Resting-State EEG Characteristics of Tardive Dyskinesia: Spectral and Network Connectivity Insights

Abstract

Background and Hypothesis

Tardive dyskinesia (TD) is an often irreversible movement disorder, mostly resulting from exposure to dopamine receptor antagonists. The electrophysiological characteristics of TD remain poorly understood. Resting-state electroencephalography (EEG) may help characterize frequency-specific cortical oscillatory and network alterations related to TD. We hypothesized that TD-related neurophysiological alterations may be reflected in cortical spectral power, hemispheric asymmetry, and large-scale functional connectivity.

Study Design

Resting-state EEG was recorded with a 32-channel system in 67 participants: 20 patients with TD, 25 patients without TD, and 22 healthy controls. Analyses focused on spectral power, hemispheric asymmetry, and functional connectivity measured using phase-lag index. Electrode-level spectral power analyses were corrected using the false discovery rate, and network-level connectivity differences were assessed using network-based statistics. Subgroup clinical association analyses within the TD group were considered exploratory.

Study Results

Compared with healthy controls, both TD and Non-TD patients showed increased theta relative power at central electrodes, including C4 and Cz, with group effects surviving false discovery rate correction. Cluster-based topographic analyses showed increased gamma2-gamma3 power over frontoparietal regions in TD patients and increased broadband gamma power over frontal-central regions in Non-TD patients relative to healthy controls. Within the TD subgroup, exploratory analyses showed nominal associations between hemispheric asymmetry indices and rapid medication withdrawal or duration of motor disorder, although these associations did not survive false discovery rate correction. Functional connectivity analyses identified frequency-specific network alterations, including reduced phase-locking value in the theta frequency band and reduced phase-lag index connectivity across alpha1, beta, and gamma frequency bands in TD-related comparisons.

Conclusion

TD patients showed altered resting-state cortical oscillations and frequency-specific connectivity changes involving central, frontoparietal, and temporo-occipital networks. These findings provide preliminary electrophysiological evidence for cortical network involvement in TD and may inform future mechanistic studies. Given the modest sample size and exploratory nature of subgroup analyses, the findings should be interpreted cautiously and validated in larger longitudinal cohorts.

Background

Tardive dyskinesia (TD) is a movement disorder, mostly resulting from exposure to dopamine receptor antagonists.¹ It is characterized by involuntary, repetitive, and often stereotyped movements that typically affect the orofacial region, trunk, and limbs.¹ Epidemiological studies have estimated that 20%–35% of patients receiving long-term antipsychotic therapy develop TD.^1,2 The disorder produces persistent and mostly irreversible motor disturbances that may continue even after discontinuation of the causative drug.³ TD severely impairs quality of life and imposes substantial economic and caregiving burdens, yet current treatment options remain limited.³ The pathophysiology of TD is incompletely understood and has been linked to multiple mechanisms, including dopamine receptor supersensitivity, GABAergic dysfunction, and oxidative stress.^2,4

Electroencephalography (EEG) is a low-cost, non-invasive method with high temporal resolution and is well suited for characterizing frequency-specific cortical oscillations and functional connectivity.⁵ Despite these advantages, EEG studies of TD remain scarce. An early clinical EEG study reported more occipital rhythm abnormalities and reduced alpha power in psychiatric inpatients with antipsychotic-induced TD.⁶ However, that study relied primarily on conventional clinical EEG assessment and did not systematically examine electrode-level spectral power, hemispheric asymmetry, or large-scale functional connectivity. More recently, a case report highlighted the importance of identifying TD-related movement artifacts in EEG recordings, but systematic quantitative EEG evidence remains limited.⁷ Therefore, a quantitative resting-state EEG study integrating spectral power, hemispheric asymmetry, and functional connectivity may provide a more comprehensive characterization of TD-related neurophysiological alterations.

Previous brain network studies of TD have largely focused on schizophrenia populations. Structural magnetic resonance imaging (MRI) studies have reported gray matter abnormalities in schizophrenia patients with TD,^8,9 while resting-state functional MRI (fMRI) studies have identified altered regional homogeneity and functional connectivity within motor and cortico-striatal circuits.^10,11 These findings provide important evidence that TD is associated with distributed brain network alterations. However, schizophrenia itself is associated with widespread abnormalities in cortical structure, neural oscillations, and functional connectivity,^12–14 and long-term exposure to antipsychotic medications may further complicate the interpretation of TD-related neurophysiological findings.¹⁵ Therefore, it remains unclear whether similar electrophysiological alterations can be observed in TD patients outside schizophrenia populations.

To address these gaps while reducing potential confounding from schizophrenia-related neural abnormalities, the present study focused on TD occurring in patients with primary mood disorders and excluded patients with schizophrenia or other psychotic disorders. Mood disorders are also commonly treated with dopamine receptor antagonists or other psychotropic agents, and TD can emerge in this context.^15,16 Therefore, investigating mood disorder-associated TD may help clarify the generalizability of previous findings derived largely from schizophrenia-associated TD and may provide complementary insight into TD-related neural dysfunction across diagnostic backgrounds.

In this study, we aimed to characterize resting-state EEG alterations in TD patients with primary mood disorders by examining cortical spectral power, hemispheric asymmetry, and functional connectivity. We hypothesized that TD-related neurophysiological alterations may be reflected in frequency-specific spectral power changes, altered hemispheric asymmetry, and disrupted large-scale functional connectivity. By comparing TD patients with both healthy controls (HC) and mood disorder patients without TD (Non-TD), this study sought to identify EEG patterns associated with TD-related clinical status and to provide preliminary electrophysiological evidence for cortical network involvement in TD.

Methods

Participants

This prospective study was conducted at the Department of Neurology, Peking University Sixth Hospital. Clinical and EEG data of all patients were collected during hospitalization from November 2023 to July 2025.

Participants were included if they met the following criteria: a clinical diagnosis of TD; right-handedness; provision of signed informed consent and the ability to cooperate with EEG acquisition; availability of complete baseline demographic and clinical information; and no exposure to dopamine receptor antagonists during the EEG recording session. Exclusion criteria encompassed a history of epilepsy, traumatic brain injury with loss of consciousness, developmental delay, brain tumor, or prior neurosurgery; exposure to neurotoxic substances; and a comorbid diagnosis of schizophrenia or neurodegenerative disorders such as Parkinson’s disease or Alzheimer’s disease.

Non-TD and HC groups were matched to the TD group based on age and sex. Participants were classified into three groups: the TD group, which included patients diagnosed with TD and with a primary diagnosis of mood disorders, excluding those with schizophrenia or other psychotic disorders; the Non-TD group, comprising patients with comparable mood disorders but without any signs of TD; and the HC group, consisting of individuals without any psychiatric or neurological disorders.

Ethics Declaration

This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Ethics Committee of Peking University Sixth Hospital (approval No. 2023 Ethics Review No. 89). Study preparation, including protocol development and preliminary organizational work, preceded formal enrollment; however, all participants included in the present analysis were enrolled from November 2023 onward, after ethics approval had been obtained. Written informed consent was obtained from all participants before study participation.

Experimental Procedure

Clinical Data

All TD patients completed a comprehensive clinical assessment battery. Demographic information, including age, sex, and years of education, was collected. TD-related clinical features included symptom subtypes, classified as orofacial or other; symptom distribution, classified as focal or generalized; history of rapid medication withdrawal; duration of motor disorder in months; and total Abnormal Involuntary Movement Scale (AIMS) score. Symptom duration was calculated based on the patient-reported onset of motor symptoms. Clinical assessments were performed by physicians or trained personnel certified in structured clinical evaluation.

Because of the modest sample size of the TD subgroup, analyses involving TD-related clinical characteristics were considered exploratory. These analyses were intended to evaluate whether EEG features were associated with potential TD-related clinical factors, including symptom subtypes, rapid medication withdrawal, and duration of motor disorder, rather than to establish causal relationships between EEG abnormalities and TD movements.

EEG Acquisition

Resting-state EEG was recorded using a 32-channel system (Thermo Nicolet Corporation, USA). Carbon-fiber electrodes were filled with conductive paste (Nihon Kohden Corporation, Japan) and placed according to the international 10–20 system. Signals were referenced to bilateral mastoids and amplified using an AC-coupled high-input-impedance amplifier with an anti-aliasing filter. The data were digitized at a sampling rate of 2000 Hz.

In addition to the standard 19 scalp electrodes, reference electrodes, and ground electrode, additional electrodes were placed as clinically indicated, including bilateral sphenoidal and mastoid electrodes and two electrocardiogram leads. Standard activation procedures, including eyes-open/eyes-closed recording, hyperventilation, and photic stimulation, were performed before the resting-state recording. Resting-state EEG was recorded for 20–60 minutes with participants’ eyes closed. Electrode impedances were maintained below 50 kΩ during recording.¹⁷

All recordings were reviewed by a trained EEG technician. A random subset of recordings was independently reviewed by a senior neurophysiologist to verify recording quality and artifact identification. To minimize medication-related confounding effects on EEG oscillations, recent exposure to benzodiazepines and anticholinergic agents was reviewed during participant screening. EEG recordings showing clear sedative-related, anticholinergic-related, or other medication-related abnormalities, including prominent generalized slowing, were excluded from quantitative analyses after visual review.

Data Analysis

EEG Preprocessing

EEG preprocessing was performed using EEGLAB 2024 in MATLAB R2024b.^18,19 Non-scalp electrodes, including sphenoidal, mastoid, and electrocardiogram electrodes, were excluded, retaining the standard 19 scalp channels for quantitative analyses. Data were band-pass filtered from 0.1 to 60 Hz, with a 48–52 Hz notch filter applied to attenuate line noise, and were then downsampled to 256 Hz. The continuous data were segmented into 2-second epochs.

Epochs containing amplitudes exceeding ±100 μV or values exceeding five standard deviations from the mean were automatically rejected. Independent component analysis (ICA) was then performed using the default runica algorithm implemented in EEGLAB. Artifact components were identified using ICLabel in combination with visual inspection. Components classified as ocular, muscle, cardiac, or other non-brain artifacts were removed before further analysis. To avoid excessive data loss and preserve neural signals, no more than 50% of independent components were rejected for any participant. For cases requiring visual judgment, artifact classification was reviewed by trained personnel, and a random subset was further checked by a senior neurophysiologist to ensure consistency of component rejection. Finally, artifact-free epochs from the eyes-closed resting-state condition were selected for subsequent spectral and connectivity analyses.

Statistical Analysis

All statistical analyses were conducted using MATLAB R2024b and SPSS 27.0. Continuous variables were presented as mean ± standard deviation if normally distributed, or as median with minimum and maximum values otherwise. Categorical variables were summarized as frequencies and percentages. Group comparisons for demographic variables were performed using one-way analysis of variance (ANOVA) for normally distributed continuous variables and the Kruskal–Wallis H-test for non-normally distributed continuous variables. Categorical variables were compared using the chi-square test.

Power spectral density (PSD) was computed using Welch’s method²⁰ across canonical frequency bands: delta, 0.5–4 Hz; theta, 4–8 Hz; alpha, 8–13 Hz (subdivided into alpha1: 8–10 Hz and alpha2: 10–12 Hz); beta, 14–30 Hz; and gamma, 30–50 Hz (subdivided into gamma1-gamma4 covering 32–48 Hz).²¹ Relative power was calculated as band-specific power divided by total spectral power and was treated as a unitless ratio, whereas absolute power was calculated from PSD estimates. For electrode-level spectral power analyses, intergroup comparisons among the TD, Non-TD, and HC groups were performed using one-way ANOVA when the homogeneity of variance assumption was met according to Levene’s test, and Welch’s ANOVA when this assumption was violated. Post hoc pairwise comparisons were conducted using least significant difference tests for equal-variance analyses and Games-Howell tests for unequal-variance analyses. Nominal two-tailed p values and FDR-adjusted q values were reported. FDR correction was applied across all scalp electrodes within each spectral power analysis. Results were considered significant after FDR correction when q < 0.05. Findings with nominal p < 0.05 but FDR-adjusted q ≥ 0.05 were interpreted as exploratory.

Hemispheric asymmetry indices were calculated for five homologous bilateral electrode pairs: F3-F4, C3-C4, P3-P4, O1-O2, and T3-T4, corresponding to frontal, central, parietal, occipital, and temporal regions, respectively. Associations between hemispheric asymmetry indices and clinical characteristics were assessed within the TD subgroup. Categorical clinical variables, including symptom subtypes, distribution, and history of rapid medication withdrawal, were analyzed using independent-sample t tests, with variance assumptions determined by Levene’s test. Continuous clinical variables were analyzed using correlation analyses according to their distribution: years of education was analyzed using Pearson correlation analysis, whereas duration of motor disorder was analyzed using Spearman rank correlation analysis because it was not normally distributed. FDR correction was applied within each clinical-variable family of asymmetry analyses. No asymmetry-clinical association remained significant after FDR correction; therefore, nominally significant findings were interpreted as exploratory.

Functional connectivity was assessed using phase-locking value (PLV) and phase-lag index (PLI).²² Before connectivity estimation, EEG data were transformed using current source density (CSD), a surface Laplacian spatial filtering approach, to reduce the influence of volume conduction and common reference effects. Functional connectivity was then computed for all pairwise combinations of the 19 scalp electrodes across frequency bands, generating symmetric connectivity matrices for each metric and frequency band per participant. Between-group comparisons of PLV and PLI were performed at the electrode-pair level. Network-based statistics (NBS) implemented in MATLAB were used to identify connected subnetworks showing group differences. The initial edge-wise threshold for defining suprathreshold connections was set at p < 0.01. Statistical significance of each network component was assessed using 5000 permutations following the default NBS procedure. Components were considered significant when their NBS-corrected component-level p value was < 0.05. Thus, p < 0.01 refers to the initial edge-wise threshold used for network construction, whereas NBS-corrected component-level p values refer to the corrected significance of the identified network components.

Given the modest sample size, particularly within the TD subgroup, subgroup and clinical association analyses were considered exploratory. The results were interpreted as hypothesis-generating rather than confirmatory.

Results

Study Demographics of TD, Non-TD, and HC Groups

A total of 67 participants were included in this study: 20 in the TD group, 25 in the Non-TD group, and 22 in the HC group. There were no significant differences in sex, χ² = 3.676, p = 0.152, or age, H = 0.222, p = 0.895, among the three groups (Table 1).

Table 1.

Demographic and Clinical Characteristics of TD, Non-TD, and HC Groups

	TD (n = 20)	Non-TD (n = 25)	HC (n = 22)	p
Age (years)	57 (18–75)	55 (18–75)	55 (32–68)	0.895 (H)
Male (n; %)	3 (15%)	10 (40%)	8 (36%)	0.152 (χ2)
Symptom Subtypes (n; %)
Orofacial	9 (45%)
Other	11 (55%)
Distribution (n; %)
Focal	13 (65%)
Generalized	7 (35%)
Rapid medication withdrawal (n; %)
Yes	12 (60%)
No	8 (40%)
Duration of motor disorder (months)	7.50 (1–72)
Admission AIMS Total score	7.80 ± 1.21

Notes: Continuous variables are shown as mean ± standard deviation or median (min-max), and categorical variables are shown as n (%). Group comparisons were performed using the Kruskal–Wallis H-test for age and the χ² test for sex.

Abbreviations: TD, tardive dyskinesia; Non-TD, patients without tardive dyskinesia; HC, healthy controls; AIMS, Abnormal Involuntary Movement Scale; SD, standard deviation; H, Kruskal–Wallis statistic; χ², chi-square statistic; p, probability value.

Spectral Power in the Awake Resting State: Group Comparisons Among TD, Non-TD, and HC Groups

Electrode-level spectral power analyses focused on theta relative power and beta absolute power. For theta relative power, significant group effects that survived FDR correction were observed at C4 and Cz (Table 2). At C4, theta relative power differed among the TD, Non-TD, and HC groups, F = 6.028, nominal p = 0.004, FDR-adjusted q = 0.038. Post hoc comparisons showed that both the TD and Non-TD groups had higher theta relative power than the HC group. At Cz, theta relative power also differed significantly among groups, F = 7.525, nominal p = 0.001, FDR-adjusted q = 0.019, with post hoc comparisons indicating higher theta relative power in both the TD and Non-TD groups relative to the HC group. The full electrode-level theta relative power results are provided in Supplementary Table S1. Several additional electrodes showed nominal group differences, mainly reflecting higher theta relative power in the Non-TD group than in the HC group; however, these findings did not survive FDR correction and were interpreted as exploratory.

Table 2.

Intergroup Comparisons of Theta Relative Power at Selected Electrodes Among TD, Non-TD, and HC Groups

Electrode	Group	Mean	SD	F	p	FDR-Adjusted q	FDR Significant	Post Hoc
C4	TD	0.109	0.041	6.028	0.004	0.038	Yes	TD > HC (p = 0.038)
	Non-TD	0.137	0.042
								Non-TD > HC (p = 0.001)
	HC	0.093	0.046
Cz	TD	0.120	0.041	7.525	0.001	0.019	Yes	TD > HC (p = 0.004)
	Non-TD	0.165	0.055
								Non-TD > HC (p = 0.001)
	HC	0.114	0.048

Notes: Theta relative power was calculated as theta-band power divided by total spectral power and is therefore unitless. Intergroup comparisons among the TD, Non-TD, and HC groups were conducted using one-way ANOVA, with results reported as F values. Nominal two-tailed p values and FDR-adjusted q values are reported. Results were considered significant after FDR correction when q < 0.05. Post hoc comparisons were performed using least significant difference tests when the overall group effect was significant. Descriptive statistics are presented as mean and standard deviation (SD) in separate columns. This table presents selected electrodes showing significant intergroup differences in theta relative power after FDR correction. The full results across all scalp electrodes are provided in Supplementary Table S1. Post hoc comparisons are shown when significant pairwise group differences were observed. p < 0.05 indicates nominal statistical significance.

Abbreviations: TD, tardive dyskinesia; Non-TD, patients without tardive dyskinesia; HC, healthy controls; SD, standard deviation; ANOVA, analysis of variance; FDR, false discovery rate; F, ANOVA F statistic; p, nominal probability value; q, FDR-adjusted probability value.

In beta absolute power analyses, nominal group differences were observed at Fz and Cz, where post hoc comparisons showed higher beta absolute power in the TD group than in the HC group. These findings did not survive FDR correction and are therefore reported in Supplementary Table S2 as exploratory results.

In addition to electrode-level spectral power analyses, cluster-based topographic analyses showed group-related gamma-band PSD differences. Compared with the HC group, the TD group exhibited increased PSD in the gamma2–gamma3 sub-bands, 36–44 Hz, over frontoparietal electrodes, including Fp1, Fp2, F7, Fz, Cz, and Pz (Figure 1a). In contrast, compared with the HC group, the Non-TD group showed increased broadband gamma PSD, 30–50 Hz, primarily across frontal and central electrodes, including Fp1, Fp2, F3, Fz, and Cz (Figure 1b). Both clusters were identified using cluster-based permutation testing with 5000 permutations and reached cluster-level significance at p < 0.05. No significant PSD differences were observed between the TD and Non-TD groups in any frequency band.

Figure 1.

Group-level PSD alterations in resting-state EEG. (a) Topographic distribution of gamma2-gamma3 band PSD differences between the TD and HC groups, showing increased frontoparietal PSD in TD participants. (b) Topographic distribution of broadband gamma PSD differences between the Non-TD and HC groups, showing increased PSD primarily over frontal and central electrodes in the Non-TD group. PSD differences were assessed using cluster-based permutation testing with 5000 permutations, and cluster-level significance was set at p < 0.05. The color bar indicates the test statistic for group differences.

Notes: gamma2-gamma3, 36–44 Hz; broadband gamma, 30–50 Hz; p, nominal probability value.

Abbreviations: TD, tardive dyskinesia; Non-TD, patients without tardive dyskinesia; HC, healthy controls; EEG, electroencephalography; PSD, power spectral density.

In summary, theta relative power differences that survived FDR correction were localized to central electrodes, whereas gamma-band PSD differences identified by cluster-based analyses were distributed over frontoparietal and frontal-central regions.

Exploratory Associations Between Rs-EEG Asymmetry Indices and Clinical Characteristics Within the TD Subgroup

After group-level spectral analyses, exploratory analyses were performed within the TD subgroup to examine associations between hemispheric asymmetry indices and TD-related clinical characteristics. No asymmetry-clinical association remained significant after FDR correction; therefore, selected nominal associations are summarized for exploratory purposes (Table 3), with the full results provided in Supplementary Table S3.

Table 3.

Selected Associations Between Rs-EEG Asymmetry Indices and Clinical Characteristics in the TD Subgroup

Variables	Band	Symptom Subtypes (Orofacial = 9/Other = 11)		Distribution (Focal = 13/ Generalized = 7)		Rapid Medication Withdrawal (Yes = 12/No = 8)		Years of Education (Years)		Duration of Motor Disorder (Months)
		t	p	t	p	t	p	r	p	ρ	p
Central Asymmetry	Delta	2.049	0.055	0.738	0.470	−2.728	0.014	−0.086	0.719	−0.054	0.820
	Theta	1.654	0.115	0.538	0.597	−2.956	0.008	0.079	0.739	0.039	0.872
Temporal Asymmetry	Delta	−0.806	0.431	−0.256	0.802	1.053	0.306	0.078	0.743	−0.491	0.028
	Theta	−1.290	0.213	−0.409	0.689	1.410	0.176	0.054	0.822	−0.521	0.018
	Alpha	−1.338	0.198	−0.134	0.895	0.977	0.342	0.087	0.716	−0.450	0.046
	Beta	−1.123	0.276	−0.215	0.832	1.363	0.190	0.060	0.802	−0.462	0.040

Notes: Hemispheric asymmetry indices were calculated for homologous electrode pairs, including F3-F4, C3-C4, P3-P4, O1-O2, and T3-T4, corresponding to frontal, central, parietal, occipital, and temporal regions, respectively. This table presents selected nominal associations observed within the TD subgroup. The full results across all asymmetry indices, frequency bands, and clinical variables are provided in Supplementary Table S3. Categorical clinical variables, including symptom subtypes, distribution, and history of rapid medication withdrawal, were analyzed using independent-sample t tests, with variance assumptions determined by Levene’s test. Continuous clinical variables were analyzed using correlation analyses according to their distribution: years of education was analyzed using Pearson correlation analysis, whereas duration of motor disorder was analyzed using Spearman rank correlation analysis because it was not normally distributed. Results are reported as t values for categorical variables, Pearson’s r values for years of education, and Spearman’s ρ values for duration of motor disorder. FDR correction was applied within each clinical-variable family of asymmetry analyses, and none of the selected associations remained significant after FDR correction; therefore, these findings should be interpreted as exploratory. EEG frequency bands were defined as delta, 0.5–4 Hz; theta, 4–8 Hz; alpha, 8–13 Hz; beta, 14–30 Hz; and gamma, 30–50 Hz. p values are nominal two-tailed values. p < 0.05 indicates nominal statistical significance.

Abbreviations: TD, tardive dyskinesia; rs-EEG, resting-state electroencephalography; EEG, electroencephalography; FDR, false discovery rate; t, t statistic; r, Pearson correlation coefficient; ρ, Spearman rank correlation coefficient; p, nominal probability value.

For categorical clinical variables, central asymmetry showed nominal associations with history of rapid medication withdrawal. Specifically, central delta asymmetry differed according to rapid medication withdrawal status, t = −2.728, nominal p = 0.014, and central theta asymmetry showed a similar nominal association, t = −2.956, nominal p = 0.008. No selected nominal associations were observed for symptom subtypes or symptom distribution.

For continuous clinical variables, duration of motor disorder showed nominal negative correlations with temporal asymmetry indices in several frequency bands. Specifically, longer duration of motor disorder was associated with lower temporal asymmetry indices in the delta, theta, alpha, and beta bands within the TD subgroup, as assessed using Spearman rank correlation: delta, ρ = −0.491, p = 0.028; theta, ρ = −0.521, p = 0.018; alpha, ρ = −0.450, p = 0.046; and beta, ρ = −0.462, p = 0.040. Years of education did not show clear associations with hemispheric asymmetry indices.

Functional Connectivity Changes in the Awake Resting State Among TD, Non-TD, and HC Groups

Group-level functional connectivity analyses revealed frequency-specific network alterations across the TD, Non-TD, and HC groups within NBS-corrected components. Compared with the HC group, individuals with TD showed reduced theta-band PLV connectivity within the identified NBS component (Figure 2a). PLI analyses also showed reduced alpha1-band connectivity in TD compared with HC (Figure 2b), consistent with findings from a connectivity metric that is less sensitive to zero-lag synchronization and volume conduction.

Figure 2.

(a) Theta-band PLV network differences between the TD and HC groups, showing reduced theta-band functional connectivity in TD participants. (b) Alpha1-band PLI network differences between the TD and HC groups, showing reduced alpha1-band connectivity in TD participants. (c) Beta-band PLI network differences between the TD and Non-TD groups, showing reduced beta-band connectivity in TD participants. (d) Gamma-band PLI network differences between the TD and Non-TD groups, showing reduced gamma-band connectivity in TD participants. (e) Theta-band PLV network differences between the Non-TD and HC groups, showing reduced theta-band connectivity in Non-TD participants. (f) Gamma-band PLI network differences between the Non-TD and HC groups, showing gamma-band connectivity differences in Non-TD participants. Network-based statistics were performed using an initial edge-wise threshold of p < 0.01 and 5000 permutations. Displayed networks represent NBS-corrected components reaching component-level significance at p < 0.05. Blue lines indicate constituent suprathreshold connections within the NBS-corrected components. Detailed component-level corrected p values, edge-level test statistics, group mean connectivity values, and directions of group differences are provided in Supplementary Table S4.

Notes: p, nominal probability value; theta, 4–8 Hz; alpha1, 8–10 Hz; beta, 14–30 Hz; gamma, 30–50 Hz.

Abbreviations: TD, tardive dyskinesia; Non-TD, patients without tardive dyskinesia; HC, healthy controls; EEG, electroencephalography; PLV, phase-locking value; PLI, phase-lag index; NBS, network-based statistics.

Direct comparisons between TD and Non-TD patients revealed additional high-frequency connectivity differences. Specifically, TD patients showed reduced beta-band and gamma-band PLI connectivity relative to Non-TD patients within NBS-corrected components (Figure 2c and d). In contrast, Non-TD patients showed a different connectivity profile relative to HC. Non-TD patients exhibited reduced theta-band PLV connectivity (Figure 2e), whereas gamma-band PLI connectivity was increased compared with HC (Figure 2f).

Overall, TD and Non-TD patients showed partly distinct resting-state connectivity patterns across theta, alpha1, beta, and gamma frequency bands. Detailed constituent suprathreshold connections within each NBS-corrected component are provided in Supplementary Table S4.

Discussion

This study examined resting-state EEG alterations in TD patients with primary mood disorders using spectral power, hemispheric asymmetry, and functional connectivity analyses. Three main observations emerged. Firstly, electrode-level analyses showed higher theta relative power at central electrodes, particularly C4 and Cz, and these group effects survived FDR correction. Cluster-based topographic analyses also revealed gamma-band PSD differences, with increased gamma2-gamma3 power over frontoparietal regions in TD patients and increased broadband gamma power over frontal-central regions in Non-TD patients relative to healthy controls. Besides, within the TD subgroup, hemispheric asymmetry indices showed nominal associations with rapid medication withdrawal and duration of motor disorder, although these associations did not remain significant after FDR correction. Meanwhile, connectivity analyses identified frequency-specific network differences, including reduced theta PLV and reduced PLI connectivity across alpha1, beta, and gamma frequency bands in TD-related comparisons. Taken together, the results indicate that TD is associated with altered resting-state oscillatory activity and functional connectivity, while the clinical association findings remain exploratory.

The increase in theta relative power at C4 and Cz indicates altered central cortical oscillatory activity in both TD and Non-TD patients compared with HC. Theta activity has been linked to arousal, sensorimotor integration, and broader neuropsychiatric state-related processes;^23–25 therefore, this finding should not be interpreted as specific to TD movements. Even so, the central topography of the theta findings is relevant to TD research, because abnormal involuntary movements are closely related to sensorimotor and cortico-basal ganglia circuits.^10,26 Prior local field potential recordings from the globus pallidus internus in patients with tardive dystonia or related tardive syndromes have also reported enhanced theta activity, which is broadly consistent with the possibility that theta-band abnormalities participate in cortico-basal ganglia dysfunction related to tardive motor phenomena.²⁷ However, scalp EEG and subcortical recordings differ substantially in signal source and spatial scale. Moreover, theta relative power was increased in both TD and Non-TD groups in the present study. The shared theta increase may therefore reflect features of the clinical mood disorder population, prior psychotropic medication exposure, altered central cortical regulation, or a combination of these factors. Studies combining scalp EEG with detailed motor ratings, cognitive assessment, medication-exposure data, and, where feasible, subcortical recordings will be needed to clarify the meaning of central theta changes in TD.

Gamma-band PSD also differed across groups, but the topographic patterns were not identical. TD patients showed increased gamma2-gamma3 PSD over frontoparietal regions, whereas Non-TD patients showed increased broadband gamma PSD over frontal-central regions. Gamma activity is commonly linked to local cortical processing, sensorimotor integration, and large-scale neural coordination. At the same time, scalp-recorded gamma activity is vulnerable to residual muscle activity and other non-neural influences, a concern that is especially relevant in movement disorder studies. In this study, ICA-based artifact rejection and visual EEG review were used to reduce artifact contamination.²⁸ Nevertheless, the gamma findings should be interpreted with caution.²⁹ Rather than being taken as evidence of TD-specific gamma pathology, these results point to group-related differences in frontoparietal and frontal-central oscillatory patterns. Replication in larger cohorts with stricter artifact and medication control will be needed to determine whether TD is associated with frequency-specific activation patterns in particular cortical regions and whether these regions reflect potential nodes of TD-related network dysfunction.

The exploratory asymmetry analyses showed nominal associations between central asymmetry and rapid medication withdrawal, and between temporal asymmetry and duration of motor disorder. These results are clinically meaningful in the sense that dopamine receptor blocking exposure is central to TD, and medication reduction may be related to dopamine-supersensitivity-related vulnerability.^30,31 These nominal associations are also consistent with clinical observations that medication reduction and longer symptom duration may influence the clinical expression of TD. However, the associations did not survive FDR correction, and the TD subgroup was small. They should therefore be treated as hypothesis-generating rather than confirmatory. It is also important not to interpret hemispheric asymmetry as a direct electrophysiological correlate of TD movements themselves. The observed nominal associations may reflect TD vulnerability, medication-related factors, illness duration, and individual neurophysiological variability. Longitudinal work will be required to determine whether asymmetry indices change with symptom progression, medication adjustment, or treatment response.

Functional connectivity analyses showed additional frequency-specific network differences across groups. Compared with healthy controls, TD patients showed reduced theta PLV and reduced alpha1 PLI connectivity. Compared with Non-TD patients, TD patients showed reduced PLI connectivity in beta and gamma frequency bands. Before connectivity estimation, CSD transformation was applied to reduce the influence of common reference and volume-conduction effects. In addition, PLI reduces the influence of zero-lag synchronization compared with conventional phase-synchronization measures.³² Therefore, the connectivity findings are less likely to be driven solely by common reference or volume-conduction effects and provide complementary information to the PLV results. The beta and gamma findings are particularly relevant to motor control, sensorimotor integration, and cortico-basal ganglia communication.^29,33–35 Nevertheless, these results should still be understood as cross-sectional associations rather than evidence that altered connectivity causes TD movements.

The present findings also need to be considered alongside the existing TD brain network literature, which has largely focused on schizophrenia-associated TD. Previous structural and functional MRI studies in schizophrenia patients with TD have reported gray matter abnormalities, altered regional homogeneity, and abnormal functional connectivity involving motor and cortico-striatal circuits.^8–11 These studies support the view that TD may involve distributed network alterations rather than isolated abnormalities of motor output. The present study differs by focusing on TD in patients with primary mood disorders and by excluding schizophrenia or other psychotic disorders. This approach reduces confounding from schizophrenia-related cortical, oscillatory, and connectivity abnormalities, but it also makes direct comparison with schizophrenia-associated TD more limited. The convergence between prior fMRI evidence of motor-circuit abnormalities and the present EEG connectivity findings suggests that network-level dysfunction may be relevant across TD populations.^10,11 However, the current data cannot determine whether the observed EEG patterns are specific to mood disorder-associated TD or generalizable to TD in other diagnostic contexts. Comparative studies including schizophrenia-associated TD, mood disorder-associated TD, psychiatric controls without TD, and healthy controls are needed to distinguish shared from diagnosis-specific neurophysiological features.

The clinical implications of these findings should be stated cautiously. At present, the results do not provide sufficient evidence to support using EEG to diagnose TD or to differentiate TD from other movement disorders in routine practice. Their value is better viewed as mechanistic: they identify candidate neurophysiological dimensions that may be useful in future TD research. Central theta power, frontoparietal gamma activity, hemispheric asymmetry, and frequency-specific connectivity may serve as intermediate phenotypes for studying TD vulnerability, medication-related risk, symptom duration, or treatment response. To establish clinical utility, future studies should examine whether these EEG features are associated with psychiatric symptom burden, cognitive impairment, functional outcome, longitudinal symptom progression, or response to interventions such as medication adjustment, VMAT2 inhibitors, or non-invasive neuromodulation.^36,37 Until such validation is available, these EEG findings should be regarded as preliminary markers of TD-related neurophysiological status rather than clinically actionable biomarkers.

Several limitations should be acknowledged. First, the cross-sectional design precludes causal inference; therefore, the observed EEG alterations should not be interpreted as electrophysiological signatures specific to TD movements alone. Second, although recent exposure to benzodiazepines and anticholinergic agents was reviewed when available and EEG recordings with clear medication-related abnormalities were excluded, detailed washout duration and cumulative psychotropic exposure were not systematically available for all participants, and residual medication-related effects therefore cannot be fully excluded. Third, the TD subgroup size limited statistical power, especially for clinical association and subgroup analyses. Although FDR correction was applied, the asymmetry-clinical associations did not survive correction and should be interpreted as exploratory. Given the modest sample size and the clinical complexity of TD, these findings may be influenced by TD-related vulnerability, prior exposure to dopamine receptor blocking agents, medication reduction, and individual differences in clinical background. Therefore, caution is needed when linking these EEG findings directly to TD motor manifestations. Fourth, the study did not include detailed assessments of cognitive function, psychiatric symptom burden, functional outcomes, or treatment response; as a result, the clinical relevance of the EEG findings beyond electrophysiological characterization remains uncertain. Finally, the standard 19-channel EEG montage limited spatial resolution, and larger longitudinal studies using high-density EEG, source localization, and multimodal imaging will be important for validating and extending these findings.

In summary, TD in patients with primary mood disorders was associated with altered resting-state spectral power and frequency-specific functional connectivity, together with exploratory associations between hemispheric asymmetry and selected clinical characteristics. These findings provide preliminary electrophysiological evidence for cortical network involvement in TD. They should, however, be interpreted cautiously because of the modest sample size, cross-sectional design, and potential medication-related confounding. Future longitudinal and independently replicated studies are needed to determine whether these EEG features can inform mechanisms of TD vulnerability, symptom progression, or treatment response.

Funding Statement

This study was supported by the National Natural Science Foundation of China, Grant/Award Numbers: 82071552 and 22376006; the Chinese Academy of Sciences, Grant/Award Number: JCTD-2021-06; the Peking University Clinical Scientist Program, Grant/Award Number: BMU2023PYJH018; the Beijing Research Ward Excellence Program, Grant/Award Numbers: BRWEP2024W074110108 and BRWEP2024W074110120; and the Beijing High-Level Innovation and Entrepreneurship Talent Support Program, Grant/Award Number: G202521097.

Data Sharing Statement

The datasets generated and/or analysed during the current study are not publicly available due to ethical and privacy restrictions related to human participant data, but are available from the corresponding author on reasonable request.

Consent Statement

All human subjects provided informed consent.

Disclosure

All authors report no conflicts of interest in this work.