Corticostriatal glutamate mechanisms underlying beta synchrony and motor deficits via striatal NMDA receptors in Parkinson’s disease

Summary

Background

Beta-band (13–30 Hz) oscillations in the cortico-basal ganglia-thalamic (CBT) network strongly correlate with motor deficits in Parkinson’s disease (PD), yet their synaptic origins remain unclear. Given that dopamine (DA) loss is necessary but not sufficient to produce sustained beta rhythms, we hypothesised that corticostriatal glutamatergic overdrive may function as a significant non-dopaminergic amplifier of pathological synchrony.

Methods

Using an integrated experimental-computational approach, we combined 6-hydroxydopamine (6-OHDA) male rat models, ex vivo striatal patch-clamp recordings, chemogenetic modulation of corticostriatal projection, and multiscale computational network modelling to examine beta oscillation dynamics in the CBT network.

Findings

Early DA denervation caused akinesia without beta elevation, while advanced degeneration triggered robust high-beta (25–40 Hz) oscillations and increased corticostriatal coherence. Ex vivo, medium spiny neurons (MSNs) exhibited heightened presynaptic glutamate release correlated with beta power. Computational modelling showed that excessive corticostriatal input under DA depletion increased MSN synchrony, disrupted striatal decorrelation, and was associated with the emergence of pathological beta rhythms, effects reversed by reducing glutamatergic input. In vivo chemogenetic silencing of corticostriatal projections suppressed beta synchrony and improved motor performance in 6-OHDA rats, whereas activation in DA-intact rats had no effect. Notably, striatal NMDA, not AMPA, receptor blockade reduced beta oscillations and motor deficits. Network simulations implicated the subthalamic → motor cortex feedback loop in the maintenance of this pathological beta state.

Interpretation

Corticostriatal glutamatergic overdrive, through NMDA receptor-dependent signalling, is linked to the amplification and propagation of beta synchronisation across the CBT circuit, highlighting it as a potential biomarker and a promising therapeutic target in PD.

Funding

This research was supported by the National Natural Science Foundation of China (32271173, 82371256) and the Natural Science Foundation of Beijing Municipality (7242214, 7252213). This study was also supported by the Swedish Research Council (VR-M-2020-01652), the Swedish e-Science Research Centre (SeRC), Science for Life Laboratory, KTH Digital Future, EU/Horizon 2020 No. 945539 (HBP 935 SGA3) and No. 101147319 (EBRAINS 2.0 Project), the European Union’s Research and Innovation Program Horizon Europe under grant agreement No. 101137289(the Virtual Brain Twin Project).

Research in context.

Evidence before this study

Motor impairment in Parkinson’s disease (PD) is strongly linked to excessive beta-band (13–30 Hz) synchronisation within the cortico–basal ganglia–thalamic circuit. Such abnormal coupling is widely recognised as a hallmark of PD-related motor dysfunction. Although nigrostriatal dopamine (DA) loss is the primary cause of PD’s motor symptoms, DA loss alone is necessary but not sufficient to produce sustained beta oscillations. This discrepancy suggests that additional non-dopaminergic mechanisms act as an amplifier that converts DA deficiency into widespread network pathology. Previous studies have shown that chronic DA deficiency enhances glutamatergic transmission from the cortex to the striatum, the main input structure of the basal ganglia. However, it has remained unclear whether this glutamatergic hyperactivity directly drives beta synchronisation across the motor network.

Added value of this study

Using integrated experimental and computational approaches, we identified the glutamatergic contribution to transforming DA loss into extensive beta synchrony. In this study, we found that enhanced glutamatergic input to the striatum does not initiate beta oscillations but functions as a permissive amplifier under DA-depleted conditions. In animals with intact dopaminergic tone, increased glutamatergic activity alone did not alter beta coherence, confirming that its pathological effect depends on DA loss. Furthermore, we demonstrated that NMDA, rather than AMPA receptors, mediate this abnormal amplification, providing a defined molecular substrate. Additionally, network simulations revealed that the subthalamic nucleus → motor cortex feedback loop is crucial for sustaining the propagation of these abnormal rhythms.

Implications of all the available evidence

These findings advance the understanding of PD pathophysiology by identifying a glutamatergic pathway that amplifies dopaminergic network dysfunction. Targeting NMDA receptor–mediated glutamatergic signalling may thus offer a promising non-dopaminergic approach to suppress pathological synchrony. Moreover, glutamatergic overactivity and related network coupling could serve as biomarkers of disease progression. Recognising that beta synchronisation is maintained through the subthalamic nucleus → cortical loop may also guide more precise deep brain stimulation (DBS) interventions. Overall, this study extends the conceptual framework of PD beyond DA deficiency and suggests new directions for therapeutic intervention and monitoring.

Introduction

Parkinson’s disease (PD) is characterised by the progressive degeneration of nigrostriatal dopamine (DA) neurons, leading to the hallmark motor symptoms of bradykinesia, rigidity, and resting tremor. Within the cortico-basal ganglia-thalamic (CBT) network, exaggerated beta-band (13–30 Hz) oscillations are correlated strongly with motor impairment severity,1, 2, 3, 4, 5 and are transiently suppressed by levodopa or deep brain stimulation (DBS).6, 7, 8 Despite this strong correlation, DA loss alone appears necessary but not sufficient for the immediate generation of sustained beta rhythms.9, 10, 11, 12 In experimental models, sustained pathological beta rhythms typically emerge only after chronic and extensive denervation, whereas acute DA depletion or pharmacological blockade fails to generate robust beta power.^9,11, 12, 13 This suggests that dopamine deficiency, while a fundamental requirement, is not the exclusive driver of pathological beta synchrony,^14,15 implying additional, non-dopaminergic factors contribute to its emergence.

Classical models have proposed several explanations for beta generation, including resonant activity within the subthalamic nucleus–external globus pallidus (STN–GPe) loop,16, 17, 18 cortical entrainment via the hyperdirect pathway (M1 → STN),17, 18, 19 and intrinsic striatal circuits involving cholinergic interneurons (ChIs)20, 21, 22 or somatostatin (SOM) interneurons interacting with D2 receptor-expressing medium spiny neurons (iMSNs).²³ Yet, these do not fully account for how beta activity becomes coherent and persists across the entire CBT network.^14,17,18

Accumulating evidence now points towards excessive glutamatergic transmission, particularly N-methyl-D-aspartate (NMDA) receptor-mediated excitation, as a crucial non-dopaminergic contributor to beta pathology. Enhanced cortical excitation has been shown to boost beta power and exacerbate motor dysfunction,²⁴ while DA loss further intensifies corticostriatal glutamatergic drive.^21,23,25,26 This promotes hyperactivity among striatal MSNs,27, 28, 29, 30 and synchrony throughout the basal ganglia circuits.^27,28,31, 32, 33 Taken together, these findings suggest that glutamatergic overdrive functions as a “secondary trigger”, transforming dopaminergic deficiency into a network-wide beta abnormality.

Nevertheless, two fundamental questions remain unresolved: (1) Does corticostriatal presynaptic glutamatergic overdrive contribute to the emergence and propagation of beta oscillations across the CBT network?17, 18, 19, 20, 21, 22, 23^,34 and (2) Through which receptor- and pathway-specific mechanisms does cortical excitation link to striatal beta synchrony?19, 20, 21, 22, 23^,26,27

To address these critical issues, we adopted an integrated experimental-computational framework. We used an in vivo 6-OHDA rat model to monitor motor deficits and network beta dynamics, ex vivo patch-clamp recordings to assess presynaptic glutamate release, chemogenetic manipulation of corticostriatal projections to examine their contribution to pathological beta activity, and multiscale computational modelling to simulate how altered synaptic transmission may be associated with the emergence and propagation of beta oscillations throughout the CBT loop. Together, these complementary approaches allowed us to delineate a presynaptically driven, NMDA receptor-dependent beta mechanism, providing a mechanistic biomarker and a potential non-dopaminergic therapeutic intervention in PD.

Methods

Animals

A total of 114 adult male Sprague–Dawley rats were initially randomised into experimental cohorts. Animals were included in the final analysis only if they satisfied the following pre-defined inclusion criteria: (i) successful PD model induction, confirmed by apomorphine-induced rotations (≥5 net contralateral turns/min); (ii) histologically verified accuracy of recording electrode tracks in M1 and STR, drug infusion cannulas, or viral expression sites; and (iii) stable LFP signal integrity free of excessive noise or system saturation. Across all cohorts, a total of 12 rats were excluded based on these criteria (5 failed model induction, 5 misplaced electrodes/cannulas, and 2 off-target viral expression), resulting in a final analysed population of 102 rats.

6-OHDA lesion surgery

Under pentobarbital (40 mg/kg, i.p.) anaesthesia, unilateral parkinsonian lesions were produced by stereotaxic injection of 6-OHDA (8 μg in 1.6 μL, with 0.02% ascorbic acid) into the right medial forebrain bundle (MFB) at AP: −4.3 mm, ML: −1.5 mm, DV: −7.6 mm from bregma. Sham controls received an equal volume of vehicle injection. For the progressive behavioural and electrophysiological experiments, a total of 42 rats were initially randomised into sham or 6-OHDA-lesioned conditions. Two rats were excluded due to electrode misplacement, leaving 40 rats (n = 8 animals per group) for the final assessments, and testing was conducted at defined post-lesion timepoints (3, 5, 7, and 14 days). For all subsequent chemogenetic or pharmacological experiments, successful PD model induction was confirmed 14 days post-surgery using the apomorphine-induced contralateral rotation test (≥5 turns/min).^29,35

Behavioural assays

Forelimb use asymmetry, indicative of akinesia, was assessed via the cylinder test and forelimb adjusting steps (FAS) test.²⁴ Results were reported as relative ratios (%) of the affected (left) to the unaffected (right) forelimb.³³ Bradykinesia was evaluated using a 30-min open-field test within a 50 × 50 × 50 cm cage. This test captured key metrics such as total distance travelled, movement duration, and mean velocity.^24,29 Motor coordination and balance were assessed using an accelerating rotarod apparatus (Stoelting Co.). Rats were placed on the rod, which accelerated from 4 to 40 rpm over 5 min. The latency to fall was recorded for three trials per session, and the mean value was used for analysis. This assay was specifically used in chemogenetic experiments to evaluate the functional rescue of motor coordination following circuit modulation.

Histology and immunofluorescence

Forty-micrometre coronal sections were immunostained for Tyrosine Hydroxylase (TH) (1:2000, Cat. No. 22941, Immunostar, RRID: AB_572268). Quantification of TH-positive cells was performed using unbiased stereology and image analysis software (ImageJ), and results were normalised to the contralateral side.³⁶ Immunofluorescence for αCaMKII (1:1000, Cat. No. SAB4503244, Sigma, RRID: AB_10694086), mCherry (1:2000, Cat. No. LS-C204825, LSBio, RRID: AB_2716246) and c-Fos (1:1000, Cat. No. 226308, SYSY, RRID: AB_2905595) were used to validate chemogenetic targeting using a Zeiss 880 confocal microscope.

Implantation of a microinfusion cannula

To assess behavioural and electrophysiological responses in rats during local infusion of neurotransmitter blockers, a microinfusion cannula was implanted into the striatum. Each rat was anaesthetised and mounted onto a stereotactic instrument (David Kopf Instruments) with the skull exposed. A stainless-steel cannula (C315G; Plastics One) was inserted perpendicularly into the striatum using the coordinates (AP: +0.15 mm, ML: −2.6 mm, DV: −3.3 mm from bregma). The cannula was fixed with bone cement on the skull, and the wound was closed with 4-0 nylon monofilament. The selective NMDA receptor blocker cis-4-[phosphomethyl]-piperidine-2-carboxylic acid (CGS, 5 mM; Tocris) and the AMPA receptor blocker 6-cyano-7-nitroquinoxaline-2,3-dione disodium (CNQX, 5 mM; Tocris) were used as glutamate receptor blockers in this study. The experiments and recordings were performed 1–2 weeks after the surgery.²⁴

Slice electrophysiology

Acute coronal striatal slices (250 μm) were prepared in ice-cold slicing solution (in mM): 110 choline chloride, 2.5 KCl, 1.3 NaH₂PO₄·2H₂O, 25 NaHCO₃, 25 glucose, 0.6 Na-pyruvate, 1.3 Na-ascorbate, 0.5 CaCl₂·2H₂O, and 7.0 MgSO₄. The slices were maintained in an artificial cerebrospinal fluid (ACSF) solution gassed with 95% O₂ and 5% CO₂ at 32 °C for 30 min and then equilibrated for at least 1 h at room temperature before use. The ACSF composition (in mM) included: 125 NaCl, 2.5 KCl, 1.3 NaH₂PO₄·2H₂O, 25 NaHCO₃, 20 glucose, 0.6 Na-pyruvate, 1.3 Na-ascorbate, 2 CaCl₂·2H₂O, and 1.3 MgSO₄. Whole-cell voltage-clamp recordings were obtained from dorsal MSNs using a HEKA EPC-10 amplifier. The signals were low-pass filtered at 1 kHz and digitised at 10 kHz. To isolate glutamatergic transmission, GABA-A receptors were blocked with 100 μM picrotoxin (Tocris). Spontaneous and miniature excitatory postsynaptic currents (sEPSCs and mEPSCs) were recorded at a holding voltage of −70 mV. Synaptic events were analysed offline using a template matching algorithm within the Mini Analysis program. The amplitude of EPSCs was defined as the mean peak value of all detected events, while frequency was calculated as the number of events per second. Data were first averaged across 3–5 neurons per animal, and group statistics were then performed using the mean values from n = 3 rats per group to ensure biological independence. For mEPSCs, 1 μM tetrodotoxin (TTX, Tocris) was added to the ACSF to block action potentials. Evoked EPSCs (eEPSCs) were elicited by stimulating corticostriatal fibres via an electrode in the corpus callosum to assess paired-pulse ratios (PPR).²⁸ To measure the AMPA/NMDA current ratio, AMPAR-mediated EPSCs were quantified by their peak amplitude at a holding potential of −70 mV. Subsequently, NMDAR-mediated EPSCs were measured at a holding potential of +40 mV, leveraging the distinct voltage-dependent properties of these receptors.³⁷

In vivo local field potential (LFP) recordings and analysis

Custom 8-channel microwire arrays were surgically implanted in the right M1 (AP: +1.5, ML: −2.8, DV: −1.3 mm) and ipsilateral dorsolateral striatum (AP: +0.2, ML: −2.8, DV: −3.8 mm). After two weeks of recovery, LFPs were recorded from awake, freely moving rats using the Cerebus system (30 kHz sampling, 0.1–250 Hz hardware filter, down-sampled to 1 kHz). For each brain region, electrodes were selected based on the highest signal-to-noise ratio (SNR). To avoid contamination by movement-related artefacts and the physiological suppression of beta power during locomotion, all quantitative analyses were restricted to alert rest epochs. These epochs were identified via synchronised video monitoring as periods of immobility while the animals remained awake and responsive. From these periods, 60-s artefact-free segments were selected for analysis. Signals were preprocessed using a 50 Hz notch filter and a zero-phase fourth-order Butterworth bandpass filter to preserve phase integrity. Segments containing high-amplitude transients or mechanical interference were identified by visual inspection and an automated threshold (excluding segments where the amplitude exceeded 4 standard deviations of the mean) and excluded prior to spectral and coherence computations.²⁹ Power Spectral Density (PSD) was computed using the multitaper fast-Fourier transform (FFT) method (NW = 3.5, 5 tapers) with a 1-s window, 50% overlap, and a ±2 Hz smoothing bandwidth.³⁸ Corticostriatal coherence was quantified using the imaginary part of coherence calculated in 2-s epochs to minimise volume conduction artefacts.

Chemogenetics

To selectively manipulate corticostriatal projection neurons, a retrograde AAV2/retro-αCaMKII-P2A-CRE-WPRE-pA (2 × 10¹² vg/mL, Cat. No. PT-9078, BrainVTA) was unilaterally injected into the striatum (AP: +0.2, ML: −2.8, DV: −3.8 mm). Concurrently, Cre-dependent AAV2/9-Ef1α-DIO-hM3D(Gq)-mCherry or AAV2/9-Ef1α-DIO-hM4D(Gi)-mCherry (both at 2 × 10¹² vg/mL, Cat. No. PT-0042/PT-0043, BrainVTA) were infused into the ipsilateral M1 (AP: +1.5, ML: −2.8, DV: −1.3 mm). rAAV-Ef1α-DIO-mCherry was used as a control virus (2 × 10¹² vg/mL, Cat. No. PT-0285, BrainVTA). Viral injections were delivered at 80 nL/min via a Hamilton syringe. After 3–4 weeks for viral expression, chemogenetic manipulation was induced with intraperitoneal clozapine N-oxide (CNO; 3 or 10 mg/kg, Cat. No. 4936, Tocris).²⁹

Computational modelling methods

Striatal microcircuit

A leaky integrate-and-fire (LIF) spiking network model of the striatal microcircuit was employed, comprising 4000 MSNs (2000 D1-MSNs, 2000 D2-MSNs) and 80 fast-spiking interneurons (FSIs). Neurons received independent Poisson-distributed cortical input (Fig. 1a). Network connectivity included sparse GABAergic MSN–MSN inhibition and feed-forward inhibition from FSIs to MSNs. The model incorporated asymmetric connectivity where D2-MSNs form more and stronger inhibitory connections onto D1-MSNs, and FSIs preferentially targeted D1-MSNs over D2-MSNs.^26,39,40 Parameters for this microcircuit are detailed in Supplementary Tables S1 and S2.

Fig. 1.

Striatal microcircuit dynamics in response to cortical glutamatergic inputs under control and DA-depleted conditions. (a and b) Schematic of striatal microcircuit in control (Ctrl) and DA-depleted (No DA) conditions. DA-depletion includes increased synaptic connections from FSIs onto D2-medium spiny neurons (D2-MSN), reduced lateral inhibition among MSNs, and loss of presynaptic connections from D1-MSN to D2-MSN. Solid lines with a red ‘X' denote the complete removal of connections, while dashed lines serve as a schematic representation of significantly weakened lateral inhibition. These changes represent the functional collapse of local decorrelation mechanisms (see Supplementary Table S6 for exact parameters). (c) Diagram of the cortical input correlation model. Parameter W denotes correlation within glutamatergic input to individual MSNs, and parameter B denotes the shared correlation among inputs to different MSNs. (d) Raster plots (top) and population activity (bottom) of D1- and D2-MSNs in control conditions, illustrating their typically low and asynchronous firing patterns. (e) Mean firing rates of D1- and D2-MSNs in response to varying cortical glutamatergic input strengths. Data are presented for four distinct conditions, including control (Ctrl), control with combined AMPA and NMDA receptor activation (Ctrl + Glu Exc), DA depletion (No DA), and DA depletion with combined AMPA and NMDA receptor blockade (No DA + Glu Inh). (f) Comparison of striatal high-beta power (25–40 Hz) across four conditions: Ctrl (black), Ctrl + Glu Exc (yellow), No DA (red), and No DA + Glu Inh (blue). (g–k) Striatal power spectral density (PSD) plots displaying changes under varying levels of shared cortical correlated inputs (B = 0.3, 0.6, and 1.0). These plots represent PSD for: (g) Ctrl; (h) Ctrl + Glu Exc; (j) No DA; and (k) No DA + Glu Inh conditions, illustrating the influence of cortical correlation on striatal oscillatory activity across different states. (i and l) Bar chart illustrating striatal beta-band power (25–40 Hz) in response to varying levels of shared cortical correlated inputs (parameter B = 0.3, 0.6, and 1.0). Panel (i) depicts data from Ctrl and Ctrl + Glu Exc conditions, while panel (l) depicts data from No DA and No DA + Glu Inh conditions. Data are presented as means ± SD and were averaged from N = 15 independent simulation seeds per condition to ensure reproducibility and robustness of the emergent network oscillations.

CBT loop

A point neuron network of single-compartment Hodgkin-Huxley type models was implemented.¹⁹ The model encompasses cortical Layer 2/3 and Layer 5 neurons, whose subtypes include intratelencephalic (IT), parvalbumin (PV), somatostatin (SOM), and pyramid-tract (PT) neurons. The model also incorporates striatal MSNs and FSIs, alongside neurons located in the subthalamic nucleus (STN), globus pallidus pars externa (GPe), globus pallidus pars interna (GPi), and thalamus. The STN-M1 feedback loop in this computational framework is supported by anatomical evidence in rats showing direct subthalamo-cortical axonal projections to cortical layers I-IV.⁴¹ In our model, this connection is further conceptualised as a functionally aggregated subcortical feedback pathway.^17,19 It represents the net oscillatory feedback from the basal ganglia to the motor cortex, encompassing multiple routes such as the STN-GPi-thalamus-M1 circuit and potential direct projections from the GPe, both of which are critical for the propagation of pathological beta rhythms.¹⁸ Cortico-striatal and thalamo-striatal projections, along with basal ganglia feedback loops, were fully implemented based on anatomical and physiological data (Fig. 2a).^42,43 Parameters for this CBT loop are detailed in Supplementary Tables S3 and S4.

Fig. 2.

Corticostriatal glutamatergic modulation of PSD and PAC under control and DA-depleted states. (a and b) Schematic illustration of the full CBT model under (a) control (Ctrl) and (b) DA-depleted (No DA) conditions. The model encompasses neurons from Layer 2/3 and Layer 5 of the primary motor cortex (M1), striatal D1-MSNs, D2-MSNs, fast-spiking interneurons (FSIs), the subthalamic nucleus (STN), globus pallidus pars externa (GPe), globus pallidus pars interna (GPi), substantia nigra pars reticulata (SNr), and the thalamus (Th). This architecture incorporates both the direct (via D1-MSNs) and indirect (via D2-MSNs) pathways, as well as the hyperdirect (M1 → STN) pathway. Red arrows denote glutamatergic projections, and the black lines indicate GABAergic projections. (c and d) PSD plots for (c) cortical and (d) striatal LFPs, illustrating spectral changes across four conditions: Ctrl (black), No DA (red), Ctrl with Glu Exc (Ctrl + Glu Exc, yellow), and No DA with Glu Inh (No DA + Glu Inh, blue). Glu Inh (blue) denotes the selective 50% reduction in the synaptic strength of the corticostriatal (M1 → STR) pathway, while other glutamatergic inputs (M1 → STN) are maintained at their pathological (DA-depleted) levels. (e–h) Modulation index (MI) plots for PAC are presented for four distinct couplings under varying conditions. Top row: striatal phase-M1 amplitude coupling (STR-M1); Second row: M1 phase-striatal amplitude coupling (M1-STR); Third row: M1 phase-M1 amplitude coupling (M1-M1); Bottom row: striatal phase-striatal amplitude coupling (STR–STR). These couplings are displayed under the four conditions: (e) Ctrl (black), (f) Ctrl + Glu Exc (yellow), (g) No DA (red), and (h) No DA + Glu Inh (blue). Dotted boxes in (g) show the range of high-beta frequencies (25–40 Hz) coupled with gamma frequencies (140–200 Hz). Data are presented as means ± SD and were averaged from N = 15 independent simulation seeds per condition to ensure reproducibility and robustness of the simulated high-beta rhythms and PAC.

Glutamatergic synaptic inputs

Glutamatergic synaptic input targeting the striatum varies across the computational models presented in this study. In the striatal microcircuit model, Poisson spike trains were used as cortical excitatory inputs to explicitly represent cortico-striatal glutamatergic projection to MSNs and FSIs.^26,27 For the CBT circuit model, AMPA/NMDA receptor-mediated synapses are incorporated into both detailed corticostriatal and thalamostriatal glutamatergic projections.^19,44 Parameters are described in Supplementary Table S5.

DA-intact versus DA-depleted states

DA depletion was simulated by altering striatal and basal ganglia parameters to represent the transition from a DA-intact (healthy) to a DA-depleted (parkinsonian) state. In the striatal microcircuit model, the DA-depleted state was implemented by reducing/removing D1-MSN lateral inhibition, weakening overall MSN lateral connectivity, and enhancing FSI inhibition onto D2-MSNs (Fig. 1b).^26,27,45 In the full CBT model (Fig. 2b), the parkinsonian state was extended from the striatum to the entire CBT loop.¹⁹ This included incorporating direct DA–glutamate (Glu) interactions on corticostriatal projections and modifying striatal parameters as described above. In addition, STN–GPe coupling was strengthened (by increasing synaptic weights for STN → GPe and GPe → STN), GPe–GPe collaterals were enhanced (by increasing GPe → GPe synaptic weight), and background inputs to both STN and GPe were reduced (Figs. 1b and 2b). Parameters for these modifications are detailed in Supplementary Table S6.

Simulation

Simulations for the striatal microcircuit model were executed using the NEural Simulation Tool (NEST) simulation environment.^39,40 The CBT loop model simulations were conducted on an Ubuntu Linux system utilising parallel NEURON and the NetPyNE Python code.¹⁹ Data were stored in Hierarchical Data Format 5 (HDF5) and analysed using MATLAB and a custom Python package.^19,22 The code for models is available for download at GitHub (https://github.com/ziruiwang0836/striatum-microcircuit). Beta power (25–40 Hz) was quantified by multi-taper FFT.^4,46 Phase-amplitude coupling (PAC) was determined by calculating the modulation index (MI).¹⁹

Ethics

Adult male Sprague–Dawley rats (2–3 months old, 220–240 g) were housed under a 12 h light/dark cycle with ad libitum access to food and water. All procedures complied with national legislation and institutional guidelines and were approved by the Animal Ethics Committee of Capital Medical University (AEEI-2018-055), and adhered to the protocols specified in the Guide for the Care and Use of Laboratory Animals (NIH). The reporting of animal studies in this research adheres to the ARRIVE guidelines.

Statistics

Data were analysed using GraphPad Prism 10. To minimise potential bias, animals were randomly assigned to experimental groups. All behavioural assessments, data acquisition, histological evaluations, and data analysis were performed by investigators blinded to the experimental conditions. The normality of all datasets was evaluated using the Kolmogorov–Smirnov test. Based on the normality assessment, appropriate parametric or non-parametric tests were applied. For small-sample cohorts (n = 6–8), parametric t-tests were employed only when the data strictly satisfied the assumptions of normality and equal variance; otherwise, equivalent non-parametric tests (e.g., Wilcoxon matched-pairs signed-rank test for Fig. 3l) were used. For single-factor experiments with two-group comparisons, two-tailed t-tests were employed. For within-subject comparisons, paired t-tests were used. Statistical significance for normally distributed data with multiple groups was assessed using one-way ANOVA, followed by a Bonferroni post-hoc test. Non-parametric data with multiple groups were analysed using the Kruskal–Wallis test, with appropriate post-hoc comparisons. Results are reported as mean ± standard error of the mean (SEM) (P < 0.05).

Fig. 3.

Chemogenetic modulation of corticostriatal glutamatergic projections alleviates motor deficits and high-beta oscillations in 6-OHDA rats. (a) Schematic of hM4Di-DREADD viral injection in 6-OHDA-lesioned rats. (b) hM4Di-mCherry expression in the motor cortex confirming specificity to αCaMKII-hM4Di DREADD for αCaMKII-positive neurons. Scale bar: 50 μm. (c) Chemogenetic inactivation of corticostriatal glutamatergic neurons improved latency to fall after CNO administration (10 mg/kg, i.p.) versus saline in the hM4Di-treated 6-OHDA-lesioned rats (paired t-test t = 2.909, df = 5, P = 0.0334, η² = 0.629; n = 6). (d) Schematic of hM3Dq-DREADD injection in sham-treated rats. (e) hM3Dq-mCherry expression in the motor cortex confirmed the specificity of αCaMKII-hM3Dq DREADD for αCaMKII-positive neurons. Scale bar: 50 μm. (f) Chemogenetic activation of corticostriatal glutamatergic neurons showed no significant effect after CNO administration (3 mg/kg, i.p.) versus saline in the hM3Dq-treated sham rats (paired t-test t = 0.7982, df = 5, P = 0.46, η² = 0.113; n = 6). (g and i) Power spectra in motor cortex and striatum after saline versus CNO in hM4Di-treated 6-OHDA-lesioned rats. (h and j) High-beta (25–40 Hz) power decreased in motor cortex (t = 3.196, df = 5, P = 0.0241, η² = 0.671) and striatum (t = 3.193, df = 5, P = 0.0242, η² = 0.671; n = 6). (k and m) Power spectra within the (k) motor cortex and (m) striatum in hM3Dq-treated sham rats after saline treatment versus CNO administration. (l) No significant high-beta (25–40 Hz) changes in the motor cortex (Wilcoxon matched-pairs signed rank test, W = 15.00, P = 0.16, r_s = 0.543) or (n) striatum (paired t-test, t = 1.985, df = 5, P = 0.10, η² = 0.441; n = 6) in hM3Dq–CNO–treated sham rats after saline treatment versus CNO administration (3 mg/kg, i.p.). The data are presented as the means ± SEM. ∗P < 0.05, ∗∗P < 0.01, versus saline treatment.

Sample size determination: The number of animals in each experimental cohort was determined based on extensive prior literature employing comparable 6-OHDA parkinsonian rat models and similar behavioural/electrophysiological paradigms. Previous high-quality studies focusing on PD network pathology^30,33,36 have demonstrated that group sizes of n = 7–10 provide sufficient statistical power to detect robust disease-related behavioural impairments and pathological beta-band synchronisation. Our final analysed population (n = 8 per group for Fig. 4) is fully consistent with these established practices. For computational models, the robustness of emergent dynamics was assessed using 15 independent simulation seeds (N = 15) per condition, which provides a reliable estimate of network stability independent of stochastic initial states.

Fig. 4.

Evolution of motor impairments, beta oscillations, and nigrostriatal dopaminergic degeneration in 6-OHDA rats. (a) Representative images showing TH⁺ fibres and TH⁺ neurons in the striatum (STR, scale bar: 1000 μm) and substantia nigra pars compacta (SNpc, scale bar: 500 μm) from control and lesioned rats on days 3, 5, 7, and 14 post-lesion. (b and c) Optical density of TH⁺ terminals in the striatum and percentage of TH⁺ neurons in the SNpc, both expressed relative to the unlesioned side. Significant reductions appeared by day 3 in STR (P = 0.0227) and by day 7 in SNpc (P = 0.0027). (d and e) Forelimb use asymmetry (d: cylinder and e: FAS tests) showed progressive deficits from day 3–14. (f and g) Open-field test revealed reduced total distance travelled and movement velocity at day 14 (P = 0.0485), showing bradykinesia. (h–m) Local field potential (LFP) spectra in motor cortex (h and i), striatum (j and k), and corticostriatal coherence (l and m). High-beta (25–40 Hz) power in M1 and STR increased significantly on days 7 and 14; corticostriatal coherence was enhanced at day 14. Data are presented as the mean ± standard error of the mean (SEM). ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, and ∗∗∗∗P < 0.0001 versus control (One-way ANOVA or Kruskal–Wallis test). Full statistical parameters including test statistics (F, H), exact P values, and effect sizes (ε², η²) are detailed in the main text and Supplementary Tables S1 and S2.

Role of funders

All the funders played no direct roles in the study design, data collection, analysis, interpretation, or the writing of the manuscript.

Results

Temporal dissociation of motor and beta oscillations in progressive 6-OHDA Parkinsonian rats

Unilateral injection of 6-OHDA into the MFB induced progressive nigrostriatal degeneration (Fig. 4a). Histological analysis revealed a significant reduction in striatal TH-positive fibres as early as day 3 (Kruskal–Wallis H = 20.12, df = 4, P = 0.0005, ε² = 0.461; Ctrl versus 3d: P = 0.0227, Dunn’s post-hoc test; Fig. 4a and b), preceding notable neuronal loss in the SNc, which became evident by day 7 (Kruskal–Wallis H = 31.11, df = 4, P < 0.0001, ε² = 0.774; Ctrl versus 7d: P = 0.0027, Dunn’s post-hoc test; Fig. 4a and c). By day 14, TH-positive fibres in the striatum were nearly absent, consistent with extensive SNc neuronal loss (Fig. 4a–c).

Motor performance deteriorated in parallel with this degeneration. Early akinesia appeared by day 3, reflected in contralateral forelimb use asymmetry in both the cylinder test (Kruskal–Wallis H = 21.90, df = 4, P = 0.0002, ε² = 0.511; Ctrl versus 3d: P = 0.0425, Dunn’s post-hoc test; Fig. 4d) and forelimb adjusting steps (FAS) tests (One-way ANOVA F_(4,35) = 19.31, P < 0.0001, η² = 0.688; Ctrl versus 3d: P < 0.0001, Dunnett’s post-hoc test; Fig. 4e). However, reductions in total distance travelled (F _(4,35) = 1.838, P = 0.14, η² = 0.174; Ctrl versus 14d: P = 0.0485) and movement velocity (F _(4,35) = 3.741, P = 0.0123, η² = 0.300; Ctrl versus 14d: P = 0.0021), markers of bradykinesia, emerged only by day 14 (Fig. 4f and g). These findings indicate that generalised motor slowing develops later and coincides with extensive nigrostriatal degeneration.

Electrophysiological recordings showed gradually rising high-beta (25–40 Hz) oscillations. In M1, beta power increased significantly from day 7 (One-way ANOVA F _(4,35) = 19.75, P < 0.0001, η² = 0.693; Ctrl versus 7d: P < 0.0001; Fig. 4h and i), followed by the striatum (Kruskal–Wallis H = 22.31, df = 4, P = 0.0002, ε² = 0.523; Ctrl versus 7d: P = 0.0080; Fig. 4j and k). This was accompanied by increased corticostriatal coherence in the high-beta band on day 14 post-lesion compared with controls (Kruskal–Wallis H = 8.422, df = 4, P = 0.08, ε² = 0.126; Ctrl versus 14d: P = 0.0248, Dunn’s post-hoc test; Fig. 4l and m). Notably, this temporal gap suggests that maximal denervation (already ∼90% by Day 3; Fig. 4b and c) is not the sole determinant of beta emergence, implying that secondary circuit remodelling is required. Collectively, these findings suggest a potential temporal dissociation: early dopaminergic fibre loss and akinesia precede the onset of bradykinesia and pathological beta oscillations, which emerge only with advanced nigrostriatal degeneration.

DA depletion enhances presynaptic glutamate transmission in 6-OHDA-lesioned rats

Whole-cell patch-clamp recordings from dorsal striatal MSNs (N = 49 cells from n = 3 rats per group) revealed no change in sEPSC frequency within 3–7 days post-lesion, but a significant increase by day 14 (Kruskal–Wallis H = 13.15, df = 4, P = 0.0105, ε² = 0.241; Ctrl versus 14d: P = 0.0059, Dunn’s post-hoc test; Fig. 5a and c), while amplitude showed no detectable change (Kruskal–Wallis H = 5.293, df = 4, P = 0.26, ε² = 0.113; Fig. 5a and d). Miniature EPSC (mEPSC), recorded in the presence of TTX (N = 34 cells from n = 3 rats per group), also exhibited elevated frequency at day 14 (Kruskal–Wallis H = 12.56, df = 4, P = 0.0136, ε² = 0.315; Ctrl versus 14d: P = 0.0043, Dunn’s post-hoc test; Fig. 5b and e), with no detectable change in amplitude (Kruskal–Wallis H = 1.667, df = 4, P = 0.80, ε² = 0.042; Fig. 5f), indicating enhanced presynaptic glutamate release. Correspondingly, PPR (N = 54 cells from n = 3 rats per group) was significantly reduced on day 14 (Kruskal–Wallis H = 7.628, df = 4, P = 0.11, ε² = 0.128; Ctrl versus 14d: P = 0.021, Dunnett’s post-hoc test; Fig. 5g and h). The evoked AMPA/NMDA current ratio (N = 54 cells from n = 3 rats per group) showed no detectable change (One-way ANOVA F _(4,49) = 0.6119, P = 0.66, η² = 0.048; Fig. 5i and j), suggesting that postsynaptic receptor composition was largely preserved. Together, the increased event frequency, decreased PPR, and stable amplitudes pinpoint a presynaptic origin of glutamatergic potentiation in the DA-depleted striatum.

Fig. 5.

Altered presynaptic corticostriatal glutamatergic transmission onto dorsal striatal MSNs in 6-OHDA rats. (a and b) Representative spontaneous excitatory postsynaptic currents (sEPSCs) and miniature excitatory postsynaptic currents (mEPSCs) traces recorded from dorsal striatal MSNs in control (pre-lesion) and 6-OHDA-lesioned rats at days 3, 5, 7, and 14 post-lesion (mEPSC recorded with 1 μM tetrodotoxin). (c and d) Mean sEPSC frequency increased on day 14 (H = 13.15, P = 0.0105; Post-hoc P = 0.0059), while amplitude showed no group difference (H = 5.293, P = 0.26; N = 49 cells from n = 3 rats per group). (e and f) Mean mEPSC frequency was elevated on day 14 (H = 12.56, P = 0.0136; Post-hoc P = 0.0043), whereas amplitude remained unchanged (H = 1.667, P = 0.80; N = 34 cells from n = 3 rats per group). (g and h) Representative paired-pulse ratio (PPR) traces (40 ms intervals) and analysis showed a significant decrease on day 14 (H = 7.628, P = 0.11; post-hoc P = 0.021). (i and j) AMPAR-mediated (−70 mV) and NMDAR-mediated (+40 mV) currents and AMPA/NMDA ratios showed no significant differences (F(4,49) = 0.6119, P = 0.66; N = 54 cells from n = 3 rats per group). Data are presented as the means ± SEM. ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, and ∗∗∗∗P < 0.0001 versus control for One-way ANOVA or Kruskal–Wallis test. Full statistical parameters including test statistics (F, H), exact P values, and effect sizes (ε², η²) are detailed in the main text and Supplementary Tables S7 and S8.

To further investigate dopaminergic modulation, pharmacological manipulations were performed. In control slices, combined blockade of D1 and D2 receptors (SCH23390 + Raclopride) significantly increased evoked EPSC amplitudes (paired t-test t = 2.466, df = 6, P = 0.0487, η² = 0.503, n = 7; Supplementary Fig. S1a and b). Conversely, in 6-OHDA-lesioned slices (day 14), application of D1 and D2 receptor agonists (SKF38393 + Quinpirole) failed to alter eEPSC amplitudes (paired t-test t = 0.9595, df = 7, P = 0.37, η² = 0.116, n = 8; Supplementary Fig. S1c and d). These results demonstrate that endogenous DA normally exerts a suppressive effect on corticostriatal transmission, whereas its depletion leads to a loss of inhibitory control and consequent hyperexcitability characterised by elevated presynaptic glutamate release.

Spearman correlation analysis further revealed strong positive relationships among cortical beta power, striatal beta power, corticostriatal coherence and sEPSC frequency across various time points (control, days 3, 5, 7, 14; Spearman’s rank correlation r = 1.000, P = 0.0167, n = 5 for all parameters; Supplementary Fig. S1e–g). These associations suggest a functional link between presynaptic glutamate overactivity and pathological beta oscillations. Nonetheless, correlation alone does not imply causation, underscoring the need for targeted manipulations to clarify the mechanistic drivers of aberrant beta dynamics in PD.

Glutamatergic hyperactivity is associated with beta oscillations in DA-depleted striatal model

To explore how striatal neuronal activity responds to varying cortical input correlations under DA-intact (Ctrl) and DA-depleted (No DA) conditions, we implemented a computational microcircuit model comprising MSNs and FSIs (Fig. 1a–c). Cortical input correlation, represented by parameterised B (ranging from 0 for independent to 1 for fully correlated), was systematically varied. In the DA-intact model, MSNs exhibited sparse firing (typically <2 Hz), consistent with in vivo observations, and displayed a uniform power spectral density (PSD) profile (Fig. 1d–f). Even with additional glutamate excitation (Ctrl + Glu Exc), their firing patterns or beta-band power did not markedly alter (Fig. 1e and f). This is consistent with the healthy striatum's intrinsic ability to actively decorrelate and gate inputs, a mechanism for filtering highly correlated inputs to maintain the sparse and asynchronous firing of MSNs.¹⁷ Conversely, the DA-depleted network exhibited heightened, imbalanced MSN firing, characterised by the hyperactivation of D2-MSNs relative to D1-MSNs, along with pronounced beta power. This indicates that excessive glutamatergic drive and lost dopaminergic gating disrupt striatal decorrelation—the intrinsic capacity of the striatal microcircuit to maintain asynchronous firing among individual MSNs despite correlated inputs. This disruption transforms sparse activity into the heightened temporal synchronisation that produces pathological beta rhythms (Fig. 1e and f). Reducing the strength of cortical glutamatergic inputs (mimicking AMPA + NMDA blockade; No DA + Glu Inh) lowered both firing rates and beta power, suggesting a partial restoration of striatal decorrelation (Fig. 1e and f).

As parameter B increased, the DA-intact network showed minor beta increase, with PSD peaks shifting towards gamma (∼43 Hz), reflecting the striatum's intrinsic capacity to decorrelate its inputs and maintain asynchronous output despite increasing input synchrony (Fig. 1g and i). Even with additional glutamate excitation (Ctrl + Glu Exc), the decorrelation mechanisms remained robust (Fig. 1h and i). By contrast, the DA-depleted network exhibited amplified beta power with increasing B, showing PSD peaks around 22 Hz at B = 1 (Fig. 1j and l). Glutamate inhibition in this state (No DA + Glu Inh) effectively reduced beta power (Fig. 1k and l). Collectively, these findings suggest that excessive correlated glutamate input under DA depletion disrupts striatal decorrelation, leading to pathological beta rhythms, whereas reducing glutamatergic transmission can restore normal network dynamics.

The robustness of these network dynamics was assessed using 15 independent simulation seeds (N = 15) per condition. The emergence of pathological beta-band activity and firing patterns was highly consistent across all runs. Furthermore, model parameters were strictly constrained by empirical biophysical and electrophysiological data (Supplementary Tables S2 and S4), indicating that the observed synchronisation arises from stable network mechanisms rather than stochastic variability or parameter over-tuning.

CBT model shows glutamatergic control of pathological beta synchrony in DA-depletion

While the striatal microcircuit model (Fig. 1) demonstrated that local remodelling can disrupt asynchronous input processing, the question remains how these local synchronisation events are propagated and sustained across the broader motor network. Therefore, to integrate these local findings into a system-wide framework, we implemented a more refined and realistic full CBT loop model (Fig. 2, Fig. 6, Fig. 7). This multiscale transition allows us to reconcile the local ‘trigger’ mechanisms identified in the striatum with the ‘maintenance’ roles of subcortical feedback loops, providing a comprehensive view of network-wide beta resonance.

Fig. 6.

Impact of postsynaptic NMDA and AMPA receptors on corticostriatal PSD and PAC in DA-depleted networks. (a and b) Power Spectral Density (PSD) plots for (a) the primary motor cortex (M1) and (b) striatum in DA-depleted networks, illustrating changes under four conditions: DA depletion (No DA, red), combined AMPA and NMDA receptor blockade (No DA + Glu Inh, blue), selective AMPA receptor blockade (No DA + No AMPA, yellow), and selective NMDA receptor blockade (No DA + No NMDA, purple). (c) Modulation Index (MI) plots illustrating striatal phase-M1 amplitude coupling (STR-M1) and M1 phase-M1 amplitude coupling (M1-M1) in DA-depleted networks for two selected conditions: selective AMPA receptor blockade (No DA + No AMPA, top row); and selective NMDA receptor blockade (No DA + No NMDA, bottom row). Glu Inh denotes a 50% reduction in M1 → STR synaptic strength, whereas ‘No NMDA’ and ‘No AMPA’ represent a selective 100% blockade of the respective receptor currents. (d) Bar chart illustrating mean MI values of STR–M1, M1–M1, M1–STR, and STR–STR coupling under four conditions: No DA (red), No DA + Glu Inh (blue), No DA + No AMPA (yellow), and No DA + No NMDA (purple). The bars represent the average MI quantified within the coupling window defined by a phase frequency of 25–40 Hz (high-beta) and an amplitude frequency of 140–200 Hz (high-gamma).

Fig. 7.

Corticostriatal inhibition and subcortical feedback disruption alter CBT circuit dynamics in parkinsonian state. (a) Schematic illustration of the complete CBT loop model, including classical basal ganglia direct (D1-MSNs) and indirect (D2-MSNs) pathways, as well as the hyperdirect (M1 → STN) pathway. Two subcortical feedback pathways, STN → M1 and Th → M1, are also depicted. Red arrows indicate glutamatergic projections, and the black lines indicate the GABAergic projections. (b and c) Spike raster plots of neurons in GPi, GPe, STN, along with voltage traces of Th neurons under simulated DA depletion (No DA) and subsequent corticostriatal glutamatergic inhibition (No DA + Glu Inh) conditions. (d) PSD plot of the STN under No DA (red) and No DA + Glu Inh (blue) conditions. (e) MI plots illustrating M1-STN, STN-M1, and STN–STN PAC under No DA (top row) and No DA + Glu Inh (bottom row) conditions. The dotted box highlights the coupling range of high-beta frequencies (25–40 Hz) coupled with gamma frequencies (140–200 Hz). (f) PSD analysis of striatal LFP illustrating changes under four conditions: No DA (red), No DA + Glu Inh (blue), DA depletion with blocking Th → M1 pathway (purple), and DA depletion with blocking STN → M1 pathway (cyan). (g) MI plots illustrating STN-M1 PAC and M1-M1 PAC under DA depletion with blocking Th → M1 (top row) and DA depletion with blocking STN → M1 (bottom row) conditions.

Our extended computational framework simulated the complete CBT loop, incorporating DA-modulated AMPA and NMDA receptor-mediated glutamatergic transmission (Fig. 2a and b). We modelled pathological “excessive glutamatergic drive” as a dual-pathway cortical phenomenon: presynaptic overdrive in M1 → STR projections (aligned with our ex vivo sEPSC data) and a strengthened hyperdirect M1 → STN pathway, while thalamostriatal (Th → STR) inputs remained at baseline. In the DA-intact (Ctrl) state, model neurons exhibited irregular, sparse firing with minimal temporal synchronisation, as illustrated by the spike rasters and firing rate profiles in Supplementary Fig. S2a, c and e. Conversely, in the DA-depleted (No DA) CBT model, neuronal firing became synchronised and shifted to rhythmic bursting (Supplementary Fig. S2d, f and f). This pathological transition in firing patterns resulted in significantly elevated beta-band power in both cortical and striatal simulated LFPs, consistent with our prior striatal microcircuit observations.²⁹

Under DA depletion, the network revealed robust high beta activity and corticostriatal coupling. Crucially, selective inhibition of the corticostriatal (M1 → STR) glutamatergic pathway (Glu Inh) significantly suppressed beta power in both the M1 and striatum (Fig. 2c and d). DA depletion also induced abnormal beta-gamma PAC (Fig. 2e–g), which was normalised by glutamate inhibition (Fig. 2g and h). Collectively, these modelling results indicate that excessive glutamatergic drive contributes to the maintenance of pathological beta synchrony in the DA-depleted network. These network-level effects were reproducible across independent simulation runs (N = 15), confirming that pathological beta synchrony and aberrant PAC in the CBT loop emerge robustly under DA depletion conditions, independent of initial stochastic states.

Chemogenetic inhibition of corticostriatal glutamate alleviates pathological beta synchrony and motor deficits in 6-OHDA rats

To identify whether direct suppression of corticostriatal glutamatergic transmission alleviates pathological beta oscillations and motor deficits, we selectively inhibited corticostriatal projection neurons in 6-OHDA rats using inhibitory chemogenetics (hM4Di-DREADD) (Fig. 3a). Immunofluorescence confirmed selective hM4Di-mCherry expression in αCaMKII-positive cortical projection neurons (Fig. 3b), with ∼80% colocalisation. Following CNO administration (10 mg/kg), 6-OHDA-lesioned rats showed significant improvement in motor coordination, as evidenced by increased latency to fall on the rotarod (paired t-test, t = 2.909, df = 5, P = 0.0334, η² = 0.629; Fig. 3c). In parallel, electrophysiology revealed a marked reduction in aberrant high-beta oscillations in both cortical (paired t-test, t = 3.196, df = 5, P = 0.0241, η² = 0.671; Fig. 3g and h) and striatal regions (paired t-test, t = 3.193, df = 5, P = 0.0242, η² = 0.671; Fig. 3i and j) following CNO administration. These results collectively suggest that suppressing corticostriatal glutamate release normalises network synchrony and motor function in parkinsonian rats.

Conversely, chemogenetic activation of corticostriatal neurons via excitatory hM3Dq-DREADDs in sham-treated rats (Fig. 3c and d) did not induce changes in motor behaviour (paired t-test, t = 0.7982, df = 5, P = 0.46, η² = 0.113; Fig. 3f) or beta oscillations in the motor cortex (Wilcoxon matched-pairs signed rank test, W = 15.00, P = 0.16, r_s = 0.543; Fig. 3k and l) or striatum (paired t-test, t = 1.985, df = 5, P = 0.10, η² = 0.441; Fig. 3m and n), despite robust hM3Dq expression (Fig. 3e) and c-Fos immunoreactivity (unpaired t-test, t = 5.273, df = 10, η² = 0.7355, P = 0.0004; Supplementary Fig. S3). This further supports the notion that pathological glutamatergic overdrive, rather than physiological activation, underlies aberrant beta oscillations in PD.

CBT model reveals NMDA receptor blockade attenuates beta synchrony in DA-depleted networks

Utilising the CBT computational model, we assessed receptor-specific mechanisms underlying pathological beta oscillations in DA-depleted networks. Selective postsynaptic NMDA receptor blockade significantly reduced beta-band power (Fig. 6a and b) and attenuated PAC between M1 and striatal signals (Fig. 6c and d). This suggests that NMDA receptor activity contributes substantially to corticostriatal beta synchrony and exaggerated PAC. By contrast, AMPA receptor blockade produced only a modest reduction in striatal beta power and negligible PAC changes (Fig. 6a–d), consistent with a previous report showing that AMPA antagonists had no discernible effect on beta oscillations.²⁴ Furthermore, combined blockade of AMPA and NMDA receptors (No DA + Glu Inh condition; Fig. 6a, b, d) resulted in a pronounced suppression of beta activity, reinforcing the importance of glutamatergic transmission in sustaining pathological oscillations. Notably, selective and complete NMDA receptor blockade (No DA + No NMDA) suppressed the Modulation Index (MI) more effectively than the 50% reduction of total corticostriatal drive (No DA + Glu Inh; Fig. 6d). This discrepancy may reflect the contribution of slow NMDA receptor kinetics in facilitating the temporal coincidence required for beta resonance; its total removal collapses the synchronous state more thoroughly than a partial reduction in overall excitatory input.^24,47 These results indicate that NMDA receptor signalling contributes to the maintenance of beta synchrony in DA-depleted networks and highlight it as a potential therapeutic target for DA depletion-related motor impairments.

Striatal NMDA receptor blockade alleviates motor deficits and beta oscillations in parkinsonian rats

In 6-OHDA-lesioned rats, intrastriatal infusion of the NMDA antagonist CGS significantly ameliorated motor impairments, as evidenced by improvements in total distance travelled (Kruskal–Wallis H = 12.06, df = 2, P = 0.0024, ε² = 0.548; post-hoc P = 0.0012, Dunn’s test; Fig. 8b), movement duration (One-way ANOVA F(2, 20) = 4.007, P = 0.0344, η² = 0.286; post-hoc P = 0.0371, Tukey’s test; Fig. 8c) and mean velocity (Kruskal–Wallis H = 10.30, df = 2, P = 0.0058, ε² = 0.468, post-hoc P = 0.0041; Fig. 8d) in the open-field test. In contrast, AMPA receptor blockade with CNQX produced no observable behavioural effect (Saline versus CNQX, P > 0.05, Fig. 8b–d). Concurrently, LFP recordings showed prominent pathological high-beta oscillations (25–40 Hz) in both the motor cortex and striatum under DA-depleted conditions. Additionally, NMDA receptor blockade markedly suppressed beta power in both regions (motor cortex: Kruskal–Wallis H = 9.829, df = 2, P = 0.0023, ε² = 0.614; post-hoc P = 0.0202, Dunn’s test; striatum: F(2,14) = 4.574, P = 0.0296, η² = 0.395; Fig. 8g). Furthermore, coherence analysis revealed that the elevated corticostriatal high-beta synchrony observed in parkinsonian rats was significantly reduced following NMDA blockade (F_(2,14) = 5.746, P = 0.0151, η² = 0.451; post-hoc P = 0.0188, Dunnett’s test; Fig. 8g). In contrast, AMPA inhibition yielded only minimal, non-significant changes in either striatal beta power (post-hoc P = 0.05, Dunnett’s test) or corticostriatal coherence (post-hoc P ≥ 0.99, Fig. 8g). These findings highlight striatal NMDA receptors as potential modulators of pathological beta rhythms and motor deficits and demonstrate that NMDA blockade effectively suppresses abnormal beta network synchrony while improving behaviour in parkinsonian rats.

Fig. 8.

Effects of glutamate receptor antagonists on locomotor activity and high-beta oscillations in hemiparkinsonian rats. (a) Representative movement traces after dorsal striatum infusion of vehicle (left), NMDA receptor blocker CGS (middle), or AMPA receptor blocker CNQX (right) into the dorsal striatum. (b–d) Quantification of locomotor activities following drug administration. In parkinsonian rats, CGS significantly ameliorated movement distance (X² = 12.06, P = 0.0024), movement duration (X² = 8.647, P = 0.0133), and mean velocity (X² = 10.30, P = 0.0058; post-hoc Dunn's test: P < 0.05 versus saline), whereas CNQX showed no discernible effects (P > 0.05; Saline: n = 6, CGS: n = 9, CNQX: n = 8). (e) Representative spectrograms showing LFP high-beta power in the motor cortex (top) and striatum (bottom) in parkinsonian rats treated with saline, CGS, and CNQX. (f) Representative linear plots show LFP spectral power in the motor cortex (left), striatum (middle), and corticostriatal coherence (right) in parkinsonian rats treated with saline, CGS, or CNQX. (g) CGS significantly suppressed normalised high-beta (25–40 Hz) power in the motor cortex (X² = 9.829, P = 0.0023), the striatum (F(2,14) = 4.574, P = 0.0296), and corticostriatal coherence (F_(2,14) = 5.746, P = 0.0151; Post-hoc P < 0.05 versus saline). CNQX suppressed cortical high-beta (X² = 9.829, P = 0.0023) but did not significantly influence high-beta power in the striatum or corticostriatal coherence (P > 0.05; Saline: n = 6, CGS: n = 6, CNQX: n = 5).

Corticostriatal glutamate inhibition desynchronises the DA-depleted CBT network and tunes feedback

To examine how inhibiting corticostriatal glutamatergic transmission affects network-wide activity, we analysed neuronal dynamics in GPe, GPi, STN, thalamus, and M1 (Fig. 7a). In the DA-depleted state, suppressing corticostriatal excitation broadly normalised electrophysiological activity by effectively reverting network dynamics towards the patterns observed in the DA-intact (Control) state. Specifically, neuronal firing in GPi, GPe, and STN shifted from pathological rhythmic bursting back to irregular, desynchronised spiking, while thalamic neurons reverted to a sparse, control-like discharge (Fig. 7b and c). Correspondingly, beta-band power decreased significantly across multiple regions, particularly within the STN (Fig. 7d). In addition, beta-gamma PAC between M1 and STN, including both M1 phase-STN amplitude coupling and STN phase-M1 amplitude coupling, was notably decreased, along with a modest decline in intra-STN coupling (Fig. 7e). These results indicate that corticostriatal glutamatergic inhibition effectively disrupts excessive network synchronisation under DA depletion, primarily impacting both inter-regional coupling between cortical and basal ganglia nuclei and intra-regional synchronisation within the basal ganglia downstream nuclei. These findings are consistent with the desynchronising effect of glutamatergic inhibition observed in prior experimental studies.^23,46

We next investigated the contribution of subcortical-to-cortical feedback in maintaining these dynamics by independently blocking the STN → M1 and Th → M1 pathways (Fig. 7a). Our results showed that blocking Th → M1 feedback did not significantly alter the striatal beta suppression achieved by glutamatergic inhibition, suggesting the thalamus is not a primary node for beta resonance in this framework (Fig. 7f). Conversely, disrupting the STN → M1 pathway largely reversed the therapeutic effects of glutamate inhibition, causing striatal beta power to re-emerge (Fig. 7f).^18,19 This indicates that the STN → M1 link may function as an important maintenance node within a resonant long-loop; its disruption decouples the subcortical nuclei from the desynchronising influence of the modified cortical drive.⁴¹ Consequently, the DA-depleted striatum reverts to its intrinsic synchronisation, highlighting that an intact subcortical feedback architecture appears to contribute to stabilising the network-wide desynchronised state.^20,26 Notably, the persistence of suppressed PAC despite the re-emergence of beta power (Fig. 7g) suggests that while feedback pathways facilitate the broad propagation of oscillatory energy, local striatal circuit interactions are the primary determinants of the specific phase-amplitude dynamics.

Discussion

Our study offers robust evidence supporting the role of excessive corticostriatal glutamatergic activity as a critical permissive amplifier of pathological beta oscillations in PD. This overactivity amplifies striatal MSN hyperactivity, strengthens intrinsic striatal beta rhythms, and facilitates their spread across the CBT loop under DA-depleted conditions. Crucially, these pathological beta oscillations depend on NMDA receptor activity within the striatal pathway. While prior studies implicated NMDA receptors in modulating beta oscillations across basal ganglia,^20,22, 23, 24^,26,47,48 our results suggest that striatal NMDA signalling contributes significantly to the initiation and maintenance of network-wide beta synchrony.17, 18, 19 By integrating in vivo, ex vivo, and computational approaches, we delineate a presynaptically driven, NMDA receptor–dependent beta signature serving as a mechanistic biomarker and therapeutic target beyond dopamine replacement.

Our 6-OHDA model reveals a temporal dissociation in PD pathophysiology. Early akinesia and dopaminergic fibre loss (day 3) precede the emergence of pathological high-beta oscillations (day 7), which are followed by severe bradykinesia and neurodegeneration (day 14). While DA depletion is the primary trigger, our data indicate it initiates a secondary, DA-loss-induced glutamatergic overdrive that acts as a permissive amplifier for beta synchrony. Notably, although significant elevations in high-beta power emerge by Day 7, the surge in presynaptic glutamate release markers (sEPSC frequency) reaches statistical significance by Day 14. This temporal sequence, supported by the strong correlation across all timepoints (Supplementary Fig. S1e–g), suggests that while early dopamine loss and associated microcircuit remodelling initiate the oscillatory state, the subsequent build-up of glutamatergic overdrive is essential for the maturation and stabilisation of these rhythms into a robust pathological hallmark. Thus, glutamate may act as a pivotal factor in the progression of PD, transforming early network fluctuations into sustained pathological synchronisation. This progression implies that robust motor impairment and strong beta synchrony depend on sustained degeneration, not simply early DA loss.^9,29,36 Indeed, DA depletion alone does not reliably induce beta rhythms: acute DA blockade often fails to elevate beta power,^11,34 and α-synuclein models may show DA loss without pronounced beta.¹² Conversely, robust beta oscillations can appear even in the absence of chronic DA loss.^21,25,34 These observations challenge a simplistic inverse relationship between DA levels and beta power,³² indicating instead a complex, nonlinear dynamic in beta generation.¹¹ While our findings highlight high-beta oscillations (25–40 Hz) as a key correlate of motor deficits in the 6-OHDA rat model, human clinical studies frequently emphasise the pathological role of low-beta activity (13–20 Hz).^11,36,47 This frequency shift may reflect species-specific differences in CBT loop architecture, including conduction delays and network scale, which can systematically influence the peak frequency of pathological synchrony.^17,18 In our experimental model, the 25–40 Hz band represents the primary synchronisation associated with bradykinesia, whereas lower-frequency power remains largely stable. Consequently, our computational model was parameterised to reproduce rodent-specific dynamics¹⁷ rather than to explicitly dissociate mechanisms between beta sub-bands. Nevertheless, these results suggest a potential circuit-level principle: corticostriatal glutamatergic overdrive may function as a permissive amplifier of pathological synchronisation, a mechanism that could potentially operate across distinct beta bands in parkinsonian networks.

Given this complexity, our findings reinforce the notion that non-dopaminergic mechanisms critically contribute to pathological beta oscillations and motor deficits. For example, optogenetic activation of striatal ChIs reliably evokes beta oscillations even under intact dopaminergic tone, producing PD-like motor deficits.^21,49 Computational models show that elevated striatal cholinergic tone promotes beta activity,²⁰ and enhanced beta synchrony emerges early in disease before overt neurodegeneration.²⁵ Prior studies have proposed striatal microcircuits as intrinsic beta generators, driven primarily by ChIs^20,21,34 and modulated by PV and SOM interneurons.^13,23,25 Meanwhile, DA depletion disrupts striatal lateral inhibition, strengthens FSI influence, and reconfigures GABAergic connectivity,^{18,20,26,27,47} all of which reshape oscillatory balance.

Building upon these findings, our microcircuit modelling reveals a central role for enhanced cortical glutamate drive to MSNs in pathological beta formation.^27,30 In the DA-depleted microcircuit, excessive corticostriatal glutamatergic input amplifies MSN firing and disrupts striatal decorrelation,^26,27 which in turn produces robust beta oscillations. Notably, when we attenuated cortical input correlation (simulated via combined AMPA and NMDA receptor blockade), both MSN firing rates and beta power declined, indicating a partial restoration of decorrelation and suppression of pathological synchrony. This mechanism aligns with our ex vivo findings: MSNs recorded at day 14 post-lesion exhibited elevated sEPSC and mEPSC frequencies, paralleling observations in Pink1-KO models with enhanced presynaptic glutamate release.^28,30 The positive correlation between sEPSC frequency and beta power further underscores glutamatergic overdrive as a significant contributor to PD progression.

At the network level, our extended CBT simulations further validated the glutamate-drive hypothesis. Under DA depletion, glutamatergic inhibition normalised both beta power and PAC, demonstrating that excessive glutamate drive contributes to the maintenance of pathological beta. These computational predictions were validated in vivo. Chemogenetic suppression of corticostriatal glutamatergic projections in 6-OHDA rats improved motor performance and attenuated aberrant high-beta oscillations in both cortex and striatum. By contrast, activation of the same pathway in DA-intact animals elicited no effect, confirming that pathological beta emerges only when dopaminergic gating fails. Altogether, these results suggest that corticostriatal glutamate acts less as a primary pacemaker and more as a permissive amplifier of beta rhythms under pathological conditions.19, 20, 21, 22^,47 Mechanistically, intrinsic ChI-driven beta can be potentiated by cortical glutamate through a polysynaptic, ChI-mediated pathway, promoting network-wide synchrony.⁵⁰ Collectively, these results support a model of sequential involvement: DA depletion disrupts corticostriatal gating, leading to glutamate-dependent amplification of beta oscillations.^27,30,51

Our models also delineate distinct roles of glutamate receptor subtypes in beta dynamics. In our computational model, selective NMDA receptor blockade sharply reduced beta power and attenuated corticostriatal PAC, whereas AMPA blockade produced minimal effects. The same pattern was observed in vivo: NMDA antagonists in the STN suppress cortico-subthalamic beta synchronisation and alleviate motor deficits, while AMPA antagonists exert negligible influence.^24,44,46,48 Similarly, acute NMDA receptor blockade in the striatum of our 6-OHDA rats reduced corticostriatal beta synchrony, whereas AMPA inhibition had little impact. While we demonstrate that blocking NMDA receptors in the striatum effectively suppresses beta power and ameliorates motor deficits,⁴⁷ prior studies have reported comparable effects following NMDA receptor antagonism in the subthalamic nucleus.^24,48 Together, these findings indicate that excessive glutamatergic signalling acts as a circuit-wide driver of pathological synchrony across the CBT loop, rather than being confined to a single anatomical node.^14,19,30

While NMDAR blockade significantly increases movement distance, this reflects a restoration of volitional exploration without inducing the stereotypical or dyskinetic behaviours typically associated with dopaminergic hypersensitivity.²⁴ Consistent with the temporal dynamics observed in our time-frequency maps (Fig. 8e), post-hoc analysis indicates that striatal NMDA receptor blockade primarily attenuates the amplitude of transient beta events and tends to shift the distribution towards shorter, more physiological durations. This further supports the role of glutamatergic overdrive as a permissive amplifier that increases the intensity of network-wide synchronisation events in the DA-depleted state.

These consistent findings across modelling and experiments suggest NMDA receptors in both the striatum and STN as plausible candidates for non-dopaminergic therapeutic targets for PD.^{14,19,24,30,47,48} A fundamental distinction exists between direct glutamatergic inhibition and indirect dopaminergic modulation. Our results indicate that direct antagonism of striatal NMDA receptors produces stable suppression of beta oscillations by interrupting the glutamatergic overdrive that acts as a permissive amplifier of pathological synchrony.⁵² In contrast, dopaminergic replacement therapy does not consistently reduce beta power, as dopamine primarily modulates oscillatory frequency rather than exerting consistent control over beta power.¹¹ Moreover, because dopaminergic control over striatal glutamate release is progressively impaired in the parkinsonian state,²⁸ indirect suppression of glutamatergic drive via dopamine is less effective than direct NMDA receptor blockade. This distinction provides a mechanistic explanation for the more reliable effects of targeting the glutamate–NMDA axis on excessive beta synchrony and associated motor deficits.

To test how feedback pathways modulate these dynamics, we disrupted the STN → M1 and Th → M1 loops under simulated DA deficiency and glutamate suppression. Interrupting the STN → M1 pathway led to the re-emergence of beta oscillations, suggesting this route facilitates pathological propagation. By contrast, blocking Th → M1 feedback had minimal effect, reinforcing STN → M1 as a principal route for transmitting abnormal basal ganglia beta to the motor cortex.¹⁹ Notably, even when beta re-emerged, coordinated gamma gating of striatal activity remained weak, implying that the elevated striatal beta likely originated from local disinhibition or altered intrinsic circuit properties, rather than cortical PAC drive.20, 21, 22^,26 Together, these findings suggest that STN → M1 feedback primarily functions as a propagation pathway under dopamine-depleted conditions.

Our framework complements existing models identifying the STN–GPe recurrent loop as a primary site of beta oscillation amplification.^16,17,51 While this subcortical circuit possesses intrinsic resonant properties, our results suggest that corticostriatal glutamatergic overdrive under DA depletion acts as a critical permissive trigger. By disrupting striatal decorrelation and increasing the structured inhibitory output to downstream nuclei, this glutamatergic imbalance—characterised by the interaction between cortical overdrive and asymmetric D1/D2 gain—facilitates the engagement of the STN–GPe loop into a pathological beta state. Furthermore, our network simulations indicate that although subcortical loops amplify beta rhythms locally, the STN → M1 feedback pathway appears to modulate the propagation and maintenance of beta synchrony across the entire motor network in a state-dependent manner,¹⁹ consistent with human electrophysiological recordings.⁵³ This hierarchical organisation, where striatal perturbations trigger subcortical resonance that is subsequently stabilised and broadcast by cortical feedback, provides a unifying framework linking local dopamine deficiency to widespread network-level pathology.

Despite these insights, several limitations warrant consideration. Our in vivo recordings captured corticostriatal glutamatergic activity at the population level but did not differentiate between D1- and D2-type MSNs in vivo, which were separately represented in our computational framework. Furthermore, consistent with the anatomy of the rodent CBT circuit,⁴¹ our model identifies the subcortical feedback to M1 as a principal route for transmitting and sustaining abnormal beta oscillations. This STN-to-M1 link functions as a compact representation of the broader subcortical feedback architecture, including the GPi-thalamic and pallido-cortical pathways.^18,53 While this projection is more prominent in rodents, subcortical-to-cortical feedback remains a shared principle across species for maintaining beta synchrony, evidenced by the efficacy of STN-DBS.^6,7 This interpretation aligns with the observation that the predicted persistence of beta activity following feedback blockade contrasts with experimental findings showing that DBS or STN inhibition suppresses beta rhythms.^6,7,24,53, 54, 55 This discrepancy, referring to the mismatch between our model's predicted re-emergence of oscillations upon feedback disruption and the clinical suppression of beta by subcortical interventions, likely reflects modelling constraints, including limited representation of intrinsic STN bursting dynamics and broader thalamocortical connectivity. Although these simplifications may limit model fidelity, our findings nonetheless underscore the central role of glutamatergic overdrive in shaping pathological beta oscillations and its potential contribution to altered PAC. Moreover, our results identify STN → M1 feedback as a modulatory pathway whose engagement appears to depend on dopamine and glutamate states, offering a refined, circuit-specific target for PD intervention. Third, while our model highlights the striatum as a permissive amplifier, clinical evidence suggests that striatal projection neurons may maintain relatively stable discharge rates without robust spike-LFP locking.⁵⁶ Our findings reconcile this by identifying the STN → M1 feedback loop as the primary sustainer of network-wide synchrony.^18,19 Future studies employing simultaneous multisite recordings across the striatum, STN, and globus pallidus are warranted to further validate the hierarchical transmission of these pathological rhythms. Finally, we acknowledge that the rapid DAergic denervation in the 6-OHDA model may not fully replicate the slow, decades-long progression of human PD. However, the observed temporal dissociation, where early akinesia at Day 3 precedes the build-up of beta synchrony at Day 7, suggests that beta power represents a secondary network transition rather than a direct neurotoxic artefact. Within our framework, ‘non-dopaminergic mechanisms' encompass broad adaptations, including striatal synaptic plasticity (e.g., dendritic spine loss and remodelling)²⁹ and the functional reconfiguration of local microcircuits involving FSI and ChI populations.²³ Corticostriatal glutamatergic overdrive thus functions as a permissive amplifier that exploits these maladaptive changes to stabilise pathological beta synchrony. Future research using progressive models will be essential to further map these slow-evolving processes.

The clinical significance of these findings relates to the “glutamatergic paradox” of current PD treatments. While NMDA antagonists like amantadine and memantine effectively reduce L-DOPA–induced dyskinesias, they offer limited improvement for “off” motor symptoms. This limitation likely stems from low receptor affinity and dose-limiting neuropsychiatric side effects that prevent systemic doses from achieving robust striatal beta suppression.^24,52 Our model reconciles this by identifying glutamate as a permissive amplifier: while NMDAR blockade can reduce the “gain” of pathological oscillations to alleviate motor complications, it does not address the fundamental dopaminergic deficit, the primary aetiological driver. Future interventions targeting specific striatal subunits, such as GluN2B, may therefore suppress pathological beta synchrony more effectively while minimising systemic side effects.

Our integrated computational and experimental approach demonstrates that striatal glutamatergic overdrive, rather than dopamine loss alone, acts as a permissive amplifier for pathological beta oscillations in PD. Acting through NMDA receptor-dependent mechanisms, excessive glutamate disrupts input decorrelation, enhances MSN excitability, and sustains beta synchrony across the CBT network. These findings reposition presynaptic glutamate release and striatal NMDA signalling as mechanistic biomarkers and plausible non-dopaminergic therapeutic targets for modulating pathological beta dynamics and restoring circuit stability in PD.

Contributors

ZW and XF contributed to the literature search, figures, study design, data collection, data analysis; YZhao, WS, XJ, HH, TX, XG, and YZhang contributed to the data collection and data analysis; YJ, TZ, JH, AK, and JHK contributed to the data interpretation; JJ contributed to literature search, figures, study design, data analysis, data interpretation, and writing. ZW and JJ accessed and verified the underlying data. All authors read and approved the final manuscript.

Data sharing statement

The data that support these findings of the study are available upon request from the corresponding authors.

Declaration of interests

We declare that we do not have any commercial or associative interest that represents a conflict of interest in connection with the work submitted.