Abstract
Childhood apraxia of speech (CAS) is conceived as an impairment of the central motor system's ability to program multiple speech movements, resulting in inaccurate transitions between and relative timing across speech sounds. However, the extant neuroimaging evidence base is scant and inconclusive, and the neurophysiological origins of these motor planning problems remain highly underspecified. In the first magnetoencephalography study of this disorder, we measured brain activity from typically developing (TD) children (N = 19, 11 males, 8 females) and children with CAS (N = 7 males) during performance of a speech task designed to interrogate function of the speech areas of the primary sensorimotor cortex. Relative to their TD peers, our sample of children with CAS showed abnormal speech-related responses within the mu-band motor rhythm, and beamformer source reconstruction analyses specify a brain origin of this speech rhythm in the left cerebral hemisphere, within or near pre-Rolandic motor areas crucial for the planning and control of speech and oromotor movements. These results provide a new and specific candidate mechanism for the core praxic features of CAS; point to a novel and robust neurophysiological marker of typical and atypical expressive speech development; and support an emerging neuroscientific consensus which assigns a central role for programming and coordination of speech movements to the motor cortices of the precentral gyrus.
Significance Statement
This study provides the first evidence that children with a rare form of developmental apraxia of speech (AOS) show abnormal functioning of the primary speech motor cortex within the left cerebral hemisphere. This finding accords with evidence from studies of acquired forms of AOS due to brain damage, indicating that the two types of AOS share common pathophysiological mechanisms. The results address a significant and widely recognized gap in the literature and provide novel support for modern models of speech motor control. This research points to a promising new biomarker for application in studies of both typical and disordered speech development.
Introduction
Childhood apraxia of speech (CAS) is a rare developmental motor speech disorder estimated to affect 0.1–0.2% of preschoolers (Shriberg et al., 1997). Individuals with CAS present with severe and highly persistent speech problems similar to those of adults with acquired apraxia of speech (AOS), characterized by inconsistent speech errors, lengthened and impaired coarticulatory transitions, and inappropriate prosody (ASHA, 2007). A key clinical feature of CAS is an absence of neuromuscular problems or weakness in the articulatory periphery (Ziegler and Ackermann, 2017). Accordingly, the core deficits of this disorder have been assigned an origin in the central nervous system, ostensibly within brain mechanisms associated with the ability to program and plan speech movements (ASHA, 2007; Morgan and Webster, 2018). Such processes would presumably be fundamental to the acquisition of skilled speech production in the typically developing (TD) human brain, developmental processes that remain largely and fundamentally unknown. Thus, insights into the neurophysiological and neuroanatomical origins of CAS have the potential to shed considerable light on the mechanisms that support and permit development of this crucial and uniquely human capacity.
Our current understanding of the brain origins of CAS remains limited by a small body of empirical neuroimaging and electrophysiological studies. To our knowledge the existing literature consists of 15 studies published over a period of about the last 25 years (Table 1). Their salient findings can be summarized as follows:
In contrast to most cases of acquired AOS, standard clinical magnetic resonance imaging (MRI) scans of individuals with CAS show no evidence of clinically significant lesions (Chilosi et al., 2015) or, in a small proportion of cases, minor abnormalities with no clear relationship to CAS features (Chilosi et al., 2022).
Brain structure and function of genotyped members of the KE family with FOXP2-related CAS phenotype (Lai et al., 2001) have been characterized in a series of studies (Vargha-Khadem et al., 1998; Watkins et al., 2002; Belton et al., 2003; Liégeois et al., 2003, 2011, 2016). Relative to unaffected family members and unrelated controls, affected KE family members exhibit abnormal volumes in distributed cortical and subcortical structures. Functional neuroimaging has similarly shown abnormalities in distributed regions of premotor, supplementary, and primary motor cortices, cerebellum and basal ganglia.
It is unclear to what extent neuroimaging results from samples with FOXP2 mutations can be extrapolated to other neurogenetic CAS phenotypes or the CAS population. For example, while bilateral abnormalities of the basal ganglia are a consistent feature of the KE family, participants from a separate multigenerational CAS family did not show any significant basal ganglia abnormalities (Liégeois et al., 2019). Complicating this picture further is the fact that only a small proportion of CAS cases present with a clear neurogenetic basis (Morgan and Webster, 2018).
Electroencephalography (EEG) studies have reported abnormal auditory mismatch responses to phonemic contrasts in speech stimuli (Froud and Khamis-Dakwar, 2012) and abnormal speech movement-related brain potentials (Preston et al., 2014).
Therapeutic interventions have been reported to result in significant volume changes in both cortical gray matter (Kadis et al., 2014) and white matter fiber tracts (Fiori et al., 2021).
Neuroimaging and neurophysiological studies of CAS
As it currently stands, this body of research constitutes solid support for both structural and functional anomalies within broadly defined speech and language brain networks but provides little further specificity for mechanisms that may directly contribute to the core praxic features of CAS. We aimed to constrain this highly underspecified picture with the first magnetoencephalographic (MEG) neuroimaging data from a well-defined sample of children with idiopathic CAS, TD controls, and healthy adults. Drawing from a larger evidence base that has recently implicated primary motor cortical mechanisms in acquired forms of AOS (see review by Miller and Guenther, 2021), we tested the hypothesis that the core features of CAS may arise from dysfunctional mechanisms in the proximity of the precentral gyrus. Primary speech motor function was interrogated using a speech task (Riecker et al., 2000; Anastasopoulou et al., 2024) with minimal requirements for linguistic and attentional/memorial processing (Frankford et al., 2021) and an emphasis on sensorimotor phonological and phonetic sequencing operations (Riecker et al., 2000).
Materials and Methods
Participants and group assignment
Nineteen TD children (11 males; mean age, 11.0 years; SD, 2.5; range, 7.5–16.7) and seven children with CAS (7 males; mean age, 8.9 years; SD, 2.2; range, 6.8–12.8) participated in the study. For statistical analyses of demographic and clinical data, seven individuals (four males, three females) of the TD group were age-matched (AM group) to the individuals of the CAS group. All children were right-handed as assessed by a short version of the Edinburgh Handedness Questionnaire (Veale, 2014). All procedures were approved by the Macquarie University Human Subjects Research Ethics Committee.
To provide a further developmental contrast for the TD children, spectrotemporal analyses were carried out in a reanalysis of data acquired in a separate experiment (Anastasopoulou et al., 2024) from a group of 10 adult participants (six males; mean age, 32.5 years; range, 19.7–61.8; all right-handed) and one additional adult participant (male, 63.2 years) not included in the dataset described in Anastasopoulou et al. (2024). All experimental tasks and procedures for the adult (AD) group were identical to those described below for the children, with the exceptions that (1) structural brain scans were obtained (and coregistered with MEG for source reconstruction) for adults but not child participants and (2) only the child groups participated in speech and motor assessments described in the following.
CAS diagnosis
A CAS diagnosis required (1) the three features established by consensus in the ASHA Technical Report (2007) and (2) a minimum of four out of the 10 features outlined in Strand's 10-point checklist (Shriberg et al., 2009; Murray et al., 2015) across at least three assessment tasks. Two certified speech–language pathologists (authors I.A. and K.B.) independently reviewed and rated audio–video recordings of speech productions obtained during administration of the Single-Word Test of Polysyllables (Gozzard et al., 2008).
Speech and motor assessments
All children were screened for pure-tone hearing thresholds. Caregivers completed developmental coordination (DCDQ; Wilson et al., 2009) and handedness questionnaires (Veale, 2014). Language, speech, and motor skills were assessed as follows: receptive and expressive language with the Clinical Evaluation of Language Fundamentals Screening Test (CELF-5 Screening Test; Wiig et al., 2013); word and sentence productions with the Sounds-in-Words and Sounds-in-Sentences subtests of the Goldman–Fristoe Test of Articulation (GFTA-3, Goldman and Fristoe, 2015); and oral motor and sequencing functions in speech and nonspeech tasks with the Verbal Motor Production Assessment for Children (Hayden and Namasivayam, 2021), with five subtests, Global Motor Control (GMC%), Focal Oromotor Control (FOMC%), Sequencing (SEQ%), Connected Speech (CSP%), and Speech Characteristics (SPC%). Nonspeech fine and gross motor skills (manual dexterity, ball skills, and balance) were assessed with the MABC-2 (Henderson et al., 2007).
MEG recordings
Neuromagnetic brain activity was recorded with a KIT-Macquarie MEG160 (Model PQ1160R-N2, KIT) whole-head MEG system consisting of 160 first-order axial gradiometers with a 50 mm baseline (Kado et al., 1999; Uehara et al., 2003). MEG data were acquired with analog filter settings of 0.3 Hz high-pass, 200 Hz low-pass, 1,000 Hz sampling rate and 16 bit quantization. Measurements were carried out with participants in supine position in a magnetically shielded room (Fujihara). Five head position indicator coils (HPI) were attached to an elastic cap on the head. HPI positions were measured at the beginning and at the end of the experiment, with a maximum displacement criterion of <5 mm in any direction.
Head shape templates
Individual structural MRI scans were not available for the child participants in this study so a surrogate MRI approach was used which warps a template brain to each subject's digitized head shape using the iterative closest point algorithm implemented in SPM8 (Litvak et al., 2011) and the template scalp surface extracted with the FSL toolbox (Jenkinson et al., 2012); see also Cheyne et al., 2014; Johnson et al. (2020) for application with child participants. Participant's head shapes and fiducial locations were digitized prior to MEG recordings with a stylus digitizer (Polhemus FastTrack). For adult participants, MEG data were coregistered with individual T1-weighted anatomical MRIs acquired in a separate scanning session using a 3 T Siemens Magnetom Verio scanner with a 12-channel head coil.
Audio speech recordings
Time-aligned audio speech recordings were recorded in an auxiliary channel of the MEG setup with the same sample rate (1,000 Hz) as the MEG recordings and were used to identify speech onset/offset events for use in MEG source reconstruction analyses.
An additional high-fidelity speech recording was simultaneously recorded with an optical microphone (Optoacoustics) fixed on the MEG dewar at a distance of 20 cm away from the mouth of the speaker and digitized with a sound card at a 48 kHz sample rate and 24 bit quantization precision. All participants were also fitted with four MEG-compatible speech movement tracking coils (Alves et al., 2016) placed on the upper and lower lips, tongue body, and jaw (Anastasopoulou et al., 2022, 2024). Analyses of the speech tracking and high-fidelity acoustic data are presented in separate reports (of data from adult participants; Anastasopoulou et al., 2022, 2024, and in forthcoming reports of data from child participants) and are not further discussed here.
Experimental design
The overall experiment consisted of a speech condition and a manual button press condition. For the speech condition, participants performed a “reiterated nonword speech task” which limits requirements for semantic, syntactic, and attentional processing (Frankford et al., 2021) and emphasizes motoric sequencing operations (Riecker et al., 2000). Previous fMRI (Riecker et al., 2000) and MEG (Anastasopoulou et al., 2024) studies of healthy adults have shown that the reiterated nonword task elicits spatially focal brain activations restricted to motor speech centers in or near the peri-Rolandic sensorimotor cortices. In contrast, commonly used expressive speech mapping protocols such as sentence reading, semantic word judgments, picture naming, and verb generation necessarily invoke memorial and cognitive/linguistic operations in addition to phonological, phonetic, and sensorimotor operations, and these tasks typically elicit widely distributed activations in broad regions of prefrontal, temporal, and parietal cortex; see Munding et al. (2016) and Agarwal et al. (2019) for reviews of fMRI and MEG expressive speech mapping studies.
The speech task protocol is illustrated in Figure 1A. Following previously published protocols (van Lieshout et al., 2007; Anastasopoulou et al., 2022, 2024), participants were required to produce utterances of the disyllabic V1CV2 sequences /ipa/ and /api/ at a constant rate during the course of a single exhalation of a deep breath intake. Two speech rate conditions (normal and faster) were used for each production, for a total of four speech conditions: /ipa/ normal rate, /ipa/ faster rate, /api/ normal rate, and /api/ faster rate. These disyllabic nonwords have been used in previous studies investigating speech motor control strategies in normal and in disordered populations (van Lieshout et al., 2002, 2007; van Lieshout, 2017) and were selected here for use in measuring intra- and intergestural coordination in the speech tracking data (see above, Audio speech recordings), as tongue and lip gestures for /ipa/ and /api/ are reversed in phase, providing a contrast of interarticulator coordination (van Lieshout et al., 2007; Anastasopoulou et al., 2022). Similarly the speech rate is used in these studies as a control variable to examine the intrinsic stability of the coordination (Kelso et al., 1986; van Lieshout, 2017).
Experimental procedures. A, Speech task. Instructions were displayed for 30 s, followed by an intertrial interval lasted for 15 and 5 s fixation cross and breath intake in preparation for the speech production trial set. During a trial set, participants produced the indicated nonword in a reiterated fashion for 12 s. Ten consecutive trial sets were performed for each nonword stimulus. B, Button press task. Instructions were displayed for ∼30 s followed by a fixation cross, during which participants performed self-paced button pressed with the index finger of their dominant (right) hand at a rate of ∼1 per 2 s for a total of ∼90 trials.
Each participant generated 10 “trial sets” (van Lieshout et al., 2007) for each speech condition, with each trial set lasting ∼12 s. For each trial set, participants were instructed to take a deep breath and, for the normal rate production, to reiteratively utter the nonwords in a comfortable, conversational rate during the course of a single breath exhalation. For the faster rate, they were instructed to produce the nonwords at a faster rate while maintaining accuracy (van Lieshout et al., 2002). Each 12 s trial set resulted in ∼10 separate nonword utterances at the normal rate and ∼15 at the faster rate. An intertrial set rest interval of 4 s was terminated by instructions (1 s) for fixation and breath intake for the next trial set. A short break was provided after performance of 10 trials sets of the same speech condition. Participants were instructed to minimize head movement and avoid blinking during speech trial sets.
For the manual task (Fig. 1B), participants performed a button press task on a fiber-optic response pad (Current Designs) with the index finger of the dominant (right) hand at a self-paced rate of ∼1 per 2 s for 180 s, resulting in a total of ∼90 button press trials. In previous studies of manual motor control, this task has been shown to elicit robust neuromagnetic responses that can be localized in a straightforward manner to the hand region of the primary sensorimotor cortex (Cheyne et al., 2014; Johnson et al., 2020; De Nil et al., 2021). This well established response provides a useful functional–anatomic reference landmark for the results of the MEG speech analyses (Anastasopoulou et al., 2024).
MEG analyses
Source reconstruction
Speech task. Speech trial set onsets were identified and marked from the speech channel of the MEG recordings. MEG data were segmented with an epoch of −10 to +5 s from the onset of each trial set, selected to encompass the final 5 s of the previous trial set, the 5 s intertrial set rest period, and the first 5 s of the current trial set (speech–rest–speech). Preliminary source analyses showed no significant difference in source location for different productions (/ipa/ vs /api/) or speech rates (faster vs slower). Accordingly, MEG data from all four speech conditions were averaged to maximize the signal-to-noise ratio, for a total of 40 trial sets/averaged epoch (4 speech conditions * 10 trial sets). All epoched data were digitally filtered with a bandpass of 0.3–100 Hz and a 50 Hz notch filter.
Source reconstruction was performed using the scalar synthetic aperture magnetometry (SAM) beamformer implemented in the BrainWave MATLAB toolbox (Jobst et al., 2018). Beamforming is a spatial filtering technique originally developed for applications in radar and communications (Godara, 1997) and subsequently applied to EEG and MEG (Van Veen et al., 1997; Robinson and Vrba, 1999; Sekihara et al., 2001). A beamformer processes signals from an array of sensors (e.g., radar antennae, EEG electrodes, MEG sensors) and applies weighted combinations of these signals to enhance activity from a given location in space while suppressing contributions from other locations. In MEG source reconstruction, a three-dimensional grid (in this case, with 4 mm spacing) is defined to encompass the brain volume and an independent set of weights is defined for each cube within the grid. Beamformer weights are computed with an optimization algorithm that maximizes contributions from a location of interest and supresses signals from other locations. A volumetric image of reconstructed source activity is then obtained by combining the output of spatial filters over the entire brain volume, and these images can be transformed to standardized space for group averaging and statistical analyses (Singh et al., 2003). Furthermore, the outputs of spatial filters at each grid volume location in the brain (termed “virtual sensors”) can be used to plot the dynamics of source activity over a given time window and within a given frequency band.
The SAM noise-normalized (pseudo-t) beamformer images were computed using a frequency range of 18–22 Hz (center of the beta frequency band), a sliding active window of 0–1.0 s (first second of current speech trial set), and a baseline window of −5 to −3 s (first 2 s of intertrial set rest period) over 10 steps with a step size of 0.2 s and pseudo-z beamformer normalization [3 ft/sqrt (Hz) RMS noise]. While beamforming analyses typically employ equal baseline and active window durations, in the present analyses we wished to use a longer baseline epoch to reduce the chance of a biased estimate of baseline power, which we reasoned is more likely to vary over time than during the active speech period (since reiterated speech is akin to a steady state). While larger-order mismatches (e.g., comparing 1.0 to 0.1 s) are likely to be problematic, simulation studies (Brookes et al., 2008) have shown that covariance errors are minimized, and therefore beamformer weights are stable, with at least 5 s of total data (a requirement that is well exceeded in the present case), and our preliminary analyses confirmed that a 2 s baseline did not affect the beamforming results relative to a 1 s baseline. The combined active and control windows were used to compute the data covariance matrix for beamformer weight calculations, while the full 15 s time window was used to compute data covariance and beamformer weights to extract single-trial time courses of localized neural activity (“virtual sensors”) for each trial.
Statistical analysis of group beamformer images was performed with cluster-based permutation testing (α = 0.05, 512–2,048 permutations, omnibus correction for multiple comparisons).
Source reconstruction results of adult group speech and button press responses have been previously described by Anastasopoulou et al. (2024) in a study using identical experimental methods and procedures to those described here for the child participants.
Manual task
MEG data were segmented into 1.5 s epochs comprising −0.5 to +1.0 s with respect to the button press onset, resulting in ∼90 button press trials. Epoched datasets were digitally filtered from 0.3 to 100 Hz and a 50 Hz notch filter. The brain activity metrics of interest in the present study were ongoing cerebral rhythms: in the context of these oscillatory phenomena, event-related responses are manifest as increases (termed “event-related synchronization”, ERS) or decreases (termed “event-related desynchronisation”, ERD) of power within a given frequency band (Pfurtscheller and Lopes da Silva, 1999). Accordingly, time windows here were chosen to encompass the known periods of maximal ERS and ERD, respectively, of MEG motor rhythms in a button press task (Cheyne et al., 2014; Johnson et al., 2020). SAM source images were computed using a frequency range of 15–25 Hz (beta band), a sliding active window of 0.6–0.8 s, and a baseline window of −0.5 to −0.3 s over 10 steps with a step size of 0.01 s and pseudo-z beamformer normalization [3 ft/sqrt (Hz) RMS noise]. Statistical analysis of group beamformer images was performed with cluster-based permutation testing (2,048 permutations, omnibus correction for multiple comparisons).
Spectrotemporal analyses
Speech task. Time–frequency plots were computed from virtual sensor locations at the center of the mean beamformer map clusters obtained as described above and shown in Figure 2. Bilateral clusters were obtained for all groups for the manual task and for the TD group for the speech task, allowing a comparison of hemispheric responses computed from a virtual sensor in each hemisphere. In order to assess hemispheric responses and lateralization where only unilateral clusters were obtained (in the speech task for AD [single left hemisphere (LH) cluster] and CAS [single right hemisphere (RH) cluster] groups), a virtual sensor location in the contralateral hemisphere was designated by reversing the X-coordinate (left–right axis) of the obtained cluster.
Group mean SAM beamformer maps (left panel) and group statistical maps (right panel) for speech and manual tasks. Statistical maps show cluster-based permutation analysis of SAM beamformer source maps, cluster α = 0.05, 512–1,024 permutations, omnibus correction for multiple comparisons. Axial views are shown here to facilitate group and task comparisons; coronal and sagittal views are shown in Extended Data Figure S1. L, left; R, right; A, anterior; P, posterior.
Time–frequency plots were generated with a time range of −10 to +5 s from the trial set onset and a frequency range of 1–100 Hz. Time–frequency data for each individual were subsequently partitioned and frequency averaged over three frequency bands: beta (15–25 Hz), mu (8–12 Hz), and theta (4–7 Hz). Based on the group-averaged time–frequency spectrograms, ERS and ERD time epochs were chosen as −5 to −3 s and −1 to 1 s, respectively, for beta and mu bands and −1 to 1 s and −5 to −3 s, respectively, for theta band.
Manual task
Voxel locations at the center of group mean clusters were used to generate virtual sensor time–frequency plots with a time range of −0.5 to +1.0 s from the button press onset and a frequency range of 1–100 Hz. Based on the group-averaged time–frequency spectrograms, ERS and ERD time epochs were chosen as −0.2 to 0 s, and 0.5 to 0.7 s respectively, for beta and mu, and 0.5 to 0.7 s and −0.2 to 0 s respectively, for theta band.
Response magnitude
For the purposes of the statistical analyses described below, a metric of “response magnitude” was defined as the average magnitude over the ERS epoch minus the average magnitude over the ERD epoch.
Statistical analyses
Between-group differences in response magnitude. For each task (button press, speech), hemisphere (left and right), and frequency band (theta, mu, beta), Kruskal–Wallis nonparametric one-way ANOVA (MATLAB kruskalwallis function) was computed to determine whether there was a significant group difference in median response magnitudes using a critical α = 0.05. Where a significant group difference was indicated by the results of the overall ANOVA, planned comparisons (CAS vs TD; TD vs AD) were performed using the nonparametric Wilcoxon rank-sum test (MATLAB ranksum function) for equal medians, using the Bonferroni’s correction for multiple comparisons. Where a significant CAS:TD group difference was obtained, a further Wilcoxon rank-sum comparison between CAS and AM groups was performed to assess whether the CAS:TD group difference still held when members of the TD group were age-matched to the CAS group.
Within-group hemispheric differences in response magnitude. For each group (CAS, TD, AD) and frequency band (theta, mu, beta), hemispheric differences were tested with nonparametric Wilcoxon signed-rank test for zero median.
Between-group differences in hemispheric lateralization. For each participant, a laterality index (LI) was computed from response magnitudes in LH and RH, according to the formula (LH − RH)/(LH + RH). LI values range from 1 to −1, where 1 indicates fully left-lateralized, −1 indicates fully right-lateralized, and 0 indicates fully bilateral. Statistical analyses were then performed as described for response magnitude above.
Effects of age within the TD group. For each task (button press, speech), hemisphere (left and right), and frequency band (theta, mu, beta), a linear regression model (MATLAB fitlm function) was used to test for linear effects of age within the TD group.
Results
CAS diagnosis
Of ten children initially enrolled in the study with suspected CAS, two who did not meet the diagnostic criteria (see Materials and Methods), and one left-handed child were excluded from further analyses. The two clinical raters agreed on a final diagnosis of CAS for seven children. Clinical evaluations agreed on the following CAS features: problems with syllable segregation (4 out of 7 children), lexical stress errors (5/7), vowel or consonant distortions (7/7), slow rate (3/7), increased difficulty with more complex words (7/7), inconsistent errors (7/7), problems with coarticulatory transitions between sounds (6/7), and inappropriate prosody (6/7). The inter-rater reliability of clinical features was 87.5%.
Language, speech, and verbal motor skills
Table 2 summarizes demographic and test performance data for the TD, AM, and CAS groups; results for individual participants are provided Extended Data Table S1. Nonparametric Wilcoxon rank-sum tests (two-sided) were employed for statistical comparisons of the CAS and AM groups. Children with CAS demonstrated significantly poorer performance than AM peers with typical development on the GFTA-3 for both words (p = 0.001) and sentences (p < 0.001). The CAS group also scored significantly lower on all verbal motor (VMPAC) subtests: GMC% (p = 0.005), FOMC% (p < 0.001), SEQN% (p = 0.002), CSP% (p < 0.001), and SPC% (p < 0.001). The results of the CELF-5 Screening Test showed no significant indication for impairment of expressive and receptive language abilities in the CAS group (p = 0.197).
Demographic data and clinical test scores by group
Co-occurring conditions
(In the present manuscript we use “co-occurring” with the same meaning as “comorbid” as conventionally employed in the medical literature. See https://waterloofoundation.org.uk/ChildDevelopmentResearch.html for discussion of this issue.) Pure-tone hearing thresholds were within normal limits for all children with CAS. All seven children with CAS exhibited articulation deficits (GFTA-3 standard score of <85), and three (43%) showed expressive and receptive language deficits based indicated by the CELF-5 Screening Test. As noted above however, the overall group comparison with AM peers did not reach statistical significance for the CELF-5 Screening Test [note that the AM subset of TD children had a lower median CELF-5 Screening Test score (8) compared with the full TD group (23). This is attributable to the younger composition of the AM group (mean age 8.53 years) relative to the full TD group (mean age 11.06 years), since the CELF-5 Screening Test criterion score increases with age]. None of the CAS children exhibited speech dysfluencies (such as stuttering) in any experimental tasks or clinical assessments, and no dysfluent behaviors were reported by parents or caregivers. The CAS group performed significantly worse on the Movement Assessment Battery for Children (MACB-2; p = 0.019), with six out of seven (86%) showing motor coordination deficits indicative of probable Developmental Coordination Disorder (DCD); caregivers of five out of these six children reported motor coordination issues in the DCDQ. These results align with previous findings highlighting a high rate of co-occurrence between CAS and DCD (And and Dodd, 1996; McCabe et al., 1998; Duchow et al., 2019; Knežević, 2019; Iuzzini-Seigel, 2021; Iuzzini-Seigel et al., 2022).
MEG data
Speech task
Source reconstruction. Group mean and corresponding statistical maps of SAM beamformer images for the speech condition are shown in Figure 2 (left panels; Fig. 2 shows axial view to facilitate group comparisons; sagittal and coronal views are also shown in Extended Data Fig. S1) and cluster coordinates are provided in Table 3. Source reconstruction indicates strikingly contrasting group results for the speech task. While source clusters for all three groups are aligned near the coronal midline (adjacent to the Rolandic fissure), adults show a single cluster in the LH, and TD clusters are bilateral, while the CAS group shows a single significant cluster in the RH. We note that caution is warranted for interpretation of statistical maps for the small N of the CAS group; however the obtained RH cluster location corresponds well to that obtained for the TD group.
Virtual sensor Talairach coordinates (mm)
Virtual sensors. Group mean virtual sensor time–frequency plots were generated from the speech task coordinates listed in Table 3 and are plotted in the left panels of Figure 3. Note here that the speech data time–frequency plots are baselined to the time epoch of −5 to −3 s (during the nonspeaking rest period) in order to enhance visualization of beta-band response over the entire 15 s epoch: as a consequence a prominent rebound ERS during the rest interval (Fig. 4) is shown scaled here to blue (ERD): ERD/ERS are better visualized in the beta frequency time course plots of Figure 4. The simpler time–frequency characteristics of the manual button press responses (Fig. 3, right panels) do not present the same scaling issues as the speech data, and both ERD and ERS are readily visualized with conventional baselining (over the entire time epoch).
Group mean virtual sensor time–frequency spectrograms. All virtual sensor plots correspond to the Talairach coordinates of Table 3. Speech task epochs encompass two 5 s speech trials beginning at times −10 and 0 s, separated by a 5 s resting epoch from times −5 to 0 s. Manual task epochs are aligned to button press onset at time zero. Manual task data are baselined to the full epoch, while speech task data are baselined −5 to −3 s (during the rest period) to optimize the visual appearance of beta-band activity over the entire epoch. Note that the rest period baselining of the speech data reverses the polarity (indicates “desynchronization”) of a strong beta-band rebound “synchronization” which is better visualized in Figure 4.
Group comparison of frequency-averaged time series showing beta-, mu-, and theta-band responses elicited by speech and manual tasks in right and LHs. Beginning and end of time series are truncated to remove edge artifacts from Fourier decomposition (Fig. 3). Shaded regions show ERS and ERD time windows. LH, left hemisphere; RH, right hemisphere; BP, button press onset; OFF, speech trial set offset; ON, speech trial set onset. Scale bar indicates 25% change from the baseline.
The following features can be observed in these source-derived temporal spectrograms: (1) substantial high-frequency broadband noise during the speech trial sets, attributable to myogenic activity that is an inevitable artifact of the overt speech task, and (2) prominent mu-/beta-band (∼10–25 Hz) desynchronization that is continuous during the speech trial sets and also for ∼2 s prior to the speech trial set onset, reflecting motor preparatory activity that is characteristic of the mu/beta bands (Cheyne, 2013). For adults, mean beta-/mu-band magnitude is notably lower in the right relative to the LH, lower in magnitude for TD relative to adults, and yet lower in magnitude for CAS relative to TD. (2) Theta-band (circa 3–7 Hz) synchronization which roughly tracks the time course of the mu/beta desynchronization. (For aid in interpretation, we note that the speech spectrograms are baselined to the first 2 s of the intertrial set rest period, −5 to −3 s. This baseline affects the visual appearance of the spectrogram during the immediately preceding time period, where the high-frequency noise and mu-/beta-band desynchronization appear to stop ∼1 s before the end of the speech trial set.)
Group mean comparisons of frequency-band averaged time series (Fig. 4) further confirm that the three groups exhibit distinct patterns of brain activity during speech movements. For the beta and mu bands, adults show strikingly larger ERS magnitudes (during the nonspeaking intertrial set rest period) than TD in both hemispheres, although TD do show speech-offset–related ERS in the RH. No comparably distinct ERS or ERD epochs are apparent in the CAS average waveforms.
Response magnitude (within hemisphere)
Group comparisons of median speech response magnitudes are summarized in Table 4 and Figure 5. Nonparametric Kruskal–Wallis one-way ANOVAs indicated significant group differences in beta-band response magnitude in the LH (AD Mdn = 23.14; TD Mdn = 6.20; CAS Mdn = 0.00; χ2(2,33) = 15; p < 0.001) but not for the RH (AD Mdn = 4.01; TD Mdn = 8.83; CAS Mdn = 4.27; χ2(2,33) = 2.64; p = 0.267). Planned comparisons for the LH confirmed significantly greater response magnitude in AD than TD (p < 0.001) but no significant difference between CAS and TD (p = 0.298).
Group comparison of response magnitudes for beta, mu, and theta bands elicited by speech and manual tasks in RH and LHs. Horizontal lines indicate median magnitude; boxes indicate interquartile range; whiskers indicate range. Note larger y-scale range for manual task than for speech task. ***p < 0.001; *p < 0.05.
Response magnitudes and lateralization index
Mu-band responses in both hemispheres show an AD > TD > CAS group ordering of median magnitudes. Kruskal–Wallis one-way ANOVA confirmed a significant overall group difference in both the left (AD Mdn = 17.05; TD Mdn = 5.55; CAS Mdn = 0.00; χ2(2,33) = 11.26; p = 0.004) and RHs (AD Mdn = 14.16; TD Mdn = 6.42; CAS Mdn = 0.00). Within the LH, planned contrasts confirmed a significant group difference between AD and TD (p = 0.023) and between TD and CAS (p = 0.018). Furthermore, the contrast between CAS and the AM subset of TD (Mdn = 22.56) was also significant (p = 0.017). In the RH, however, neither planned contrast achieved statistical significance (CAS vs TD, p = 0.078; TD vs AD, p = 0.280).
For the theta band, Kruskal–Wallis one-way ANOVAs showed no significant group differences in response magnitude in either hemisphere.
Hemispheric differences (within group)
Wilcoxon signed-rank tests showed significant hemispheric differences in beta-band responses for adults (LH Mdn = 23.14; RH Mdn = 4.01; p = 0.012), TD (LH Mdn = 6.20; RH Mdn = 8.83; p = 0.045), and the AM subset of TD (LH Mdn = 3.15; RH Mdn = 11.67; p = 0.031) but not for CAS (LH Mdn = 0.00; RH Mdn = 4.27; p = 0.875).
No significant hemisphere differences in mu- or theta-band response magnitudes were obtained for any of the three groups.
Lateralization index (between group)
For assessment of between-group differences in hemispheric lateralization, LI (as described in Materials and Methods) was computed from response magnitudes in each hemisphere. Figure 6 (left panel) summarizes LI for the speech task. Kruskal–Wallis one-way ANOVA indicated a significant group difference for beta lateralization index (AD Mdn = 0.50; TD Mdn = −0.17; CAS Mdn = 0.00). Wilcoxon rank-sum planned contrasts confirmed the AD group was significantly more left-lateralized than TD (p = 0.006), but there was no significant difference for TD and CAS groups (p = 0.652).
Group comparison of hemispheric lateralization index for beta-, mu-, and theta-band responses to speech and manual tasks. *p < 0.05.
ANOVAs showed no significant overall LI group differences for mu band (AD Mdn = 0.020; TD Mdn = −0.32; CAS Mdn = 0.000; p = 0.40) or theta band (AD Mdn = 0.06; TD Mdn = 0.00; CAS Mdn = 0.14; p = 0.537).
Effects of age in TD children
Linear regression modeling indicated significant linear effects of age within the TD group only in the theta-band response for the speech task (Extended Data Fig. S2). For the speech task, theta response magnitude showed a significant linear increase with age in the RH (F(1,17) = 0.015; R2 = 0.300; p = 0.015). There was no significant effect of age in the LH (p = 0.375), and theta LI showed a significant decrease (i.e., increasing right lateralization) with age (F(1,17) = 6.528; R2 = 0.277; p = 0.021). Overall, within the 7–16 years range of the TD group, significant (albeit small) age effects were obtained only in theta-band speech responses, with no indication of any (linear) maturational trajectory that could account for the prominent differences between the TD and adult mu-/beta-band speech responses.
Manual task
Source reconstruction. Group mean and corresponding statistical maps of SAM beamformer images for the manual condition are shown in Figure 2 (right panel). Consistent with the previous literature, all groups showed maximal responses in the regions of the left precentral gyrus (contralateral to the right-handed button press) with center coordinates (Table 3) located at or near the hand region of the hand knob of the precentral gyrus. Also consistent with previous work, all groups also showed mirrored RH activations in homologous regions of the RH motor cortex (data not shown), although these clusters were smaller in magnitude and did not reach statistical significance for any of the groups.
Virtual sensors
Group mean virtual sensor time–frequency plots were generated from the button press coordinates listed in Table 3 and are plotted in the right panels of Figure 3. The following features can be observed in the plots for all three groups: (1) beta-band (circa 13–30 Hz) desynchronization beginning ∼300 ms before the button press and persisting for ∼300–400 ms after the button press; (2) beta-band synchronization (beta “rebound”) beginning ∼500–600 ms after the button press and persisting for 300–400 ms; theta-band (circa 3–7 Hz) synchronization beginning ∼200–300 ms prior to the button press and persisting until 500–600 ms postbutton press. Comparable but lower magnitude spectral perturbations are also evident in the RH spectrograms. All of these temporal–spectral features are known characteristics of neuromagnetic brain responses in a self-paced button press task for both adults and children and have been described in previous publications (Cheyne et al., 2014; Johnson et al., 2020; Fung et al., 2022). Group comparisons of frequency-band averaged time series (Fig. 4) further confirm that the three groups exhibit entirely comparable brain activity during manual movements.
Response magnitude (within hemisphere)
Further statistical analyses of event-related spectral perturbations were carried out using the response magnitude metric (see Materials and Methods). Group comparisons of median response magnitudes are summarized in Table 4 and Figure 5. Nonparametric Kruskal–Wallis one-way ANOVAs showed no significant group differences in response magnitude within either hemisphere, for any of the three frequency bands.
Hemispheric differences (within group)
Hemispheric differences were assessed within each group using nonparametric Wilcoxon signed-rank tests (two-sided). Results showed that beta-band response magnitudes were significantly higher in the LH for AD (N = 11; L Mdn = 61.77; R Mdn = 12.85; W = 34; p = 0.023) and TD (N = 19; L Mdn = 42.49; R Mdn = 3.12; W = 153; p < 0.001). While median LH response magnitudes were also greater in the CAS group, the hemisphere difference did not reach statistical significance (N = 7; L Mdn = 28.78; R Mdn = 6.70; W = 15; p = 0.438) but did for the AM subset of TD (N = 7; L Mdn = 21.31; R Mdn = 15.39; W = 28; p = 0.016).
In the mu band, no significant hemispheric differences in response magnitude were obtained for any group, while in the theta band, a significant hemispheric difference in response magnitude was obtained only for the TD group (N = 19; L Mdn = 13.55; R Mdn = 5.54; z = 2.656; W = 161; p = 0.008).
Lateralization index (between group)
Nonparametric Kruskal–Wallis one-way ANOVA showed no significant group differences for beta-band LI, with all groups showing a left-lateralized profile (AD Mdn = 0.61; TD Mdn = 0.48; CAS Mdn = 0.35; χ2(2,33) = 2.67; p = 0.263). Kruskal–Wallis one-way ANOVA indicated a significant group difference for mu LI, with only adults showing a left-lateralized response (AD Mdn = 0.73; TD Mdn = 0.00; CAS Mdn = 0.01; χ2(2,33) = 6.44; p = 0.040). Wilcoxon rank-sum planned comparisons confirmed significantly greater left lateralization for AD compared with TD groups (p = 0.013), but there was no significant difference between TD and CAS groups (p = 0.952). No significant group differences were obtained for theta-band LI, and all groups showed a left-lateralized profile (AD Mdn = 0.57; TD Mdn = 0.51; CAS Mdn = 0.26; χ2(2,33) = 2.94; p = 0.230).
Discussion
The present results provide the first evidence for dysfunction in the neurophysiological generators of the mu-band speech motor rhythm in a sample of children with clinically diagnosed CAS. Our analyses specify a brain origin of this speech rhythm in the left cerebral hemisphere, within or near pre-Rolandic motor areas crucial for the planning and control of speech and oromotor movements. We further refine this anatomical specification by demonstrating the proximity of speech-elicited beamformer maps to clusters associated with manual button press responses known to originate in the hand area of the peri-Rolandic sensorimotor cortex. Taken together, these results provide clear functional–anatomical delineation of neural activity as a candidate biomarker of childhood speech apraxia.
Summary of speech task source reconstruction results
The three groups showed distinct patterns of brain activity for the speech task. While SAM beamformer clusters indicated activations in proximity to the hand task clusters near the peri-Rolandic sensorimotor cortex, adults were significantly more (left) lateralized than TD. We obtained no significant difference in overall hemispheric lateralization between TD and CAS for any of the three frequency bands; however while TD showed smaller magnitudes of speech-related beta/mu responses than AD, these were virtually absent in the CAS children. A group ordering of response magnitudes was apparent for the left hemispheric mu-band response, which was significantly lower in magnitude relative to both the overall TD group and the subset of AM TD.
Relative to the adults, the child groups showed speech-elicited clusters that were deeper and more ventral. Group discrepancies in the precise locations of clusters are difficult to interpret from the present data. On the one hand, good between-group agreement of source locations for the manual task supports adequacy of source modeling for all groups: in this case all groups showed good agreement for x and y coordinates, while the z discrepancy is explainable from the more extensive cluster configuration in the adults. On the other hand, group discrepancies may be expected to arise from three sources in the present experiment: first, adult source modeling was based on individual MRI scans, while the child models were derived from template brains; second, the cluster-based analysis indicates both more extensive and stronger (larger magnitude) sources for the adults than for the child groups, a difference in signal-to-noise ratios that may affect model comparisons; and third, it is known that adaptive beamformers perform suboptimally in the case of correlated bilateral sources (as in the TD group), since linear dependencies between the neuronal source time series are utilized by these algorithms to minimize the output power (Kuznetsova et al., 2021).
Overall, these considerations preclude any clear inferences concerning adult and child group differences in the precise locations of speech-related sources. Within the stated limitations, the results of the statistical cluster-based analyses suggest the following: for adults, a differential SAM beamformer effect (p < 0.05) corresponds to a single cluster in the observed data at a spatial location in the precentral gyrus of the LH immediately inferior to the known anatomic location of the hand region of sensorimotor cortices and coextensive with the region of the medial frontal gyrus that is immediately anterior to this middle region of the precentral gyrus; in contrast, for the TD group, the differential SAM beamformer effects (p < 0.05) correspond to bilateral clusters in both cerebral hemispheres, at relatively lateralized locations on a midline (in the sagittal plane) roughly corresponding to the location of the central sulcus/pre- and postcentral gyri in both hemispheres, while for the CAS group, the significant beamformer effects correspond to a single cluster in the RH corresponding roughly to the location of the RH cluster of the TD group.
Summary of manual task source reconstruction results
In contrast to the speech task results, all three groups showed entirely comparable patterns of brain activity during the manual movement task, with SAM beamformer clusters indicating activation at or near the contralateral hand region of the sensorimotor cortex and with smaller (and statistically nonsignificant) activations in homologous regions of the RH. Spectrotemporal patterns computed from these regions were prominently characterized by a robust pattern of event-related beta–band desynchronization prior to and following the button press movement and followed after ∼500–600 ms by a “rebound” beta-band synchronization. Within the theta band, the spectral response followed a comparable time course with opposite polarity (i.e., theta ERS during mu/beta ERD and theta ERD during mu/beta ERS). As the only significant group difference, adults showed greater LH lateralization of mu-band activity than TD, although mu-band activity was notably small in magnitude in all groups and both hemispheres. Overall, we conclude that the cluster-based permutation analyses show a significant differential SAM beamformer effect for all three groups, with clusters in the observed data at a spatial location corresponding to the known anatomic location of the hand region of the sensorimotor cortex in the LH, contralateral to the right index finger used for the button press task. Furthermore, virtual sensor time spectrograms generated from the cluster center locations for each group are entirely representative of and in accord with known and well-characterized neuromagnetic brain responses in a self-paced button press task (Cheyne et al., 2014; Johnson et al., 2020; Fung et al., 2022).
Apraxia of speech
CAS is a clinical subtype of AOS and is differentiated from acquired forms of AOS by a developmental etiology and an absence of frank lesions from acquired brain injuries. Despite differences in clinical presentation due to the etiological dissimilarities, the two subtypes share many apraxic features likely to be attributable to commonalities in their causal neuropathologies (Miller and Guenther, 2021). While differential diagnosis of CAS is an area of active and ongoing research (Murray et al., 2015), features identical to those used for acquired AOS—including problems with syllable segregation, syllable stress, and speech sound distortions—have been proposed and are increasingly used for diagnosis of CAS (Yoss and Darley, 1974; Shriberg et al., 2009, 2017; Murray et al., 2015; Miller and Guenther, 2021). Psycholinguistic and motor control models have long attributed these apraxic features to disruptions of motor programming due to dysfunction at the interface between otherwise intact phonological and motor systems, an interface wherein abstract linguistic codes are transformed into movement commands for interpretation by the motor system (ASHA, 2007).
The current results dovetail neatly with evidence implicating the left precentral gyrus in acquired AOS. In particular, cases of “pure” acquired AOS (i.e., AOS features without other speech–language deficit) have consistently been associated with lesions of the left precentral gyrus (Robin et al., 2008; Graff-Radford et al., 2014; Basilakos et al., 2015; Itabashi et al., 2016; Moser et al., 2016; Takakura et al., 2019). Further evidence for the role of the left precentral gyrus is provided by reports that neurostimulation of this region results in transient disruptions of speech with the main features of AOS (Tandon et al., 2002; Duffau et al., 2003). Finally, neurocomputational lesion modeling studies (Guenther, 2016; Miller and Guenther, 2021) have reported that AOS speech symptomology can only be produced in these models through damage to a putative speech sound map located in the left ventral precentral gyrus and surrounding portions of posterior inferior frontal gyrus and anterior insula. In contrast, lesions to most other components of the model resulted in dysarthria rather than AOS. According to this computational model, AOS symptomology is dependent specifically on damage to the speech sound map in the “left” ventral precentral gyrus which is tasked with feedforward control of speech motor programs (while its RH homolog is proposed to be specialized for “feedback” motor control of corrective motor commands; Guenther, 2016).
The hemispheric division of feedforward and feedback motor control proposed by the DIVA model is further supported by recent evidence from neurophysiological measurements in healthy adults. Mantegna et al. (2025) used MEG measures of mu (8–12 Hz) and low (18–22 Hz) and high (28–32 Hz) beta-band oscillations to assess lateralization of functional connectivity between auditory and motor brain regions. Participants produced the nonwords /pa/, /ta/, and /ka/ covertly. The researchers analyzed a phase-lag metric of functional connectivity during two time windows: one associated with feedforward commands (−200 to 400 ms relative to the expected onset of production) and another corresponding to feedback processing (400–600 ms after the expected onset). Consistent with the DIVA model's predictions, their findings revealed a distinct shift in hemispheric lateralization of mu-band functional connectivity—left-lateralized before and during word production, transitioning to right-lateralized afterward. Notably, and in alignment with the present results showing mu-band–specific deficits in CAS, this alternating lateralization pattern was observed only in the mu band (8–12 Hz) and not in the two beta frequency bands. Thus, the results of Mantegna et al. (2025) specifically implicate the left hemispheric mu-band motor rhythm in feedforward aspect speech motor control: in agreement with the predictions of the DIVA computational model discussed above, the results of the present study further show that dysfunctioning of this rhythm is associated with AOS.
Speech-elicited brain activities
Relative to the extended and distributed patterns of activations that are typically obtained with lexical speech production tasks (Munding et al., 2016; Agarwal et al., 2019), our results show spatially restricted clusters associated with peri-Rolandic regions of the sensorimotor cortex. These loci and their relative focality concur with the fMRI results using a comparable reiterated speech task reported by Riecker et al. (2000). These authors suggested that overlearned, automatic, and repetitive utterances can be effectively “chunked” at a planning level in a manner that places far fewer demands on neural resources than are required for single “individualized” speech gestures. In addition, the lack of lexical content in the nonword utterances removes the need for any access to higher language centers for syntactic processing or lexical retrieval.
Our results and those of Riecker et al. (2000) are entirely supportive of an emerging neuroscientific consensus which assigns a central role in speech motor planning and coordination to speech motor regions of the precentral gyrus and immediately adjacent regions of the prefrontal cortex (Glasser et al., 2016; Silva et al., 2022; Gordon et al., 2023; Jensen et al., 2023; Willett et al., 2023). This new conceptualization is exemplified in the recent “dual motor system” model of expressive speech processing (Hickok et al., 2023), which posits a hierarchical control system consisting of dorsal and ventral portions of the precentral gyrus and the posterior portion of the middle frontal gyrus.
Limitations and future directions
Two important limitations of our CAS data are the small group size, common for published studies of this rare disorder (Table 1), and the presence of several co-occurring neurodevelopmental features, including indications for possible DCD in 6/7 children and indications for expressive and receptive language impairments in 3/7 children. While caution is warranted when interpreting findings from small sample sizes, the present results are supported by a number of important methodological and analytical strengths: core CAS features were characterized with a comprehensive battery of clinical tests and diagnosed independently on the basis of overt speech behaviors by two experienced speech–language pathologists (Chenausky and Tager-Flusberg, 2022); the locus of the speech cluster in the CAS group is in accord with that of the RH cluster of TD children, and the neurophysiological responses of all seven individuals show the absence or very low magnitudes of LH speech-related mu–band activity that is a prominent feature of both the TD and AD groups (Extended Data Fig. S2). Therefore, despite the small sample size, group mu-band magnitude differences were statistically significant for both the CAS-TD group comparison and the CAS-AM comparison. These considerations are indicative of a robust and sensitive neurophysiological marker which has substantial potential for application in studies of typical and apraxic speech development.
Any interpretation of neurophysiological and neuroanatomical anomalies in CAS is complicated by the known prevalence of co-occurring neurodevelopmental disorders (such as autism spectrum disorder) and communication-related problems (most prominently, expressive and receptive language impairments, cognitive impairments, and articulation impairments). Indeed, a retrospective examination of medical records of 375 children with CAS (Chenausky et al., 2023) found that only one of these children had no co-occurring conditions. These authors point out that many of the genes that have been associated to date with CAS act via molecular pathways that can have broad effects on brain development and conclude that CAS is better conceptualized as a component of a wider ranging set of neurodevelopmental problems than an isolated impairment of speech. Notwithstanding this caveat, the central apraxic features of CAS also point to a fairly specific dysfunction of brain mechanisms associated with the ability to program and plan speech movements. As such, this disorder represents an important and currently understudied line of lesion evidence for ongoing investigations of speech neuromotor control. As discussed above, this developmental evidence will strongly complement evidence from acquired forms of AOS in adults, which also come with inherent (but different) inferential limitations including the variable extent of acquired brain lesions and a significant prior history of intact speech.
Our MEG results for the manual task indicate developmentally typical hand motor cortex function in the CAS group, at least for the simple movements required in this task. This provides an intriguing contrast to the neurophysiological anomalies associated with the speech task, particularly in light of our behavioral data indicating significant deficits for nonspeech motor skills. Fine and gross nonspeech motor deficits suggestive of DCD are estimated to occur in some 50–80% of children with CAS (Duchow et al., 2019; Iuzzini-Seigel, 2021; Iuzzini-Seigel et al., 2022), a co-occurrence that has invited considerable speculation about shared mechanisms within the central motor system (And and Dodd, 1996; McCabe et al., 1998; Duchow et al., 2019; Knežević, 2019; Iuzzini-Seigel, 2021; Iuzzini-Seigel et al., 2022). While the present data from a simple button press task are not readily comparable to the sequential and coordinated articulatory movements required in the speech task, the basic experimental framework can be readily adapted for future studies wherein the relative functioning of speech and hand motor cortices can be probed using tasks where movement complexity is better equated (De Nil et al., 2021).
While the primary aim of the present study was to contrast brain function in CAS and TD controls, we have also included analyses of data previously obtained from healthy adults using the same experimental tasks and methods (Anastasopoulou et al., 2024). The significantly larger magnitudes of adult mu- and beta-band speech–related brain responses provide insights into the normative developmental trajectory of the immature activities and further ground these neurophysiological activities within two additional sets of scientific literatures: first, our results confirm those of previous studies reporting that nonspeech-related motor rhythms increase in magnitude over development (Cheyne et al., 2014; Heinrichs-Graham, 2018; Johnson et al., 2020; Rempe et al., 2023) and extend these observations to the speech motor domain. Second, our observation that the immature beta-band response is bilateralized while the adult response is prominently left-lateralized is consistent with previous reports of comparable developmental changes in language-related brain functions (Olulade et al., 2020). The speech-related mu rhythm, in contrast, showed no significant lateralization change across age groups, indicating that individual components of speech neuromotor control may follow distinct developmental trajectories. These observations set the stage for future studies examining speech-related motor rhythms in adults with AOS, investigations which may shed light on the neurophysiological mechanisms that account for shared features of the developmental and acquired types of speech apraxia.
Footnotes
This work was supported by a Waterloo Foundation Child Development Fund Research Grant (Ref. No. 2532-4758) and a Discovery Project Grant (DP170102407) from the Australian Research Council.
The authors declare no competing financial interests.
↵*I.A.’s present address: Ioanna Anastasopoulou, Hospital for Sick Children Research Institute, Toronto, Ontario M5G 0A4, Canada.
- Correspondence should be addressed to Ioanna Anastasopoulou at ioanna.anastasopoulou{at}sickkids.ca or Blake W. Johnson at blake.johnson{at}mq.edu.au.
