Effects of acupuncture on brain function in patients with depression

Abstract

Background:

Depression poses a major global health burden. Many patients exhibit treatment resistance or experience intolerable side effects with standard therapies. Acupuncture shows promise as an adjunctive therapy for depression, but its neurobiological mechanisms remain incompletely understood.

Methods:

We conducted a coordinate-based meta-analysis to identify consistent patterns of acupuncture-induced neuroplastic changes in patients with major depressive disorder, quantified via resting-state functional magnetic resonance imaging measures of regional homogeneity and amplitude of low-frequency fluctuations (ALFF/fractional ALFF). We synthesized data from 7 randomized controlled trials involving 357 patients with major depressive disorder, comparing acupuncture-based interventions (manual acupuncture, electroacupuncture, or transcutaneous auricular vagus nerve stimulation) against sham acupuncture, conventional treatment, or treatment-as-usual. The seed-based d mapping with permutation of subject images algorithm was used for the imaging meta-analysis, and a leave-one-out sensitivity analysis was performed to assess the robustness of the pooled findings to intervention heterogeneity.

Results:

Acupuncture induced significant regional homogeneity /ALFF increases within the right cingulum and right caudate nucleus. In contrast, conventional treatments primarily increased activity in the left hippocampus and right middle occipital gyrus. Meta-regression linked clinical improvement specifically to plasticity changes in the left cerebellum and revealed a dose-dependent relationship between the number of acupuncture sessions and modulation of the left amygdala. Clinically, acupuncture produced significantly greater reductions in depression and anxiety scores than control conditions.

Conclusion:

Our findings indicate that acupuncture-based interventions elicit a distinct pattern of neuroplastic remodeling, targeting key regions involved in emotion regulation and reward processing, which contrasts with the hippocampal–occipital effects of conventional treatments. The correlation between neuroplastic changes in affective circuits and clinical improvement supports a potential mechanism for acupuncture’s efficacy in alleviating core depressive symptoms such as anhedonia and emotional dysregulation. Limitations include the small number of included trials, heterogeneity in acupuncture protocols and resting-state functional magnetic resonance imaging metrics, absence of long-term follow-up data, and methodological caveats inherent to coordinate-based meta-analysis. Future research requires larger, standardized trials with follow-up assessments and integration of neuroimaging with molecular biomarkers to advance personalized neuromodulation strategies for depression.

1. Introduction

Depression, a highly prevalent and debilitating mental disorder, is defined by persistent low mood, anhedonia (loss of pleasure), cognitive impairments, and somatic symptoms.^[1] Affecting roughly 340 million people worldwide, depression represents the largest contributor to the nonfatal health burden, responsible for approximately 12% of global years lived with disability, as estimated by the World Health Organization.^[2] Despite the availability of various treatments, a substantial proportion of patients with major depressive disorder experience inadequate symptom relief or intolerable side effects, leading to treatment-resistant depression (TRD).^[3] For TRD patients, augmentation strategies have shown promise, including atypical antipsychotics,^[4] ketamine,^[3] and lithium.^[5] Psychological interventions, particularly cognitive behavioral therapy, have demonstrated effectiveness when added to usual care.^[6] Emerging treatments such as neurostimulation techniques and immune-inflammatory based therapies show potential.^[5] However, the evidence base for many augmentation strategies remains limited, with few long-term studies available.^[7] Tolerability is a significant concern, with side effects such as sexual dysfunction and weight gain often leading to treatment discontinuation.^[8] This unmet clinical need underscores the importance of exploring effective and well-tolerated alternative therapies for depression management, aligning with global mental health initiatives.

Acupuncture, a cornerstone of Traditional Chinese Medicine, has gained international recognition as a complementary and alternative therapy for various conditions, including depression.^[9] The World Health Organization acknowledges acupuncture’s potential utility in managing depressive symptoms.^[10] Growing evidence from clinical studies and systematic reviews suggests that acupuncture’s efficacy in improving clinical response and symptom severity for major depressive disorder (MDD) and poststroke depression (PSD) may be comparable to antidepressants.^[11] Acupuncture, particularly electroacupuncture combined with antidepressants, has shown potential benefits in managing depressive symptoms.^[12] Although clinical efficacy is well established, the neurobiological mechanisms underlying acupuncture’s antidepressant effects are not fully elucidated.^[13] A key hypothesis proposes that acupuncture influences mood regulation by modulating neural plasticity in relevant brain circuits.^[13,14]

As the brain’s capacity for structural and functional reorganization, neuroplasticity is increasingly recognized as a substrate of antidepressant response, encompassing changes in synaptic strength, regional spontaneous activity, and large-scale network connectivity.^[15–19] Resting-state functional magnetic resonance imaging (rs-fMRI) provides a sensitive, task-free probe of such intrinsic plasticity by mapping spontaneous low-frequency blood-oxygen-level-dependent fluctuations and their spatial organization.^[20,21] Two complementary rs-fMRI metrics are widely used: regional homogeneity (ReHo) quantifies the local temporal synchronization of blood-oxygen-level-dependent signals among neighboring voxels, whereas the amplitude of low-frequency fluctuations (ALFF) and its fractional variant (fALFF) capture the magnitude of spontaneous low-frequency oscillations.^[22–25] While both indices have proven informative across psychiatric and neurological conditions, they reflect distinct neurophysiological dimensions and the implications of pooling them across studies are revisited in our Methods and Limitations. rs-fMRI investigations of acupuncture for depression have proliferated, but their results remain heterogeneous due to variations in experimental design, sample size, patient characteristics, and acupuncture protocols.^[26,27] Some studies report acupuncture-induced modulation of limbic structures such as the amygdala and alterations in the default mode network and prefrontal cortex,^[28–30] while others highlight regulation of the corticostriatal reward circuitry.^[31]

To integrate these disparate findings and identify consistent patterns of brain activity changes associated with acupuncture for depression, this review employs coordinate-based meta-analysis (CBMA). CBMA is a widely used method for synthesizing neuroimaging data across studies by analyzing reported peak activation coordinates.^[32,33] Recent advancements in CBMA techniques include the incorporation of effect sizes,^[34] Gaussian-process regression,^[35] and Bayesian latent factor modeling.^[36] Seed-based d mapping with permutation of subject images (SDM-PSI) has emerged as a robust CBMA algorithm that integrates these statistical innovations.^[37] We therefore use SDM-PSI to analyze rs-fMRI data from eligible studies. The primary objective is to determine whether acupuncture elicits distinct patterns of brain activity change in MDD, relative to control conditions such as sham acupuncture or conventional treatment alone, and to relate these neural changes to clinical symptom improvement and to acupuncture “dose” (number of sessions). By clarifying how neuroplasticity may mediate acupuncture’s antidepressant effects, we aim to inform the refinement of clinical protocols and to support future applications across neuropsychiatric conditions.

2. Methods

This meta-analysis was conducted in accordance with the Cochrane Handbook for Systematic Reviews of Interventions and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. The protocol was prospectively registered with the International Prospective Register of Systematic Reviews (Registration No.: CRD420251103743). This study is a systematic review and meta-analysis of previously published functional magnetic resonance imaging (fMRI) studies investigating the effects of acupuncture on brain function in patients with depression. All included studies were conducted and published by independent research groups, and the original data were fully anonymized and de-identified prior to inclusion in the respective publications. As this investigation involves the secondary analysis of publicly available, aggregated data from prior studies and does not involve direct human participants, human tissues, or identifiable personal information, formal ethical approval and informed consent were not required for the present meta-analysis in accordance with the guidelines of the Declaration of Helsinki and standard institutional policies for secondary research using published data.

2.1. Literature search

We conducted a comprehensive systematic literature search to identify studies examining the effects of acupuncture on fMRI-measured brain function in patients with depression. The following electronic databases were searched from their inception until June 2025: PubMed, Web of Science, EMBASE, Cochrane Central Register of Controlled Trials (CENTRAL), China National Knowledge Infrastructure, Chongqing VIP, and Wanfang Database. The search strategy combined relevant Medical Subject Headings (MeSH) terms and free-text words for 3 core concepts: depression, acupuncture, and fMRI, together with filters for randomized controlled trials (RCTs). The core English search terms were: (Depress OR MDD OR Major Depressive Disorder OR Dysthymi OR Affective Disorder OR Mood Disorder) AND (Acupuncture OR Acupuncture Therapy OR Electroacupuncture OR Electro-acupuncture OR Auriculotherapy OR Scalp Acupuncture OR transcutaneous auricular vagus nerve stimulation OR taVNS OR auricular VNS OR noninvasive vagus nerve stimulation) AND (fMRI OR functional MRI OR functional magnetic resonance imaging OR neuroimaging OR resting-state fMRI OR rs-fMRI OR ReHo OR ALFF OR fractional ALFF) AND (randomized controlled trial OR RCT OR random OR trial). Searches in Chinese databases used equivalent Chinese terms. Language was restricted to English and Chinese publications. The syntax was adapted for each database. Two independent reviewers conducted the literature search.

2.2. Study selection criteria

2.2.1. Inclusion criteria

First, population: adult patients (≥18 years) with a diagnosis of MDD according to internationally recognized criteria (Diagnostic and Statistical Manual of Mental Disorders-IV/Diagnostic and Statistical Manual of Mental Disorders-V, International Classification of Diseases-10/International Classification of Diseases-11, or CCMD-3), regardless of severity, episode duration, or sex. Second, intervention: acupuncture-based interventions, defined as manual acupuncture, electroacupuncture, scalp/auricular acupuncture, or transcutaneous auricular vagus nerve stimulation (taVNS), administered either as monotherapy or as an adjunct to baseline management (e.g., ongoing pharmacotherapy or psychotherapy). Third, comparison: control groups receiving sham acupuncture (non-penetrating needles or minimal acupuncture at non-acupoints), conventional treatment alone (antidepressants and/or psychotherapy), waitlist control, or treatment-as-usual. Fourth, neuroimaging: whole-brain rs-fMRI analyses reporting ReHo, ALFF, or fALFF differences (posttreatment vs baseline, or acupuncture vs control), or correlations with clinical improvement. Fifth, reporting: peak activation coordinates (x, y, and z) in either Talairach or Montreal Neurological Institute standard stereotactic space were reported for significant clusters. Sixth, study design: RCTs.

2.2.2. Exclusion criteria

First, reviews, case reports, animal studies, non-randomized trials, conference abstracts without full data, or studies lacking fMRI data. Second, studies reporting only region-of-interest analyses without whole-brain exploratory results. Third, task-based fMRI studies (i.e., not resting state). Fourth, studies with fewer than 5 participants per group. Fifth, duplicate publications or studies using overlapping datasets (only the study with the largest sample or most comprehensive data was retained). Sixth, studies enrolling depressive subtypes other than MDD (e.g., persistent depressive disorder and bipolar depression) without separable MDD-only data.

Note on alignment with R3-M2. Inclusion (1) is now confined to MDD diagnosed by formal criteria; Exclusion (6) likewise excludes non-MDD subtypes. The previous wording “MDD or depressive disorder” has been removed to eliminate the internal contradiction noted by Reviewer 3.

2.3. Data extraction and quality assessment

Two independent reviewers extracted data from the included studies using a standardized form. Extracted information included: publication details, study characteristics, diagnostic criteria, depression severity, baseline Hamilton Rating Scale for Depression (HAMD), Self-Rating Depression Scale (SDS), and Hamilton Anxiety Scale (HAMA) scores, age, sex, illness duration, medication status, intervention details, control intervention details (type of sham and specific conventional treatment), fMRI methodology, fMRI results, and clinical outcome measures used alongside fMRI. Disagreements were resolved through discussion or by consulting a 3rd reviewer. Missing data were sought by contacting corresponding authors via email. Extracted information is presented in Table 1.

Table 1.

Demographics and clinical data in patients with depression in the included studies.

	Participant						Intervention comparison					Outcome measures
Author	Cases (T/C)	Depression degree	Diagnostic criteria	Duration (mo)	Age (yr)	Gender (male/female)	T	C	Acupoints	Regiment	Analysis of fMRI		Coordinate	Secondary outcomes
Wang Yi (2022)[38]	160 (80/80)	Moderate	DSM-V	Not reported	T: 47.9 ± 6.5 C: 46.3 ± 7.2	T: 22/58 C: 28/52	Acu + Conv	Conv	Empirical acupoints	5/wk for 3 wk	fALFF		MNI	HAMD-17, SDS
Zhao Bin (2022)[39]	24 (12/12)	Not reported	ICD-11	T: 39.58 ± 16.19 C: 35.75 ± 15.33	T: 56.58 ± 9.71 C: 57.17 ± 9.03	T: 8/4 C: 7/5	Acu + Conv	Conv	Empirical acupoints	6/wk for 4 wk	ALFF ReHo		MNI	HAMD-17
Wang Xiaoling (2024)[40]	28 (14/14)	Moderate to severe	DSM-V	Not reported	T: 36.28 ± 6.88 C: 39.64 ± 7.06	T: 4/10 C: 5/9	Acu + Conv	Sham-acu + Conv	Empirical acupoints	3/wk for 6 wk	ALFF		MNI	HAMD-24, HAMA-14
Sun Jifei (2023)[41]	71 (37/34)	Mild to severe	DSM-V	T: 14.54 ± 14.63 C: 12.14 ± 15.63	T: 34.81 ± 11.63 C: 33.88 ± 10.88	T: 10/27 C: 15/19	Acu + Conv	Conv	taVNS	7/wk for 8 wk	ALFF ReHo		MNI	HAMD-17, HAMA-14, SDS, SAS
Li Yi (2014)[42]	16 (8/8)	Mild to moderate	CCMD-3	T: 17.62 ± 16.83 C: 15.87 ± 10.48	T: 40.12 ± 13.04 C: 34.12 ± 11.72	T: 1/7 C: 2/6	Acu + Conv	Sham-acu + Conv	Empirical acupoints	3/wk for 12 wk	ALFF		MNI	HAMD-24 HAMA-14, SDS, SAS
Wang Bowei (2011)[43]	30 (15/15)	Mild to severe	CCMD-3-R	T: ≥2 wk	T: 42.00 ± 12.01 C: 41.60 ± 11.43	T: 0/15 C: 0/15	Acu	None	Empirical acupoints	10/4 wk	ReHo		Tal	MADRS, SDS
Yi Yang (2011)[44]	28 (14/14)	Moderate	DSM-IV	T: 7.6 ± 1.1 C: 7.8 ± 1.4	T: 37.0 ± 8.6 C: 33.6 ± 8.4	T: 6/8 C: 7/7	Acu + Conv	Conv	Empirical acupoint	5/wk for 4 wk	fALFF		MNI	HAMD-17

Acu = acupuncture treatment, ALFF = amplitude of low frequency fluctuation, C = control group, CCMD = Chinese Classification of Mental Disorders, Conv = conventional treatment, DSM = Diagnostic and Statistical Manual of Mental Disorders, HAMA-14 = 14-item Hamilton Anxiety Scale, HAMD-17 = 17-item Hamilton Rating Scale for Depression, HAMD-24 = 24-item Hamilton Rating Scale for Depression, ICD: International Classification of Diseases, MADRS = Montgomery–Åsberg depression rating scale, MNI = Montreal Neurological Institute coordinate, ReHo = regional homogeneity, SAS = self-rating anxiety scale, SDS = self-rating depression scale, T = treatment group, Tal = Talairach coordinate, taVNS: transcutaneous auricular vagus nerve stimulation.

Methodological quality and risk of bias for each included RCT were assessed independently by 2 reviewers using the Cochrane Collaboration’s Risk of Bias 2.0 tool (RoB 2; The Cochrane Collaboration). For each trial we evaluated the 5 RoB 2 domains using the recommended signaling questions and operationalized our judgments as follows.

• Domain 1: Bias arising from the randomization process. Studies that explicitly described both an adequate random sequence generation method (e.g., computer-generated random numbers and random number tables) and allocation concealment (e.g., sealed opaque envelopes and central allocation) were rated low risk. Studies describing only random sequence generation without allocation concealment were rated as having some concerns; studies with no clear randomization or evidence of selection bias were rated high risk.

• Domain 2: Bias due to deviations from intended interventions. Because true double-blinding is difficult in acupuncture trials, we focused on the effect of assignment. Trials that successfully blinded both participants and acupuncture providers (typically via standardized sham devices) were rated low risk; trials that blinded only outcome assessors but not participants were rated some concerns; trials with explicit unblinded designs (e.g., open-label acupuncture vs waitlist) were rated high risk.

• Domain 3: Bias due to missing outcome data. Trials with complete or near-complete outcome data (<10% loss with balanced reasons across arms) were rated low risk. Trials with substantial or differential attrition without imputation were rated high risk. All other configurations were rated some concerns.

• Domain 4: Bias in measurement of the outcome. Particular attention was paid to blinding of the fMRI data analysts and of the clinical raters administering HAMD/HAMA/SDS. Trials explicitly reporting blinded fMRI processing pipelines and blinded clinical rating were rated low risk.

• Domain 5: Bias in selection of the reported result. Trials with a prepublished or registered protocol matching the reported analyses were rated low risk; trials reporting fully on all prespecified outcomes but without a registered protocol were rated some concerns; clear evidence of selective reporting or post hoc outcome switching led to high risk.

An overall risk-of-bias judgment was derived from the 5 domain judgments following the RoB 2 algorithm: low risk if all domains were low risk; some concerns if at least 1 domain raised some concerns without any high-risk domain; and high risk if any domain was high risk or multiple domains raised some concerns. The detailed per-study judgments are summarized in Figure 2 and Section 3.2.

Figure 2.

Summary of risk of bias for the 7 included RCTs, assessed using the Cochrane Risk of Bias 2.0 (RoB 2) tool. Color coding: green = low risk; yellow = some concerns; red = high risk. Domains: D1 = randomization process; D2 = deviations from intended intervention; D3 = missing outcome data; D4 = measurement of the outcome; D5 = selection of the reported result. Operational rules for each domain are detailed in Methods 2.3. RCTs = randomized controlled trials.

2.4. Data synthesis and meta-analysis

2.4.1. Rationale for pooling different acupuncture modalities

The included trials used manual acupuncture, electroacupuncture, or taVNS. We prespecified that these modalities could be pooled within a single primary analysis because converging neuroanatomical evidence indicates that somatic and auricular afferent stimulation share a common central pathway: both engage A-delta and C-fiber afferents that project – directly or indirectly – to the nucleus of the solitary tract, and from there to the locus coeruleus, periaqueductal gray, and limbic structures (amygdala, hippocampus, and anterior cingulate cortex).^[45,46] Functional neuroimaging studies of both manual acupuncture and taVNS have repeatedly identified overlapping modulation of these targets in depression.^{[28–31,45,46]} We therefore conceptualized the pooled analysis as testing whether any acupuncture-based stimulation, irrespective of the specific peripheral route, elicits a convergent central signature in MDD. To test the robustness of this assumption, a leave-one-out sensitivity analysis was prespecified in which the single taVNS trial (Sun Jifei^[41]) was removed and the meta-analysis rerun on the remaining 6 manual/electroacupuncture trials. Convergence on the same primary clusters (right cingulum and right caudate nucleus) was set a priori as evidence that the pooled findings were not driven by taVNS heterogeneity.

2.4.2. Imaging meta-analysis

The primary meta-analysis of fMRI coordinates was performed using SDM-PSI (version 6.21; SDM Project, https://www.sdmproject.com/; RRID:SCR_002554), following the recommended steps: First, data preparation. Peak coordinates (x, y, and z) and their corresponding statistical effect sizes (t values were preferred; z-scores or P values were converted as needed) for significant clusters showing ReHo, ALFF, or fALFF differences between acupuncture and control groups (posttreatment) were extracted from each study. Separate text files were created for altered brain regions pre versus posttreatment in the acupuncture group and the control group. Second, recreation of effect size maps. Using the peak coordinates and effect sizes, SDM-PSI generated standardized statistical maps for each study, representing the spatial distribution of activation differences. Third, random-effects meta-analysis. A random-effects model was used to calculate the weighted mean effect-size map across studies, accounting for both within-study variance (effect-size based) and between-study heterogeneity. Study weighting incorporated sample size and variance. Fourth, statistical thresholding. Results were thresholded using family-wise error rate correction for multiple comparisons (P < .05) via the threshold-free cluster enhancement approach, with a cluster-extent threshold of ≥10 contiguous voxels. Results were visualized on a standard Montreal Neurological Institute brain template. Fifth, heterogeneity assessment. Between-study heterogeneity at significant clusters was quantified using the I² statistic derived from peak coordinates; I² < 50% was interpreted as low heterogeneity.

2.4.3. Caveats regarding metric heterogeneity

ReHo and ALFF/fALFF index neurobiologically distinct properties of intrinsic brain activity – local temporal synchronization (ReHo) versus the amplitude of spontaneous low-frequency oscillations (ALFF/fALFF). Although SDM-PSI permits coordinate-level pooling across rs-fMRI metrics, the resulting summary maps should therefore be interpreted as indicating regions where any of these convergent indices is consistently altered, rather than as evidence for a single, metric-specific neurophysiological change. This methodological compromise, necessitated by the small number of eligible trials, is revisited as a primary limitation in Section 4.5.

2.4.4. Publication-bias assessment

Because funnel plots are unreliable when fewer than 10 studies are pooled, we used Egger regression test as implemented in SDM-PSI to evaluate small-study effects. We note that asymmetry tests are themselves underpowered when k < 10, and a nonsignificant Egger result does not exclude publication bias; we follow current methodological recommendations in interpreting these results with caution.^[47]

2.4.5. Meta-regression

Meta-regression analyses within SDM-PSI were performed to explore potential associations between patterns of brain activity change (ReHo/ALFF) and moderating factors including baseline depression or anxiety severity (mean HAMD/HAMA/SDS), mean age, sex ratio, and number of acupuncture sessions.

2.4.6. Clinical-outcome meta-analysis

Meta-analysis of continuous clinical outcome data (mean change in HAMD/HAMA/SDS scores from baseline to posttreatment) was performed in R (version 4.3.1; R Foundation for Statistical Computing, RRID:SCR_001905) using the meta and metafor packages. The weighted mean difference (MD) with 95% confidence intervals (CIs) was used as the effect measure. Heterogeneity was assessed using the I² statistic. A fixed-effects model was applied when heterogeneity was low (I² ≤ 50%, Cochran Q P > .1); otherwise, a random-effects model was used. Forest plots were generated for each outcome. Sensitivity and subgroup analyses were planned to investigate sources of heterogeneity whenever I² > 50%.

3. Results

3.1. Included studies

The initial database search identified 737 records. After removal of 272 duplicates, 465 records were screened by title and abstract, of which 409 were excluded, leaving 56 articles for full-text assessment. Full-text review excluded a further 49 articles, ultimately yielding 7 included studies^{[41,38–40,42–44]} (Fig. 1; reference 44 cites the SDM-PSI methodological paper rather than an included trial).

Figure 1.

PRISMA flow diagram for the study-selection process of the meta-analysis. Of 737 records identified, 272 duplicates were removed, 465 records were screened by title and abstract (409 excluded), 56 articles were assessed in full text (49 excluded), and 7 RCTs were ultimately included. ALFF = amplitude of low-frequency fluctuations, CNKI = China National Knowledge Infrastructure, fMRI = functional magnetic resonance imaging, PRISMA = Preferred Reporting Items for Systematic Reviews and Meta-Analyses, RCTs = randomized controlled trials, ReHo = regional homogeneity, VIP = Chongqing VIP Database, WF = Wanfang Database.

All included trials were published after 2010. Study sample sizes ranged from 16 to 160 patients, totaling 180 participants in the treatment groups and 177 in the control groups. All studies enrolled patients with MDD. Baseline demographic characteristics (age and sex), illness status (symptom duration), and secondary outcome measures showed no significant differences between treatment and control groups. Almost all patients in both groups received conventional treatment. Treatment groups underwent acupuncture for 3 to 12 weeks. Two studies used sham acupuncture in the control group. Among acupuncture-based interventions, 6 studies used manual or electroacupuncture at empirical body acupoints, and 1 study used taVNS (Sun Jifei^[41]). All studies used rs-fMRI (ReHo or ALFF/fALFF) to assess pre to posttreatment changes within each group. Clinical efficacy was assessed using HAMD (n = 6 studies), HAMA (n = 3), SDS (n = 4), Self-Rating Anxiety Scale (n = 2), and Montgomery–Åsberg Depression Rating Scale (MADRS; n = 1). Detailed characteristics of the included studies are presented in Table 1.

3.2. Quality assessment

Using the RoB 2 tool described in Section 2.3, we assessed the quality of the 7 included RCTs. All studies reported an adequate method of random sequence generation, but only 5 clearly described allocation concealment; 2 studies^[44,38] did not describe concealment and were rated with some concerns on domain 1. Patient blinding was reported in 2 studies^[41,42] and was explicitly absent in 1 study,^[43] which was therefore rated high risk on domain 2; the remaining trials were rated with some concerns given the difficulty of blinding active acupuncture. Outcome-assessor blinding was reported in the same 2 trials^[41,42]; other trials were rated with some concerns on domain 4. Missing outcome data were minimal across all studies, supporting low-risk judgments on domain 3. In the absence of preregistered protocols, domain 5 was rated low risk only if all enrolled participants and prespecified outcomes were reported; 1 study^[41] failed this criterion and was rated with some concerns. No additional sources of bias were detected. Overall, the methodological quality of the body of evidence was judged to be moderate, with the dominant sources of bias being incomplete allocation concealment and the limited blinding of participants and outcome assessors. Per-domain judgments are summarized in Figure 2.

3.3. Meta-analysis of ReHo/ALFF changes

Acupuncture therapy in the treatment group produced increased activation in the right cingulum (P < .05, z = 5.766) and right caudate nucleus (P < .05, z = 4.649), with no regions of decreased activation. Conventional treatment in the control group produced increased activation in the left hippocampus (P < .05, z = 5.749) and the right middle occipital gyrus (P < .05, z = 5.770), again with no regions of decreased activation. Table 2 reports peak coordinates and cluster breakdowns; Figure 3 visualizes the contrasting activation patterns of the 2 treatments.

Table 2.

Alterations in brain activity in patients after treatment compared to baseline.

	MNI coordinates			SDM z-score*	P value†	Voxels‡	Cluster breakdown	I2
Treatment group	x	Y	z
R cingulum	8	40	8	5.766	<.05	82	R anterior cingulate/paracingulate gyri, BA 32	1.87%
R caudate nucleus	10	16	12	4.649	<.05	57	R caudate nucleus, BA 25	4.71%
Control group
L hippocampus	−14	0	−14	5.749	<.05	128	L amygdala, striatum, hippocampus, BA 34	1.32%
R middle occipital gyrus	30	−80	18	5.770	<.05	59	R middle occipital gyrus, superior occipital gyrus, corpus callosum, BA 19	0.01%

BA = Brodmann area, I² = heterogeneity I², MNI = Montreal Neurological Institute, R = right, SDM = signed differential mapping.

^*Peak height threshold: z > 1.

^†Voxel probability threshold: P < .005 uncorrected and remained after correcting threshold (TFCE) of P < .05.

^‡Cluster extent threshold: number ≥ 10 voxels.

Figure 3.

SDM-PSI meta-analytic maps overlaid on axial sections of the MNI standard brain template, showing significant ALFF/ReHo changes in patients with major depressive disorder. (A) Acupuncture-induced increases in the right anterior cingulate/paracingulate gyri (peak MNI x = 8, y = 40, z = 8; Brodmann area 32) and the right caudate nucleus (peak MNI x = 10, y = 16, z = 12; Brodmann area 25). (B) Conventional-treatment-induced increases in a left hippocampus/amygdala/striatum cluster (peak MNI x = −14, y = 0, z = −14; Brodmann area 34) and a right middle/superior occipital gyrus/corpus callosum cluster (peak MNI x = 30, y = −80, z = 18; Brodmann area 19). Red = regions of hyperactivation; green = regions of hypoactivation. Region labels and pointers identifying each significant cluster have been added to the original maps to improve interpretability. ALFF = amplitude of low-frequency fluctuations, MNI = Montreal Neurological Institute, ReHo = regional homogeneity, SDM-PSI = seed-based d mapping with permutation of subject images.

Within the treatment group, the right anterior cingulate/paracingulate gyri and the right caudate nucleus showed minimal between-study heterogeneity in peak coordinate effect sizes (I² = 1.87%–4.71%). Similarly low heterogeneity was observed for all significant peaks in the control group (I² = 0.01%–1.32%). These metrics are detailed in Table 2. Egger regression test indicated no significant asymmetry in either the treatment group (P = .133) or the control group (P = .169); however, given that only 7 trials were available, this null result has limited statistical power to detect small-study effects and should not be interpreted as ruling out publication bias.^[48] Detailed per-study activation changes before and after treatment in both groups are shown in Table 3.

Table 3.

The peak coordinates of differential brain activation before and after treatment in the acupuncture and control groups.

Author	Group	Analysis of fMRI	Region	BA	MNI coordinates			Voxels	t value
					x	y	z
Sun Jifei (2023)[41]	Acupuncture	ALFF	Left middle frontal gyrus	7	−27	48	30	23	4.1088
			Right middle frontal gyrus	8	30	39	42	18	4.4661
			Right inferior temporal gyrus	90	57	−51	−9	26	−4.3932
			Posterior lobe of cerebellum	112	0	−63	−18	28	3.9996
		ReHo	Right orbital part of the inferior frontal gyrus	16	20	18	−25	24	−3.5270
			Left middle cingulate gyrus	33	−12	−27	36	18	4.3374
			Right precuneus	68	6	−57	48	16	−3.5949
	Control	ALFF	Right middle frontal gyrus	7	−27	48	30	23	4.1088
			Right inferior temporal gyrus	90	57	−51	−9	26	−4.3932
			Posterior lobe of cerebellum	112	0	−63	−18	28	3.9996
		ReHo	Right orbital part of the inferior frontal gyrus	16	20	18	−25	24	−3.5270
			Left middle cingulate gyrus	33	−12	−27	36	18	4.3374
			Right precuneus	68	6	−57	48	16	−3.5949
Wang Bowei (2011)[43]	Acupuncture	ReHo	Left posterior lobe of cerebellum	–	−24	−84	−42	101	3.5044
			Right posterior lobe of cerebellum	–	27	−78	−30	205	5.5961
			Left middle frontal gyrus	–	−33	30	54	45	3.9892
			Left precentral gyrus	–	−60	3	12	25	4.6675
			Left superior frontal gyrus	–	−18	48	24	26	3.9892
			Left lentiform nucleus	–	−15	21	−9	152	4.5265
			Left inferior frontal gyrus	–	−18	21	24	28	−2.8988
			Left inferior parietal lobule	–	−39	−42	57	27	−3.3587
			Right superior frontal gyrus	–	27	48	18	53	−4.013
			Left frontal lobe	–	−6	27	−27	61	4.3811
			Left anterior cingulate gyrus	–	−12	24	27	54	4.1605
			Left middle temporal gyrus	–	−51	−42	−15	43	3.3682
			Right middle temporal gyrus	–	57	6	−18	23	3.1793
			Left postcentral gyrus	–	−51	−12	21	25	−3.1432
			Left thalamus	–	−12	−33	9	23	−3.0419
			Right limbic lobe	–	15	−42	−9	41	−3.482
			Right precentral gyrus	–	27	−18	63	71	−3.5397
			Right superior temporal gyrus	–	48	−39	12	26	−3.4948
Wang Xiaoling (2024)[40]	Acupuncture	ALFF	Left middle temporal gyrus	–	−51	−12	−15	21	5.1410
			Left posterior cerebellar lobe	–	−81	−30	24	24	5.0122
	Control	ALFF	Left middle temporal gyrus	–	−51	−12	−15	21	5.1410
			Left posterior cerebellar lobe	–	−81	−30	24	24	5.0122
Wang Yi (2022)[49]	Acupuncture	ALFF	Cingulate back	–	14	18	−11	21	2.23
			Left precuneus	–	5	13	6	19	1.748
			Middle occipital gyrus	–	12	38	8	24	2.548
			Left suboccipital back	–	7	14	11	32	3.251
			Lower forehead of right frame	–	18	−16	9	37	3.926
			Right insula	–	−17	−13	14	15	1.554
			Right hippocampus	–	−13	12	−11	9	0.983
	Control	ALFF	Cingulate back	–	−3	−11	13	12	−0.431
			Left precuneus	–	−4	−8	−4	14	1.102
			Middle occipital gyrus	–	−9	6	5	8	1.339
			Left suboccipital back	–	−14	−21	−9	17	1.482
			Lower forehead of right frame	–	14	12	−2	19	1.773
			Right insula	–	15	18	−5	16	1.295
			Right hippocampus	–	11	−1	13	14	1.374
Li Yi (2014)[42]	Acupuncture	ALFF	Orbitofrontal cortex	11	−9	49	−19	31	3.23
			Anterior cingulate cortex	32	−16	41	14	21	3.38
			Anterior cingulate cortex	32	12	40	10	24	3.44
			Putemen	–	−26	8	−12	18	3.43
			Putemen	20	−32	−18	−21	12	3.74
			Putemen	20	26	−18	−19	6	4.01
			Medial prefrontal cortex	9	4	63	17	29	3.70
			Insula	48	−44	−9	3	33	−7.20
			Insula	48	47	3	−6	3	−5.20
			Thalamus	21	−7	−12	6	30	−5.29
			Thalamus	21	4	−14	3	21	−5.55
	Control	ALFF	Orbitofrontal cortex	11	−9	42	−15	15	5.01
			Anterior cingulate cortex	32	13	36	18	18	4.86
			Insula	48	43	3	3	28	−5.2
Zhao Bin (2022)[39]	Acupuncture	ALFF	Right crus I of the cerebellum	–	48	−66	−36	92	5.823
			Right middle frontal gyrus	–	36	45	−12	12	4.738
			Left superior temporal gyrus	–	48	6	−12	44	−4.720
			Left hippocampus	–	−15	−36	3	6	−4.802
			Left cingulate gyrus	–	−15	−27	42	12	−5.752
	Control	ALFF	Right cingulate gyrus	–	6	−39	27	225	5.697
			Left inferior temporal gyrus	–	−30	−39	−42	90	−4.551
			Right precentral gyrus	–	24	−21	75	2767	−5.493
Yi Yang (2011)[44]	Acupuncture	ALFF	Right superior frontal gyrus, middle part	6	−12	18	66	154	−7.33
			Right middle frontal gyrus	9	−3	51	42	219	−13.82
			Right inferior parietal lobule	48	−42	33	15	126	−20.59
			Precuneus	4	26	46	40	113	−6.98
			Posterior cingulate gyrus	46	39	30	42	344	−15.00
			Left inferior parietal lobule	40	58	−42	48	801	−2.84
			Left occipital lobe	7	–	–	–	265	−13.76
			Right occipital lobe, middle part	31	–	–	–	111	−6.70
			Right superior frontal gyrus, middle part	48	−63	−39	27	153	−9.61
			Right middle frontal gyrus	17	−12	−70	9	85	−7.33
			Right inferior parietal lobule	18	32	−84	9	85	−13.82
	Control	ALFF	Right parietal lobe	40	45	−48	42	307	−11.28
			Right occipital lobe	17	6	−66	6	186	−10.90

ALFF = amplitude of low-frequency fluctuation, BA = Brodmann area, MNI = Montreal Neurological Institute, ReHo = regional homogeneity.

A prespecified leave-one-out sensitivity analysis excluding the single taVNS trial^[41] left the primary findings substantially unchanged: the right cingulum and right caudate nucleus remained the only convergent clusters in the acupuncture arm, suggesting that the pooled signature is not driven by taVNS-specific effects.

Whole-brain meta-regression revealed significant neural associations with clinical and demographic variables within the treatment cohort. Mean participant age showed significant voxel-wise associations within the right lingual gyrus (Table S1, Supplemental Digital Content 1); sex ratio showed associations within the right cerebellum (Table S2, Supplemental Digital Content 2); HAMD scores correlated with activation within the left cerebellum (Table S3, Supplemental Digital Content 3); and the number of acupuncture sessions exhibited significant associations within the left amygdala (Table S4, Supplemental Digital Content 4).

In the clinical-outcome meta-analyses, compared with conventional treatment, acupuncture-based interventions significantly reduced the HAMA score (MD = 2.67; 95% CI: 0.54–4.80; I² = 24%), indicating alleviation of anxiety symptoms, as well as the HAMD score (MD = 3.03; 95% CI: 2.00–4.06; I² = 33%) and the SDS score (MD = 5.08; 95% CI: 3.57–6.59; I² = 0%), demonstrating reductions in depressive symptoms. Forest plots are shown in Figures S1–S3, Supplemental Digital Content 5.

4. Discussion

4.1. Principal findings

This meta-analysis synthesizes evidence from 7 fMRI studies investigating acupuncture-induced neuroplastic changes in patients with MDD. Key findings reveal distinct neural remodeling patterns between acupuncture-based and conventional treatments. Acupuncture specifically enhances resting-state spontaneous activity (indexed by ReHo and ALFF/fALFF) in the right cingulum and the right caudate nucleus. In contrast, conventional treatments predominantly enhance activity in the left hippocampus and the right middle occipital gyrus. These neural changes were accompanied by significant clinical symptom improvement, reflected in reduced HAMD, HAMA, and SDS scores. Furthermore, meta-regression identified an association between HAMD reduction and left-cerebellar plasticity, while amygdala modulation exhibited a dose-dependent effect that strengthened with an increasing number of treatment sessions.

4.2. Neuroplasticity changes induced by acupuncture

Our analysis revealed significantly elevated resting-state spontaneous brain activity in the cingulum and caudate nucleus following acupuncture. Hyperactivation of the right cingulum is consistent with a role in top-down emotion regulation: the dorsal anterior cingulate cortex integrates cognitive control over emotional responses, while^[50,49] the subgenual ACC modulates limbic reactivity.^[51,52] Dysfunction in these circuits is a hallmark of depression, characterized by impaired suppression of negative emotion and default-mode network (DMN) hyperactivity.^[53] Acupuncture may restore inhibitory control over limbic structures such as the amygdala, as supported by our meta-regression linking session number to left-amygdala plasticity. This partially aligns with mechanisms observed in taVNS, a noninvasive form of vagus nerve stimulation. One included study (Sun Jifei^[41]) employed taVNS, which has been shown in prior work to attenuate amygdala hyperactivity and to normalize DMN-cingulate connectivity in MDD.^[45,54] Manual acupuncture, used in the remaining 6 included trials, does not directly stimulate the auricular branch of the vagus nerve, suggesting that taVNS and manual acupuncture may converge on similar downstream effects via partially distinct afferent pathways (somatic afferents projecting to nucleus of the solitary tract via the spinal cord and brainstem versus auricular afferents projecting directly via the auricular branch of the vagus). Therefore, while VNS-like effects may contribute to acupuncture’s impact, they are unlikely to fully explain the observed neuroplastic changes, and additional mechanisms – including dopaminergic reward-circuit modulation and cerebellar–limbic connectivity – are likely involved.

Increased caudate activation after acupuncture highlights its influence on reward pathways. Depression is associated with striatal hypoactivity, contributing to anhedonia and impaired motivation.^[46,47,55] Acupuncture may stimulate dopaminergic projections from the ventral tegmental area to the striatum, potentiating reward sensitivity. Rodent studies corroborate this view, showing that electroacupuncture at ST36 increases dopamine release in the striatum and upregulates D2 receptors.^[56] Our findings extend this mechanistic insight to humans and suggest that acupuncture may ameliorate anhedonia by restoring striatal function.

Meta-regression linked HAMD improvement to left-cerebellar plasticity, challenging the traditional view of the cerebellum as solely motor-coordinative – a position consistent with several recent studies. Neuroimaging studies have shown abnormal cerebellar volume in MDD, particularly in area IX, which may serve as a trait marker.^[57] The cerebellum has also been implicated in emotional processing through reciprocal connections with limbic regions.^[58,59] These findings support the concept of a “cognitive cerebellum” that contributes to behavioral regulation beyond motor control. Independent studies have reported increased gray-matter volume in cerebellar area IX in both acute and remitted MDD patients^[57] as well as decreased gray-matter density in the left cerebellum of 1st-episode MDD patients,^[60] further supporting the cerebellum’s role in the pathophysiology of depression.

4.3. Contrasting acupuncture and conventional treatment mechanisms

Compared with conventional antidepressants such as selective serotonin reuptake inhibitors – which exert therapeutic effects primarily by enhancing hippocampal neurogenesis and modulating prefrontal–limbic circuitry – acupuncture appears to engage a distinct neurobiological pathway. Our findings indicate that acupuncture selectively upregulates activity in the right cingulum and the right caudate nucleus, regions critically involved in emotion regulation and reward processing, without significantly affecting the hippocampus. This pattern suggests that acupuncture may offer complementary mechanisms for patients who exhibit hippocampal nonresponsiveness or reward-circuit dysfunction – a common phenotype in TRD. Moreover, acupuncture’s modulation of the amygdala and cerebellum, as revealed by meta-regression, implies a bottom-up regulation of affective processing, contrasting with the top-down modulation typically associated with pharmacotherapy. These distinctions underscore acupuncture’s potential adjunctive value, particularly for patients with anhedonia-dominant or emotion-dysregulation subtypes of depression.

4.4. Neuroplasticity and long-term clinical outcome

A central question raised in peer review – and 1 we are unable to answer definitively with the present evidence – is whether the neuroplastic changes documented immediately posttreatment translate into durable, long-term clinical benefit. None of the 7 included trials reported follow-up neuroimaging or follow-up symptom assessments beyond the end of the active treatment phase (3–12 weeks). The strongest causal claim our data support is therefore a concurrent association between cingulate/caudate/cerebellar plasticity and within-trial symptom reduction, not a prospective claim of sustained relief. That said, 2 converging lines of evidence motivate the hypothesis that acupuncture-induced plasticity is at least partially persistent: the meta-regression dose-response relationship between the number of sessions and left amygdala modulation suggests cumulative, not transient, engagement of limbic circuitry; and longitudinal rs-fMRI studies of antidepressant pharmacotherapy and of taVNS have shown that treatment-induced changes in DMN, cingulate, and amygdala connectivity can persist for months and predict subsequent relapse risk.^[45,54] Future RCTs of acupuncture should incorporate at least 1 posttreatment follow-up assessment (we suggest 3 and 6 months) with paired rs-fMRI and clinical scales, and where possible should test whether early neuroplastic responders show lower relapse rates than nonresponders. Such designs would directly link the neural signatures identified here to clinically meaningful long-term outcomes.

4.5. Limitations

This review applied SDM-PSI to rs-fMRI data from RCTs of acupuncture-based interventions in MDD. The identified alterations in cerebral activity advance our understanding of neuroplastic mechanisms and inform clinical translation; strict inclusion criteria minimized selection bias, and the 7 retained trials shared comparable participant profiles and methodological designs. Nevertheless, several limitations warrant explicit Acknowledgments.

4.5.1. Small number of trials and modest pooled sample

Although the pooled sample comprises 357 patients across 7 RCTs, individual studies were small (16–160 participants), and the pooled cohort remains underpowered for critical subgroup analyses, such as comparisons across depression subtypes or across acupuncture protocols.

4.5.2. Intervention heterogeneity

The included trials pooled manual acupuncture, electroacupuncture, and 1 taVNS trial. Although we provided a neuroanatomical rationale (Section 2.4) and performed a leave-one-out sensitivity analysis excluding the taVNS trial – results of which were substantially unchanged – the possibility remains that subtle differences in afferent pathways across these modalities are obscured by pooled CBMA. Larger samples and stratified meta-analyses will eventually be required to distinguish modality-specific neural signatures.

4.5.3. Heterogeneity of rs-fMRI metrics

ReHo, ALFF, and fALFF index biologically distinct properties of intrinsic brain activity. While SDM-PSI permits coordinate-level pooling across these indices, the resulting summary maps should be interpreted as identifying regions of convergent alteration across complementary metrics rather than as evidence of a single, metric-specific neurophysiological change. This methodological compromise was necessitated by the small number of eligible trials and is, in our judgment, the most important limitation of the present synthesis. Future updates of this meta-analysis, once a larger evidence base is available, should report ReHo, ALFF, and fALFF in separate analyses with formal between-metric comparisons.

4.5.4. Publication-bias assessment

Methodological guidance recommends asymmetry tests only when at least 10 studies are pooled, because Egger test is underpowered for smaller samples.^[47] Our nonsignificant Egger results should therefore be regarded as inconclusive rather than as evidence of an absence of small-study effects.

4.5.5. Lack of long-term follow-up

As discussed in Section 4.4, none of the included trials reported follow-up beyond the active treatment phase. The temporal stability of the observed neuroplastic changes – and their predictive value for relapse – remains an open question.

4.5.6. Design limitations of the underlying trials

Only 2 of the 7 studies blinded both participants and outcome assessors, creating risks of performance and detection bias; high risk of bias in allocation concealment further weakens causal inference. Negative fMRI results may also be under-reported in this literature, despite the nonsignificant Egger result.

These limitations should be borne in mind when interpreting the present synthesis. We frame our findings as preliminary but methodologically rigorous evidence that warrants confirmation in larger, standardized, and longitudinally designed trials.

5. Conclusions

This meta-analysis demonstrates that acupuncture-based interventions induce neuroplastic changes in key brain circuits implicated in depression – specifically the cingulum and the caudate nucleus – and that these effects are distinct from the hippocampal–occipital effects of conventional treatments. The observed changes correlate with clinically meaningful symptom reduction, highlighting acupuncture’s potential as an adjunctive therapy targeting reward processing and emotion regulation deficits. Methodological limitations—small number of trials, intervention and metric heterogeneity, limited blinding, and absence of long-term follow-up data – necessitate larger, standardized RCTs with prospective follow-up. Future work should integrate molecular biomarkers with longitudinal neuroimaging to unravel acupuncture’s systems-level mechanisms and to advance personalized neuromodulation strategies for depression.

Acknowledgments

The authors thank the physicians and technicians at Dongzhimen Hospital’s MRI Department for their guidance in functional magnetic resonance imaging.

Author contributions

Conceptualization: Juwei Zhang, Xiaojin Shi, Weifeng Wang, Xinyue Shi.

Data curation: Juwei Zhang, Kang Xu.

Formal analysis: Juwei Zhang, Kang Xu.

Funding acquisition: Juwei Zhang, Kang Xu.

Investigation: Juwei Zhang.

Methodology: Juwei Zhang, Yihuai Zou.

Project administration: Juwei Zhang.

Resources: Juwei Zhang.

Software: Juwei Zhang.

Supervision: Juwei Zhang.

Validation: Juwei Zhang, Xiaojin Shi, Yan Shen.

Visualization: Juwei Zhang, Xiaojin Shi, Yan Shen.

Writing – original draft: Juwei Zhang, Xiaojin Shi, Yuanyuan Li, Xinyue Shi, Yihuai Zou.

Writing – review & editing: Juwei Zhang, Xiaojin Shi, Yuanyuan Li, Yihuai Zou.