Effect of theta-transcranial alternating current stimulation on working memory performance among healthy adults: A systematic review and meta-analysis

Abstract

BACKGROUND

Working memory serves as a fundamental cognitive function that substantially impacts performance in various cognitive tasks. Extensive neurophysiological research has established that theta oscillations (4-8 Hz) play an essential role in supporting working memory operations. Theta-band transcranial alternating current stimulation (tACS) offers a potential mechanism for working memory enhancement through direct modulation of these fundamental neural oscillations. Nevertheless, current empirical evidence shows substantial variability in the observed effects of theta-tACS across studies.

AIM

To conduct a systematic review and meta-analysis evaluating the effects of theta-tACS on working memory performance in healthy adults.

METHODS

A systematic literature search was performed on PubMed, EMBASE, and Web of Science up to March 10, 2025. Effect sizes were computed using Hedges’ g with 95% confidence intervals (CIs), with separate meta-analyses for all included studies and for distinct working memory paradigms [n-back and delayed match-to-sample (DMTS) tasks] to examine potential task-specific effects. Subgroup analyses and meta-regression were performed to evaluate the influence of key moderating variables.

RESULTS

The systematic review included 21 studies (67 effect sizes). Initial meta-analysis showed theta-tACS moderately improved working memory (Hedges’ g = 0.405, 95%CI: 0.212-0.598). However, this effect became nonsignificant after correcting for publication bias (trim-and-fill adjusted Hedges’ g = 0.082, 95%CI: -0.052 to 0.217). Task-specific analyses revealed significant benefits in n-back tasks (Hedges’ g = 0.463, 95%CI: 0.193-0.733) but not in DMTS tasks (Hedges’ g = 0.257, 95%CI: -0.186 to 0.553). Moderator analyses showed that performance in n-back tasks was influenced by stimulation frequency (P = 0.001), concurrent status (P = 0.014), task modality (P = 0.005), and duration (P = 0.013), whereas only the region of targeted stimulation (P = 0.012) moderated DMTS tasks.

CONCLUSION

Theta-tACS enhances working memory in healthy adults, with effects modulated by the task type and protocol parameters, offering dual implications for cognitive enhancement and clinical interventions.

Key Words: Working memory; Transcranial alternating current stimulation; Theta; Healthy adults; N-back; Delayed match-to-sample task

Core Tip: In this meta-analysis, we evaluated theta-band transcranial alternating current stimulation for working memory enhancement in healthy adults. Although initial analyses showed moderate efficacy, correction for publication bias rendered effects nonsignificant. Notably, task-specific effects emerged, specifically showing significant improvement in n-back tasks vs null effects in delayed match-to-sample tasks. Key moderators of performance included stimulation frequency, concurrent status, task modality, and stimulation duration for n-back tasks and the target region for match-to-sample tasks. Although these findings primarily guide cognitive enhancement protocols for healthy individuals, they also provide mechanistic insights for potential clinical translation to populations with working memory deficits.

INTRODUCTION

Working memory is a fundamental cognitive system responsible for the active maintenance, transformation, and recall of information related to ongoing tasks[1]. As a core component of executive function, working memory plays a critical role in determining performance across a wide range of cognitive tasks, including stimulus discrimination[2], learning[3,4], language comprehension[5], reasoning[6], and problem-solving[7]. Of particular clinical significance, working memory impairment has been consistently identified as a robust transdiagnostic phenomenon manifesting across major psychiatric disorders, particularly major depressive disorder[8], attention deficit hyperactivity disorder[9], and schizophrenia spectrum disorders[10]. This well-established pathophysiological association, combined with the fundamental role of working memory in cognition, has driven considerable research efforts aimed at developing targeted interventions for working memory enhancement.

Extensive neurophysiological research has established neural oscillations as fundamental mechanisms of cognitive processing, and converging evidence from multiple methodologies has highlighted the particular importance of theta (4-8 Hz) activity in working memory operations[11-13]. Decades of electrophysiological studies using electroencephalography and magnetoencephalography have consistently shown that frontal midline theta oscillations show robust amplitude increases during working memory tasks[14-16], with these increases showing a strong correlation with improved task performance. The critical role of theta rhythms in working memory is further supported by the finding that individual differences in theta power predict working memory performance, suggesting theta oscillations represent not just an epiphenomenon but a fundamental mechanism for working memory[17].

Building upon these established neural mechanisms, the interest in the use of noninvasive brain stimulation techniques as potential interventions for working memory enhancement has increased[18,19]. Among these approaches, transcranial alternating current stimulation (tACS) has emerged as a particularly promising approach given its unique ability for frequency-specific entrainment of endogenous neural oscillations[20]. This technique’s potential is particularly relevant because of the well-documented association between working memory mechanisms and neural oscillation patterns[21]. Unlike neurofeedback approaches[22], tACS can directly interact with ongoing brain rhythms through principles of phase alignment and resonance[23], making it ideal for targeting the theta oscillations that underlie working memory function.

The extant literature shows considerable heterogeneity in the reported effects of theta-tACS on working memory performance, and this variability can be systematically attributed to stimulation and task parameters that collectively shape stimulation outcomes[24]. First, the brain region stimulated emerges as a fundamental determinant, with neuroanatomical specificity critically influencing efficacy. Empirical findings show that theta tACS targeting parietal regions (P3/P4) consistently enhances working memory performance, whereas analogous stimulation targeting the left prefrontal cortex (F3) yields less robust effects[25,26]. Notably, concurrent bifocal tACS application to both left prefrontal and temporal cortices has recently been reported to offer significant working memory improvements in healthy aging populations, whereas isolated stimulation of either region alone reportedly fails to elicit cognitive enhancement[27]. These findings collectively suggest that the neurocognitive effects of tACS are significantly site-specific. Moreover, apart from the stimulation site, frequency parameters also contribute to the observed heterogeneity. Theta tACS studies targeting working memory have used a range of frequencies from 4 Hz to 7 Hz[28-30], with heterogeneous outcomes reported across studies. Comparative analyses between 4 Hz and 7 Hz theta-tACS have shown superior working memory enhancement with 4 Hz stimulation[31]. Although some investigations have used personalized theta frequencies for intervention, these approaches have similarly yielded inconsistent results[32-34]. Notably, a critical gap remains: The literature lacks systematic comparisons between fixed-frequency and personalized theta-tACS protocols, leaving undetermined whether individualized approaches offer substantive advantages. The complexity of parameter-dependent effects is compounded by additional methodological variables. The neuromodulatory efficacy of tACS on working memory is influenced by: (1) Variability in current intensity across studies[24,35,36]; (2) Lack of standardization in stimulation duration protocols[24,25,33,36]; and (3) Differential timing of cognitive task administration relative to stimulation (online vs offline paradigms)[24,35,37,38]. Moreover, some inconsistencies may be attributable to technological limitations: Conventional tACS approaches exhibit inferior spatial resolution and target engagement compared to relatively advanced techniques like high-definition tACS, which incorporates computational current flow modeling[39].

Importantly, task-specific factors introduce another layer of complexity. Studies have implemented diverse working memory paradigms, such as the n-back task[26,33], visual-array comparison task[33], delayed match-to-sample (DMTS) task[31,40], and change detection tasks[41]; however, emerging evidence suggests that cognitive demands of each paradigm may differentially interact with tACS effects[42]. This paradigm-dependent variability underscores a key interpretive challenge; without systematic analysis of task characteristics as moderators, it remains difficult to disentangle whether divergent findings reflect true parameter effects or artifacts of task design.

The existing literature on tACS and working memory reveals several critical gaps in our current understanding. While few meta-analyses have examined tACS effects on working memory in healthy populations, their findings remain inconsistent and fail to establish a clear consensus regarding its efficacy[43-45]. More importantly, despite growing evidence suggesting theta-tACS as a particularly promising approach for working memory enhancement, there have been no comprehensive quantitative studies that have specifically focused on theta-band stimulation effects in healthy individuals or systematically investigated how different stimulation parameters may influence outcomes. This oversight is particularly notable given that previous reviews have primarily emphasized clinical applications[46], leaving a significant knowledge gap regarding optimal protocols for cognitive enhancement in healthy populations. Consequently, the lack of parameter-specific analysis and standardized protocols has hindered both theoretical advances in understanding tACS mechanisms and their practical applications for working memory improvement.

The present meta-analysis attempts to fill these critical gaps by pursuing two primary objectives: (1) Providing the first quantitative synthesis of theta-tACS efficacy for working memory improvement in neurotypical adults; and (2) Systematically elucidating how key stimulation parameters, such as frequency, intensity, duration, and montage, moderate these effects. Our findings will establish evidence-based guidelines for protocol optimization while advancing theoretical understanding the role of theta oscillations in working memory processes, thus bridging fundamental research with translational applications in both cognitive augmentation and clinical rehabilitation. Importantly, clarifying these parameter-dependent effects may inform optimized protocols for other theta-mediated cognitive functions, such as attention control, decision-making, and learning. Furthermore, the methodological framework may facilitate tACS translation to clinical populations with working memory deficits.

MATERIALS AND METHODS

To ensure methodological transparency and reproducibility, this meta-analysis was conducted in strict accordance with the PRISMA 2020 guidelines[47]. This study has been registered with International Prospective Register of Systematic Reviews, No. CRD20251046725.

Systematic literature search strategy

The search strategy incorporated both keywords and controlled vocabulary terms [e.g., Medical Subject Headings (MeSH) in PubMed] to maximize retrieval sensitivity. For instance, the PubMed search strategy was as follows: [“Executive Function” (Mesh)] OR [Working Memory (Title/Abstract)] OR [Working Memories (Title/Abstract)] OR [Executive Functions (Title/Abstract)] OR [Function, Executive (Title/Abstract)] OR [Functions, Executive (Title/Abstract)] OR [Executive Control (Title/Abstract)] OR [Executive Controls (Title/Abstract)] OR [n-Back (Title/Abstract)] AND [transcranial alternating current stimulation (Title/Abstract)] OR [tACS (Title/Abstract)]. To maintain focus on widely accessible literature, the search was restricted to English-language publications. However, to ensure a thorough examination of the available evidence, no date restrictions were imposed. More details regarding the search strategy are available in Supplementary Figure 1.

Eligibility criteria

For inclusion, studies were required to meet the following eligibility criteria according to the “PICOS” principle detailed below: (1) Population (P): Participation restricted to healthy adult populations; (2) Intervention (I): Implementing tACS in single or multiple sessions, either during (online) or preceding (offline) behavioral evaluations; (3) Comparison (C): Inclusion of a sham-controlled design, where tACS was compared with a placebo condition (sham tACS) with identical parameters except for active stimulation; (4) Outcome (O): Availability of quantitative working memory performance data suitable for effect size calculations (e.g., mean with standard deviations/standard errors, or precise statistical values such as F, P, t, or z); and (5) Study design (S): Randomized controlled trials using either parallel-group or crossover methodologies. Studies were excluded if they: (1) Involved non-human subjects; (2) Were review publications; (3) Constituted case studies or anecdotal reports; or (4) Involved implementing tACS in combination with adjunct interventions (e.g., concurrent working memory training protocols).

Study selection and data extraction

The systematic literature search for relevant papers published anytime before March 10, 2025 was conducted during March 10-15, 2025. Consequently, we retrieved 702 records from PubMed (n = 116), EMBASE (n = 218), and Web of Science (n = 368). The records were exported from these databases and imported into Endnote 20 for further management. Within Endnote, an automated duplicate detection process was conducted, followed by manual review to eliminate any remaining duplicates. After deduplication, 454 articles remained for screening, which were screened based on their titles and abstracts using predefined inclusion and exclusion criteria. As a result, 40 publications were selected for full-text review. Studies with ambiguous eligibility were independently assessed by a second reviewer to ensure accuracy. The final selection of studies was determined through consensus between the two reviewers. A PRISMA flowchart was used to illustrate the study selection process (Figure 1).

Figure 1  PRISMA flowchart of the study selection procedure.

Data extraction

Two authors (Hou TY and Zhang RK) independently screened the titles, abstracts, and full texts of the publications. Disagreements regarding study inclusion were resolved through iterative discussions until consensus was achieved. For qualifying articles, the following variables were systematically extracted: Author(s), publication year, study design (e.g., within- vs between-subjects), sample size, participant demographics (age and sex), working memory task type, cognitive load (e.g., the 2-back task), task modality (verbal/spatial/object), stimulation parameters (duration, frequency, number of sessions, target brain region, and phase), and finally, task performance (e.g., mean and standard deviations) at baseline, during tACS, or after stimulation (for change score calculation). Missing numerical data were addressed in the following sequence: (1) First, direct data requests were sent to the corresponding authors via email; (2) If authors were unresponsive and results were graphically presented, data were extracted using GetData Graph Digitizer; and (3) Finally, reported inferential statistics (e.g., t- or F-values) were converted to standardized effect sizes after established methodologies[48]. The studies that were finally included are listed in Table 1[24,25,28-38,40-42,49-53].

Table 1 Characteristics of the included studies.

Ref.	Study design	Target region	Duration	Frequency	Concurrent task	Task	Task modality
Alekseichuk et al[49], 2017	Within subject	Frontoparietal	18	6	Online	N-back	Spatial
Bender et al[35], 2019	Within subject	Parietal		4	Online	Delayed match-to-sample task	Spatial
Biel et al[24], 2022	Between-subject, within subject	Frontoparietal	14	6	Online	Delayed letter recognition task	Verbal
Chander et al[50], 2016	Between-subject	Frontoparietal	4	Personalized	Online	N-back	Verbal
Debnath et al[36], 2024	Within subject	Frontal	20	5	Online	Delayed match-to-sample task	Objective
Draaisma et al[34], 2022	Within subject	Frontoparietal	30	Personalized	Online	N-back	Objective
Jaušovec and Jaušovec[25], 2014	Within subject	Frontal, parietal	15	Personalized	Online	Visual-array comparison task	Objective
Jaušovec et al[33], 2014	Within subject	Frontal, parietal	15	Personalized	Online	N-back	Verbal
Jones et al[51], 2019	Within subject	Frontoparietal	15	4.5	Online	N-back	Objective
Kleinert et al[40], 2017	Within subject	Frontoparietal	26	5	Online	Delayed match-to-sample task	Spatial
Ociepka et al[42], 2024	Within subject	Parietal	20	5.5	Online	Antisaccade, recall, graph mapping task	Spatial
Pahor and Jaušovec[32], 2017	Within subject	Parietal, frontoparietal, frontal	15	Personalized	Offline	N-back	Verbal
Röhner et al[37], 2018	Within subject	Frontoparietal	15	6	Offline	N-back	Verbal
Tseng et al[41], 2018	Within subject	Parietal	22	6	Online	Change detection task	Spatial
Violante et al[30], 2017	Within subject	Frontoparietal	26.5	6	Online	N-back	Verbal
Wolinski et al[31], 2018	Within subject	Parietal	12	4	Online	Delayed match-to-sample task	Objective
Xiao et al[29], 2023	Within subject	Frontoparietal	20	6	Offline	Delayed match-to-sample task	Spatial
Yang et al[28], 2024	Within subject	Frontal	16	5	Offline	N-back	Spatial
Zeng et al[52], 2022	Between-subject	Frontal	20	8	Offline	N-back	Verbal
Zhang et al[53], 2022	Within subject	Frontoparietal	25	Personalized	Online	Spatial forward span task	Spatial
Zhang et al[38], 2022	Between-subject	Frontoparietal	15	6	Offline	N-back	Verbal

Risk-of-bias assessment

The included randomized controlled trials underwent rigorous quality evaluation using the Cochrane risk-of-bias (RoB) 2.0 tool[54]. This comprehensive assessment framework critically examines key dimensions of trial methodology, including study design, implementation procedures, and reporting practices. Two independent reviewers evaluated each study, assigning categorical ratings (low risk/some concerns/high risk) to all domains. To ensure consistent judgments, discrepancies were reconciled through panel discussion or third-party arbitration.

Statistical analysis

The meta-analyses were conducted in R (R Core Team, 2018) with RStudio[55]. The present meta-analysis used robust variance estimation through the “robumeta package” in RStudio[56] to calculate pooled effect sizes, particularly addressing the statistical interdependencies arising from multiple effect sizes reported within individual studies. Effect sizes were quantified using Hedges’ g, wherein positive values indicated enhanced working memory performance after active tACS intervention. To ensure analytical robustness, we implemented a multi-tiered quality control protocol. First, Grubbs’ test[57] identified potential outliers that may distort conclusions. Subsequently, leave-one-study-out sensitivity analyses systematically evaluated the influence of each study by iteratively excluding individual studies and recomputing effect estimates. This approach precisely quantified the stability of our findings against potential outlier effects.

For moderator analysis, we subsequently conducted two types of supplementary analyses using the meta package: (1) Subgroup analyses for categorical variables (e.g., task modality and stimulation region); and (2) Meta-regression for continuous variables (e.g., stimulation duration). This methodological choice was based on the superior ability of the meta package to quantify between-subgroup heterogeneity and determine the proportion of variance explained (R²) in regression models. Importantly, comparative analyses confirmed strong concordance between the effect size estimates generated by these two approaches, with the meta package’s random-effects models[58] and the restricted maximum likelihood method[59] producing results showing high consistency with the results of robumeta’s correlated effects model.

Publication bias was examined by visual examination of the funnel plot[60], a graphical representation plotting effect sizes against their respective standard errors. Owing to the inherent properties of tACS interventions, complete blinding of experimenters was methodologically not feasible because active administration required direct experimenter involvement in device operation. Consequently, experimenter blinding was excluded from the RoB evaluation criteria. The symmetry of this distribution indicates publication bias, with asymmetry suggesting potential systematic omissions in the literature. In addition, Egger’s regression provided a quantitative assessment of funnel plot asymmetry. To further validate these findings, we implemented the trim-and-fill algorithm[61], which statistically estimates the number of hypothetical “missing” studies required to restore plot symmetry, assuming observed asymmetry reflects publication bias.

RESULTS

RoB of included studies

Among the 21 included studies, five showed moderate overall RoB, primarily due to insufficient methodological detail regarding random sequence generation, allocation concealment, participant blinding, and outcome assessment procedures. The remaining 16 studies were classified as having high overall RoB, given that each had at least one domain (e.g., randomization, blinding, or selective reporting) that was judged to be at high risk. A comprehensive visualization of these RoB assessments is shown in Figure 2.

Figure 2  Summary of risk-of-bias for each study and domain.

Effects across all studies

From the initial database search, 21 articles meeting the predefined inclusion criteria were identified, contributing a total of 67 effect sizes (k = 67) for inclusion in the meta-analysis. The results showed a moderate yet significant overall effect of tACS on working memory performance compared to sham stimulation, with a pooled effect size of Hedges’ g = 0.405 [P < 0.001, 95% confidence interval (CI): 0.212-0.598]. Substantial heterogeneity was observed among the studies (I² = 73.68%, τ² = 0.170), indicating considerable variability in effect sizes. Grubbs’ test identified one outlier with a Hedges’ g of 2.191; however, its removal did not significantly alter the overall effect size (Hedges’ g = 0.396, P < 0.001, 95%CI: 0.205-0.587), and the heterogeneity remained significant (I² = 72.49%, τ² = 0.159).

The results of the sensitivity analysis revealed that no individual study exerted undue influence on the overall effect size, showing that the findings of the meta-analysis were robust and stable (Supplementary Figure 2). This suggests that the pooled effect size is not disproportionately driven by any single study. Study-specific effect sizes and the overall pooled effect are visually summarized in the forest plot (Figure 3).

Figure 3 The overall meta-analysis effect size combines the results from all included studies. The corresponding forest plot illustrates the comparative effects, with values greater than 0 indicating a preference for transcranial alternating current stimulation and values less than 0 indicating a preference for sham stimulation. CI: Confidence interval.

Effects on n-back tasks

A meta-analysis was conducted on 12 studies comprising 41 effect sizes that used the n-back task to assess working memory performance. The pooled effect size across studies was statistically significant (Hedges’ g = 0.463, P = 0.003, 95%CI: 0.193, 0.733). Heterogeneity analysis revealed moderate variability among effect sizes (I² = 45.422%, τ² = 0.119). Subgroup analyses were performed to assess potential moderating variables (Table 2). A notable difference was identified between tACS administered at fixed vs personalized frequencies. In particular, fixed-frequency tACS resulted in a significantly greater enhancement in n-back performance (Hedges’ g = 0.608) compared with personalized frequency stimulation (Hedges’ g = 0.133), with a mean difference of 0.475. In addition, the moderating effect of concurrent task modality was significant. Online n-back tasks showed a stronger response to tACS (Hedges’ g = 0.490) than offline tasks (Hedges’ g = 0.153). Furthermore, differential effects were observed across task types: Active tACS showed the greatest improvement in spatial n-back performance (Hedges’ g = 1.168), followed by visual n-back (Hedges’ g = 0.405) and verbal n-back tasks (Hedges’ g = 0.189). None of the tested moderators, study design, target brain region, concurrent task, or cognitive load, yielded statistically significant effects (all, P > 0.05). Meta-regression results revealed that the stimulation duration significantly moderated the effect of tACS on working memory (P = 0.013), accounting for 19.56% of the between-study variance. The regression coefficient indicated a positive linear relationship, with each additional minute of stimulation duration associated with an increase in effect size of β = 0.0437.

Table 2 N-back subgroup analyses.

Covariables	Subgroups	Number of effect sizes	Hedges’ g	95%CI	I² (%)	τ²	P value
Study design							0.296
	Within-subject	33	0.260	0.136-0.386	57.20	0.079
	Between-subject	8	0.504	0.064-0.945	64.80	0.268
Target region							0.681
	Frontoparietal	18	0.324	0.096-0.552	72.50	0.181
	Parietal	9	0.200	-0.012 to 0.412	39.30	0.039
	Frontal	14	0.306	0.128-0.485	41.20	0.035
Frequency							0.001
	Fixed	15	0.608	0.336-0.881	72.30	0.204
	Personalized	26	0.133	0.039-0.227	11.70	< 0.001
Concurrent task							0.014
	Online	15	0.490	0.239-0.742	67.3	0.167
	Offline	20	0.153	0.057-0.250	36.80	0.004
Load							0.726
	1	4	0.532	-0.032 to 1.097	69.80	0.227
	2	18	0.274	0.041-0.508	73.60	0.192
	3	15	0.251	0.131-0.371	26.50	< 0.001
	4	2	0.549	0.005-1.094	21.60	0.033
	5	2	0.187	-0.283 to 0.657	0.00	0.000
Task modality							0.005
	Spatial	3	1.168	0.529-1.807	77.90	0.245
	Verbal	36	0.189	0.096-0.281	35.40	0.014
	Objective	2	0.405	0.143-0.666	0.00	0.000

CI: Confidence interval.

Effects on the DMTS task

A random-effects meta-analysis was performed on 6 studies containing 17 effect sizes examining tACS effects on DMTS tasks. The pooled effect size across studies was not significant (Hedges’ g = 0.257, P = 0.003, 95%CI: -0.186 to 0.553), suggesting no robust overall enhancement of DMTS performance through tACS intervention. Substantial heterogeneity was observed among effect sizes (I² = 76.247%, τ² = 0.258), indicating considerable variability in stimulation outcomes across studies.

Subgroup analyses by the stimulation region revealed significant regional differences (P = 0.012). Parietal stimulation showed the most pronounced behavioral enhancement (Hedges’ g = 0.574), whereas frontal (Hedges’ g = 0.077) and frontoparietal (Hedges’ g = -0.082) stimulation did not show significant effects on DMTS performance. In subsequent analyses, no other moderating variables (e.g. study design and task modality) showed significant influences on effect sizes (Table 3).

Table 3 Delayed match-to-sample task subgroup analyses.

Covariables	Subgroups	Number of effect sizes	Hedges’ g	95%CI	I² (%)	τ²	P value
Study design							0.564
	Within-subject	15	0.092	-0.197 to 0.380	76.00	0.255
	Between-subject	2	-0.090	-0.636 to 0.456	0.00	0
Target region							0.012
	Frontoparietal	10	-0.082	-0.490 to 0.326	78.10	0.346
	Parietal	3	0.574	0.277-0.871	0.00	0
	Frontal	4	0.077	-0.173 to 0.328	0.00	0
Concurrent task							0.295
	Online	15	0.105	-0.185 to 0.395	75.40	0.252
	Offline	2	-0.168	-0.588 to 0.253	0.00	0
Task modality							0.433
	Spatial	7	0.005	-0.642 to 0.652	87.60	0.680
	Verbal	4	0.020	-0.241 to 0.280	0.00	0
	Objective	6	0.226	0.022-0.430	0.00	0

CI: Confidence interval.

Publication bias

Figure 4 shows a funnel plot of the distribution of effect sizes plotted against their corresponding standard errors. The results of the Egger regression test showed no risk of publication bias (β = -0.427, SE = 0.233, P = 0.085). However, the funnel plot showed a slight skewing of the data. Further analysis using the trim-and-fill method estimated that 13 studies were missing on the left side of the funnel plot. After imputing these missing effects, the overall effect size of tACS on working memory was no longer statistically significant (Hedges’ g = 0.082, P = 0.228, 95%CI: -0.052 to 0.217).

Figure 4  Funnel plot for the transcranial alternating current stimulation effect on working memory.

DISCUSSION

The present meta-analysis, focusing exclusively on theta-band tACS for working memory enhancement in healthy adults, yielded a statistically significant and theoretically meaningful moderate effect size (Hedges’ g = 0.405, 95%CI: 0.212 to 0.598), underscoring the efficacy of theta-frequency stimulation in cognitive enhancement. Several noteworthy patterns were identified upon comparisons with previous comprehensive meta-analyses examining tACS across all frequency bands. Although our observed effect size is marginally smaller than that of Nissim et al[44] (Cohen’s d = 0.514), it substantially exceeds the minimal effect (Hedges’ g = 0.076) reported by Chuderski and Chinta[43]. Crucially, both comparative studies examined tACS effects across multiple frequency bands without focusing on theta-specific effects. This suggests two key points. First, the modest discrepancy with Nissim et al’s findings[44] may stem from methodological differences in effect size computation (Hedges’ g vs Cohen’s d). Second, and more significantly, the robust effect observed in our theta-specific analysis, particularly when contrasted with Chuderski and Chinta’s weak omnibus effect, provides compelling evidence suggesting that theta-band stimulation engages distinct neurophysiological mechanisms particularly conducive to working memory enhancement[43].

However, the trim-and-fill adjustment attenuated the effect to nonsignificance (adjusted Hedges’ g = 0.082, 95%CI: -0.052 to 0.217), indicating potential overestimation in our initial analysis caused by unpublished null findings, a methodological caution consistent with Chuderski and Chinta’s findings[43]. Although primary analyses supported intervention efficacy, the reduced evidentiary strength following bias adjustment necessitates cautious interpretation. Notably, sensitivity analyses (e.g., exclusion of high-RoB studies) maintained consistent effect directionality, suggesting some robustness of the observed trend. Nevertheless, the current findings should be interpreted with consideration of these potential biases.

Different effects of tACS on nback

The meta-analysis showed a statistically significant enhancement of n-back performance after theta-tACS administration. Moderation analyses revealed that fixed-frequency tACS protocols produced significantly larger effect sizes than individualized-frequency approaches (Δg = 0.475, P = 0.001), a pattern superficially inconsistent with previous literature advocating personalized parameters[43]. This discrepancy may be attributed to two key factors. First, the frequency-band specificity of our analysis should be considered. Our analysis exclusively examined theta-band studies, whereas previous meta-analyses favoring individualized protocols typically included broader frequency ranges (e.g., alpha/gamma), where frequency customization may yield greater benefits[43]. Second, the substantial methodological heterogeneity in individualized frequency determination methods (e.g., peak vs center-of-gravity approaches and resting-state vs task-based derivations) across studies may have introduced noise, obscuring potential advantages of personalized protocols in the theta band[32-34,50].

Although both online and offline stimulation protocols showed measurable improvements in working memory, which is consistent with existing neuromodulation literature, our moderation analysis identified a notable efficacy advantage for online protocols (Δg = 0.337, P = 0.014)[45]. This observed superiority likely reflects fundamental differences in their underlying mechanisms. Online theta-tACS appears to exert its effects through immediate neurophysiological integration, particularly through cross-frequency coupling mechanisms[62]. The exogenous theta entrainment may dynamically modulate endogenous oscillatory activity across multiple frequency bands, with low-frequency phase-amplitude coupling potentially optimizing high-frequency neural synchronization during task performance. This real-time interaction between applied stimulation and ongoing neural processing may account for the protocol’s enhanced behavioral effects. In contrast, the effects of offline theta-tACS are presumably mediated by spike-timing dependent plasticity mechanisms that induce longer-term synaptic modifications in targeted networks[20]. Spike-timing dependent plasticity-dependent neuroplastic changes typically require repeated stimulation sessions for consolidation, whereas most of the included studies used single-session designs, an approach that may be suboptimal for eliciting robust and long-lasting neuroplastic effects. These findings suggest that although both approaches show promise, online theta-tACS may offer more immediate and pronounced cognitive enhancement.

In addition, our meta-analysis revealed graded efficacy of active tACS across different n-back modalities, with the largest improvement in spatial tasks (Hedges’ g = 1.168), followed by visual (Hedges’ g = 0.405) and verbal tasks (Hedges’ g = 0.189), suggesting that task-specific neurocognitive processes determine stimulation effects. Notably, we observed no differential effects across load conditions, which is consistent with evidence that theta oscillations (4-8 Hz) mediate temporal coordination of working memory items rather than capacity-limited maintenance, which relies more on gamma-band mechanisms[63,64]. These findings collectively demonstrate that tACS effects depend primarily on the neural processes recruited by different task modalities rather than absolute working memory load or difficulty. This dissociation underscores the need for modality-specific protocol optimization and suggests that addressing capacity limitations may require complementary gamma-band or cross-frequency stimulation approaches.

Our meta-analysis identified a significant positive duration–response relationship for theta-tACS effects on working memory performance, which is in contrast with previous null findings reported by Nissim et al[44]. This observed dose–response pattern suggests potential cumulative neuroplastic effects, emphasizing the need for careful optimization of stimulation duration in protocol design. However, these findings should be interpreted with caution because the included studies were limited to durations of ≤ 30 minutes, leaving open the critical question of whether the relationship between stimulation duration and working memory enhancement follows an inverted U-shaped curve.

Different effects of tACS on DMTS task

The meta-analysis found no significant enhancement of DMTS task performance after theta-tACS administration. However, subsequent subgroup analyses revealed significant regional differences in stimulation efficacy (P = 0.012), with parietal stimulation showing robust enhancement (Hedges’ g = 0.574, P < 0.001) compared with negligible effects for frontal (Hedges’ g = 0.077) and frontoparietal (Hedges’ g = -0.082) protocols. These findings align with established neuroanatomical divisions of labor: Parietal regions predominantly support working memory storage buffers, whereas frontal areas mediate higher-order executive processes like attention and inhibitory control[65,66]. The selective benefit of parietal stimulation likely reflects the core mnemonic demands of DMTS tasks, which primarily assess item maintenance rather than executive manipulation[67]. This dissociation underscores the significance of target precision in tACS applications, suggesting that stimulation sites should be carefully matched to the specific cognitive components being evaluated.

Implications

To our knowledge, this meta-analysis provides the first comprehensive synthesis of evidence on the effects of theta-band tACS on working memory performance in healthy adults. Our study addresses a key gap in the neuromodulation literature by specifically targeting the theta band. Although theta oscillations have been strongly linked to working memory processes, prior tACS meta-analyses have combined data across multiple frequency bands, potentially masking frequency-specific effects. Our exclusive focus on theta-tACS isolates unique contribution of the theta band, helping clarify whether theta-band stimulation has specific, frequency-dependent effects on working memory beyond general neuromodulation. Furthermore, by analyzing how potential moderators influence theta-tACS effects, we provide evidence-based recommendations for future research and potential clinical applications. This is particularly timely given the increasing use of tACS in cognitive enhancement studies. This work not only advances our fundamental understanding of the role of the theta band in working memory but also guides the development of more targeted, effective tACS protocols for both research and translational purposes. Importantly, although our findings are derived from healthy populations, they carry important translational implications. The robust modifiability of working memory through theta-tACS observed in our analysis establishes proof-of-concept for clinical applications. This is because the same neurophysiological mechanisms (particularly frontoparietal theta synchronization) underlie working memory function across both healthy and clinical populations[68]. This dual perspective not only advances the fundamental understanding of theta oscillations in working memory but also informs the development of targeted intervention protocols with broader applications.

Limitations

The current meta-analysis has several limitations that need to be acknowledged. First, in the present study, numerous studies included in the analysis were identified as having a high RoB regarding the blinding of outcome assessment. This introduces a potential risk of detection bias, particularly for subjective working memory measures wherein the assessor’s expectations could influence scoring. However, we note that: (1) The included studies used objective, computerized working memory tasks with automated scoring, which may mitigate this concern; and (2) Sensitivity analyses excluding studies with high RoB did not substantially alter our primary findings (Supplementary Figure 2). Nevertheless, this limitation suggests that our results should be interpreted with appropriate caution regarding potential overestimation of effect sizes.

Second, the funnel plot asymmetry observed in our study (Figure 4) warrants careful consideration as it suggests potential publication bias. Egger’s test yielded a statistically significant result (P = 0.006)), indicating that smaller studies with null or negative findings may be underrepresented in the published literature. This interpretation is further supported by the trim-and-fill analysis, which identified 13 potentially missing studies and reduced the overall effect size to nonsignificance (Hedges’ g = 0.082, P = 0.228), highlighting the fragility of our primary findings. This phenomenon could be attributed to the exclusion of unpublished materials, commonly referred to as grey literature, which represents a notable limitation within the scope of the present investigation. To enhance the robustness of future meta-analyses in this field, we recommend: (1) Implementing exhaustive search strategies encompassing multiple clinical trial registries and grey literature sources; and (2) Establishing prospective protocols for unpublished data acquisition. These methodological refinements should substantially mitigate publication bias and improve the validity of effect size estimates in subsequent research syntheses.

Third, the interpretation of certain moderator variables (e.g., task modality, brain region) was limited by their reliance on a small subset of studies, which may reduce the robustness and generalizability of subgroup analyses. This limitation is exemplified by our counterintuitive finding that fixed-frequency tACS protocols showed superior efficacy than personalized theta-tACS approaches. To resolve this discrepancy, a dedicated comparative study should be conducted to directly contrast fixed-frequency vs personalized theta-frequency stimulation parameters.

Fourth, although the current study focused on the effects of theta-tACS on behavioral task performance, investigating its impact on neurophysiological biomarkers (e.g., frontal midline theta power and frontoparietal connectivity via functional magnetic resonance imaging/electroencephalography) may markedly advance our understanding of the neural substrates underlying working memory enhancement. Such multimodal approaches would not only elucidate target engagement mechanisms but also optimize stimulation parameters for maximal therapeutic efficacy.

Finally, although this proof-of-concept study focused on healthy adults to investigate core cognitive mechanisms, our findings offer valuable preliminary evidence for clinical translation. Key priorities for future research include: (1) Systematic dose-response studies to optimize training parameters; and (2) Longitudinal evaluations of multi-session protocols to assess treatment durability. Such investigations are especially crucial for clinical populations with significant working memory deficits, particularly schizophrenia spectrum disorders and attention-deficit/hyperactivity disorder, wherein effective cognitive remediation strategies remain an urgent clinical need. In addition, combining cognitive training with targeted neuromodulation approaches may provide a deeper mechanistic understanding while potentially improving therapeutic outcomes. By addressing these research gaps, we can both advance theoretical understanding and achieve meaningful progress in clinical cognitive rehabilitation.

CONCLUSION

In conclusion, theta-band tACS shows a moderately heterogeneous yet statistically significant effect on working memory enhancement in healthy adults; however, these findings should be interpreted with caution because of evident publication bias and significant between-study heterogeneity. In addition, the present study systematically reviewed how stimulation parameters and task characteristics moderate effects. By integrating these multifaceted results, our analysis serves the dual purpose of validating theta-tACS as a promising neuromodulation approach and providing an evidence-based framework for optimizing protocols and guiding subsequent research directions, particularly regarding generalizability to clinical populations and synergistic combination with other interventions. Ultimately, these findings not only synthesize current knowledge but also transform isolated observations into a cohesive evidence base to inform both theoretical models of theta-mediated cognition and practical applications in cognitive enhancement.