Passive OPM-MEG paradigm reveals altered brain network organization in children with DLD

Summary

Optically pumped magnetometer magnetoencephalography (OPM-MEG) enables the study of pediatric neurodevelopment during naturalistic tasks. Here, we investigated functional connectivity (FC) changes in 30 typically developing (TD) children and 10 children with developmental language disorder (DLD), aged 4–7 years, during a rest task (“Inscapes” video) and a story task (short narrative videos). Both groups exhibited similar power modulation (δ enhancement, α, β, and low-γ suppression) and decreased network modularity when transitioning from rest to story processing. However, children with DLD displayed increased high-γ network modularity, indicating a more segregated neural architecture. During the story task, TD children displayed enhanced low-γ FC within a left-hemisphere network with hubs in temporal, parietal, central, and occipital regions, whereas children with DLD showed weaker connectivity. These findings suggest that altered γ-band network organization may be a core neural signature of DLD and highlight the potential of OPM-MEG in pediatric neurodevelopmental disorders.

Graphical abstract

Highlights

• •OPM-MEG enables child-friendly imaging of neurodevelopment using passive paradigms
• •Passive paradigms reveal functional connectivity deficits in developmental disorders
• •Altered gamma-band networks may index impaired language processing in DLD

Health sciences; Internal medicine; Medical specialty; Medicine; Neurology

Introduction

Passive paradigms in language processing research of typical development children

In the study of language processing in children, passive and naturalistic paradigms, such as resting-state scans and listening to stories, have gained significant attention. Compared to traditional task-based experiments, these paradigms better approximate real-world communication, allowing for a more authentic capture of the dynamic processes underlying phonetic, syntactic, and semantic processing.¹

Resting-state electroencephalography (EEG) has been instrumental in developmental neuroscience, revealing the characteristics of spontaneous oscillations in language-related networks.² Different frequency bands are associated with distinct cognitive functions: the theta band with speech rhythm and working memory, the alpha band with attention and sensory inhibition, and the beta and gamma bands with semantic integration and prediction.³ Longitudinal studies have shown that frontal γ power levels in 2- and 3-year-old children can predict their language abilities at ages 4 and 5.⁴

Naturalistic stimuli, such as stories, elicit robust neural tracking of speech rhythms in both children and adults. For example, both infants and adults show significant δ and θ rhythm tracking when passively listening to nursery rhymes.⁵ Similarly, children aged 5–10 years effectively use context to process words during story listening, indicating the use of top-down semantic mechanisms.⁶ An fMRI study suggested that the similarity of neural activity patterns (a measure of neural maturity) in Broca’s area between children and adults during naturalistic movie viewing can predict language ability.⁷ These findings underscore the importance of naturalistic paradigms in revealing the neural dynamics of language processing.

In contrast, magnetoencephalography (MEG)-based studies using passive paradigms in children remain limited. Existing work has shown that left-lateralized theta connectivity correlates with language performance in preschoolers aged 3–4 years⁸ and that listening to a familiar reader enhances alpha-band global integration.⁹ However, the majority of studies have focused on either resting-state or single-frequency-band activity, lacking comprehensive connectivity analyses of brain networks.

Language processing brain network deficits in preschool children with DLD

Developmental language disorder (DLD) is a common neurodevelopmental condition characterized by significant language difficulties without a clear cause, impacting learning and social adaptation.¹⁰ Growing evidence suggests that children with DLD have deficits in processing speech rhythm and prosody, linked to atypical neural oscillations and cortical tracking.^11,12

Most EEG studies on DLD have utilized event-related potentials, identifying abnormal lexical and syntactic processing.^13,14 However, research using naturalistic paradigms is emerging. A systematic review showed that, at rest, children with learning disabilities, including DLD, often exhibit an excess of slow-wave (theta) activity, particularly in left fronto-temporal regions for children with DLD, suggesting a maturational delay in language networks.¹⁵ A naturalistic-stimulus EEG study has shown that while phase locking in the δ and θ bands in children with DLD is similar to that of control groups, there are significant deficits in the γ band, suggesting an impairment in tracking faster acoustic modulations.¹¹ In that task, children watched a female face on a screen and listened to pure-tone stimuli, pressing a key when a syllable was out of time. In a story-listening task, children with DLD exhibited atypical δ-θ cross-frequency phase-amplitude coupling, a phenomenon not observed in children with dyslexia. This effect was mainly observed in electrode clusters over the bilateral temporal regions, possibly indicating deficits in the sensory representation and integration of sound features in children with DLD.¹⁶

MEG studies, although limited, have provided crucial insights for understanding the language processing mechanisms in DLD. One MEG study found that when passively listening to common words or their corresponding sounds, children with DLD had a significantly poorer representation of the speech amplitude envelope in the 200- to 300-ms time window, suggesting a deficit in retaining acoustic-phonetic information.¹² Another study recorded the N400m response using MEG in 11 children with DLD and 10 typically developing (TD) children (aged 9–10) as they listened to real and pseudo-words.¹⁷ It was found that TD children showed reduced activation in the left temporal lobe upon hearing repeated words, whereas children with DLD lacked this repetition effect, suggesting a deficit in sensitivity to lexical processing. Further analysis of this MEG data shows that children with DLD showed enhanced auditory responses in the left hemisphere but lacked the rightward lateralization seen in typical children, which might reflect a compensatory mechanism.¹⁸

However, these studies have not yet characterized the broader brain network features of DLD from a connectivity perspective. Recent fMRI research has highlighted that children with DLD exhibit altered functional connectivity (FC) patterns,¹⁹ particularly in regions related to language processing.²⁰ However, the high temporal precision of MEG offers a unique advantage in capturing dynamic FC associated with language processing. A multi-frequency band approach is crucial, as different neural oscillations are understood to support distinct linguistic functions. Slower rhythms, such as delta and theta, are primarily involved in tracking prosodic and syllabic structures in speech.^3,21,22 In contrast, higher-frequency beta and gamma rhythms are more closely associated with higher-level processes, including semantic and syntactic integration and prediction.^3,21 Combining resting-state and naturalistic-stimulus MEG, passive paradigms may, therefore, offer a new perspective for revealing how these network-level deficits manifest across different frequency bands in DLD.

OPM-MEG for pediatric neuroimaging

Traditional task-driven experiments are often challenging for young children, especially those with DLD. Optically pumped magnetometer (OPM)-MEG is a transformative technology well suited for this population. Compared to conventional Superconducting Quantum Interference Device (SQUID)MEG, OPM sensors do not require cryogenic cooling and can be placed directly on the scalp, dramatically improving signal-to-noise ratio and accommodating any head size. As a non-invasive and silent neuroimaging tool, MEG is ideal for pediatric research. The enhanced capabilities of OPM-MEG make it particularly promising for investigating complex neurodevelopmental disorders like DLD, where subtle abnormalities in neural oscillations and connectivity are thought to be core features.

The present study

Children aged 4–7 are in a crucial stage of developing top-down language processing, including syntax, semantics, and prosody.²³ The use of naturalistic paradigms, such as movie viewing, offers significant advantages for collecting brain development data in young children.²⁴ However, the application of advanced neuroimaging techniques, such as OPM-MEG, to investigate the FC alterations associated with DLD is still in its early stages. This study aims to investigate FC differences in children with DLD using a 128-channel OPM-MEG system in passive paradigms. We compare brain network organization between 10 children with DLD and 30 TD children (ages 4–7) during a rest task and a story task. The focus is on identifying group differences in FC across frequency bands linked to language processing, such as gamma and theta. This research aims to provide insights into the neural signatures of DLD, potentially advancing early diagnostic markers for intervention.

Results

Task-induced power modulations

Analysis of the power spectral density (PSD) in the TD group revealed that MEG power changed significantly across multiple frequency bands during story task compared to the rest task (Table S1 and Figure 1).

Figure 1.

Within-group differences in power between the story task and rest task

∗p_cluster < 0.01, x: p_cluster < 0.05.

As shown in Table S1 and Figure 1, for the TD group, the cluster-based non-parametric permutation test identified two significant clusters of power enhancement in the low-frequency δ band. The main effect cluster was extensive, covering 45 channels primarily distributed across the midline, parietal, and temporal regions and extending to the occipital and central areas, with high statistical significance (p_cluster = 0.001). A smaller cluster (p_cluster = 0.031) involved 14 channels concentrated in the temporal and frontal regions. In contrast to the low-frequency enhancement, higher-frequency bands generally showed power suppression. The α band formed a widespread suppression cluster (p = 0.009) involving 55 channels, mainly covering the parietal, central, and midline areas. The suppression in the β band was even more significant and comprehensive, with a negative cluster covering almost the entire brain (p_cluster < 0.001), involving 107 channels and indicating strong global power suppression. A similar pattern appeared in the low-gamma band, where a significant suppression cluster (p_cluster = 0.020) involved 30 channels, primarily distributed in the central, frontal, and temporal regions.

The DLD group showed a similar overall trend to the TD group, but the strength and spatial extent of the effects were notably weaker (Table S1 and Figure 1). For instance, in the δ band, the DLD group also exhibited power enhancement, but its significant cluster covered only 28 channels and was more localized to the parietal and central regions (p_cluster = 0.002). In the α band, the DLD group’s suppression cluster (p_cluster = 0.014) involved only 24 channels, a significantly smaller area than in the TD group. In the β band, the DLD group’s suppression effect was weaker than the TD group’s (p_cluster = 0.009 vs. p_cluster < 0.001) and covered only 40 channels, whereas the TD group showed global suppression (107 channels). In the low-gamma band, the DLD group only showed a marginally significant suppression effect (p_cluster = 0.051), forming a small cluster (12 channels) mainly in the parietal region.

An independent samples t test was conducted on the change scores between groups, with false discovery rate (FDR) correction applied across six frequency bands and 128 channels (Table 1). At the p_FDR < 0.05 level, the power increase in the δ band was significantly greater in the DLD group than the TD group for bilateral central region (MZC4, MC20, MC22, MC17, MC14, MC5, MC6, MC10, MC18, and MC24) channels. Conversely, for the right temporoparietal region (MT30, MT4), the TD group showed a power increase, while the DLD group showed a decrease. In the α band, channels located over the right central and frontal regions (MF16, MZC1, MF20, MC4, MC12, MF8, and MC6) showed a power decrease, while the DLD group showed an increase tendence. In the β band, a significant group difference was observed only at channel MT11. In the low-γ band, both groups showed a decrease in power over the left temporoparietal region (MT7, MT17, MT15, MT21, and MP5) and bilateral parietal sites (MP7, MP8, and MZP2), although the decrease was more pronounced in the DLD group. No significant effects were observed at the p_cluster < 0.05 level (Family-Wise Error, FWE correction), which may reflect that the group differences in change scores did not form clusters robust enough to survive the cluster-based permutation correction.

Table 1.

Between-group differences in power between the story task and rest task

Frequency	Channel	DLD_diff	TD_diff	DLD_Story	DLD_Rest	TD_Story	TD_Rest	T_value	p_value
δ	MZC4	96.966 (87.067)	29.763 (64.162)	337.216 (111.781)	240.250 (63.553)	321.146 (123.660)	291.383 (106.087)	2.636	0.003
δ	MC20	38.115 (33.260)	5.947 (37.618)	191.480 (80.952)	153.366 (57.308)	196.784 (77.537)	190.837 (68.487)	2.413	0.009
δ	MC22	42.310 (63.086)	4.078 (36.077)	206.060 (77.140)	163.750 (48.324)	185.638 (83.750)	181.561 (69.292)	2.399	0.016
δ	MC17	78.605 (34.434)	33.684 (61.371)	281.716 (85.762)	203.111 (72.908)	258.368 (111.008)	224.684 (80.290)	2.194	0.011
δ	MC14	29.423 (44.407)	1.428 (34.986)	180.539 (72.355)	151.115 (48.030)	174.943 (74.787)	173.515 (59.997)	2.060	0.027
δ	MC5	59.156 (69.604)	19.740 (47.862)	236.358 (118.735)	177.203 (60.383)	208.808 (109.762)	189.067 (85.797)	2.020	0.036
δ	MP7	89.604 (44.930)	49.953 (59.594)	335.702 (61.023)	246.098 (64.910)	344.167 (118.406)	294.213 (95.286)	1.928	0.027
δ	MC6	30.541 (89.370)	−2.114 (33.040)	188.926 (87.840)	158.385 (72.846)	202.931 (81.212)	205.045 (74.916)	1.734	0.050
δ	MC10	35.108 (52.520)	6.232 (43.765)	241.640 (116.560)	206.532 (83.670)	239.009 (100.437)	232.777 (82.183)	1.729	0.046
δ	MC18	38.470 (36.879)	13.008 (41.587)	214.224 (87.445)	175.754 (72.760)	225.726 (98.424)	212.718 (79.782)	1.727	0.043
δ	MC24	45.820 (42.184)	16.618 (48.431)	250.032 (141.685)	204.212 (114.941)	189.674 (106.395)	173.055 (83.428)	1.706	0.043
δ	MC19	−73.955 (44.987)	−27.282 (96.370)	307.608 (129.385)	381.563 (140.413)	294.588 (122.013)	321.871 (116.422)	−1.471	0.036
δ	MT4	−12.906 (86.892)	38.256 (78.598)	98.778 (67.636)	111.684 (83.735)	143.896 (124.015)	105.640 (69.260)	−1.746	0.038
δ	MT30	−33.173 (95.287)	12.265 (31.777)	151.610 (65.347)	184.783 (117.398)	152.967 (82.755)	140.702 (76.958)	−2.331	0.023
θ	MT11	0.363 (0.807)	−0.215 (0.617)	2.376 (2.192)	2.013 (2.314)	1.418 (0.614)	1.633 (0.739)	2.390	0.015
θ	MT12	−0.564 (1.318)	0.054 (0.531)	2.298 (2.798)	2.862 (2.993)	1.400 (0.662)	1.346 (0.733)	−2.164	0.027
α	MT11	0.132 (0.343)	−0.072 (0.158)	1.064 (0.477)	0.932 (0.593)	0.829 (0.238)	0.900 (0.272)	2.603	0.011
α	MF16	0.029 (0.086)	−0.026 (0.063)	0.458 (0.197)	0.430 (0.149)	0.408 (0.095)	0.434 (0.102)	2.164	0.023
α	MZC1	0.027 (0.141)	−0.037 (0.064)	0.544 (0.360)	0.517 (0.355)	0.424 (0.106)	0.460 (0.125)	2.001	0.043
α	MF20	0.020 (0.106)	−0.033 (0.065)	0.449 (0.153)	0.428 (0.120)	0.412 (0.097)	0.445 (0.111)	1.900	0.038
α	MC4	0.021 (0.120)	−0.039 (0.078)	0.543 (0.308)	0.521 (0.299)	0.441 (0.116)	0.481 (0.147)	1.855	0.045
α	MC12	0.017 (0.105)	−0.054 (0.104)	0.587 (0.277)	0.571 (0.252)	0.507 (0.143)	0.561 (0.176)	1.852	0.032
α	MF8	0.023 (0.086)	−0.018 (0.054)	0.400 (0.123)	0.377 (0.101)	0.360 (0.079)	0.378 (0.082)	1.786	0.049
α	MC6	0.012 (0.103)	−0.043 (0.083)	0.513 (0.184)	0.501 (0.163)	0.474 (0.125)	0.517 (0.143)	1.705	0.049
β	MT11	0.034 (0.100)	−0.019 (0.033)	0.416 (0.091)	0.381 (0.134)	0.344 (0.062)	0.363 (0.068)	2.629	0.013
Low γ	MT11	0.016 (0.043)	−0.001 (0.010)	0.189 (0.032)	0.173 (0.050)	0.164 (0.016)	0.165 (0.018)	2.111	0.028
Low γ	MP7	−0.005 (0.005)	0.001 (0.010)	0.145 (0.030)	0.150 (0.033)	0.128 (0.019)	0.128 (0.015)	−1.580	0.038
Low γ	MZP2	−0.006 (0.005)	−0.000 (0.010)	0.138 (0.028)	0.144 (0.031)	0.124 (0.017)	0.124 (0.013)	−1.711	0.038
Low γ	MP5	−0.005 (0.006)	−0.001 (0.007)	0.146 (0.019)	0.151 (0.021)	0.130 (0.013)	0.131 (0.012)	−1.934	0.026
Low γ	MT7	−0.007 (0.013)	−0.001 (0.006)	0.142 (0.015)	0.149 (0.023)	0.127 (0.014)	0.129 (0.014)	−1.976	0.037
Low γ	MP8	−0.007 (0.007)	−0.001 (0.010)	0.153 (0.023)	0.160 (0.028)	0.140 (0.016)	0.141 (0.014)	−1.980	0.026
Low γ	MT18	−0.018 (0.028)	−0.005 (0.014)	0.168 (0.029)	0.186 (0.042)	0.155 (0.018)	0.160 (0.020)	−1.992	0.040
Low γ	MT21	−0.005 (0.005)	0.000 (0.007)	0.138 (0.015)	0.143 (0.018)	0.125 (0.015)	0.125 (0.014)	−2.026	0.020
Low γ	MT17	−0.006 (0.013)	0.001 (0.006)	0.164 (0.016)	0.170 (0.024)	0.149 (0.016)	0.148 (0.015)	−2.202	0.024
Low γ	MT15	−0.008 (0.007)	−0.001 (0.006)	0.161 (0.013)	0.169 (0.017)	0.148 (0.014)	0.149 (0.014)	−3.124	0.002

Only p_FDR < 0.05 level results were reported.

Group and condition effects on network modularity

We conducted a 2 (Group: DLD, TD) × 2 (Condition: story, rest) mixed-design ANOVA on the network modularity index. The results are shown in Figure 2 and Table 2.

Figure 2.

Network modularity for both groups across tasks in six frequency bands

Data are represented as mean ± SEM.

Table 2.

Results of the mixed-design ANOVA on modularity

Frequency band	FGroup	PGroup	FCondition	PCondition	FInteraction	PInteraction
δ	0.443	0.508	15.031a	<0.001	0.828	0.366
θ	0.119	0.731	20.911a	<0.001	0.137	0.713
α	2.547	0.115	45.690a	<0.001	0.817	0.369
β	1.518	0.222	5.771a	0.019	1.028	0.314
Low γ	2.926	0.091	20.857a	<0.001	0.166	0.685
High γ	7.448a	0.008	14.207a	<0.001	0.031	0.860

^ap_FDR<0.05.

The mixed-design ANOVA on modularity revealed a significant main effect of Condition across all six frequency bands (all p_FDR < 0.05; Table 2). This indicates that the story task robustly decreased network modularity compared to the rest task for all children, suggesting a shift toward a more globally integrated processing state.

Crucially, a significant main effect of Group was found exclusively in the high-γ band (F(1, 41) = 7.45, p_FDR < 0.05). The DLD group exhibited consistently higher modularity than the TD group in this band, irrespective of the task. No significant Group × Condition interactions were found. In the other five frequency bands (δ, θ, α, β, low γ), the main effect of group was not significant (all p_FDR > 0.05).

For the interaction between Group and Condition, the F values were not significant for any frequency band (all p_FDR > 0.05, Table 2). This suggests that, from an overall brain activity perspective, the effect of the task condition on modularity (i.e., the difference between story and rest) did not show a statistically significant difference between the DLD and TD groups.

Altered low-gamma functional connectivity in DLD

A cluster-based permutation test on the coherence change scores (story − rest) revealed a marginally significant negative cluster in the low-gamma band (p_cluster = 0.074). This cluster comprised 214 functional connections and generally showed that the change was lower in the DLD group than in the TD group (Figure 3). Topologically, this cluster exhibited a hub structure centered around the left temporal lobe (MT27, MT25, and MT29), left parietal lobe (MP9, MP7, MP3, MP13, MP1, MP11, MP15, MP5, etc.), central region (MC18, MZC6, MZC24, MZP1, MZP2, etc.), and left occipital lobe (MO5, MO7, MO9, etc.), primarily involving short-range connections between these hubs.

Figure 3.

Topography of the low-gamma connectivity cluster showing a reduced task-related change in the DLD group compared to the TD group

To further evaluate the functional relevance of this low-gamma cluster, we extracted the mean coherence within this cluster for each subject and computed the task-related change (story − rest). A non-parametric permutation test (5,000 permutations) revealed a significant group difference in this task-related modulation (p = 0.003), with TD children showing an increase in coherence during story listening, whereas children with DLD showing a decrease (Figure 4). This confirms a robust Group × Condition interaction independent of the cluster-based statistics.

Figure 4.

Mean low-gamma connectivity within the significant cluster

(A) Mean low-gamma connectivity within the significant cluster for each group and task.

(B) The task-induced change (story − rest) in mean connectivity for each group.

Data are represented as mean ± SEM.

Discussion

This study demonstrates the utility of a 128-channel OPM-MEG systemfor investigating neural dynamics in children with DLD using passive paradigms(Figures 5 and 6). The oscillatory patterns observed in our data are highly consistent with well-established findings from the MEG/EEG literature on language and naturalistic auditory processing. In particular, increases in low-frequency (δ) activity during narrative comprehension reflect enhanced tracking of speech rhythms and linguistic structure,²⁵ whereas decreases in α/β power are commonly associated with increased lexical-semantic processing demands and top-down prediction.^26,27,28 Suppression in the γ band has also been linked to higher-level linguistic computation and cognitive control during complex tasks.²⁹ And the observed decrease in network modularity during the story task aligns with previous studies showing that complex cognitive tasks promote brain network integration.^30,31 Together, these convergent findings indicate that OPM-MEG successfully captures canonical neural signatures of auditory language processing and large-scale network reconfiguration, supporting the validity and reliability of this approach in preschool-age children with DLD.

Figure 5.

Schematic of the procedure and materials for the two passive paradigms

Figure 6.

MEG system and recording environment

(A) 128-sensor layout.

(B) Mock testing scenario with a child seated.

(C) External monitoring.

(D) Internal recording environment.

However, the study also uncovers novel findings regarding the neural signatures of DLD. First, we observed a marginal alteration in low-γ FC, which did not reach the corrected significance threshold but may indicate a potential trend warranting further investigation in larger samples. Compared to the resting state, during the story task, TD children exhibited enhanced FC within a left-hemisphere network crucial for language, spanning temporo-parieto-occipital regions. In contrast, children with DLD showed weakened connectivity in the same network. This suggests inefficiencies in auditory language processing at a fundamental stage, where low-γ oscillations are involved in encoding fast-varying acoustic information.^3,32 The observed failure to recruit this local network in DLD children provides neurophysiological evidence for inefficient communication at a fundamental stage of auditory language processing. This finding contrasts with a recent fMRI study on children with DLD aged 7–13, which reported widespread reductions in FC across the brain, particularly in sensorimotor, cognitive control, and default mode networks.¹⁹ In contrast, the current study found enhanced overall FC in the left temporo-parieto-occipital regions during the resting state in children with DLD. This discrepancy may arise from the fact that MEG is more sensitive to fast-frequency neural oscillations, whereas fMRI tends to focus on different brain networks (such as the default mode network and sensorimotor network) and captures longer-term neural activity changes. Additionally, the difference could be related to the younger age range of participants in this study (ages 4–7). At this developmental stage, brain networks may still be maturing, and children with DLD may exhibit compensatory mechanisms through strengthened connectivity in specific regions to address language processing deficits. Over time, as children age, these compensatory patterns may change or become less pronounced.

Second, we found that children with DLD exhibited increased high-γ network modularity compared to TD children, regardless of the task. This finding points to a “hyper-segregated” brain network, which may reflect a rigid, fragmented processing architecture that hinders the integration of information across different brain regions critical for language processing. High-gamma activity is associated with high-level computations in brain regions supporting language and cognition.^3,32 Abnormally high modularity can be a marker of poor network integration and has been reported in other neurodevelopmental disorders like autism.^33,34 Elevated modularity in DLD suggests impaired network integration, which may underlie the language deficits observed in these children. This new insight adds to the growing body of literature on network-level abnormalities in DLD.

It is noteworthy that we did not find significant group-by-task interactions in our macro-level measures (PSD and modularity). This suggests that the core neural deficits in DLD may not lie in the overall magnitude of brain responses, but rather in the more subtle patterns of inter-regional communication (i.e., FC), which passive paradigms are well-suited to detect.

In conclusion, this study demonstrates that passive OPM-MEG paradigms are a sensitive and feasible tool for pediatric cognitive neuroscience. Using this approach, we identified a robust finding of persistently higher high-gamma network modularity in children with DLD. In addition, we observed a trend toward altered recruitment of a left-hemisphere low-gamma functional network during a naturalistic story task, which warrants further investigation in larger samples. These results provide preliminary evidence that disordered network communication may be a key feature of DLD and highlight the potential of passive OPM-MEG paradigms for detecting subtle neural markers in clinical populations. The successful application of this technology opens a promising new avenue for identifying early biomarkers and assessing intervention outcomes in neurodevelopmental.

Limitations of the study

While the study provides valuable insights into the neural signatures of DLD, it has limitations. First, our analysis was conducted in sensor space. Future work incorporating source reconstruction³⁵ would provide better anatomical specificity. Second, the sample size for the DLD group for MEG study was relatively small and predominantly male (9/10). This imbalance may be due to the characteristics of recruitment at a single clinical center. As a result, the statistical power of our analyses may be limited and the generalizability of the findings to the broader DLD population should be interpreted with caution.³⁶ Future studies with larger cohorts are needed to confirm these findings and explore potential heterogeneity within the DLD population.^37,38

Resource availability

Lead contact

Further information and requests for resources and materials should be directed to and will be fulfilled by the lead contact, Xiaolin Ning (ningxiaolin@buaa.edu.cn).

Materials availability

This study did not generate new unique reagents or materials.

Data and code availability

• •All data reported in this article will be shared by the lead contact upon reasonable request.
• •No custom code or software was generated or used in this study.
• •Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request.

Acknowledgments

We sincerely thank all the children and their parents for their invaluable participation in this study. We are also grateful to the staff at the Health Center of Beijing Children’s Hospital for their support. Funding: This study was supported by Fundamental and Interdisciplinary Disciplines Breakthrough Plan of the Ministry of Education of China (JYB2025XDXM606), by a grant from the Beijing Municipal Science and Technology Commission and Zhongguancun Science Park Administrative Committee (no. Z251100000425001), and by Beihang University and Beijing Children’s Hospital under the project titled “Development and Evaluation of EEG/MEG Diagnostic Models for Developmental Language Disorder in Children” (YPT202002).

Author contributions

Conceptualization, A.L; methodology, Z.C. and Y.S.; software, Z.C. and J.W.; validation, Z.C.; formal analysis, Z.C., J.W., and J.L.; investigation, H.W., Y.S., and C.Q.; resources, J.F., X.N. (Xin Ni), and A.L.; data curation, H.W. and J.W.; writing – original draft, A.L. and Z.C.; writing – review & editing, X.N. (Xiaolin Ning), and A.L.; visualization, Z.C.; supervision, J.F., X.N. (Xin Ni), X.N. (Xiaolin Ning), and A.L.; project administration, A.L.; funding acquisition. J.F., X.N. (Xin Ni), X.N. (Xiaolin Ning), and A.L. All authors have read and agreed to the published version of the manuscript.

Declaration of interests

J.W. is an employee of Hangzhou Lingci Medical Equipment Co., Ltd, which is involved in the development of MEG technology.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
Dataset	This paper	–
Software and algorithms
MATLAB R2024b	The Mathworks	https://mathworks.com
FieldTrip toolbox	Robert Oostenveld, Pascal Fries, Eric Maris, and Jan-Mathijs Schoffelen	https://www.fieldtriptoolbox.org/
Brain Connectivity Toolbox (BCT)	Franco Pestilli	https://sites.google.com/site/bctnet/

Experimental model and study participant details

This study recruited 30 TD children (mean age 5.51 ± 0.64 years, range 4.0–7.0 years, 7 females) and 10 children with DLD (mean age 5.70 ± 0.80 years, range 4.0–7.0 years, 1 females). The age distribution between the two groups of children did not differ significantly (t = 0.66, p = 0.52). All children were right-handed, native Mandarin Chinese speakers with normal or corrected-to-normal vision. No participant had a history of neurological disorders, hearing impairments, or autism spectrum disorder. Children with parent-reported late talking or expressive difficulties were evaluated for DLD. Those aged ≥6 years completed the Wechsler Intelligence Scale for Children – Fourth Edition (WISC-VI)³⁹ and a standardized Mandarin language assessment⁴⁰ based on local norms. Children with IQ > 70 and a language delay of at least one year in grammar, comprehension, expression, or communication according to the standardized Mandarin language assessment were considered for DLD. Children under 6 years completed the Children Neuropsychological and Behavior Scale (CNBS).⁴¹ Those with nonverbal DQ > 70 and a language DQ ≤ 70, or a clear delay on the standardized Mandarin language assessment, were considered for DLD. A senior developmental-behavioral pediatrician and a research psychologist made the final diagnosis based on developmental history, parent interview, clinical observation, and DSM-5 criteria for language disorder. TD children were recruited from routine health checks. They had no parental concerns about speech or language, completed WPPSI (≥6 years) or CNBS (<6 years), and had IQ or nonverbal DQ > 70. Their final status was also confirmed by the same clinical team using developmental history, parent interview, clinical observation.

Ethical approval for the experiment was issued by the Ethical Committee of Beijing Children Hospital ([2021]-E−236-Y). Written informed consent was obtained from the parents or legal guardians of all child participants.

Method details

The experimental procedure was explained to the guardian, and the child was invited to choose a gift to enhance cooperation. A parent was allowed inside the magnetically shielded room during the experiment to reduce the child’s anxiety.

The experiment followed a fixed sequence. First, after head registration using binocular optical imaging, participants completed a 7-min resting-state task (hereafter, rest task). Next, participants completed an approximately 6-min audiovisual story task (hereafter, story task). Following this, participants completed a sentence-level semantic judgment task lasting about 10 min (data from this task were not used in the present analysis). Instructions were provided before each task. This study focuses on passive, no-response naturalistic paradigms; therefore, only data from the rest and story tasks were included in the subsequent analysis. Figure 5 illustrates the experimental design and materials employed in the two passive paradigms.

To improve cooperation and reduce head motion during the rest task, we played a 6-min animated video called “Inscapes”.⁴² This animation, featuring abstract geometric shapes and calming music, has been shown to effectively capture children’s attention, significantly reduce head motion, and maintain a low cognitive load. Compared to narrative films, the neural synchrony induced by the Inscapes paradigm is closer to that of traditional resting states. During the experiment, children sat comfortably in a chair with their heads inside the MEG helmet, watching the video with sound. The volume was set to 60%, and no button responses were required.

The story task consisted of three short stories adapted from Aesop’s fables or Chinese myths (Story 1: The Fisherman and the Genie; Story 2: Nüwa Mends the Heavens; Story 3: The Monkey and the Crocodile). Each story was narrated by a voice actor and accompanied by scene images, creating videos approximately 2 min in length. Before the task, instructions were played: “Hello, you are about to hear a story. Please do not talk or move, and listen quietly.” During the task, children listened to the stories and watched scene images without making any response.

The OPM-MEG system (LCMEG-CH-128A, Hangzhou Lingci Medical Equipment Co., Ltd., China) included an OPM sensor array, a rigid helmet, a magnetically shielded room, and a data acquisition and control computer (Figure 6). Brain magnetic signals were recorded using the OPM-MEG system at a sampling rate of 1000 Hz. Single-axis OPM sensors that were designed to be sensitive specifically along the direction perpendicular to the scalp. The sensor array was arranged based on the international 10-5 system. The whole distribution of our 128 OPM sensors is shown in Figure 6. The dimensions of each sensor are 10∗19∗20 mm, and crosstalk between sensors was not detected. The automatic co-registration of the head, helmet and OPM sensors was achieved by a stereo vision approach, which was to detect helmet markers and to reconstruct the 3D facial structure of the subject.⁴³ The system’s noise level in the 2–30 Hz frequency band did not exceed 15 fT/√Hz, ensuring high sensitivity for detecting weak neural magnetic fields. To minimize external magnetic interference, the experiment was conducted in a specially designed magnetically shielded room, where the residual magnetic field in the working area was strictly controlled below 3 nT. The ample space inside the shielded room (approximately 9 m²) could accommodate a parent, providing a comfortable experimental environment for the children.

Quantification and statistical analysis

All MEG data were preprocessed offline in MATLAB R2024b (The MathWorks, Natick, MA, USA) using the FieldTrip toolbox.⁴⁴ The preprocessing pipeline was designed to maximize the signal-to-noise ratio and effectively remove physiological and physical artifacts. First, the data were segmented based on trigger signals. Next, bad channels were repaired using interpolation algorithms. The data were filtered by a 6-order band-pass filtered between 0.1 and 100 Hz, and spike noise in specific frequency bands was removed using a Fast Fourier Transform. Subsequently, the Signal Space Separation (SSS) algorithm was applied to eliminate environmental noise originating outside the sensor measurement range. To minimize motion artifacts, periods marked as containing body movements during acquisition were visually inspected in the raw data and excluded from further analysis. Finally, Independent Component Analysis (ICA) was used to identify and remove residual physiological artifacts, including eye blinks, muscle activity, heartbeats, and any remaining movement-related components.

To investigate the characteristics of different neural oscillatory activities, we calculated the Power Spectral Density (PSD) for each channel in various frequency bands (δ: 0.1–4 Hz, θ: 4–7 Hz, α: 7–12 Hz, β: 12–30 Hz, low γ: 30–55 Hz, high γ: 55–80 Hz). The frequency band boundaries were chosen based on developmental considerations for preschool children and previous EEG and MEG studies in pediatric populations,^45,46,47 reflecting the typical maturation of θ and α oscillations in early childhood. This analysis aimed to quantify the energy distribution of neural activity across different brain regions under various experimental conditions.

To assess the patterns of FC between different brain regions, we performed a comprehensive network topology analysis. This was central to understanding how connectivity between brain regions differs in children with DLD compared to their TD peers.

First, at the functional connectivity level, this study used coherence, a widely used index in MEG research.⁴⁷ Coherence quantifies the stability of the phase relationship between two neural signals at a specific frequency to assess the strength of linear coupling between brain regions, with values ranging from 0 to 1. For the calculation, a multitaper Fast Fourier Transform (FFT) frequency domain analysis was performed using the FieldTrip toolbox with a Hanning window. The frequency range covered 0.1–80 Hz, and Fourier coefficients were output. Based on this, we calculated the inter-channel phase synchrony using coherence to evaluate functional connectivity. Finally, for subsequent statistical analysis, we averaged the coherence results within common frequency bands for children (δ, θ, α, β, γ) to obtain a channel × channel functional connectivity matrix for each band.

Coherence values (ranging between 0 and 1) were used directly as weighted edges in the functional connectivity matrices. Because coherence is a normalized measure, no additional normalization or thresholding was applied prior to modularity analysis. Using the full weighted coherence matrix avoids arbitrary sparsity settings and preserves the complete connectivity profile across participants. After obtaining the functional connectivity matrices, we computed the whole-brain functional network modularity index using the Louvain algorithm implemented in the Brain Connectivity Toolbox (BCT).⁴⁸ For each frequency band and each participant, the modularity optimization was repeated 100 times to ensure stable estimates despite the algorithm’s randomness, and the average Q value was taken as the final modularity estimate. An increase in this index suggests that the brain is shifting toward a highly “specialized” and relatively segregated state. A decrease in the index indicates a shift toward an “integrated” and globally cooperative state.

We first identified brain regions (for PSD) or connections (for Coherence) that showed significant changes due to the task condition (story vs. rest) within each group (DLD and TD). For this, we used a cluster-based non-parametric permutation test implemented in the FieldTrip toolbox. This method involved the following steps: (1) performing a paired-samples t test for each sensor (or connection); (2) clustering adjacent spatial or topological units where the t-value exceeded a predefined threshold (cluster alpha = 0.05); (3) generating a null distribution of the cluster statistic through 5000 random permutations of the condition labels; and (4) comparing the observed cluster statistic to the null distribution to calculate a corrected p-value. This method effectively controls the family-wise error rate from multiple comparisons while accurately identifying spatiotemporal patterns with significant differences. Next, we calculated the change scores (story listening minus resting-state) for each participant and performed an independent samples t test using 5000 random permutations.

For the network modularity index, we directly used a repeated-measures ANOVA to test for the main effect of condition, the main effect of group, and their interaction. The model included one within-subjects factor, “condition” (Story, Rest), and one between-subjects factor, “group” (DLD, TD). This analysis aimed to test the main effect of “group” (whether there was an overall group difference) and the “condition × group” interaction effect (whether the pattern of change induced by the task condition differed between the two groups).

Published: March 12, 2026