A Wireless Cortical Surface Implant for Diagnosing and Alleviating Parkinson's Disease Symptoms in Freely Moving Animals

Abstract

Parkinson's disease (PD), one of the most common neurodegenerative diseases, is involved in motor abnormality, primarily arising from the degeneration of dopaminergic neurons. Previous studies have examined the electrotherapeutic effects of PD using various methodological contexts, including live conditions, wireless control, diagnostic/therapeutic aspects, removable interfaces, or biocompatible materials, each of which is separately utilized for testing the diagnosis or alleviation of various brain diseases. Here, a cortical surface implant designed to improve motor function in freely moving PD animals is presented. This implant, a minimally invasive system equipped with a graphene electrode array, is the first integrated system to exhibit biocompatibility, wearability, removability, target specificity, and wireless control. The implant positioned at the motor cortical surface activates the motor cortex to maximize therapeutic effects and minimize off‐target effects while monitoring motor activities. In PD animals, cortical motor surface stimulation restores motor function and brain waves, which corresponds to potentiated synaptic responses. Furthermore, these changes are associated with the upregulation of metabotropic glutamate receptor 5 (mGluR5, Grm5) and D5 dopamine receptor (D₅R, Drd5) genes in the glutamatergic synapse. The newly designed wireless neural implant demonstrates capabilities in both real‐time diagnostics and targeted therapeutics, suggesting its potential as a wireless system for biomedical devices for patients with PD and other neurodegenerative diseases.

A wireless cortical surface implant employing graphene electrode arrays diagnoses and alleviates Parkinson's disease symptoms in freely moving animals. The device continuously monitors cortical activity and delivers targeted stimulation, restoring beta–gamma and delta oscillations. It improves motor function by promoting synaptic plasticity in glutamatergic and dopaminergic circuits, demonstrating a promising closed‐loop neurotherapeutic approach.

1. Introduction

Parkinson's disease (PD) is a common neurodegenerative disease that occurs due to the degeneration of the dopaminergic system. Aging is the primary etiology of PD. In societies with advanced healthcare and nutrition, life expectancy increases substantially, leading to a rise in the number of patients with PD. Approximately 2% of adults have severe motor impairments, such as tremors, bradykinesia or akinesia, muscle stiffness, and stooped posture, necessitating immediate medical intervention. Tonic dopamine release from the substantia nigra pars compacta, a dopamine reservoir, into the putamen in the striatum is crucial for modulating motor planning and execution. The degeneration of the dopaminergic pathway reduces the activity of the thalamocortical motor circuit, causing difficulty in motor planning.^{[ 1 ]} Moreover, the reduction in dopamine release alters brain oscillations in the motor cortex, leading to abnormal motor command and control. PD is often comorbid with nonmotor symptoms such as dementia, depression, anxiety, insomnia and/or rapid‐eye‐movement sleep behavior disorder, difficulty in communication, and olfactory dysfunction, thus leading to social isolation.^{[ 2 ]} These factors markedly reduce the quality of daily life for both patients and caregivers. To date, an effective treatment for patients with PD is yet to be developed.

Although pharmacological interventions are effective in the initial treatment of PD, they can lead to drug tolerance and nonspecific consequences, such as adverse, global, and wearing‐off effects.^{[ 3 ]} Alternatively, electrotherapeutic approaches, such as vagus nerve stimulation (VNS), deep brain stimulation (DBS), repetitive transcranial magnetic stimulation (rTMS), and magnetic resonance imaging (MRI)‐guided focused ultrasound (MRgFUS), have received attention for their specificity and effectiveness. VNS, which involves an electrical pulse generator implanted along the vagus nerve in the neck, has shown effectiveness in some animal and human studies; however, its clinical application is limited because of its off‐target effects.^{[ 4 ]} DBS, which involves implanting electrodes in deep brain structures, such as the striatum, subthalamic nucleus (STN), and thalamic centromedian nucleus, has demonstrated therapeutic efficacy in alleviating PD symptoms.^{[ 3} , ^{5 ]} However, DBS is associated with a risk of substantial brain damage during and after surgery, highlighting the need to develop more noninvasive methods. As a noninvasive method, rTMS performed over the head induces signal interference in the motor cortex, often leading to motor recovery in patients with PD.^{[ 6 ]} MRgFUS is another noninvasive method that uses MRI and ultrasound to locate and stimulate a target area, respectively.^{[ 7 ]} These noninvasive devices are considered complementary treatment options to drug treatments. However, they often produce ambiguous effects because of their non‐specificity, leading to stimulation‐induced adverse complications such as headache, local pain, burning sensation, gait worsening, tinnitus, and cognitive impairments.^{[ 2} , ⁶ , ^{8 ]} To prevent these unwanted side effects, a minimally invasive neurostimulator that simultaneously enhances on‐target specificity and effectiveness is necessary.

Graphene electrode arrays recently developed by our groups are biocompatible, conductive, scalable, thermally stable, and mechanically flexible, providing stable recordings even on nonrigid surfaces.^{[ 9 ]} These properties make them suitable for biomedical applications in treating brain disorders. Our previous studies have demonstrated that microfabricated graphene electrode arrays placed on the subdural cortex detect brain signals and deliver electrotherapy without causing brain edema or inflammation.^{[ 9} , ^{10 ]} We observed that the wireless microprocessor connected to a graphene electrode array (named wireless neural implant) enabled the measurement of brain activity and the delivery of electrical stimulation to freely moving animals, thereby eliminating the limitation of movement constraints and providing a real‐time evaluation of therapeutic effects on motor function. Moreover, we demonstrated that graphene electrode‐based subdural motor stimulation could alleviate PD symptoms by potentiating the neural activity of the motor cortex. Cortical surface stimulation in a PD animal model modulated dominant brain rhythms, such as beta, delta, and gamma rhythms, thereby alleviating motor function and behavior. This electrotherapeutic effect was likely attributed to enhanced synaptic plasticity in the glutamatergic pathway through the upregulation of mGluR‐AMPARs and D₅R‐AMPARs. Our wireless neural implant can offer clinical and marketing benefits because of its biocompatibility, target specificity, removability, minimal invasiveness, and wireless control.

2. Results

To investigate the electrotherapeutic effects of high‐frequency stimulation (HFS) on the motor cortex of hemi‐parkinsonian rats, we used our wireless neural implant equipped with a microfabricated graphene microelectrode array (Figure S1 and Information S1, Supporting Information) and its connected headstage, NeuroStim, for this study (refer to Table 1 for a comparison with other wireless headstages). Surface stimulation at the motor cortex significantly alleviated the behavioral symptoms of PD (Figure 1A). The NeuroStim, which integrates a stimulation feature into a previous recording system,^{[ 11 ]} is compact and lightweight. These characteristics make it suitable for freely moving animals, showing diagnostic and therapeutic potential when combined with an onboard edge‐computed system. This study involves a wide range of bioassays to evaluate the therapeutic and diagnostic efficacy of the wireless, cortical surface implant in PD animal models, including behavioral, neural network, cellular, synaptic, molecular, and genetic assays. In a PD animal model, cortical surface stimulation enhanced motor gait (Figure 1B), leading to a decrease in high‐frequency brain waves (i.e., beta–gamma waves) and an increase in low‐frequency brain waves (i.e., delta waves) (Figure 1C). The stimulation‐induced enhancement in motor function was mainly involved in altered synaptic plasticity and related gene expression in the glutamatergic synapse pathway (Figure 1D).

Table 1.

Specifications of NeuroStim (this work) and comparison with other wireless headstages. Abbreviation: N/A, Not available; BLE, Bluetooth low energy; †, Commercially available.

Company	Multichannel system	White matter LLC	WAND System	Université Laval	CBRAIN	NeuroStim
/Organization	(Germany) †	(US) †	(UC Berkeley)[ 23 ]	(Canada)[ 24 ]	(our previous work, KR)[ 11 ]	(This work, KR)
Dimension	15.5 × 15.5 × 9.2 (mm3)	12 × 10 × 15 (mm3)	33 × 36 × 15 (mm3)	17 × 18 × 10 (mm3)	16 × 28 × 6 (mm3)	16 × 28 × 6 (mm3)
Weight	4.5 g (w/o battery)	7.0 g (total)	7.4 g (board)/17.95 g (total)	4.9 g (total)	2.6 g (total)	3.8 g
						(total, bigger Battery)
Power consumption	180 mW (50 mA × 3.7 V)	–	172 mW (46.5 mA × 3.7 V)	175 mW	31 mW (8.5 mA × 3.7 V)	39 mW (10.5 mA × 3.7 V)
Power source	Rechargeable battery	Rechargeable battery	Rechargeable battery	Rechargeable battery	Rechargeable battery	Rechargeable battery
			(500 mAh)	(100 mAh)	(40 mAh)	(150 mAh)
Duration of	2 h	3.5 h	11.3 h	1.2 h	4 h	10 h
continuous operation
# Recording channels	32	Recording (≈320),	128 (64 × 2)	32	Recording (8),	Recording (16),
		Wireless monitor (2)			Wireless monitor (3)	Stimulation (16)
Input signal bandwidth	1.0 Hz–5.0 kHz	0.1 Hz–25 kHz	1.0 Hz–500 Hz	0.1 Hz–500 Hz or	0.1 Hz–500 Hz	0.1 Hz–120 Hz
				0.1 Hz–20 kHz
Input range	12.4 mV	± 5 mV	≈100 mV	± 5 mV	± 5 mV	± 5 mV
ADC resolution (bit)	16	16	15	16	16	16
Embedded processor	N/A	MCU	ARM Cortex M3 + FPGA	FPGA	ARM Cortex M4	ARM Cortex M4
Computing performance	–	No information	Real‐time spectral power	Spike detection and	Real‐time gamma burst detection (spectral power)	Low‐latency closed‐loop neuromodulation
			Threshold trigger	Wavelet compression		(spectral power)
Wireless link	BLE (2.4 GHz)	BLE (2.4 GHz)	BLE (2.4 GHz)	BLE (2.4 GHz)	BLE (2.4 GHz)	BLE (2.4 GHz)
Wireless data transfer rate	–	–	1.96 Mbps	1.4 Mbps	2 Mbps	2 Mbps
Distance of	5 m	3 m	–	3 m	3 m	6 m
wireless data
Comments	No on‐board computing	†Commercially available	For primate	For rodent	LED neuroreporting	Neuromodulation
						For rat

Figure 1.

Strategies for correcting the behavior and brain waves of freely living hemi‐parkinsonian rats through wireless cortical recording and stimulation. A) Graphene‐based electrode configuration and the wireless device mounted on the motor cortex of a hemi‐parkinsonian rat for recording and stimulation. B) Schematic representation of the gait test designed to assess PD‐induced gait abnormalities (left) and improved gait (right) after high‐frequency cortical surface stimulation. C) Changes in brain waves before and after cortical surface stimulation, with spectrograms showing corrections in beta–gamma and delta frequencies following cortical stimulation. D) Molecular changes in stimulus‐induced synaptic plasticity after cortical surface stimulation. Abbreviations: FPCB, flexible printed circuit board; NMDAR, N‐methyl‐D‐aspartate receptor; VGCC, voltage‐gated calcium channel; AMPAR, α‐amino‐3‐hydroxy‐5‐methyl‐4‐isoxazolepropionic acid receptor; mGluR 5, metabotropic glutamate receptor 5; AC, adenylyl cyclase; cAMP, cyclic adnosine monophosphate; PLC, phospholipase C.

2.1. HFS of the Motor Cortex of Hemi‐Parkinsonian Rats Improves Motor Function

To examine the electrotherapeutic effect of cortical surface stimulation on PD, we delivered electrical stimulation through our wireless implantable device to the motor cortex of 6‐hydroxydopamine (6‐OHDA) lesion rat model of hemi‐parkinsonism (refer to Figure 2A for experimental timeline). After 6‐OHDA injections into the medial forebrain bundles (MFBs) of one cortical hemisphere (i.e., ipsilateral cortex), the apomorphine rotation test was conducted to confirm the successful generation of rat PD models. Rats displaying a contralateral rotation of at least 3 rpm were considered to have PD symptoms (Figure S2; Ctrl: −1.50 ± 3.22, n = 5; 6‐OHDA: 171.18 ± 13, n = 17; independent sample t‐test). Subsequently, either DBS or graphene cortical stimulation was applied to the STN or motor cortex, respectively (Figure 2B). We measured the gait of the hind paws of the PD rats (Figure 2C). The PD rats exhibited significant abnormalities in the step length ratio (right–left hind paw step difference) and velocity (Figure 2D and refer to Table 2 for statistical details). Our findings indicated that 6‐OHDA‐lesioned PD rats displayed significantly impaired gait. Upon administering HFS for 2 weeks, significant differences in the step length ratio were observed among the Cortical stim (stimulation), DBS, and PD groups. Both the DBS and cortical stim groups showed a balanced step length ratio with improved gait from the first to the second week of stimulation (Figure 2E, and refer to Table 3 for statistical details). An increased velocity ratio was also observed for both the DBS and cortical stim groups from the first week of stimulation (Figure 2F, and refer to Table 3 for statistical details).

Figure 2.

High‐frequency stimulation of the motor cortex of hemi‐parkinsonian rats improves motor function. A) Overview of the experimental timeline. B) A schematic indicating the location of the cortical electrode array and DBS electrodes in the motor cortex or STN, respectively. C) Illustration showing the gait indices of a rat. Hind paw lesion was induced by 6‐OHDA injection into the contralateral cortex. D) 6‐OHDA‐lesioned PD rats exhibited significantly impaired gait (step length and velocity ratio). E) Both the DBS and cortical stim groups displayed a balanced step length ratio, with improved gait after stimulation. F) Both the DBS and cortical stim groups showed an increased velocity ratio from the second week of stimulation. G) Electrical stimulation at the cortical surface and deep brain structures did not result in a loss of dopaminergic neurons. Representative images of TH‐positive staining in SN and STR slices. No significant differences were observed between the groups. All data are presented as the mean ± standard error of the mean. *p < 0.05, **p < 0.01, ***p < 0.005, ****p < 0.001. Abbreviations: w, week; stim, stimulation; SN, substantia nigra; STR, striatum; DBS, deep brain stimulation; TH, tyrosine hydrolase.

Table 2.

Statistical summary related to Figure 2D. Data are presented as the mean ± standard error of the mean. Independent sample t‐tests were used for comparisons.

Figure	Group (n)	Step length ratio	Velocity [cm s−1]
2D	Ctrl	1.05 ± 0.09	20.98 ± 2.45
	n	5	6
	6‐OHDA	0.80 ± 0.74	15.37 ± 0.71
	n	4	5
	Independent sample t	t = 2.023, p = 0.041	t = 2.205, p = 0.035

Table 3.

Statistical summary related to Figure 2E,F. Data are presented as the mean ± standard error of the mean. Statistical significance was determined using a two‐way analysis of variance (ANOVA).

Figure	Group (n)	pre	1 week	2 week	ANOVA
2E	PD	0.56 ± 0.07	0.52 ± 0.05	0.56 ± 0.04	F = 21.019, p = 0.001
	n	6	6	6
	DBS	0.60 ± 0.08	0.83 ± 0.07	0.80 ± 0.05
	n	7	7	7
	Cortical stim	0.61 ± 0.06	0.95 ± 0.05	1.08 ± 0.06
	n	11	11	11
2F	PD	1.00 ± 0.00	0.65 ± 0.14	0.64 ± 0.10	F = 4.582, p = 0.014
	n	6	6	6
	DBS	1.00 ± 0.00	1.41 ± 0.32	1.36 ± 0.20
	n	7	7	7
	Cortical stim	1.00 ± 0.00	1.32 ± 0.21	1.30 ± 0.28
	n	11	11	11

Next, we assessed the density of dopaminergic neurons and their fibers in the nigrostriatal pathway to determine whether electrical stimulation at the cortical surface and deep brain affects the survival of dopaminergic neurons. We measured the loss of tyrosine hydrolase (TH)‐positive cells in the SN and striatum after 2 weeks of stimulation. No significant difference in the loss of TH‐positive cells was observed between the groups (SN‐PD: 89.04% ± 3.68%, n = 10; DBS: 82.74% ± 1.64%, n = 12; Cortical stim: 77.41% ± 4.23%, n = 10; STR‐PD: 98.67% ± 0.72%, n = 4; DBS: 98.78% ± 0.48% n = 6; Cortical stim: 93.60% ± 4.76%, n = 4; one‐way ANOVA with Bonferroni's post hoc test; Figure 2G).

2.2. HFS of the Motor Cortex of Hemi‐Parkinsonian Rats Decreases High‐Frequency Brain Waves and Increases Low‐Frequency Brain Waves

We visually observed real‐time changes in brain waves before and after cortical stimulation (Movie S1, Supporting Information). Sequential 2‐week stimulation resulted in altered brain waves across all graphene channels. The waveforms are shown in the entire frequency range (Figure 3A, top row) with a specific focus on the beta range (13–30 Hz, middle row) and gamma range (30–50 Hz, bottom row). The spectrogram analysis of the entire frequency range confirmed the change in brain waves from low‐ to high‐frequency waveforms (Figure 3B). After 2 weeks of stimulation, we noted an increase in delta waves (0.1–3.5 Hz), a decrease in beta–gamma waves, and no substantial and consistent changes in theta (4–7 Hz) and alpha (9–11 Hz) waves over 2 weeks (Figure 3C and refer to Table 4 for statistical details). Each graphene channel covered the motor area of a small, mapped body part (referred to as the homunculus; Figure 3D).^{[ 12 ]} A Spearman correlation coefficient heatmap indicated a correlation of 32 channels (x‐axis) with each body part and five frequency bands (y‐axis) after 2 weeks of stimulation (Figure 3E). This finding also revealed that the change in brain wave frequencies was proportional to the homunculus size (Figure S3 and Table S1 for statistical details, Supporting Information). This result demonstrates that cortical stimulation leads to changes in brain waves, which are correlated with their ethological use in motor control.

Figure 3.

High‐frequency stimulation of the motor cortex of hemi‐parkinsonian rats reduces high‐frequency brain waves and increases low‐frequency brain waves. A) Representative example of brain wave alterations following sequential 2‐week cortical stimulation. Each row represents the raw signal (top row), beta wave (13–30 Hz), and gamma wave (30–50 Hz). B) Spectrogram analysis of the raw signal confirmed the change in brain waves from low‐ to high‐frequency waveforms. C) Following cortical surface stimulation, the delta wave (0.1–3.5 Hz) increased, the beta–gamma waves decreased, and the theta (4–7 Hz)/alpha waves (9–11 Hz) did not change with temporary increments. Data are presented as the mean ± standard error of the mean (SEM). *p < 0.05, **p < 0.01, ***p < 0.005, ****p < 0.001. D) Each graphene channel covered the motor area of a small, formed body part. E) A Spearman correlation coefficient heatmap shows a correlation across 32 channels (x‐axis) corresponding to each body part, represented with five frequency bands (y‐axis) after 2 weeks of stimulation, indicating that HFS‐induced changes in brain waves are correlated with functional use for motor control.

Table 4.

Statistical summary related to Figure 3C. Animal numbers are in parentheses for each group of experimental status: control (Ctrl), PD (before stimulation), 1 week after stimulation, and 2 weeks after stimulation. Data is presented as the mean ± standard error of the mean. Statistical significance was determined using ANOVA.

Figure	Group	# of channel	Delta	Theta	Alpha	Beta	Gamma
3C	Ctrl	30	0.60 ± 0.02	0.22 ± 0.01	0.80 ± 0.00	0.07 ± 0.00	0.03 ± 0.00
	PD	29	0.44 ± 0.01	0.23 ± 0.01	0.10 ± 0.01	0.16 ± 0.01	0.07 ± 0.01
	1 week	29	0.47 ± 0.03	0.28 ± 0.01	0.12 ± 0.01	0.10 ± 0.01	0.03 ± 0.00
	2 week	28	0.53 ± 0.03	0.24 ± 0.01	0.10 ± 0.01	0.09 ± 0.01	0.04 ± 0.00
	ANOVA		F(3112) = 10.224, p = 0.001	F(3112) = 5.638, p = 0.001	F(3112) = 4.406, p = 0.006	F(3112) = 35.178, p = 0.001	F(3112) = 21.918, p = 0.001

2.3. HFS of the Motor Cortex of Hemi‐Parkinsonian Rats Enhances mGluR‐Mediated, Long‐Term Synaptic Plasticity

Our previous studies, along with others, have shown that altered synaptic strength and plasticity are closely involved in the pathological symptoms of various brain disorders.^{[ 10} , ^{13 ]} Accordingly, we examined whether synaptic transmission and plasticity contribute to the cellular mechanism underlying the alleviation of behavior symptoms of PD induced by cortical surface stimulation, using extracellular field recordings with primary motor cortex slices. Two stimulations were applied at two locations: one stimulus (Stim1) at the cortical surface (layer 1/2) and the other (Stim2) at cortical layers 3/4 (Figure 4A). These stimuli mimicked cortical surface and thalamocortical stimulation (e.g., DBS‐driven inputs), respectively. Stimulus‐induced synaptic responses consisted of fast tetrodotoxin (TTX)‐dependent presynaptic volley and slow 2,3‐dihydroxy‐6‐nitro‐7‐sulphamoyl‐benzo(F)quinoxaline (NBQX)‐sensitive postsynaptic responses to both layer 1/2 (Stim1) and 3/4 (Stim2) stimulation (Ctrl: 100.00 ± 0.00, n = 6; NBQX: 48.71 ± 7.52, n = 6; NBQX + TTX: 8.70 ± 3.77, n = 6; independent sample t‐test; Figure 4B). In both the naive and PD groups, the synaptic input–output function on the cortical surface and thalamocortical responses demonstrated a proportional relationship between stimulation intensity and field excitatory postsynaptic potentials (fEPSPs). Moreover, the NBQX‐sensitive synaptic response of layer 1/2, when stimulated in the PD group (PD Stim1), exhibited a remarkable increase compared with other groups, suggesting the hyperexcitability of layer 1/2 motor neurons in PD (Figure 4C and refer to Table 5 for statistical details). To examine whether presynaptic responses are altered by cortical surface stimulation, we examined paired‐pulse responses at various inter‐pulse intervals (1000, 500, 250, 100, and 50 ms) in both the Ctrl and PD groups. No significant differences were observed in the paired‐pulse ratio (a known measurement for presynaptic activity) between the groups (Figure 4D and refer to Table 6 for statistical details). These results indicate that cortical surface stimulation preferentially induces postsynaptic modification, enhancing motor activity, particularly in PD.

Figure 4.

High‐frequency stimulation of the motor cortex of hemi‐parkinsonian rats enhances mGluR/D₅R‐mediated, long‐term synaptic plasticity. A) Schematic illustration of electrode locations for stimulation and recording. Two stimulations were applied at two locations: one stimulus (Stim1) at the cortical surface (layer 1/2) and the other (Stim2) at cortical layer 3/4. B) Representative field excitatory postsynaptic potentials (fEPSPs) with NBQX and TTX at each layer. C) Cortical surface stimulation in the PD group enhanced synaptic transmission compared with the other groups. Input–output curves showing the amplitude of fEPSPs as a function of stimulation intensity D) The paired pulse ratio across different stimulus intervals showed no significant alterations in any groups. E,F) Synaptic responses to three consecutive 130‐Hz high‐frequency stimuli were measured using the field potential amplitude. HFS induced long‐term synaptic potentiation in all four groups, with a salient increase in the PD Stim1 group (PD Stim1, n = 17, PD Stim2, n = 12, Ctrl Stim1, n = 12, Ctrl Stim2, n = 9, one‐way ANOVA with Bonferroni's post hoc test). G) Schematic showing stimulation and recording locations under drug application. aCSF, artificial cerebrospinal fluid; D‐AP5, an NMDAR blocker; MPEP, an mGluR5 blocker; LY 367385, an mGluR1 blocker, SCH 22390; a dopamine receptor 5 (D₅R) blocker. H) The PD groups, regardless of the presence of D‐AP5, showed enhanced responses compared with the control groups, both of which showed no D‐AP5‐induced alterations. I) The paired‐pulse ratio did not exhibit significant changes even in the presence of D‐AP5, indicating no involvement of presynaptic modification in cortical surface stimulation. J–L) D‐AP5 completely blocked long‐term synaptic potentiation in control rats, whereas it partially blocked long‐term synaptic potentiation in PD rats. Additional application of MPEP and LY 367385 completely eliminated HFS‐induced synaptic potentiation in PD rats. Moreover, the application of SCH 22390 partially blocked synaptic potentiation in PD rats. All data are presented as the mean ± standard error of the mean. *p < 0.05, ** p < 0.01, ***p < 0.005, ****p < 0.001. Abbreviations: Rec, recording; stim, stimulation; fEPSP, field excitatory postsynaptic potentials; NBQX, 2,3‐dihydroxy‐6‐nitro‐7‐sulphamoyl‐benzo(F)quinoxaline, an AMPA receptor blocker; TTX, tetrodotoxin, a sodium channel blocker.

Table 5.

Statistical summary related to Figure 4. Data are presented as the mean ± standard error of the mean. Statistical significance for Figure 4C was determined using two‐way analysis of variance (ANOVA). Independent sample t‐tests for Figure 4H were used for comparisons.

		Stimulus intensity [µA]
Figure	Group (n)	50	100	150	200	250	300	350	400	ANOVA
4C	Ctrl Stim1 (28)	0.35 ± 0.03	0.79 ± 0.05	0.99 ± 0.06	1.11 ± 0.06	1.22 ± 0.06	1.31 ± 0.06	1.39 ± 0.06	1.46 ± 0.63	F = 31.601, p = 0.001
	Ctrl Stim1 (19)	0.40 ± 0.05	0.76 ± 0.05	0.95 ± 0.05	1.07 ± 0.06	1.14 ± 0.05	1.22 ± 0.05	1.29 ± 0.06	1.35 ± 0.06
	PD Stim2 (31)	0.43 ± 0.05	0.91 ± 0.07	1.17 ± 0.07	1.35 ± 0.08	1.50 ± 0.08	1.61 ± 0.09	1.72 ± 0.10	1.78 ± 0.10
	PD Stim2 (28)	0.41 ± 0.03	0.78 ± 0.03	1.17 ± 0.07	1.1 3 ± 0.04	1.30 ± 0.03	1.34 ± 0.05	1.29 ± 0.06	1.35 ± 0.06
4H	Ctrl Stim1 (29)	0.52 ± 0.04	0.97 ± 0.48	1.19 ± 0.05	1.33 ± 0.06	1.43 ± 0.06	1.50 ± 0.06	1.58 ± 0.06	1.64 ± 0.06	–
	Ctrl Stim1 (26)	0.60 ± 0.03	0.98 ± 0.04	1.16 ± 0.04	1.31 ± 0.06	1.42 ± 0.07	1.50 ± 0.07	1.56 ± 0.07	1.62 ± 0.07
	Independent sample t	t = 1.741, p = 0.044	t = 0.160, p = 0.437	t = −0.319, p = 0.375	t = −0.209, p = 0.418	t = −0.042, p = 0.483	t = −0.035, p = 0.486	t = −0.171, p = 0.432	t = −0.226, p = 0.411
	PD Stim2 (39)	0.64 ± 0.44	1.08 ± 0.05	1.31 ± 0.06	1.46 ± 0.06	1.59 ± 0.06	1.68 ± 0.06	1.77 ± 0.05	1.82 ± 0.05	–
	PD Stim2 (29)	0.69 ± 0.06	1.12 ± 0.07	1.33 ± 0.08	1.50 ± 0.08	1.60 ± 0.08	1.69 ± 0.09	1.78 ± 0.09	1.83 ± 0.09
	Independent sample t	t = 0.646, p = 0.260	t = 0.480, p = 0.316	t = 0.202, p = 0.197	t = 0.390, p = 0.349	t = 0.185, p = 0.427	t = 0.092, p = 0.463	t = 0.156, p = 0.438	t = 0.021, p = 0.492

Table 6.

Statistical summary related to Figure 4D and Figure 4I. Data are presented as the mean ± standard error of the mean. Statistical significance was determined using one‐way analysis of variance (ANOVA).

		Interpulse duration [ms]
Figure	Group (n)	1000	500	250	100	50
4D	Ctrl Stim1 (19)	95.70 ± 1.48	93.41 ± 1.70	90.58 ± 2.38	103.98 ± 2.89	119.96 ± 3.37
	Ctrl Stim1 (15)	93.74 ± 2.14	88.77 ± 3.58	82.30 ± 4.67	99.22 ± 3.98	116.92 ± 4.55
	PD Stim2 (31)	95.3 ± 1.29	92.57 ± 1.27	90.60 ± 1.87	109.74 ± 2.29	129.52 ± 3.97
	PD Stim2 (22)	93.57 ± 1.39	89.66 ± 2.24	85.15 ± 3.44	94.41 ± 4.21	116.26 ± 4.21
	ANOVA	F (3,83) = 0.947, p = 0.422	F (3,83) = 1.066, p = 0.368	F (3,83) = 1.813, p = 0.151	F (3,83) = 5.133, p = 0.003	F (3,83) = 2.622, p = 0.056
4I	Ctrl (30)	96.82 ± 1.21	95.79 ± 1.49	95.39 ± 2.08	108.88 ± 2.73	108.88 ± 3.63
	Ctrl AP5 (26)	97.56 ± 0.64	96.34 ± 0.97	95.24 ± 1.46	107.99 ± 2.24	120.14 ± 2.78
	PD (39)	96.57 ± 0.92	94.47 ± 1.33	92.97 ± 1.62	105.79 ± 2.04	118.93 ± 2.63
	PD AP5 (25)	98.64 ± 1.09	95.62 ± 1.50	94.11 ± 1.53	107.23 ± 2.00	116.62 ± 2.43
	ANOVA	F (3116) = 0.818, p = 0.422	F (3116) = 0.496, p = 0.686	F (3116) = 0.475, p = 0.700	F (3116) = 0.373, p = 0.773	F (3116) = 0.231, p = 0.875

Next, we examined whether cortical stimulation induces long‐term synaptic potentiation, which appears to be associated with behavioral changes in the long term. Three consecutive 130‐Hz high‐frequency stimulations induced long‐term synaptic potentiation in all four groups (Figure 4E). The synaptic modification in the PD Stim1 group exhibited markedly greater fEPSPs than that in the other groups (Ctrl Stim1: 117.11% ± 4.07%, n = 12, Ctrl Stim2: 118.03% ± 7.42%, n = 9, PD Stim1: 141.43 ± 6.26, n = 17, PD Stim2: 114.18% ± 3.71%, n = 12; one‐way ANOVA with Bonferroni's post hoc test; Figure 4F). This finding is likely due to compensatory mechanisms for decreased thalamocortical input resulting from dopaminergic degeneration (see Discussion for cellular mechanisms). To elucidate the molecular mechanisms underlying the long‐term synaptic modification of the cortical surface, we performed an antagonism test using several receptor blockers involved in long‐term synaptic plasticity (Figure 4G). The input–output relationship at the synapse of the cortical surface was examined using an N‐methyl‐D‐aspartate (NMDAR) antagonist, D‐2‐amino‐5‐phosphonopentanoic acid (D‐AP5), in the control and PD groups. Regardless of the presence of D‐AP5, the PD groups showed enhanced responses compared with the control groups (Figure 4H and refer to Table 5 for statistical details). We also examined the paired‐pulse ratio in all four groups in the presence of D‐AP5. As anticipated, in the presence of D‐AP5, the paired‐pulse ratio did not exhibit significant changes, indicating little or no involvement of presynaptic modification in cortical surface stimulation (Figure 4I and refer to Table 6 for statistical details). However, D‐AP5 completely blocked long‐term synaptic potentiation in control rats, but it only partially blocked long‐term synaptic potentiation in PD rats. Subsequently, we examined a molecular source of the remaining potentiation in the PD groups. Additional application of 2‐methyl‐6‐phenylethynyl‐pyridine (MPEP), a mGluR5 blocker, and (S)‐(+)‐α‐amino‐4‐carboxy‐2‐methylbenzeneacetic acid (LY 367385), a mGluR1 blocker, completely eliminated the HFS‐induced synaptic potentiation in PD, suggesting a compensatory activation of mGluRs in PD (Ctrl: 119.53% ± 5.02%, n = 18; Ctrl_AP5: 100.04% ± 1.97%, n = 11; independent sample t test) (PD: 141.66% ± 6.68%, n = 19; PD_AP5: 118.51% ± 9.32%, n = 18; PD_AP5+MPEP+LY: 97.01% ± 2.56%, n = 24; one‐way ANOVA with Bonferroni's post hoc test). Furthermore, 7‐chloro‐3‐methyl‐1‐phenyl‐1,2,4,5‐tetrahydro‐3‐benzazepin‐8‐ol (SCH 22390), a dopamine receptor 5 (D₅R) blocker, partially blocked synaptic potentiation, suggesting the involvement of D₅Rs in PD (PD: 141.66% ± 6.68%, n = 19; PD_{SCH 22390}: 125.41% ± 4.58%, n = 17; independent sample t‐test; Figure 4J–L).

2.4. HFS of the Motor Cortex of Hemi‐Parkinsonian Rats Alters Gene Expression in the Glutamatergic Synapse Pathway

We investigated whether gene expression associated with glutamatergic and dopaminergic synapse pathways changed after 2‐week cortical surface stimulation using next‐generation sequencing. We used the heatmap to illustrate the significant gene expression profile in the glutamatergic synapse pathway (Figure 5A). The PD group receiving cortical stimulation exhibited increased gene expression for various glutamatergic receptors/channels when compared with the PD group not receiving cortical stimulation. For example, D₅R (Drd5), mGluR5 (Grm5), VGCC (Cacna1a, Cacna1c, and Cacna1d), AMPAR (Gria2,3,4), and NMDAR (Grin2a) were altered in the PD plus cortical stimulation group (Figure 5B and refer to Table 7 for statistical details). Furthermore, gene expression associated with intracellular signaling pathways, such as adenylate cyclase (Adcy4 and Adcy7), G‐protein (Gng11), calcium signaling (Itpr2), and phospholipase (Pla2g4a and Plcb2), also increased after cortical stimulation (Figure 5C; refer to Table S2 for statistical details, Supporting Information). These findings suggest that cortical surface stimulation can enhance the gene expression of many cation channels associated with glutamatergic/dopaminergic synaptic transmission and plasticity through G‐protein‐mediated intracellular signaling.

Figure 5.

High‐frequency stimulation of the motor cortex of hemi‐parkinsonian rats alters gene expression in the glutamatergic/dopaminergic synapse pathways. A) A heatmap illustrating the top seven differences in the expression patterns of significant glutamatergic/dopaminergic synapse‐associated genes after cortical surface stimulation (PD, n = 3, cortical stim, n = 3). The color keys of each box indicate gene expression levels in the direction of upregulation (reddish) or downregulation (bluish). Gene expression differences were analyzed using the “edgeR” R package. B,C) Bar plots representing the relative fold changes in mRNA expression of genes associated with the glutamatergic/dopaminergic synapse pathway before and after cortical surface stimulation, when the mRNA expression before cortical surface stimulation was normalized to 1. *p < 0.05, ** p < 0.01, ***p < 0.005, ****p < 0.001. Abbreviations: cortical stim., cortical stimulation; abs, absolute; log2FC, log2‐fold change.

3. Discussion

This study demonstrated 1) real‐time monitoring of brain waves in the motor cortex for PD, 2) the electrotherapeutic effectiveness of cortical surface stimulation in improving motor activity in PD, and 3) the role of glutamatergic/dopaminergic receptors and their intracellular signaling with cortical surface stimulation, revealing molecular substrates for the alleviation of PD symptoms. By using our newly developed cortical surface implant, we delivered electrical stimulation directly to the motor cortex and observed significant improvements in the behavioral symptoms of PD. This cortical surface stimulation was as effective as DBS for PD and more biocompatible than DBS, demonstrating the clinically adaptable, minimal invasiveness of our implant (Table S3, Supporting Information). Cortical surface stimulation restored brain waves and promoted synaptic transmission and plasticity in layer 1/2 through the upregulation of the mGluR‐AMPAR and D₅R‐AMPAR pathways. Thus, our neural implant enabled both cortical activity tracking for diagnostic purposes and cortical surface stimulation for therapeutic purposes, demonstrating an integrated neurotherapeutic interface for patients with PD.

Our fully implantable device is currently under development for clinical implementation. The wearable multichannel electrode can be inserted into the subdural or epidural layers of the motor cortex and is connected to a microprocessor positioned between the skull and the skin. The wireless Bluetooth communication system records brain waves for diagnosis and delivers electrical pulses for treatment. A small rechargeable battery adjacent to the microprocessor can be charged through inductive coupling from external sources. Once a PD‐specific brain wave pattern is detected during the patient's daily activities, stimulation can be immediately administered until PD symptoms are alleviated. Additional research and clinical trials are necessary to validate this biomedical device and explore the long‐term benefits of cortical surface stimulation for patients with PD. The ability of cortical surface stimulation to shift brain waves from high frequency (i.e., gamma and beta rhythms) to low frequency (i.e., delta rhythm) enables the development of an effective neurotherapeutic device (Table S4, Supporting Information).

In our study, we systematically analyzed four frequency bands of local field potentials in the motor cortex—delta (0.1–3.5 Hz), theta (4–7 Hz), alpha (9–11 Hz), and beta–gamma (13–50 Hz)—to assess the effects of cortical surface stimulation on PD. PD is known to be associated with excessive synchronization in the beta–gamma band, which is linked to motor impairments such as rigidity and bradykinesia.^{[ 14 ]} Consistently, we observed that cortical surface stimulation in the PD animal model significantly reduced the beta–gamma power, suggesting that electrical intervention effectively subdued the pathological high‐frequency oscillations. In contrast, the delta band, which contributes to motor command and control,^{[ 15 ]} increased upon cortical surface stimulation in the PD animal model, coinciding with a restoration of motor activity. Meanwhile, the theta and alpha bands did not show consistent long‐term changes after cortical stimulation, reflecting the variable involvement of these frequencies in PD as reported in previous studies.^{[ 16 ]} Together, these observations suggest that the high‐to‐low frequency shift induced by cortical surface stimulation can serve as a quantitative biomarker for therapeutic efficacy.

Cortical surface stimulation also exerts long‐term effects on synaptic plasticity and gene expression in the glutamatergic/dopaminergic synapse pathways. Dopaminergic degeneration in PD leads to a reduction in thalamocortical input to the motor cortex, reducing the neurocircuit activity of the columnar motor network (Figure 6 ). As suggested in our and other studies, such sensorimotor deprivation enhances cortical activity as a homeostatic adjustment.^{[ 17 ]} Sensorimotor deprivation reduces synaptic plasticity at the cortical surface and alters behavior to compensate for the reduced thalamic input to the cortex, which, in turn, can increase the intrinsic activity of neurons in the motor column. Under these conditions, cortical surface stimulation can enhance synaptic plasticity (i.e., long‐term potentiation [LTP]) and promote the expression of LTP‐related proteins, such as mGluR/D₅ receptors and G‐protein‐mediated intracellular molecules. For instance, the mGluR5 signaling pathway, reciprocally influenced by dopamine receptor activity, plays a vital role in potentiating synaptic transmission and motor‐related plasticity, as supported by previous studies.^{[ 18 ]} The modulation of the mGluR5 signaling pathway aligns with the therapeutic approach for alleviating motor dysfunction induced by PD. While altered synaptic plasticity and gene expressions can be the consequence of cortical surface stimulation, it is unclear whether the alteration of synaptic plasticity and its related gene expression is a cause of PD alleviation. The generation of beta–gamma rhythms is closely associated with AMPAR activity in excitatory and inhibitory neurons,^{[ 19 ]} and alterations in these rhythms are linked to motor activity malfunction,^{[ 19} , ^{20 ]} Conversely, HFS‐induced increases in delta waves might enhance attention because delta‐band oscillations synchronously increase with attention.^{[ 21 ]} Further research is needed to elucidate how HFS‐induced molecular changes affect the ratio of delta to beta–gamma waves. In addition, as our study was conducted only in a PD animal model, these findings might not be directly applicable to nonhuman primates or humans due to the more complex neural network of their motor cortices. However, given the DBS neurostimulator, which has been validated in animal and clinical settings,^{[ 14} , ^{22 ]} our electrocorticography neurostimulator warrants clinical attempts on PD patients as in our PD animal model.

Figure 6.

Schematic illustration showing a circuit mechanism of how cortical surface stimulation implanted on the motor cortex can alleviate PD symptoms. The therapeutic effect of cortical stimulation on PD is compared with that of DBS. DBS indirectly enhances the activity of the thalamocortical network by dopaminergic degeneration, whereas cortical stimulation directly enhances cortical network activity.

4. Conclusion

This study demonstrates the potential of our wireless cortical surface implant based on graphene electrode arrays for diagnosing and alleviating PD symptoms in freely moving animal models. The cortical surface stimulation via the implant significantly restored motor behaviors, such as the gait pattern and velocity, by correcting abnormal brain wave oscillations in the motor cortex. These changes were engaged with enhanced synaptic plasticity, particularly in glutamatergic and dopaminergic pathways, and were further supported by the upregulation of genes associated with glutamatergic and dopaminergic signaling. Compared to DBS, cortical surface stimulation achieved comparable therapeutic outcomes with superior biocompatibility, minimal invasiveness, and precise targeting. These results highlight the promise of this implant as a minimally invasive neurotherapeutic device for managing PD and advancing treatments for other neurodegenerative diseases.

5. Experimental Section

System Architecture

The innovative feature of the proposed neuroprosthetic device (NeuroStim) is its ability to deliver short‐latency, closed‐loop electrical stimulation in response to computations performed on measured physiological signals. Closed‐loop stimulation based on spectrum power can operate independently in the device once the appropriate parameters are uploaded using wireless telemetry.^{[ 32 ]} Figure S1A (Supporting Information) displays the simplified schematic design of the system architecture. The complete 3.8‐g neuroprosthetic device included components for recording, stimulating, telemetry, and power management and a microprocessor for edge computing. The device was designed using a system‐on‐module architecture for a compact form factor. Table 1 summarizes the proposed system specifications compared with those of other commercialized telemetry and custom‐designed edge‐computing neuroprosthetic devices.^{[ 11} , ³² , ^{33 ]} The developed edge computing had an integrated closed‐loop system with superior recording, stimulation, digital signal processing, telemetry, low power consumption, and real‐time edge computing capabilities. The custom‐designed prosthetic devices met desirable specifications for practical use, including small size, low weight, and low power consumption, suitable for applications in freely moving animals. Given the weight of a rat, the device is sufficiently light to be tested without prior habituation and/or training.

Analog Front‐End

For programmable signal input amplifiers, digital electrophysiology interface chips from INTAN Technologies, LLC (Los Angeles, CA, USA) were used. The RHS2116 chip, a low‐power 32‐channel single‐ended amplifier integrated with a 16‐bit analog‐to‐digital converter, was used for recording local field potentials of M1. The output from the amplifier chip was filtered using variable R–C elements. Different filter parameters were selected and programmed through the wireless interface. The bandpass filter's high‐pass and low‐pass cutoff frequencies were set at 0.5 Hz and 4 kHz, respectively. After filtering, the signals were transmitted to the onboard microprocessor via an 8‐Mbit s⁻¹ serial peripheral interface (SPI).

Computing Unit

The nRF52832 System‐on‐Chip (SoC, Nordic Semiconductor) was selected for computing and wireless data transmission on the device because of its low power consumption and flexibility. It was built around an ARM cortex‐M4 processor (64 MHz clock) with single‐cycle multiply and accumulates MAC instructions. In addition, the single‐precision floating‐point unit of the processor is suitable for the real‐time analysis of incoming neural signals. The device was designed to invoke a 512‐Hz interrupt service routine (ISR) from a 32 768‐Hz crystal oscillator. All ISRs captured physiological signals from 32 channels of graphene electrodes.

Stimulator

The proposed device could deliver independent stimulation across 32 channels through the Intan RHS2116. Biphasic stimulation was delivered to a common ground electrode independently for each of the 32 electrodes. The voltage regulator of the stimulation circuit could produce a constant pulse up to a maximum intensity of 7 V with a minimum pulse width of 60 µs. The stimulator delivered free‐running, periodic pulses at frequencies between 10 and 130 Hz or delivered individual pulses triggered by a signal from the SoC within 2 ms of the operational service routine, demonstrating low‐latency operation due to the edge computing scheme in our device.

Wireless Interface and Data Loss

The BLE 4.0 protocol was implemented in the edge computing device to reduce the complexity of wireless data transmission. With a wireless data bandwidth of 2 Mbit s⁻¹ and a carrier frequency of 2.4 GHz, the maximum data payload was 240 bytes. One of the most severe issues with telemetry was data loss due to disconnections. To address this, a sample sequence value with neural signals was included, ensuring that the neural signals could be logged with the exact timing even if data were lost. The data transfer rate was measured in terms of the bit error rate by performing five iterations of 1‐h recordings at a 143, 360‐bps transmission rate with 32 channels at a 512‐Hz sampling rate to characterize data loss. No data loss was observed up to 6 m under static conditions.

Graphical User Interface

The graphic user interface (GUI) was operated using customized software. Through BLE 4.0, the receiver (nRF52832) received data from the edge computing neuroprosthetic device and converted that data to a format suitable for the GUI through UART communication. In the GUI, data was saved from the device, including raw neural activity and processed motor cortex stimulation timing. In addition, parameters were transmitted to the device to specify the recording channel, signal filter band, and the duration, frequency, and intensity of stimulation for each channel.

Simultaneous Control of Multiple Devices

A multi‐device platform was developed to facilitate wireless data transmission to a single computer from multiple devices operating simultaneously in the same room. In addition, external signals, such as behavioral event markers, were wirelessly transmitted to the same computer and integrated with the data stream produced by the device.

Power Management

A3.2‐g, 150‐mA lithium polymer rechargeable battery was used for the device's small form factor design. Both theoretical and experimental assessments indicated that the device consumed 10.5 mA for recording and computing, allowing it to last up to 10 h: INTAN recording consumed 1.8 mA, edge computing for closed‐loop operations required 3.7 mA on average (over a 512‐Hz sampling rate), and BLE data communication used 1.8 mA for basic manipulations. However, the stimulation current for DBS was not constant but variable, estimated to reach up to 3.2 mA depending on the intensity and frequency of therapeutic interventions.

Connectors

Dual head sockets (Samtech Co., USA) with a 1.27 mm pitch and 36 channels (32 channels, 2 references, and 2 grounds) were used on the NeuroStim side of the interaction with the 32 channel electrode. A customized connection adapter (CN32) with 36 channel NSD‐AA connectors (Omnetics, MN, USA) and 1.27 mm pitch 36 channel dual head pins from Samtech is manufactured in order to compare with a conventional wired system.

Animals

Adult male Sprague–Dawley rats weighing 260–300 g were used. The rats were housed in a room maintained at a temperature of 25°C and a 12‐h light/dark cycle. All animal procedures were approved by the Institutional Animal Care and Use Committee at Incheon National University (INU‐ANIM‐2017‐08) in accordance with their guidelines.

6‐OHDA‐Induced Parkinsonian Rat

Degeneration of dopaminergic neurons was induced by stereotactically administering 6‐OHDA into dopaminergic neurons located in the MFB to generate a drug‐induced PD animal model. Rodents were anesthetized with isoflurane before the unilateral creation of a burr hole at −3.84 mm anterior and −1.4 mm lateral to the bregma for rats. 6‐OHDA dissolved in saline with 0.02 % ascorbic acid was injected using a Hamilton syringe with a micro pump injector (26 G needle) at a depth of 8.5 mm from the dura and an infusion rate of 0.5 µL min⁻¹ for 8 min. The injection targeted the MFB unilaterally for a comparison with the unaffected hemisphere as the control. After PD induction over 2 weeks, the rats received a subcutaneous injection of apomorphine (0.5 mg kg⁻¹) and were considered lesioned when they exhibited a rotation of at least 3 rpm. Unilateral implementation was performed on the cortical surface or in the subthalamic nucleus with either a graphene electrode array or a handmade platinum–iridium DBS electrode, respectively.

Electrode Implantation

Unilateral implantation was performed on the cortical surface or in the STN using either a graphene electrode array or a handmade platinum–iridium DBS electrode, respectively. Animals were anesthetized with isoflurane (2–2.5 %) and placed in a stereotaxic frame, followed by scalp preparation and craniotomy. For cortical implantation, the dura mater was carefully removed, and the graphene electrode array was positioned onto the motor cortex and secured with dental cement. For STN implantation, a burr hole was drilled above the target site, and the platinum–iridium electrode was stereotaxically inserted and fixed with dental cement. Post‐surgical recovery included monitoring until full recovery and providing analgesics as needed. The implanted electrodes were connected to either wired or wireless headstages for subsequent recordings and stimulation experiments. Figure S4 (Supporting Information) provides a detailed illustration of the electrode placement procedure for the cortical region.

Gait Test

To assess locomotor function, a gait test was performed. The walking task was repeated until a satisfactory walk of at least five steps without pause was achieved. The bottom view of the walking track was video‐recorded to obtain the spatiotemporal parameters of gait patterns. During the measurement, the rats were allowed to walk freely on the track at their own pace. The gait pattern parameters of the hind paws were analyzed using ImageJ (National Institutes of Health, MD, USA). The analysis involved calculating the step length ratio, which was achieved by dividing the step length of the lesioned hind paw by that of the intact hind paw. The velocity ratio was determined by dividing the velocity at the measured time point by the velocity immediately after disease induction. Velocity was calculated as the distance covered by the hind paw from the first to the last step, divided by the time taken.

Immunohistochemistry

After the behavioral tests, the rats were sacrificed for TH immunostaining to evaluate dopaminergic cell loss. The rats were deeply anesthetized with isoflurane and perfused transcardially with phosphate‐buffered saline (PBS) and 4% paraformaldehyde to fix the brains. The brains were postfixed in 4% paraformaldehyde for 24 h and cryoprotected in 30% sucrose solution until they sank. Subsequently, the brains were embedded in the optimal cutting temperature medium (Thermo Fisher Scientific, USA) and frozen using dry ice. The frozen tissues were sectioned using a cryostat (Leica CM1520) at a thickness of 50 µm, and the areas of the substantia nigra and striatum were collected. For permeabilization, the sections were washed with 0.3% Triton X‐100 solution for 30 min and then incubated with a primary rabbit anti‐TH antibody (1:1000, AB152, Millipore, USA). After three washes with PBS, the sections were incubated with a secondary anti‐rabbit antibody for 1 h (1:200, MP‐7401, Vector Labs, USA). Finally, the binding sites of the primary antibody were visualized using 3,3‐diaminobenzidine for 3 to 5 min.

Graphene Electrode Fabrication and Stimulation

The method employed in the previous study to fabricate a graphene electrode array was used.^{[ 9b ]} Briefly, monolayer graphene grown on Cu foil through chemical vapor deposition was multi‐stacked into four layers following the chemical doping process. A 1.2‐µm‐thick layer of polyimide was spin‐coated onto a Cu carrier substrate. Interconnects of Cr/Au (3 nm/40 nm) were formed on the PI film through thermal evaporation and photolithography. Four‐layer graphene was transferred onto the PI substrate, in contact with the Au tracking metal, and patterned by photolithography and oxygen plasma. The undesired area of the Au tracking metal was insulated by SU‐8. The insulated graphene was patterned using photolithography and oxygen plasma etching. Nitric acid was used for the chemical doping of graphene (Figure S5, Supporting Information). This electrode configuration achieved an optimal balance of high signal‐to‐noise ratio (SNR) and multichannel scalability, outperforming various alternative materials in neural recording systems (Figure S6 and Table S5, Supporting Information).

Electrotherapy According to Stimulation Location

Three groups of rats were randomly assigned to the cortical surface stimulation group (n = 11), the deep brain stimulation group (n = 7), and the PD group (n = 6). In the graphene electrode stimulation and DBS groups, electrical stimulation was administered to awake rats (starting 1 week after electrode implantation, 1 h day⁻¹, consecutively 5 days per week) for 2 weeks. The RHS stim/recording controller was connected to the electrode implanted with the SPI headstage for stimulation and recording. The stimulation (biphasic pules, 130 Hz, 66.6 µs, 300 µA) was applied. Local field potentials in the motor cortex or subthalamic nucleus were recorded before and after stimulation to investigate the effects of cortical surface stimulation. To evaluate the effects of cortical surface stimulation on gait, the step length and BOS length of the hind paws were measured in the gait test for 3 weeks.

Wireless Device

The proposed wireless neuroprosthetic device delivered short‐latency, closed‐loop electrical stimulation for brain signals through telemetry. The compact 3.8‐g headstage of the device included components for recording, stimulating, telemetry, and power management, as well as microprocessor components for effective edge computing. Edge computing enabled low‐latency, multiple closed‐loop therapies, surpassing the capabilities of traditional telemetry functions. Custom‐designed neuroprosthetic devices are necessary to meet the requirements of small size, light weight, and low power consumption for applications in animals. The device used BLE 4.0 for wireless data transmission, minimizing data loss with sample sequence values for accurate timing even in the event of a short break of disconnections. Detailed information is provided in Figure S7 and Information S1 (Supporting Information).

Brain Wave Analysis

The brain wave following 2‐week cortical surface stimulation was visually observed in an active condition for 30 min. To determine the channels and frequency bands that changed the most significantly before and after electrical stimulation, we used the Kendall rank correlation coefficient (τ).^{[ 34 ]} First, the pre‐stimulation data and 2‐week post‐stimulation data were divided into nonoverlapping 10‐s samples for each of the 32 channels. This resulted in 44 samples for pre‐stimulation data and 73 samples for post‐stimulation data. Second, brain wave features were extracted for each sample by calculating the band power (delta, theta, alpha, beta, and gamma) for each channel. Thus, each sample contained 80 neural features. Finally, the statistical correlation coefficient τ between each neural feature and the electrical stimulation status (label) was computed as follows:

$$\tau =\frac{2}{n\left(n-1\right)}\sum _{i=1}^{n-1}\sum _{j>i}^n\mathrm{sgn}\left(x_i-x_j\right)\mathrm{sgn}\left(y_i-y_j\right)$$ (1)

Here, n represents the number of samples, x_i denotes the feature value, y_i indicates the label (0 for pre‐stimulation and 1 for post‐stimulation), and sgn(·) is the sign function. The value of τ ranges from −1 to 1, and a large absolute value indicates a strong association between the feature and the label.

Brain Slice Preparation

Primary motor cortical slices were obtained from the hemi‐parkinsonism rats. According to the established method,^{[ 9c ]} briefly, animals were deeply anesthetized using 2 % isoflurane, and their brains were rapidly removed and placed in a chilled, oxygenated dissection buffer. The buffer consisted of 75.0 mm sucrose, 25.0 mm glucose, 87.0 mm NaCl, 2.5 mm KCl, 1.3 mm NaH₂PO₄, 25 mm NaHCO₃, 7.0 mm MgCl₂, and 0.5 mm CaCl₂. Using a Leica VT1200S vibratome, coronal motor cortical slices were prepared with a thickness of 400 µm and transferred them to a holding chamber with artificial cerebral spinal fluid (ACSF) composed of 25.0 mm glucose, 125.0 mm NaCl, 2.5 mm KCl, 1.3 mm NaH₂PO₄, 25.0 mm NaHCO₃, 1.0 mm MgCl₂, and 2 mm CaCl₂ and saturated with 5 % CO₂ and 95 % O₂. The slices were incubated for 12 min at 32 °C and allowed to recover for 1 h at room temperature. Individual slices were then transferred to a submersion‐type recording chamber and continuously superfused with oxygenated ACSF at a temperature of 31 to 32 °C. The left hemisphere slices were designated as the control (Ctrl) group, whereas the right hemisphere slices were used as the PD group.

Extracellular Field Recording

The extracellular post‐synaptic potentials (fEPSPs) were recorded in the field using glass electrodes filled with ACSF. Synaptic responses were elicited through a stimulating electrode (concentric bipolar microelectrode, FHC, USA) by applying single electric stimulation to two locations (motor cortex layer 1/2 or 3/4), with the field potentials recorded only in the motor cortex layer 1/2. All responses were acquired using an Axon Digidata 1550B 8‐Channel Digitizer (Molecular Devices, San Jose, CA) and amplified using a MultiClamp 700B Microelectrode amplifier (Molecular Devices, San Jose, CA, USA). The maximum field potential amplitudes were measured to establish the input–output (I/O) relationship. For the paired‐pulse test and LTP, a 40%–50% half‐response was used. Paired pulse intervals included 1000, 500, 250, 100, and 50 ms. After stable postsynaptic responses were maintained for 20 min, three tetanic stimulations (130 Hz, 1‐s duration, and 5‐min intervals) were applied. The amplitudes of the evoked synaptic responses were measured and represented relative to the normalized preconditioned baseline. Synaptic response changes were measured for 1 h, representing the average response over the last 5 min.

RNA Sequencing

The preparation of whole transcriptome libraries and sequencing were conducted by Macrogen Inc. The total RNA concentration was calculated by Quant‐IT RiboGreen (Invitrogen, Carlsbad, CA, USA). TapeStation RNA ScreenTape (Agilent Technologies, Santa Clara, CA, USA) was used to assess the integrity of a total RNA. Only high‐quality RNA preparations, with RNA integrity number (RIN) over 7.0, were used for RNA library construction. A library was independently prepared using 1 µg of total RNA for each sample by using the Illumina TruSeq Stranded mRNA Sample Prep Kit (Illumina, Inc., San Diego, CA, USA). The libraries were quantified using KAPA Library Quantification kits for Illumina Sequencing platforms (KAPA BIOSYSTEMS). An Illumina NovaSeq (Illumina) was used for paired‐end sequencing (2 × 100 bp).

Differential Expression Analysis and Data Visualization

Reads with low‐quality and adapter sequences were trimmed using TrimGalore.^{[ 35 ]} The STAR 2‐pass mode was used to align raw sequencing reads to the reference genome of Rattus norvegicus (mRatBN7.2).^{[ 36 ]} A list of genes associated with the glutamatergic/dopaminergic synapse pathway was obtained from the Kyoto Encyclopedia of Genes and Genomes rat pathway “Glutamatergic Synapse”^{[ 37 ]} with the “msigdbr” R package.^{[ 38 ]} Differential expression analysis was examined using the “DESeq2R” package.^{[ 39 ]} Genes with a p < 0.05 and |log2FC| ≥ 0.5 were considered significant. The expression of genes was visualized using a heatmap with the “pheatmap” R package.

Statistics

All statistical analyses were performed using SPSS 28.0 (IBM Corp., Armonk, NY, USA). All graphs were generated using GraphPad Prism 8 (GraphPad Software Inc., La Jolla, CA, USA), and final arrangement and labeling were completed using Adobe Illustrator CC 2019 (Adobe Inc., San Jose, CA, USA). Extracellular field recording data were numerically presented using Axon pCLAMP11 Electrophysiology Data Acquisition and Analysis Software (Molecular Devices, San Jose, CA).

Conflict of Interest

The authors declare no conflict of interest.

Author Contributions

S.Y. led the overall project strategy and device development. S.Y., S.Q.L., and S.C.Y. served as corresponding authors and jointly conceptualized the study. H.S., K.K., J.L. contributed equally to this work. H.S., K.K., C.Y., and M.K. conducted the experiments. H.S. and K.K. acquired data. J.L., J.N., E.B., H.C., M.H., and J.K. analyzed data. S.Y., S.Q.L., and S.C.Y. drafted and revised the manuscript. S.Y., S.C.Y., J.H.A., J.G.K., and C.K.C. reviewed and edited the manuscript.

Supporting information

Supporting Information

Supplemental Movie 1

Shin H., Kim K., Lee J., Nam J., Baeg E., You C., Choi H., Kim M., Chung C. K., Kim J. G., Ahn J. H., Han M., Kim J., Yang S., Lee S. Q., Yang S., A Wireless Cortical Surface Implant for Diagnosing and Alleviating Parkinson's Disease Symptoms in Freely Moving Animals. Adv. Healthcare Mater. 2025, 14, 2405179. 10.1002/adhm.202405179

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.