showed both lower and higher alpha power in this population depending on age and condition, such as eyes closed vs. open (e.g., Bellato et al. 2020; Cantor et al. 1986; Chan et al. 2007; Cornew et al. 2012; Dawson et al. 1995; Edgar et al. 2015b, 2019; Larrain-Valenzuela et al. 2017; Leno et al. 2018; Levin et al. 2017; Liang and Mody 2022; Mash et al. 2020; Mathewson et al. 2012; Murias et al. 2007; Neuhaus et al. 2021; Sheikhani et al. 2012; Shepard et al. 2018; Takarae et al. 2022; Tierney et al. 2012; Wang et al. 2013; Zhao et al. 2023). In general, reduced alpha power during rest is associated with increased neural activation (Chapeton et al. 2019; Jensen and Mazaheri 2010), pointing to a higher level of excitability in the cortical networks in ASD. Some authors proposed that the etiology of such neural profile in ASD can be related to abnormal functioning of gamma-aminobutyric acidergic (GABAergic) inhibitory system (see, Wang et al. 2013). These alterations in resting-state / baseline or pre-stimulus alpha activity were linked to the core and co-occurring conditions in ASD such as deficit in social functioning, intellectual delay, impairments in language and auditory processing, attention to details, etc. (e.g., Dickinson et al. 2018; Edgar et al. 2015a, 2019; Larrain-Valenzuela et al. 2017; Leno et al. 2021; Mathewson et al. 2012; Neuhaus et al. 2021). Remarkably, some studies showed that the relationship between resting-state alpha power and cognitive functioning of children with ASD depended on the children’s age (e.g., Edgar et al. 2019). It is important to note that although resting-state and pre-stimulus alpha activity can reflect different phenomena, some authors consider them as a similar measure of the same state (e.g., Bai et al. 2016).

Summarizing, most previous studies revealed atypicality in alpha-band ERD and resting-state / baseline alpha power in ASD, which hypothetically can be related to the imbalance between E and I. This abnormal oscillatory functioning was reflected in core and co-occurring conditions of ASD. However, although a number of studies have focused on the alpha-band activity in ASD, most of them addressed visual, frontal or somatosensory systems, whereas the knowledge on auditory alpha oscillations (both ERD and baseline) is limited (see Weisz et al. 2011). At the same time, given the functional importance of alpha-band neural activity in the basic auditory perception, its pathological state in the auditory cortex can be related to language abilities of children with ASD.

Previous studies with other paradigms have also shown that the low-level auditory perception is altered in ASD and can be associated with language and communication impairments in these children (e.g., Berman et al. 2016; Bloy et al. 2019; Matsuzaki et al. 2019; Roberts et al. 2008, 2010, 2011, 2019, 2021). For example, Roberts et al. (2011), using mismatch negativity paradigm in MEG (MMF), have shown that a delayed latency of this response was related to language abilities of children with ASD. In a group of minimally verbal autistic individuals, it has been revealed the delayed latencies of M50 and M100 components in comparison to verbal individuals with ASD (Roberts et al. 2019). Using 40 Hz Auditory Steady-State Response (ASSR) paradigm, Stroganova et al. (2020) have demonstrated the specific left-hemispheric deficit in sustained field (SF) in the ‘core auditory area’ in children with ASD and proposed a possible association between this response and pitch processing. A direct link between 40 Hz steady-state gamma response in the language-dominant left auditory cortex and the language abilities of children with ASD has been shown in Arutiunian et al. (2023b) study. A similar relationship between low-level gamma activity and language skills in autistic children was also revealed by Roberts et al. (2021). Therefore, previous studies have identified that the primary and secondary processing difficulties in the auditory cortex can be associated with language impairment in ASD.

The goal of this study was to investigate stimulus-induced alpha ERD and baseline alpha activity (both periodic and aperiodic components) in the auditory cortex of children with ASD, using MEG. First, we aimed to estimate the cortical sources of the sustained auditory response and to assess the stimulus-induced and baseline alpha activity in the same region. We expected that similar to the findings in other sensory systems, alpha-band ERD and baseline alpha activity would be altered in the auditory cortex of children with ASD. Second, the study addressed the relationships between alpha-band parameters and individual characteristics of children with ASD (language abilities, non-verbal IQ, the severity of autism, age). The novelty of the study is that it focused on the auditory alpha activity which is largely unexplored compared to the alpha-band activity in other cortical systems (Weisz et al. 2011). Moreover, the study addressed not only the raw baseline power in the auditory cortex but also periodic (oscillatory) and aperiodic (arrhythmic) components of a neural field that are associated with different neurobiological mechanisms (see Levin et al. 2020). This will help to highlight specific mechanisms that could be altered in children with ASD and related to their language functioning.

In the previous study with the same group of participants (Arutiunian et al. 2023a), we analyzed stimulus-induced auditory and visual alpha oscillations in children with ASD and their relationships with the severity of autistic traits. The present study aimed to reanalyze the data from the previous work and to present a comprehensive neural parameterization as well as association of different neural parameters with a number of phenotypic characteristics of children with ASD.

Methods

Participants

Forty children were included in this study: 20 children with ASD (5 female, age range 8.02–14.01 years, Mage = 10.03, SD = 1.7) and 20 TD children (9 female, age range 7.02–12.03 years, Mage = 9.11, SD = 1.3). All participants had normal hearing (based on the parental reports as well as the screening with Audiogramm version 4.6.1.3, Professional Audiometric System; Sennheiser HAD 280 audiometry headphones) and normal or corrected-to-normal vision (based on the parental reports).

Children with ASD met the criteria of the International Classification of Diseases – 10 (World Health Organization 2016), and 18 out of 20 children were additionally assessed with Autism Diagnosis Observation Schedule – Second Edition, ADOS-2 (Lord et al. 2012). Also, parents of both groups of children completed the Autism Spectrum Quotient: Children’s Version, AQ (Auyeung et al. 2008). Exclusion criteria for the ASD group were co-occurring neurological disorders (e.g., epilepsy) and the presence of known single-gene conditions (e.g., Fragile X syndrome). The information of medication status of children was not available.

The non-verbal IQ of children with ASD was measured with the Kaufman Assessment Battery for Children, NVI index (Kaufman and Kaufman 2004), or the Wechsler Intelligence Scale for Children – Third Edition, performance IQ score (Wechsler 1991); the non-verbal intelligence of TD children was assessed with the Raven’s Colored Progressive Matrices (Raven 2000). Language abilities were screened with the Russian Child Language Assessment Battery, RuCLAB (Arutiunian et al. 2022b); mean language score (MLS) was calculated for each child (Table 1). RuCLAB consists of 11 subtests assessing phonology, vocabulary, morphosyntax and discourse in both production and comprehension; MLS is a standard average score (from 0.0 to 1.0) across all subtests. Additional information about each subtest as well as scoring is freely available online: (1) https://osf.io/uaxrd and (2) https://osf.io/x8hty.

Table 1 The demographic information, M ± SD (range)
Full size table

The study was approved by the ethics committee of Moscow State University of Psychology and Education (for ASD group) and the HSE University Committee on Interuniversity Surveys and Ethical Assessment of Empirical Research (for TD group) and was conducted according to the Declaration of Helsinki.

Materials and procedure

The stimuli were 40 Hz amplitude-modulated tones with 1000 ms duration. The presentation of stimuli (N = 90) was binaural with 2000 ms inter-trial interval. The duration of the experiment was ∼ 5 min. The experiment was programmed and run with the PsychoPy software (Peirce 2007). Stimuli were delivered via plastic ear tubes with foam tips inserted into the ear canals, and the intensity level was set at 83.7 dB sound pressure level. Prior the experiment, the children were instructed to sit as still as it is possible, listen to the sounds, and look at the fixation cross on the screen in front of them during MEG recording. During the experiment, lab assistant was monitoring if children pay attention to the task.

MEG acquisition and pre-processing

MEG was collected with a whole-head 306-channel MEG system (Vectorview, Elekta Neuromag) which consists of 204 orthogonal planar gradiometers and 102 magnetometers. The data were acquired in a sitting position in a magnetically shielded room. The head position in the MEG helmet was monitored every 4 ms during the recording via four head position indicator (HPI) coils which were digitized together with fiducial points with the 3D digitizer ‘Fastrak’ (Polhemus). To remove external interference signals and to compensate for head movements, we applied the temporal signal space separation (Taulu and Simola 2006) and the movement compensation approaches implemented in MaxFilter software. We used an electrooculogram (EOG) to detect the blinks (the electrodes were placed above and below the left eye) and horizontal eye movements (the electrodes were placed at the left and right outer canthi). An electrocardiography (ECG) was monitored with ECG electrodes to compensate for cardiac artifacts.

MEG was acquired at 1000 Hz sampling rate, band-pass filtered offline in the 0.1–330 Hz frequency range, and a notch filter of 50 Hz was applied. We used only gradiometers for the analysis because of the current debates over mixing both magnetometers and gradiometers, which have different levels of noise (see Seymour et al. 2020). Ocular and cardiac artifact removal was performed with EEGLAB’s (Delorme and Makeig 2004) Independent Component Analysis (ICA) implemented in Brainstorm (Tadel et al. 2011). Cleaned MEG data were segmented in 3000 ms epochs with 1500 ms before and after stimulus onset. The DC offset correction of from − 100 ms to − 2 ms was applied. Epochs were inspected visually and those affected by muscular artifacts (± 3000 fT) were rejected. The number of artifact-free epochs did not differ between groups of children: TD group, Mnumber = 84, SD = 3.30, range 77–87; ASD group, Mnumber = 84, SD = 3.85, range 75–87; t(37.13) = 0.17, p = 0.86.

MEG source analysis

To build individual brain models, we collected structural MRI data. T1 MRI scans were acquired with a 1.5 T Siemens Avanto scanner with the following parameters: repetition time = 1900 ms, echo time = 3.37 ms, flip angle = 15°, matrix size = 256 × 256 × 176, voxel size = 1.0 × 1.0 × 1.0 mm3. The segmentation of the individual MRIs and the reconstruction of cortical surface were performed with the FreeSurfer software (Dale et al. 1999). The surface was down-sampled to 15,000 vertices for each participant. The co-registration between MRI and MEG was performed with the Brainstorm toolbox based on the six reference points (left and right pre-auricular points, nasion, anterior and posterior commissure, and interhemispheric point) and the additional digitized head points (N = ∼ 150).

T1 MRIs were obtained for all TD children and for 15 out of 20 children with ASD. For those children with ASD who were not able to lie still in the MRI scanner, we used the template anatomy (‘MRI: ICBM152’) applying the special warping procedure implemented in Brainstorm. This algorithm allowed to build a pseudo-individual brain based on the head points digitized before the MEG data collection and represented the real head shape of each child.

To compute the individual head models, we used the ‘Overlapping spheres’ method (Huang et al. 1999). The inverse problem was solved with the depth-weighted linear L2-minimum norm estimate method (Lin et al. 2006), with the dipole orientation constrained to be normal to the cortical surface. A common imaging kernel was computed and then applied to obtain single epoch cortical reconstructions. A noise covariance was estimated from a 2 min empty room recording, taken after each participant’s recording session. To provide a between-subject comparison, the individual MNEs were projected to the ‘MRI: ICBM152’ template brain.

To estimate the cortical sources of the sustained auditory potential, we selected the regions of interest (ROIs) that corresponded to the ‘core auditory area’ (primary auditory cortex and the vicinity of this region) in both hemispheres: transverse temporal gyrus, transverse temporal sulcus, planum temporale of the superior temporal gyrus, lateral superior temporal gyrus, superior temporal sulcus, planum polar of the superior temporal gyrus, and inferior part of the circular sulcus of the insula based on the Destrieux parcellation cortical atlas (Destrieux et al. 2010).

Time-frequency (TF) analysis at the source level was performed with the Morlet wavelets (central frequency = 10 Hz, time resolution = 0.5 s) and the power was calculated. TF maps were normalized with an Event-Related Synchronization/Desynchronization approach considering the time window between − 500 ms and − 200 ms as a baseline. The normalized power was averaged in the standard alpha-band frequency range (8–12 Hz) in the interval from 200 ms to 1000 ms after stimulus onset. We chose this time window based on the previous study (Arutiunian et al. 2022a) which demonstrated that sustained auditory potential on the amplitude-modulated tones is evoked at around 200 ms and stays stable during auditory presentation. To localize the cortical sources of auditory alpha desynchronization, we estimated MNI coordinates of 30 vertices with the highest alpha power decrease values in the defined ROIs (15 vertices per hemisphere), and for the further analysis, extracted power averaged over these 15 vertices in the time interval between 200 ms and 1000 ms in each hemisphere for each child. This approach was successfully used previously because it accounts for the inter-individual variability of auditory response in the ‘core auditory area’ (defined ROIs) which contributes to a more precise source estimation (Arutiunian et al. 2022a; Stroganova et al. 2020).

The analysis of the baseline alpha activity was performed in the same ROIs defined for the auditory alpha desynchronization. Power spectral density (PSD) was calculated in the − 1400 ms and − 100 ms pre-stimulus time interval, and PSDs were normalized using Log10. The normalized PSDs were averaged in frequency (8–12 Hz) and in the time window from − 1400 ms to − 100 ms. Also, to model the periodic and aperiodic components of power spectrum (Donoghue et al. 2020), we used the specparam toolbox (Ostlund et al. 2022) in Python v3.10 with the following settings: peak_width_limit = [1.5, 14], n_peaks = 2, peak_height = 0.10, and frequency_range = [1,35]. For each child, the model provided two parameters (aperiodic offset and aperiodic exponent) to describe the aperiodic 1 / f signal. Additionally, to determine an aperiodic-adjusted alpha power, the aperiodic signal from the model was subtracted from the raw power spectrum, resulting in a flattened spectrum. For the statistical analysis, raw and aperiodic-adjusted power were calculated for alpha-band frequency range, whereas aperiodic offset and aperiodic exponent were calculated for the broader range (1–35 Hz).

Statistical analysis

First, for between-group comparisons of stimulus-induced neural activities (alpha ERD), we applied linear mixed-effects models including a main effect of hemisphere, the effect of group (ASD vs. TD) nested within the left and right hemispheres as fixed effects, baseline power as a covariate, and participants as a random intercept. The structure of the models was as follows: lmer(neural_activity ∼ 1 + baseline power + Hemisphere / Group + (1 | ID), data = data, control = lmerControl(optimizer = “bobyqa”)). For the baseline neural activity (baseline alpha power, aperiodic-adjusted alpha power, aperiodic offset, and aperiodic exponent), we used linear mixed-effects models with main effects of hemisphere, group, and hemisphere × group interaction as fixed effects, and participants as a random intercept, as follows: lmer(neural_activity ∼ 1 + Hemisphere * Group + (1 | ID), data = data, control = lmerControl(optimizer = “bobyqa”)). Second, to analyze the association between alpha-band ERD and baseline oscillatory alpha activity (aperiodic-adjusted alpha power), we used linear mixed-effects models including a main effect of hemisphere, main effect of group, and the effect of baseline alpha activity nested within the left and right hemispheres as fixed effects, and participants as a random intercept, according to the formula: lmer(alpha_ERD ∼ 1 + Hemisphere / alpha_baseline + Group + (1 | ID), data = data, control = lmerControl(optimizer = “bobyqa”)). Finally, the relationships between neural activities and the behavioral measures of children with ASD were assessed with the Spearman’s correlation analysis (as some of the variables were not normally distributed based on the Shapiro–Wilk test in R) and mediation models. The correction for multiple comparisons was used where it was needed (see details further).

All models were estimated in R (R Core Team 2019) with the lme4 (Bates et al. 2015), sem (Fox 2006), lavaan (Rosseel 2012), and Hmisc (Harrell 2023) packages. The data were plotted with ggplot2 package (Wickham 2016), and the figures for neural responses were created using Brainstorm and python data visualization library matplotlib (Hunter 2007).

Results

Sample description

We compared children with and without ASD in different domains of functioning (the severity of autistic traits and language abilities). The ASD group had a significantly higher presence of autistic symptoms compared to the TD group: MASD = 83.6 (SD = 18.8) vs. MTD = 50.2 (SD = 14.2), t(35.4) = 6.23, p < 0.001. There was also a significant difference in language abilities, indicating the lower scores in children with ASD: MASD = 0.75 (SD = 0.23) vs. MTD = 0.95 (SD = 0.02), t(19.37) = − 4.04, p < 0.001 (see Table 1). The ASD group was a highly heterogeneous in the cognitive level, including children both with intellectual disability (e.g., IQ = 41) and with a normal non-verbal cognition (e.g., IQ = 118).

Stimulus-induced alpha power decrease

The source estimation of alpha-band ERD in the auditory cortex showed that the highest values of alpha desynchronization were in the primary auditory cortex (A1) and adjacent regions of the superior temporal gyrus in both hemispheres in children with and without ASD (see the time-courses of the estimated sources in both hemispheres, Fig. 1). Between-group comparisons revealed a larger ERD in the TD group in comparison to the ASD group (even when accounting for the baseline power (averaged in − 500 to − 200 ms time interval) which can be different in two groups, see further), evident in the magnitude of the alpha power suppression: left auditory ROI, MASD = − 16.54 (SD = 8.79) vs. MTD = − 23.26 (SD = 8.00), β = − 6.66, SE = 2.82, t = − 2.36, p = 0.02; right auditory ROI, MASD = − 14.89 (SD = 8.44) vs. MTD = − 21.10 (SD = 10.29), β = − 6.38, SE = 2.83, t = − 2.26, p = 0.03 (Table 2; Fig. 2).

Fig. 1
figure 1

The grand average time-courses of the estimated sources in the left and right auditory ROIs in children with and without ASD (amplitude, z-score)

Full size image
Table 2 Between-group difference in stimulus-induced auditory alpha power decrease in children with and without ASD.
Full size table
Fig. 2
figure 2

The comparison of stimulus-induced alpha-band Event-Related Desynchronization (ERD) between children with and without ASD: (A) Time-frequency maps for the left and right auditory ROIs (normalized power, % change from the baseline); (B) Between-group differences in alpha-band ERD (the significance is labeled as *)

Full size image

Baseline periodic and aperiodic neural activity

To investigate baseline alpha-band activity in both groups of children, we calculated PSD in the pre-stimulus time interval from − 1400 ms to − 100 ms in the same ROIs as were estimated for the stimulus-induced auditory alpha oscillations.

Baseline raw alpha power did not differ between groups of children: main effect of hemisphere, β = − 0.00, SE = 0.03, t = − 0.03, p = 0.97, main effect of group, β = 0.04, SE = 0.09, t = 0.46, p = 0.69, and hemisphere × group interaction, β = 0.03, SE = 0.04, t = 0.69, p = 0.49 (Table 3; Fig. 3).

Table 3 Between-group differences in the baseline alpha-band activity in children with and without ASD.
Full size table
Fig. 3
figure 3

The comparisons of baseline alpha-band activity between children with and without ASD: (A) Log10 transformed power for a raw spectrum (upper), periodic component of a spectrum (middle) and aperiodic component of a spectrum (lower); (B) Boxplots representing group differences in the metrics of the baseline neural activities (* = statistically significant difference, ns = non-significant)

Full size image

To determine differences in the aperiodic and oscillatory power spectrum components in the alpha-band frequency range, we used parameterization algorithm (Ostlund et al. 2022), estimating aperiodic 1 / f and periodic / oscillatory components of the power spectrum. To obtain the aperiodic-adjusted alpha power, the aperiodic signal was subtracted from the raw power spectrum, resulting in a flattened spectrum. For the aperiodic-adjusted alpha power, the results revealed a significant main effect of group, β = 0.10, SE = 0.05, t = 1.99, p = 0.05, but no main effect of hemisphere, β = 0.02, SE = 0.01, t = 1.57, p = 0.12, or hemisphere × group interaction, β = 0.02, SE = 0.06, t = 1.48, p = 0.15 (see Table 3; Fig. 3).

Aperiodic exponent was also different between groups of children, main effect of group, β = 0.08, SE = 0.04, t = 2.06, p = 0.04, but there was no main effect of hemisphere, β = 0.01, SE = 0.02, t = 0.71, p = 0.48, and hemisphere × group interaction, β = 0.01, SE = 0.02, t = 0.52, p = 0.60. No significant effects were observed for aperiodic offset, main effect of hemisphere, β = − 0.01, SE = 0.03, t = − 0.27, p = 0.79, main effect of group, β = 0.01, SE = 0.08, t = 0.09, p = 0.93, and hemisphere × group interaction, β = 0.01, SE = 0.05, t = 0.25, p = 0.81 (see Table 3; Fig. 3).

The relationships between stimulus-induced and baseline alpha activity

The analysis revealed significant relationships between neural responses, indicating that the higher alpha-band ERD was associated with the higher baseline aperiodic-adjusted alpha power in both hemispheres, however, no main effects were identified (Table 4; Fig. 4).

Table 4 The relationships between alpha-band Event-Related Desynchronization and baseline aperiodic-adjusted alpha power
Full size table
Fig. 4
figure 4

The relationships between stimulus-induced Event-Related Desynchronization (ERD) in the alpha-band (%, change from the baseline) and the aperiodic-adjusted baseline alpha activity (log10 transformed power) in children with and without ASD (the cases of negative power could be explained by the 1 / f subtraction)

Full size image

Brain-behavior correlations in children with ASD

First, to investigate the relationships between neural activity and behavioral measures (non-verbal IQ, language abilities, children’s age, the severity of autism) in children with ASD, we used Spearman’s correlation analysis. We applied a Bonferroni correction for the total number of correlations (N = 50), so that the predictors were significant at the α = 0.001 level.

The results demonstrated two statistically significant correlations, i.e., between baseline alpha power / aperiodic offset in the language-dominant left auditory cortex and language abilities of children with ASD, r = − 0.71, p = 0.0003 and r = − 0.77, p < 0.0001 (Fig. 5; Table 5). This indicated that the lower power and lower offset were associated with better language skills.

Fig. 5
figure 5

The relationships between neural metrics in the left auditory cortex and language abilities of children with ASD

Full size image
Table 5 The relationships between neural activities and behavioral measures in children with ASD (the results of Spearman’s correlation analysis)
Full size table

Second, as previous studies have shown that non-verbal IQ was frequently associated with language skills of children with ASD, we provided an additional mediation modeling to assess whether non-verbal IQ plays a mediation role in the relationship between baseline alpha power / aperiodic offset in the left auditory cortex and language skills in the ASD group. The model assesses the direct effect between brain responses and language skills of children with ASD as well as a mediation role of non-verbal IQ in these relationships. The results of mediation modeling are presented in Table 6.

Table 6 The output of the mediation models
Full size table

The results indicated the significant associations between language skills and both non-verbal IQ and brain responses. However, there is no significant direct relationship between non-verbal IQ and baseline alpha power / aperiodic offset as well as indirect mediation effect of non-verbal IQ.

Finally, we provided Spearman’s correlation analysis between measures of alpha-band neural activity (alpha ERD, baseline alpha power, and aperiodic-adjusted alpha power) and AQ subscores related to attention, i.e., attention to details and attention switching domains (as the previous studies have reported that alpha power can be associated with the engagement of attention in the task). The analysis revealed no significant effects in any hemisphere for both AQ domains: attention to details, alpha ERD (right), r = − 0.13, p = 0.58, alpha ERD (left), r = − 0.00, p = 0.97; baseline alpha power (right), r = − 0.11, p = 0.64, baseline alpha power (left), r = − 0.33, p = 0.14; aperiodic-adjusted alpha power (right), r = 0.02, p = 0.92, aperiodic-adjusted alpha power (left), r = 0.25, p = 0.29; attention switching, alpha ERD (right), r = − 0.09, p = 0.71, alpha ERD (left), r = 0.08, p = 0.75; baseline alpha power (right), r = − 0.21, p = 0.36, baseline alpha power (left), r = − 0.30, p = 0.19; aperiodic-adjusted alpha power (right), r = − 0.22, p = 0.35, aperiodic-adjusted alpha power (left), r = − 0.07, p = 0.74.

Discussion

In this MEG study, we focused on the stimulus-induced alpha-band ERD and the baseline alpha activity (both periodic and aperiodic) in the auditory cortex of primary-school-aged children with ASD. Additionally, we aimed to investigate the relationships between these neural activities and behavioral measures in children with ASD. In general, we revealed abnormalities in both stimulus-induced and baseline neural activity in ASD and also the associations between the lower baseline alpha power and aperiodic offset in the left auditory cortex and the higher language scores of children with ASD.

MEG localization analysis demonstrated that the cortical sources of alpha-band ERD were in the primary auditory cortex (A1) and its vicinity in both hemispheres in children with and without ASD. This corresponded to the previous findings, which showed the sources of sustained auditory response to be in the A1 and adjacent regions (Arutiunian et al. 2022a; Pellegrino et al. 2019; Stroganova et al. 2020).

As we hypothesized, between-group comparisons revealed a bilateral reduction of alpha-band ERD in children with ASD, indicating that the suppression of alpha power in the auditory cortex was higher in the TD group. Our findings are in line with some of the previous studies that demonstrated abnormal alpha-band ERD in different cortical systems in ASD (Cañigueral et al. 2021; Ewen et al. 2016; Keehn et al. 2017; Martínez et al. 2019). However, it is important to note that these abnormalities may have different directions: for instance, Martínez et al. (2019) have shown the elevated alpha ERD in the visual cortex whereas we observed a reduced alpha ERD in the auditory cortex, pointing to a possible difference in the alterations of alpha ERD depending on sensory modality. See also Bloy et al. (2019) where no difference in alpha ERD in children with ASD has been reported. The baseline alpha-band oscillatory activity (aperiodic-adjusted power) also differed between groups of children, so as the ASD group had a lower power. As well, this corresponded to other findings, indicating that the power of resting-state / baseline or pre-stimulus alpha oscillations in different cortical areas was decreased in children with ASD (Chan et al. 2007; Neuhaus et al. 2021; Wang et al. 2013). It is important to highlight, however, that most studies addressed resting-state rather than pre-stimulus alpha power and although we consider them as the same phenomena (based on Bai et al. 2016), they can reflect different neurophysiological states. In general, abnormalities in both stimulus-induced alpha-band ERD and the baseline alpha oscillations in ASD can be hypothetically associated with the imbalance between E and I (Edgar et al. 2016; Levin and Nelson 2015; Rubenstein and Merzenich 2003; Seymour et al. 2020; Sohal and Rubenstein 2019; Tang et al. 2021; Yizhar et al. 2011); a lower stimulus-induced alpha-band suppression as well as a lower baseline alpha power are related to a higher level of excitability in the cortical networks in ASD (Chapeton et al. 2019; Jensen and Mazaheri 2010). In addition to these between-group differences, our sample of ASD participants had flattened aperiodic exponent (a measure of E / I balance) in comparison to TD controls with the evidence of increased E relative to I in the auditory cortex of children with ASD, which is consistent with the previous findings (see, for example, Levin et al. 2020; Manyukhina et al. 2022; Wilkinson and Nelson 2021).

Studies in both humans and animals have suggested that the extent of desynchronization of the neuronal activity in the alpha-band during information processing and perception is primarily related to the extent of the involvement of neural tissue in the activity (e.g., Bastarrika-Iriarte and Gaudes-Caballero 2019; Bastiaansen et al. 2001; Weisz et al. 2011); importantly, increased alpha-band synchronization during task may signify active suppression / filtering of irrelevant information (see, e.g., Händel et al. 2011). Similarly, a higher level of synchronization of neuronal activity during a pre-stimulus / resting period in the same frequency range is associated with more rest condition of the cortical area (e.g., Klimesch et al. 2007). This means that altered neural activity in the alpha frequency band in the ASD group can be linked to the less involvement of the area to the processing / perception as well as ‘less resting’ condition of the neural tissue during rest. One of the possible functional explanations of lower alpha suppression and lower baseline alpha power in the auditory cortex of children with ASD is the involvement of attention to the task. Previous studies showed that when the attention is engaged in the perception and processing of stimuli, the power of alpha (both ERD and resting) decreases (see Cona et al. 2020; Händel et al. 2011; Hanslmayr et al. 2011). Therefore, lower ERD in children with ASD can be associated with less involvement of attention to the task. Note, however, that our sample of children with ASD had not been screen for ADHD or attention problems, and we did not find significant relationships between neural metrics and attention skills (measured with AQ) in the ASD group.

The difference between raw and aperiodic-adjusted baseline alpha powers in the ASD group could shed some light on the specific neurobiological mechanisms that are impaired in our sample of participants. Raw power consists of both oscillatory activity itself and aperiodic activity that refers to the arrhythmic component of neural field data (see He 2014). Neural oscillations (aperiodic-adjusted activity / rhythmic / periodic aspect of the spectra) registered with MEG / EEG reflect a synchronized activity of the large neuronal populations and a rhythmic fluctuation of excitability in neuronal ensembles associated with the dynamics of the circuits (e.g., Bauer et al. 2020; Gray et al. 1994). At the same time, aperiodic components hypothetically related to E / I balance and average firing rate (He 2014; Ostlund et al. 2022). This could mean that the abnormal auditory alpha-band activity that was observed in our sample of children with ASD was related to both rhythmic and arrhythmic (aperiodic exponent) components of MEG signal.

In our sample of participants, we found a relationship between alpha-band ERD and baseline alpha oscillations in both auditory cortices, indicating that the higher baseline aperiodic-adjusted alpha power was associated with better stimulus-induced alpha-band suppression. Similar results were shown by Keehn et al. (2017) in a mixed group of children with and without ASD. All these findings indicated that the pre-stimulus / baseline (or reference) alpha power is greatly associated with the stimulus-induced alpha-band suppression, interpreting the phenomenon in terms of more vs. less cortical excitability preceding a task performance / stimulus perception (e.g., Doppelmayr et al. 1998; Klimesch et al. 2006). Remarkably, although children with ASD had reduced alpha ERD and baseline aperiodic-adjusted alpha power, the pattern of relationship between them was similar to those as in TD children.

Our hypothesis that alpha-band neural activity in the auditory cortex would be related to the language abilities of children with ASD was partly confirmed. To the best of our knowledge, only one study has addressed the relationship between auditory alpha ERD and the language abilities of children with ASD (Bloy et al. 2019). The authors reported that the greater ERD was associated with better language skills. However, in our study, we did not find this relationship. Perhaps, the difference in results can be explained by the type of stimuli. Bloy et al. (2019) used high-order stimulus items (words and pseudowords), whereas we used low-level amplitude-modulated tones. Some studies demonstrated that the stimulus complexity influenced the extent of alpha-band ERD (Boiten et al. 1992; Sharma et al. 2019) and the difference of processing was found even between vowels and tones (Krause et al. 1995). Moreover, although the previous studies did not focus on the low-level auditory alpha-band activity and its relation to language in ASD, there is an evidence of the association between the low-level auditory perception and language skills in autistic individuals using other paradigms, such as presenting simple pure tones, phonemes, amplitude-modulated tones, click trains, and amplitude-modulated sweeps (e.g., Arutiunian et al. 2023b; Roberts et al. 2011, 2019, 2021; Stroganova et al. 2020). Thus, words / pseudowords used in (Bloy et al. 2019) were spectrally and semantically more complex than simple amplitude-modulated tones used in our study, and this can be one of the potential explanations for the difference between results.

However, we found that the raw baseline alpha power and aperiodic offset in the language-dominant left auditory cortex were negatively related to the language abilities of children with ASD, so that the lower power / aperiodic offset were associated with higher language skills. The link between baseline alpha power and language skills, at the first glance, demonstrates that more altered neural circuits in children with ASD (higher cortical excitability during rest) is related to better language skills and it is in contrast to most of the previous studies showed that higher E / I ratio was associated with worse phenotypic characteristics (e.g., see Bozzi et al. 2018; Nelson and Valakh 2015; Rivell and Mattson 2019; Sohal and Rubenstein 2019). However, this effect considered in the framework of the findings on aperiodic offset can be interpreted differently. Aperiodic offset is associated with a neuronal population spiking (Manning et al. 2009; Miller et al. 2014) and the blood oxygenated level dependent signal from functional MRI (Winawer et al. 2013); and to the best of our knowledge this is the first evidence of the relationship between this neural marker and language skills in children with ASD. Our results suggested that lower average firing rate (lower offset) and, thus, more rest condition of a neural tissue during rest is related to higher language scores. As there is no relationship between aperiodic-adjusted alpha power and language skills in the same brain area and a high correlation between raw baseline power and offset (r = 0.83, p < 0.0001), we hypothesize that the link between raw power and language skills is driven by the broadband shift of the spectrum rather than by the baseline alpha power itself. Our findings contributed to the previous evidence showing that not only stimulus-induced but also the baseline alpha oscillations in the auditory cortex can be related to language in ASD. Similarly to Bloy et al. (2019), we did not find any relationships between alpha-band neural activity and general cognitive functioning or the severity of autistic symptoms.

In conclusion, the study focused on largely unexplored alpha-band activity in the auditory cortex of children with ASD. We revealed both abnormal suppression of alpha power during sound perception and atypically low baseline aperiodic-adjusted alpha power as well as flattened aperiodic exponent in ASD. Moreover, the results revealed the association between the lower raw baseline alpha power and aperiodic offset in the language-dominant left auditory cortex and the better language scores of children with ASD. Future studies would benefit from addressing the changes in specifically the aperiodic offset in the auditory cortex and language skills of children with ASD during speech therapy. Moreover, as there is a difference in alpha-band ERD between complex and simple auditory stimuli, it is needed to use both types of stimuli in the same group of children. This will help to understand how different types of stimuli influence alpha-band ERD and how this neural response is related to language in ASD.

Neural markers of excitation / inhibition balance: methodological considerations

Recent neurobiological studies in both humans and animals have proposed that E / I imbalance is considered as one of the core pathophysiological mechanisms in ASD (e.g., Levin and Nelson 2015; Sohal and Rubenstein 2019). Previous studies have identified a number of neural markers that reflect or are associated with E / I balance, such as gamma-band activity (Rubenstein and Merzenich 2003; Seymour et al. 2020), alpha-band activity (Weisz et al. 2011), and aperiodic exponent or 1 / f slope of arrhythmic component of the neural activity (Donoghue et al. 2020; Gao et al. 2017; Molina et al. 2020). In our study, we observed the alterations in both alpha-band neural activity and aperiodic exponent in children with ASD. These findings suggest that different metrics of E / I balance can complement each other as they reflect the same phenomena: all alterations in the neural activity in the ASD group (reduced alpha power suppression during sound perception, low baseline alpha power during rest, and flattened aperiodic exponent) are associated with increased E relative to I.

Limitations and future directions

The study has some limitations which should be mentioned. First, a sample size of participants is relatively small given a large age-range and imbalance in sex (25% and 45% of females in TD and ASD groups respectively). To generalize the findings, it is necessary to include larger sex-balanced sample sizes. Second, we used the constant inter-stimulus interval and, therefore, did not manipulate with that parameter; this, hypothetically, can cause adaptation to the stimulus presentation. To reduce that effect, future studies need to use different time intervals between stimuli. Third, we do not provide any measures of sensory processing. Future studies would benefit from addressing the relationships between sensory processing, auditory alpha-band activity and language abilities of children with ASD. Finally, we did not account for individual differences in hearing thresholds and loudness perception which can play a role in strengths of auditory response.