Electroacupuncture alters brain network functional connectivity in subacute stroke: A randomised crossover trial

Abstract

Background:

Electroacupuncture (EA) is a promising rehabilitation treatment for upper-limb motor recovery in stroke patients. However, the neurophysiological mechanisms underlying its clinical efficacy remain unclear. This study aimed to explore the immediate modulatory effects of EA on brain network functional connectivity and topological properties.

Methods:

The randomized, single-blinded, self-controlled two-period crossover trial was conducted among 52 patients with subacute subcortical stroke. These patients were randomly allocated to receive either EA as the initial intervention or sham electroacupuncture (SEA) as the initial intervention. After a washout period of 24 hours, participants underwent the alternate intervention (SEA or EA). Resting state electroencephalography signals were recorded synchronously throughout both phases of the intervention. The functional connectivity (FC) of the parietofrontal network and small-world (SW) property indices of the whole-brain network were compared across the entire course of the two interventions.

Results:

The results demonstrated that EA significantly altered ipsilesional parietofrontal network connectivity in the alpha and beta bands (alpha: F = 5.05, P = .011; beta: F = 3.295, P = .047), whereas no significant changes were observed in the SEA group. When comparing between groups, EA significantly downregulated ipsilesional parietofrontal network connectivity in both the alpha and beta bands during stimulation (alpha: t = −1.998, P = .049; beta: t = −2.342, P = .022). Significant differences were also observed in the main effects of time and the group × time interaction for the SW index (time: F = 5.516, P = .026; group × time: F = 6.892, P = .01). In terms of between-group comparisons, the EA group exhibited a significantly higher SW index than the SEA group at the post-stimulation stage (t = 2.379, P = .018).

Conclusion:

These findings suggest that EA downregulates ipsilesional parietofrontal network connectivity and enhances SW properties, providing a potential neurophysiological mechanism for facilitating motor performance in stroke patients.

1. Introduction

Stroke is a leading cause of death worldwide, and upper limb impairment is a common complication that significantly affects the daily lives and quality of life of patients with stroke. Electroacupuncture (EA), a technique that combines traditional acupuncture with electrical stimulation, has shown promise in enhancing upper-limb motor recovery in patients.^[1,2] However, the neural substrates underlying EA efficacy in upper-extremity motor recovery are still not well understood.

Recent research has demonstrated that acupuncture can regulate neural plasticity in stroke patients both functionally and structurally. For instance, one study using voxel-based morphometry analysis found that acupuncture can induce pronounced structural reorganization in the frontal areas and “default mode” network, suggesting its potential role in improving motor recovery.^[3] Functional magnetic resonance imaging (fMRI) studies have also shown that acupuncture can stimulate bilateral brain regions, particularly the gyrus, which may be associated with motor recovery in stroke patients.^[4] Additionally, acupuncture has been found to restore abnormal functional connectivity and modulate disrupted whole-brain network patterns.^[5] However, most of these studies have primarily focused on the cerebral effects of acupuncture at lower-limb acupoints. Moreover, previous research has primarily focused on the functional reorganization of widespread brain cortex regions using fMRI, with less emphasis on targeted cerebral regions. Specifically investigating the modulatory effects of acupuncture on specific brain regions can enhance our understanding of brain plasticity and potentially offer valuable clinical biomarkers. When compared to fMRI, electroencephalography (EEG) presents advantages in terms of portability, patient accessibility, high temporal resolution, and cost-effectiveness,^[6] rendering it an attractive neuroimaging technique for studying EA modulation of brain plasticity. Particularly, within EEG metrics, functional connectivity (FC) analysis, which measures the information exchange between different neuronal groups in spatially distributed regions at given frequency bands,^[7] offers valuable insights into understanding the functional reorganization of the brain following a stroke.

EA is widely recognized as a treatment involving various forms of sensory stimulation.^[8] During EA, activation of sensory neurons leads to signals traveling through peripheral nerves, spinal cord pathways, and central subcortical transmission to reach the sensory areas of the cerebral cortex. Notably, the somatosensory cortex in the parietal lobe and motor cortex in the frontal lobe are closely interconnected anatomically and functionally,^[9] providing a neural basis for the promotion of motor recovery through sensory stimulation. Moreover, deficits in sensory post-stroke are closely associated with motor impairment and recovery.^[10] Tu-Chan’s study^[11] demonstrated that sensory electrical stimulation can enhance finger fractionation and hand function in individuals with chronic acquired brain injuries, and cortical oscillations can serve as important neurophysiological markers to observe individual responses to sensory electrical stimulation. It is reasonable to consider that additional sensory stimulation in the form of EA may potentially modulate the activation of the somatosensory cortex, motor cortex,^[12] and regulate the excitability of the corticospinal tract,^[13] which is highly relevant to motor recovery. However, the specific role of EA in post-stroke motor function recovery through the modulation of communication between the somatosensory and motor cortices from a systematic perspective, as well as the time-dependent nature of EA’s regulation on brain networks, remains uncertain.

The reciprocal information exchange between the somatosensory and motor cortices, known as the parietofrontal network, which encompasses the motor cortex in the frontal lobe and the somatosensory cortex in the parietal lobe, plays a substantial role in motor planning, motor learning, and the integration and processing of sensorimotor information.^[14] Studies have shown that the parietofrontal network undergoes reorganization and evolution after stroke, and its functional rearrangement is closely related to residual motor impairments and motor recovery post-stroke.^[15,16] In our previous research,^[17] we have demonstrated that EA modulation of parietofrontal network connectivity can serve as a biomarker for predicting upper limb motor recovery post-stroke. To deepen our understanding of the neural mechanisms of EA in promoting motor recovery after stroke, we integrated graph theory-based brain network properties. This integration not only allowed us to observe EA’s regulatory effects on the local parietofrontal network but also provided insight into how EA modulates the brain network at a larger scale.

In the current study, we conducted a randomized crossover trial to investigate the potential modulatory effects of EA on brain networks. Initially, we investigated the FC differences between the EA and sham electroacupuncture (SEA) groups within the parietofrontal network, seeking to identify the specific modulatory effects of EA stimulation. Subsequently, we explored differences in small-world (SW) properties between the two groups to investigate how EA affects functional integration and segregation of whole-brain networks. By elucidating the immediate modulatory effects of EA on brain network, we aim to establish a theoretical foundation for understanding EA’s role in enhancing motor function recovery after a stroke, ultimately leading to improved clinical and rehabilitation interventions.

2. Methods

2.1. Study design

This study was conducted from November 2022 to March 2023 at Shiyan Taihe Hospital in Hubei Province, China. The study protocol was approved by the Medical Ethics Committee of Taihe Hospital and followed the Declaration of Helsinki. The trial was registered in the Chinese Clinical Trial Registry with ID ChiCTR2200065399. All participants provided informed consent prior to enrollment.

2.2. Sample size

The primary outcome considered was the difference in the parietofrontal network FC between the two groups during and after the intervention. The sample size was determined based on the differences observed in this primary outcome in a pre-experimental study involving five patients with subacute stroke. The calculation was performed using G. Power 3.1.9.6 software (HHU, Düsseldorf, Germany). The F-test (ANOVA: repeated measures, within-between interaction) was employed with a power of 0.80, alpha of 0.05, two groups, three measurements, correlation among repeated measures of 0.5, non-sphericity correction of 1, and effect size f of 0.2. The calculated sample size was 42 patients. Accounting for 20% data loss, the total required sample size was determined to be 52 participants.

2.3. Participants

The inclusion criteria were as follows: age between 30 and 80 years with first-time stroke; confirmed stroke diagnosis by clinical CT or MRI; subcortical stroke type; existing upper limb motor impairment (upper extremity of Fugl-Meyer assessment, UE-FMA < 66); stroke occurring between 15 days and 90 days prior to the first assessment; ability to follow instructions (Mini-Mental State Examination score ≥ 23); right-handedness and ability to sit in a comfortable chair for more than 30 minutes; and patients or their family members were willing to sign informed consent.

The exclusion criteria were as follows: communication or attention maintenance disorders that may affect experimental participation; contraindications to EEG collection or EA; fear of acupuncture; history of epilepsy or use of antiepileptic drugs; and unstable medical conditions or other serious illnesses that may affect experimental participation.

2.4. Randomization and blinding

Participants were assigned to two groups in a 1:1 ratio using an independent randomizer. The randomization scheme was generated using the IBM SPSS 26 software (IBM Corporation, Armonk, NY). Group assignment was concealed using opaque sealed envelopes. Each envelope contained a grouping scheme with codes written outside the envelope. Once the participants were enrolled, the sealed envelopes were opened to disclose the assigned intervention sequences, which were then implemented. The patients and assessors remained blinded to the treatment sequence, whereas the acupuncturist was aware of the assigned interventions. The data analysts analyzing the results were also blinded to patient grouping.

2.5. Interventions

Participants underwent two interventions, EA and SEA, with concurrent EEG signal acquisition. Both EA and SEA interventions were administered by an experienced acupuncturist with over 5 years of practice. The entire experimental procedure lasted for 40 minutes. Based on previous literature,^[8] a 24-hour washout period was established between the two interventions, which were scheduled during the same period on alternate days.

Four verum acupoints were selected on the paralyzed upper limb: LI4 (Hegu), LI10 (Shousanli), LI11 (Quchi), and LI13 (Shouwuli).^[18] These acupoints were chosen from the Large Intestine Meridian of Hand Yangming, following the traditional Chinese medicine theory of “Treating Flaccid Paralysis by taking Yangming Alone” which is frequently used in clinical treatment for upper limb paralysis after stroke.^[19] Control acupoints positioned 1 cun distal and lateral to the verum acupoints were also selected. These control points did not correspond to any established traditional acupoints or meridians.^[20] Figure 1 illustrates the localization of verum acupoints and non-acupoints.

Figure 1.

A schematic procedure of interventions. Four verum acupoints and control points (CP) were used in the procedure of electroacupuncture (EA) and shame electroacupuncture (SEA) stimulation.

In the EA intervention, the acupuncturist inserted needles (0.30 × 40 mm; Suzhou Medical Supplies Factory Co. LTD., Suzhou, China) perpendicularly into the four acupoints, reaching a depth of approximately 10 to 15 mm. After reporting a deqi sensation using the DeQi Sensation Questionnaire VAS score,^[21] two wires were wrapped around the needle handle to apply the EA stimulation (Dalla electroacupuncture apparatus 6805-D, Shanghai Hanfei Medical Devices Co., LTD., Shanghai, China). The EA parameters consisted of an intermittent frequency of 2 Hz over a period of 20 minutes, as in previous studies.^[22] The current was gradually increased until a slight vibration was felt without visible muscle contraction.

In the SEA intervention, the four control points were superficially punctured at a shallower depth of 2 to 3 mm using needles (0.30 × 13 mm). After ensuring the absence of a significant deqi sensation, two short-circuited wires were connected to the needle handles to simulate the EA stimulation. The SEA stimulation parameters matched those of the EA group, but no actual electrical current was delivered despite the power light of the EA apparatus being turned on. The SEA setup followed an approach similar to that of a previous study,^[23] with a stimulation time of 20 minutes. To maintain blinding, all patients were informed that the electric current would be below the perceivable threshold for the human body.

2.6. EEG recording

EEG recordings were conducted in a quiet and soundproof room. Participants’ ongoing EEG was recorded with their eyes open in 3 stages: before stimulation (5 minutes), during EA or SEA stimulation (20 minutes), and after stimulation (5 minutes). A 16-channel EEG cap with a wireless EEG amplifier (Shanghai Idea-Interaction Tech. Co., Ltd., Shanghai, China) was used according to the international 10 to 20 system. EEG signals were sampled at a rate of 500 Hz. The ground lead was placed at AFz, and the reference electrode was placed at FCz. Throughout the experiment, the impedance of the electrodes was maintained at below 20 kΩ.

2.7. EEG preprocessing

Offline EEG preprocessing was performed using the EEGLAB 2020 Toolbox implemented in MATLAB (R2021a) software (The MathWorks Inc., Natick, MA). The raw EEG data underwent the following preprocessing steps: bandpass filtering in the range of 0.5 to 40 Hz; followed by a notch filter at 50 Hz to remove powerline noise; reference to average reference; removal of muscle artifacts through visual inspection of EEG amplitudes exceeding ± 100 uV; segmentation of EEG epochs into non-overlapping 5-second segments; and removal of ocular and cardiac artifacts using the runica plug-in for independent component analysis. The detailed procedure for the EEG data processing is shown in Figure 2.

Figure 2.

A schematic diagram of electroencephalography (EEG) data processing.

2.8. Functional network analysis

FC analyses were performed using a custom MATLAB script. The phase-locking value (PLV) metric^[24] was employed to evaluate the FC among different brain regions. Notably, the alpha and beta frequency bands, recognized for their physiological relevance to motor function and extensively validated in association with motor recovery in subacute stroke patients.^[25–27] In line with this, PLVs were computed within these frequency bands: alpha (8–13 Hz) and beta (13–30 Hz).^[28] The sensorimotor cortex leads (FC3, C3, CP3, P3, FC4, C4, CP4, and P4) within the parietofrontal network were analyzed, focusing on motor-related regions (M1, C3, C4, PMC, FC3, and FC4) and somatosensory-related areas (S1, CP3, CP4, PPC, P3, and P4). Previous studies have provided a reference for selecting specific leads corresponding to distinct cortical areas.^[29,30] In cases where the participants had right hemisphere injuries, the electrode arrays were flipped across the midline to ensure that the left hemisphere was aligned with the ipsilesional hemisphere for subsequent analysis. Thus, the parietofrontal network FC was characterized by intra-ipsilesional hemispheric connectivity, intra-contralesional hemispheric connectivity, and interhemispheric connectivity. The distribution of leads of interest is shown in Figure 3.

Figure 3.

Electrode positions were utilized in the experiment. CMC = contralesional motor cortex, CPC = contralesional parietal cortex, arrows denote interconnection between brain network, IMC = ipsilesional motor cortex, IPC = ipsilesional parietal cortex, L = left, R = right.

2.9. Network topology analysis

In this study, a graph theoretical network analysis was conducted using the MATLAB Brain Connectivity Toolbox (brain-connectivity-toolbox.net). An undirected binarized network was created based on PLV matrices, with each channel representing a node and the connections between different channels defined as edges. To determine a reasonable threshold for network construction, a range of sparsity thresholds (0.1, 0.2, 0.3, 0.4, and 0.5) was considered. After evaluating various thresholds, a threshold value of 0.3 was selected. This threshold was set to achieve balance, avoiding isolated nodes or excessive spurious connections in the brain network.^[31]

The clustering coefficient (CC), characteristic path length (PL), and SW Index were computed to quantify the local and global properties of the brain network.

CC measures the functional separation of the brain network by calculating the ratio of the number of existing edges between adjacent nodes to all possible connected edges.

$${\displaystyle \begin{array}{l}{\displaystyle t_i=\frac{1}{2}{\displaystyle \sum _{j,h\in N}}{a}_{ij}{a}_{ih}{a}_{jh}}\end{array}}$$ (1)

$t_i$ represents the number of triangles around node i. where nodes j and h are all possible pairs of neighboring nodes that create triangles with node i.

The local-level CC is defined as:

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{C}\mathrm{C}=\frac{1}{n}{\displaystyle \sum _{i\in N}}\frac{2t_i}{k_i(k_i-1)}}\end{array}}$$ (2)

where the denominator represents the maximum number of edges that node i can have in the network and $k_i$ represents the degree of node i. n is the number of nodes in the specific network. In this study, the intra-parietofrontal network consisted of four nodes; therefore, n was 4.

PL quantifies the functional integration of the entire brain network. It is calculated as the average of the shortest path lengths between all possible nodes in the network.

The PL is defined as:

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{P}\mathrm{L}=\frac{1}{n(n-1)}{\displaystyle \sum _{i,j\in N,i\ne j}}{d}_{ij}}\end{array}}$$ (3)

where $d_{ij}$ indicates the shortest distance between two nodes i and j and $n$ is the set of all nodes within the whole-brain network.

The SW measures how efficiently brain networks transfer information from one region to another compared to random networks. It is computed as the ratio of normalized CC to normalized PL.

The SW is defined as:

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{S}\mathrm{W}=\frac{C/C_r}{L/L_r}}\end{array}}$$ (4)

where $C_r$ and $L_r$ represent the CC and PL of equivalent random networks with the same degree distribution.

2.10. Statistical analysis

Statistical analyses were conducted using IBM SPSS 26 software. A per-protocol analysis approach was used. A box plot was plotted to identify potential outliers within the dataset. Outliers were identified using the following criteria: values below the lower quartile minus 1.5 times the interquartile range or values above the upper quartile plus 1.5 times the interquartile range. Outliers with a significant degree of deviation were removed from the dataset. However, it is important to note that the original dataset, prior to the removal of outliers, was also subjected to sensitive analysis to ensure the robustness of the results. Normality of the data was assessed using the Shapiro–Wilk test. Data were reported as mean ± standard deviation if the data satisfied the assumption of normality. Otherwise, the data were described using quartiles. To assess the differences between groups before stimulation, either Wilcoxon’s signed-rank test or paired t test was used, depending on the normality of the data. For the analysis of the PLV values of the parietofrontal networks and graph theory indicators, repeated-measures ANOVA was performed using linear mixed-effects models. The PLV values and graph theory indicators served as the dependent variables, whereas time, group, random sequence, and time × group interaction were the independent variables. The covariance structure was set to unstructured. Fixed effects included group, random sequence, time, and group × time interaction, whereas subject ID was included as a random effect. To adjust for target effects in the between-group analyses, pre-stimulation stage indicators and random sequences were included as covariates. The group × time interaction term was tested initially. If it was found to be significant, differences between groups at different time points were analyzed further. If not significant, the main effects of the group and time were tested. Bonferroni correction for multiple comparisons was used for time differences within the groups. The significance level was set at P < .05 for all analyses.

3. Results

3.1. Participants

As illustrated in the flowchart of this study in Figure 4, 44 patients completed two interventions with simultaneous EEG assessments. Table 1 presents the demographic and clinical profiles of the participants. The mean age of the participants was 55.8 ± 7.91 years and the median time after stroke was 22.5 days. The median UE-FMA score at enrollment was 12.5, indicating that the majority of the participants had severe upper extremity impairment. Furthermore, 75% of the subjects suffered an ischemic stroke, and 63% had injuries in the basal ganglia, corona radiata, or centrum semiovale.

Figure 4.

A flowchart of the experiment design.

Table 1.

Demographical and clinical characteristics of participants (n = 44).

Measures	Value
Age (yr)	55.8 ± 7.91
Sex (female/male)	16/28
Stroke type (ischemic/hemorrhagic)	33/11
Lesion side (left/right)	21/23
Time post-stroke (d)	22.5 (17, 41.5)
UE-FMA (0–66 points)	12.5 (6, 46)
Lesion location
EC	1
BG/CR/CS	28
TH	3
BS	10
No MRI	2

Values are presented as mean ± standard deviation or median (lower quartile, upper quartile).

BG = basal ganglia, BS = brain stem, CR = corona radiata, CS = centrum semiovale, EC = external capsule, MRI = Magnetic Resonance Imaging, TH = thalamus, UE-FMA = Upper Extremity of Fugl-Meyer assessment.

3.2. Altered parietofrontal network connectivity by EA stimulation

In the alpha and beta bands, the results of the linear mixed-effects models showed no significant group × time interaction in the contralesional and interhemispheric parietofrontal networks (all P > .05), However, a significant group × time interaction in the ipsilesional parietofrontal network (alpha: F = 4.814, P = .031; beta: F = 6.066, P = .016). Specifically, during the intervention, the FC of the ipsilesional parietofrontal network in the EA group significantly decreased in both the alpha and beta bands compared to the SEA group (alpha: t = −1.998, P = .049; beta: t = −2.342, P = .022; Table 2). Moreover, for the time main effects, the FC in the ipsilesional parietofrontal network in the EA group revealed significant differences in both the alpha and beta bands (alpha: F = 5.05, P = .011; beta: F = 3.295, P = .047; Table 2), whereas no such differences were observed in the SEA group (P > .05).

Table 2.

Comparisons of ipsilesional parietofrontal network connectivity between the two groups (n = 44).

	Group	Before	During	After	F	P
Alpha	EA Group	0.53 (0.47, 0.57)	0.50 (0.44, 0.55)	0.53 (0.47, 0.59)	5.05	.011*
	SEA Group	0.51 (0.46, 0.60)	0.53 (0.47, 0.58)	0.53 (0.46, 0.58)	0.203	.817
	Mean difference (95% CI)	–	−0.04 (−0.07 to 0.00)	0.01 (−0.03 to 0.04)
	Z/t	−0.338	−1.998	0.353
	P	.735	.049*	.725
Beta	EA Group	0.48 (0.42, 0.51)	0.44 (0.39, 0.52)	0.48 (0.41, 0.54)	3.295	.047*
	SEA Group	0.44 (0.40, 0.55)	0.45 (0.43, 0.54)	0.48 (0.40, 0.53)	1.027	.367
	Mean difference (95% CI)	–	−0.04 (−0.08 to −0.01)	0.01 (−0.03 to 0.05)
	Z/t	−0.128	−2.342	0.681
	P	.898	.022*	.498

Data described as median (P25, P75).

CI = confidence interval.

^*P < .05.

3.3. Altered network properties by EA stimulation

3.3.1. Clustering coefficient.

In the alpha band, there were no significant differences in the main effects of time, group, and group × time interaction in both the ipsilesional and contralesional parietofrontal networks (all P > .05). In the beta band, a significant group × time interaction was observed in the ipsilesional parietofrontal network (F = 4.301, P = .041). However, there were no significant differences in the main effects of time and group in this band (P > .05).

3.3.2. Characteristic path length.

In both the alpha and beta bands, there were no significant differences in the main effects of time, group, and group × time interaction in the PL (P > .05).

3.3.3. Small-world Index.

In the alpha band, there were no significant differences in SW for the main effects of time, group, and group × time interaction (all P > .05). In the beta band, the SW index exhibited some outliers. With the outliers present, there was a significant difference in group × time (F = 8.646, P = .004) and a notable difference in SW between the two groups after EA stimulation (t = 2.741, P = .006, EA > SEA, Figure 5A). Although there was no significant difference in the main effect of time (P > .05), both groups exhibited significant differences in the time effects of the SW index (EA: F = 6.149, P = .004; SEA: F = 4.442, P = .018). After the removal of outliers, significant differences were also found in the main effect of time and the group × time interaction for the SW index (time: F = 5.516, P = .026; group × time: F = 6.892, P = .01). For between-group comparisons, the SW index was significantly higher in the EA group than in the SEA group at the post-stimulation stage (t = 2.379, P = .018, see Figure 5B). Specifically, during the post-intervention stage, the SW index in the EA group increased, whereas the SW index in the SEA group continuously decreased.

Figure 5.

Comparisons between the two groups of small-world indexes in the beta band with and without outlier throughout the intervention. *P < .05, compared with the SEA group.

4. Discussion

This study provides evidence regarding the immediate modulatory effects of EA stimulation on the parietofrontal network in stroke survivors. The findings demonstrated that EA could alter the FC of the ipsilesional parietofrontal network in patients with stroke. Notably, EA significantly reduced FC in both alpha and beta frequency bands during the stimulation phase. Additionally, in the poststimulation phase, EA significantly enhanced the SW properties of the entire brain network. These findings suggest that EA stimulation may play a role in modulating communication between parietofrontal regions, potentially facilitating motor recovery in stroke patients.

4.1. EA specifically modulates the ipsilesional parietofrontal network connectivity

The parietofrontal network was specifically targeted for EA stimulation, as it serves as a command center, involving in the higher-level cognitive control of motor behavior, such as motor planning, decision-making, and sensorimotor integration.^[32] FC measures the information communication between different neuronal populations involved in the regulation of motor behavior. Therefore, the EA-induced modulation of the parietofrontal network’s FC may reflect its influence on higher-level cognitive motor functions, which could contribute to understanding of EA’s mechanisms in motor control within clinical settings.

The brain’s ability to reorganize itself, known as neuroplasticity, comprises the neural foundation for stroke-induced motor function recovery. Changes in the parietofrontal network’s FC may be part of this neuroplasticity process, and its alterations correlated with motor recovery post-stroke. Previous studies have demonstrated that the early upregulation of the ipsilesional parietofrontal network during the subacute phase of stroke.^[15] Moreover, increased connectivity within this network in the acute phase has been linked to persistent deficits in the late subacute recovery phase.^[16] Additionally, Wu’s research^[33] indicated a negative correlation between high beta band ipsilesional M1-parietal connectivity in subacute stroke and upper limb motor recovery. However, the direct link between the ipsilesional parietofrontal FC and motor recovery post-stroke remains to be established. From the perspective of maladaptive compensation, enhanced connectivity of the parietofrontal network might hinder motor recovery, while decreasing it could facilitate motor recovery. The results of this study indicate that EA specifically downregulates the ipsilesional parietofrontal network connectivity in the alpha and beta frequency bands of stroke patients, and this effect is time-dependent. However, no significant changes were noted with SEA, underscoring EA’s capacity to modulate brain network connectivity beyond placebo effects. Supporting these findings, fMRI studies have also demonstrated acupuncture’s immediate impact on cortical networks in stroke patients.^[34]

Previous research demonstrated that EA could enhance the ipsilesional sensorimotor cortex FC.^[35] However, our study indicates that EA specifically reduces parietofrontal network FC in subacute stroke patients. This finding may stem from the different acupoints selection and methodological differences in brain function assessment tools. Evidence suggests that the FC of brain networks and their contribution to motor function recovery may differ among individuals, influenced by factors such as time post-stroke and corticospinal tract (CST) damage severity. In our cohort, participants had severe upper limb impairment, as indicated by a median UE-FMA score of 12.5 points. This may result in an exaggerated reliance on the neural resources around the injured primary motor cortex, leading to abnormally enhanced parietofrontal connectivity. In support, it has been reported that stroke patients with significant CST damage display higher FC in the ipsilesional parietofrontal network compared to those with intact CST.^[36] This present study suggests that EA’s downregulation of parietofrontal network FC may play a role in modulating the “abnormal” brain networks post-stroke. To elaborate the temporal dynamics of EA’s effects on parietofrontal network FC and its correlation with motor recovery, longitudinal EA intervention with concurrent EEG recording studies are required. If the causal relationship between the downregulation of parietofrontal network connectivity by EA and upper limb motor recovery is confirmed, this would provide compelling evidence for using EEG oscillations as biomarkers to guide therapeutic strategies.

Compared with fMRI, EEG provides unique insights into the frequency regulation of neuronal oscillations. Specifically, the alpha and beta bands, which serve as prominent background rhythms in the brain, play pivotal roles in mediating communication among widespread cortical networks, facilitating top-down information processing, and supporting sensorimotor functions.^[7,37] Previous research has suggested that alterations in alpha and beta oscillations could reflect the intricate mechanisms of cortical excitatory-inhibitory balance, potentially serving as indicators of motor recovery post-stroke.^[38] For instance, Cassidy utilized lasso regression to evaluate the extent of motor recovery during subacute stroke, identifying the high beta (20–30 Hz) and alpha bands (8–12 Hz) as significant contributors to the model, with both positive and negative associations with motor outcomes.^[7] Our study confirms the specific modulatory effects of EA stimulation within the alpha and beta bands of the brain networks. Similar instantaneous modulatory effects on alpha and beta frequencies have also been reported in subacute stroke patients.^[39] Given that EA incorporates diverse sensory stimuli such as needling and electrical currents, the observed specific reactivity of the alpha and beta bands may indicate enhanced sensorimotor cortical coupling facilitated by top-down control mechanisms and sensorimotor integration, crucial elements in the motor recovery process. This concept aligns with previous research suggesting that additional sensory stimuli may promote motor relearning and enhance cortical plasticity.^[40] Although it would be premature to conclusively assert that EA-induced oscillation modulation directly entails sensorimotor integration, our findings underscore the vital role of neural oscillations in regulating brain network functionality.

Indeed, the specific modulation of the parietofrontal network by EA is influenced by several factors. These factors include the selection of the verum acupoints, presence of deqi sensation, depth of needle insertion, and output intensity of the electrical current. Previous studies have demonstrated that stimulating the verum acupoints produces more stable brain responses than sham acupoint stimulation.^[41] The deqi sensation, widely acknowledged as a key factor in the therapeutic effects of acupuncture, has been linked to specific alterations in the coherence of the theta and alpha bands, with its clinical effectiveness intricately tied to the central nervous system response.^[42] In our study, the absence of deqi sensation in the SEA group could be a critical contributing factor to the significant differences observed between the two groups. Additionally, the manipulation of superficial needling in the SEA group may have influenced the brain’s response, potentially unveiling the blinding of certain subjects, subsequently impacting their internal psychological experiences and affecting the treatment’s efficacy and associated brain response. Another important consideration is the intensity of electrical current stimulation. The output of the EA current may distinctly influence the FC of the brain network when compared to the absence of a current in the SEA. An fMRI study suggested that EA may activate more brain areas related to motor activity than manual acupuncture.^[43] Collectively, these multifaceted factors contribute to the specific modulation of the brain networks induced by EA stimulation. Further investigation of these factors could provide deeper insights into the clinical applications and mechanisms of EA in stroke recovery.

4.2. EA specifically alters network properties

We investigated how acupuncture modulates brain networks by examining the SW properties of whole-brain networks in patients with stroke. SW networks are known for efficient information processing within the brain, providing a balance between global integration and the local segregation of information transmission. After a stroke, these properties tend to be disrupted, shifting the network characteristics to a more random configuration.^[44] However, our study found that, across the entire intervention phase, stroke patients exhibited SW attributes in their functional brain networks.

Interestingly, although there were no significant differences in the comparison of the PL and CC parameters between the two groups, the post-stimulation phase revealed a significantly higher SW in the EA group than in the SEA group. This finding provides evidence that EA can effectively regulate the efficiency of global information transmission within the brain networks. These findings diverge from those reported by Han et al,^[5] who found that acupuncture could modify the disrupted whole-brain network post-ischemic stroke by improving the CC and local efficiency while not influencing the SW. Collectively, our findings suggest that acupuncture’s impact on brain networks is not limited to specific attributes, but extends to comprehensive network rearrangements. Furthermore, studies have highlighted the role of acupuncture in promoting neuroplasticity, which refers to the brain’s ability to reorganize and adapt following an injury. Acupuncture can enhance neuroplasticity by promoting the growth and survival of neurons, stimulating the release of neurotrophic factors, and modulating synaptic plasticity.^[45] These neuroplastic changes may contribute to the improvement of SW properties in the brain networks of patients with stroke. Overall, our study provides additional evidence supporting the hypothesis that EA can enhance SW properties in the brain networks of patients with stroke. Further research is needed to elucidate the optimal parameters and protocols for EA treatment that optimize the modulation of SW attributes in brain networks, ultimately improving motor outcomes in stroke patients.

4.3. The clinical application

The current study provides valuable insights into the neurophysiological mechanisms of EA as a motor recovery intervention following stroke. It provides evidence of EA’s modulation of local and global brain networks, shedding light on its potential effects on stroke rehabilitation. These findings carry significant implications for inspiring innovative rehabilitation strategies for stroke patients. Additionally, as a form of peripheral intervention through sensory stimulation, EA demonstrates the potential to modulate the reorganization of neural connections within the brain. This also underscores the importance of incorporating sensory modalities into comprehensive stroke rehabilitation programs. Moreover, the study emphasizes the role of EEG data in guiding EA intervention by providing real-time insights into brain activity and network connectivity. These insights can inform the optimal parameters and timing of EA stimulation, including frequency, intensity, and duration, to modulate specific brain networks associated with motor recovery after stroke. The ability to monitor changes in brain activity and connectivity in response to EA stimulation enables the tailoring of intervention program to accommodate individual differences in neural responses, potentially enhancing the effectiveness of EA in stroke rehabilitation.

5. Limitations

Despite the interesting results obtained in the current study, it is important to acknowledge some limitations that need to be acknowledged. First, we did not include healthy subjects, which may limit the confirmation of the specificity of the brain modulatory effect of EA in patients, as the brain modulation effects of acupuncture can differ between healthy individuals and patients. Another limitation is that we focused on the immediate effects of a single session of EA rather than investigating the cumulative and long-term effects of EA over an extended period. This limitation prevented us from establishing a causal relationship between EA-induced modulation in the parietofrontal network and long-term motor benefits. Future studies should include long-term EA interventions and consider EEG recordings at different time points to elucidate this relationship. Furthermore, with a limited number of channels in EEG recordings, the spatial coverage and resolution of the brain activity are restricted. Further studies should consider using more advanced neuroimaging techniques, such as high-density EEG or functional MRI with higher channel coverage, to provide a more complete understanding of brain network interactions. Finally, the fixed threshold we employed for comparing brain network attributes between groups may oversimplify the intricate nature of brain networks. Future studies should consider implementing individualized thresholds and explore more sophisticated network analysis methods to enhance the accuracy and validity of network analysis.

6. Conclusions

In conclusion, our findings demonstrate that EA can downregulate ipsilesional parietofrontal network connectivity, particularly in the alpha and beta bands. Furthermore, it can enhance the SW properties of functional brain networks in the post-stimulation phase. These findings provide neurophysiological evidence for the therapeutic effects of EA in promoting upper-limb motor recovery. By shedding light on the modulation of specific network connectivity and global network properties, our study highlights the potential of EA as a valuable intervention for improving motor outcomes in patients with stroke. Further research is warranted to explore the long-term effects and to optimize the parameters of EA treatment to maximize its therapeutic benefits.

Acknowledgments

We thank all the medical experts, including statisticians, for their support and assistance with the project. We also thank all participants for their interest and participation in the study.

Author contributions

Conceptualization: Mingfen Li, Fei Zou.

Data curation: Mingfen Li, Fei Zou, Tingting Zheng.

Formal analysis: Mingfen Li.

Funding acquisition: Mingfen Li, Haifeng Li, Su Zheng.

Investigation: Mingfen Li, Fei Zou, Tingting Zheng, Weigeng Zou.

Methodology: Su Zheng.

Project administration: Su Zheng.

Resources: Haifeng Li.

Software: Yifang Lin.

Supervision: Li Peng.

Writing – original draft: Mingfen Li.

Writing – review & editing: Su Zheng.