Subgenual anterior cingulate cortex in major depression: a systematic review and meta-analysis of MRI studies with patients and healthy controls

Abstract

Background

This systematic review aims to synthesize anatomical and functional magnetic resonance imaging (MRI) findings in the subgenual anterior cingulate cortex (sgACC) and associated network among medication-free patients with major depressive disorder (MDD) compared to healthy controls.

Methods

A search for published studies was performed through Ovid (MEDLINE, Embase, APA PsycINFO) on 18 April 2024. The JBI Critical Appraisal Tools and a modified version of the specialist protocol for MRI research were used for the quality assessment. Random-effects and activation likelihood estimation (ALE) meta-analyses were applied to studies reporting numerical and stereotactic data, respectively.

Results

Forty-two publications were reviewed. There was an equal split between the number of studies (n = 5) reporting reduced sgACC GM volume vs. no difference in MDD, with both findings supported by significant ALE clusters with peaks in the right sgACC (p ≤ .0005). Four task-based activity studies reported overactive sgACC in MDD, whereas 6 reported no difference. Both findings were supported by two significant ALE clusters with peaks in the right sgACC (p ≤ .00001). Eleven studies highlighted the global reduced resting-state network coherence of the sgACC in MDD, and reduced coherence with ventromedial prefrontal cortex (p = .00004) and right insula (p = .00002) was replicated across several reports.

Conclusions

Results highlight laterality and voxel-wise differences across subregions of the sgACC, displaying non-uniform GM volume, activity, and network coherence patterns in MDD. High-resolution parcellation of the sgACC should be considered in future MRI investigations.

Introduction

The subgenual anterior cingulate cortex (sgACC) is a caudal region of the ventromedial prefrontal cortex (vmPFC) situated inferior to the genu and rostrum of the corpus callosum in the human brain. The activity of the sgACC has been linked with several specific manifestations of negative emotion [1], notably sadness [2–4] and threat generalization [5, 6], but also with autonomic control, such as maintaining physiological functions including heart rate variability (HRV) and vagal tone [7, 8]. Reduced HRV is generally interpreted as reflecting diminished parasympathetic (vagal) influence and reduced autonomic flexibility, often accompanied by relative sympathetic predominance [9–11]. Studying the anatomy and function of the sgACC has long been of interest to the clinical and neuroimaging communities. It has been proposed that major depressive disorder (MDD), an increasingly prevalent psychiatric illness [12], may be associated with structural and functional abnormalities in the sgACC. More specifically, it is thought that dysfunction of the sgACC broadly contributes to symptoms of negative affect in MDD [4, 13–15].

Reduced gray matter (GM) volume [15–17] and inappropriately elevated activity [4, 13–15, 18] of the sgACC have been consistently reported in individuals with MDD relative to healthy controls (HC). It is also known that the sgACC displays altered functional connectivity (FC) with nodes of the central executive (CEN), reward (RN), and central autonomic (CAN) networks. In this paper, this network-level FC is referred to as network coherence and reflects symptomatic domains of MDD beyond negative affect, such as cognitive impairment, anhedonia, and reduced HRV [19–21]. Beyond blood-oxygen-level-dependent (BOLD)-based activity and FC measures, alterations in regional cerebral blood flow (CBF) within the sgACC have also been reported in MDD, consistent with the region’s dense vascularization and role in integrating affective and autonomic signals [13, 22]. CBF measures provide a complementary physiological index of sgACC function that is closely related to metabolic demand and may capture aspects of regional dysfunction not fully reflected in the BOLD signal alone [23, 24]. Importantly, these structural and functional abnormalities are unlikely to be independent. A limited body of work has examined the structure and function of the sgACC in tandem, suggesting that reductions in GM volume may co-occur with altered regional activity or FC, although findings are heterogeneous and do not support a simple linear relationship between structure and function [13, 22, 25–27]. Whether structural alterations directly constrain or bias sgACC activity, therefore, remains unresolved.

In the context of therapeutics, deep brain stimulation (DBS) of the sgACC (sgACC-DBS) has been shown to exert an antidepressant effect in patients with treatment-resistant depression (TRD) [28]. Moreover, the resting-state FC between the sgACC and dorsolateral prefrontal cortex (dlPFC) is currently used in brain stimulation clinics to guide the selection of target sites for repetitive transcranial magnetic stimulation (rTMS) [29–32]. Notably, converging evidence suggests that sgACC structural and functional abnormalities are most consistently observed in individuals with TRD or recurrent, severe forms of MDD, whereas such alterations are often absent or attenuated in milder or well-controlled depression [14, 15, 25]. This clinical heterogeneity underscores the importance of illness severity and chronicity when interpreting sgACC findings across studies.

As routine evaluation of sgACC structure and function through magnetic resonance imaging (MRI) becomes more clinically relevant, several problems remain unaddressed. First, anatomical findings regarding reduced sgACC GM volume in MDD remain unconsolidated, as anatomical MRI studies have reported heterogeneous results [15]. Large-scale efforts have further highlighted substantial between-study variability, suggesting that effect sizes may depend on clinical and methodological factors [25]. Most of these studies date back to the 1990-2000s, when MRI with lower field and gradient strengths was typically used for research. Second, the consensus around a functionally overactive sgACC in MDD is largely based on task-based positron emission tomography (PET) literature, much of which was published over two decades ago. Advances in MRI now enable voxel-wise assessment of sgACC structure, activity, perfusion, and network coherence within a unified stereotactic framework, facilitating systematic and coordinate-based syntheses across studies. It remains unknown whether more recent resting-state or task-based functional MRI (fMRI) studies support this view. Third, it is unclear which brain regions among those displaying network coherence with the sgACC (such as nodes of the CEN, RN, or CAN) are most relevant to MDD as a potential neurobiological signature of affective symptoms.

The goals of this systematic review and meta-analysis are two-fold. First, it seeks to discern consistently reported abnormalities in the anatomy and function of the sgACC to assess the validity of the prevailing view that sgACC in MDD shows reduced GM volume and inappropriately elevated activity. Second, it aims to identify trends in network coherence abnormalities of the sgACC in MDD and highlight relevant, functionally connected brain regions. The present review, therefore, focuses on MRI-based modalities to enable systematic and coordinate-based synthesis of sgACC findings across studies.

Methods

This systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [33].

Search strategy

A comprehensive search of the published literature was conducted on 18 April 2024 using the Ovid platform (MEDLINE, Embase, APA PsycINFO). The search strategy combined four concepts using the Boolean operator AND: 1) depressive disorder/; 2) [subgenual OR subcallosal OR (Brodman* area ADJ “25” OR “32” OR “12” OR “24”)]; 3) neuroimaging/ OR magnetic resonance imaging/; and 4) (health* OR control*). No search limits were applied with respect to publication year, language, study design, or population. Duplicate records across databases were removed using the Ovid automated deduplication function. A supplementary manual search of reference lists from key review articles was performed using PubMed and Scopus on 15 May 2024. The full search strategy is provided in Supplementary Methods. Two authors (R.S. and V.B.) independently performed the search.

Eligibility criteria and screening process

Original research studies, corrigenda and errata, case reports, and case series were eligible for inclusion if they involved a comparison of adults (18-65 years old) with MDD or TRD vs. HC. Eligible studies were required to include at least one MDD/TRD group and at least one HC comparator group. When specified, TRD was defined according to study-specific criteria that were largely consistent with standard definitions, most commonly failure to respond to two or more adequate antidepressant trials [34]. Studies were excluded if participants had depression in the context of other medical, neurological, or psychiatric comorbidities, with the exception of anxiety disorders due to their high co-occurrence with depression. Participants with depression following head injury, postpartum depression, or depression in remission were excluded to reduce clinical heterogeneity associated with distinct physiological and neuroendocrine states. Participants were required to be unmedicated at the time of scan acquisition [35]; medication-naïve status was not required.

Eligible studies were required to report MRI outcomes for the sgACC as a region of interest at baseline, prior to any intervention or experimental manipulation. This review was intentionally restricted to MRI-based modalities to ensure methodological homogeneity and compatibility with coordinate-based meta-analysis (CBMA). Included modalities comprised anatomical MRI and advanced functional MRI techniques, including echo-planar imaging (EPI) and perfusion arterial spin labelling (ASL). ASL was included as an MRI perfusion sequence that provides voxel-wise estimates of CBF in standard stereotactic space. MRI outcomes of interest included anatomical measures (GM volume or cortical thickness), activity metrics (e.g., BOLD signal or perfusion-derived CBF), and network coherence metrics (e.g., FC) within sgACC voxels. Molecular imaging modalities (e.g., PET) and diffusion MRI studies were considered beyond the scope of this MRI-focused synthesis due to differences in physiological signals, analytical pipelines, and limited compatibility with CBMA procedures.

Studies involving pediatric ( < 18 years) and older adult ( > 65 years) populations were excluded, as well as non-English language studies. Reviews, systematic reviews, meta-analyses, books, book chapters, theses and dissertations, conference abstracts, and editorials/letters were also excluded. A detailed list of eligibility criteria is provided in Supplementary Methods. Title/abstract and full-text screening was independently performed by two authors (R.S. and V.B.), with disagreements resolved by consensus with a third party (I.D.).

Variable extraction

Full texts of included studies were examined, and data were independently extracted by two authors (I.D. and V.B.), with discrepancies resolved by consensus with a third author (R.S.). Extracted variables included study characteristics (title, authors, year, country); participant characteristics (sample sizes, diagnoses, age, sex, and baseline depression severity measures); MRI methodology (modality, resting-state or task-based condition, scanner and acquisition parameters, and outcomes of interest); and study outcomes such as statistically significant between-group differences in sgACC anatomy, activity, and network coherence; GM volume values for each group when available; and Montreal Neurological Institute (MNI) coordinates for left and right sgACC when available.

Quality assessment

Study quality was assessed using the Joanna Briggs Institute (JBI) Checklist for Analytical Cross-Sectional Studies [36]. MRI methodological quality was additionally evaluated using a modified version of a specialist protocol for MRI research [37, 38] based on published best-practice guidelines for neuroimaging data analysis and sharing [39]. Quality assessment (QA) was performed independently by two authors (general QA: R.S. and I.T.; MRI QA: I.T. and N.R.), with disagreements resolved by consensus with a third author (I.D.).

Classical meta-analyses

Classical meta-analyses were performed using JASP v.0.18.3 [40]. Separate univariate random-effects meta-analyses were conducted for age, standardized depression percent of maximum possible (POMP) scores [41], and GM volume metrics across anatomical studies reporting numerical values for the left and right sgACC. POMP scores rescale depression severity to a common 0-100 metric to enable comparison across instruments. Meta-analyses were conducted separately for left, right, and total sgACC volume. For each outcome, the dependent variable was the study-level standardized mean difference (Cohen’s d) between MDD/TRD and HC groups. An intercept-only random-effects model was specified, estimating a pooled effect size while allowing true effects to vary across studies. Heterogeneity was estimated using the restricted maximum-likelihood estimator and the Q-test. Funnel plot asymmetry was checked using the rank correlation test and Egger’s tests, with the standard error of the effect size as a predictor. To assess whether between-study heterogeneity in sgACC GM volume could be related to methodological differences across studies, exploratory random-effects meta-regressions were conducted using publication year and scanner field strength ( ≤ 1.5 T vs ≥3 T) as moderators. Moderator analyses were restricted to total sgACC GM volume and were estimated using restricted maximum likelihood.

Coordinate-based meta-analyses: activation likelihood estimation

CBMA were conducted using the activation likelihood estimation (ALE) method implemented in GingerALE v.3.0.2 [42]. Coordinates reported in Talairach space were converted to MNI space using the Lancaster transformation (icbm2tal) integrated into GingerALE [43]. Separate ALE analyses were performed for studies subgrouped into those reporting increases, decreases, or no differences in sgACC GM volume, activity, or network coherence between MDD and HC. Analyses were restricted to datasets reporting left or right sgACC coordinates and used a 2 voxel resolution. Statistical significance was assessed using 1 000 permutations with a cluster-forming threshold of p < .001 (uncorrected). To enhance test sensitivity for false positives [44], significance was corrected with a cluster-level family-wise error (FWE) threshold of p < .05 [45], which is a gold-standard method employed in other meta-analytic ALE studies [46–48]. ALE maps were visualized using Mango v.4.1 [49] overlaid onto the high-resolution T1-weighted MNI template publicly available on the GingerALE website (http://brainmap.org/ale/; courtesy of Simon Eickhoff).

Assessment of robustness against noise studies: Fail-Safe N

To assess robustness against publication bias, an adapted Fail-Safe N (FSN) was computed for each ALE run [50]. Adapted FSN quantifies the number of noise experiments with randomly distributed foci across the brain, which is necessary to alter the ALE statistical thresholds to the extent that a significant cluster becomes non-significant (i.e., fails to converge). To determine the FSN, noise experiments with sample sizes and numbers of foci matching those of the original studies were generated using an open-source R script https://github.com/NeuroStat/GenerateNull). Noise foci were incrementally added to the original dataset until significant clusters failed to converge. As the minimum amount of injected noise validated for behavioral studies (5k + 10) is expected to be too high for MRI-based CBMA, we used k/2 as the lower bound and 3k as the upper bound, similar to previous literature [51, 52].

Results

Search results

The search of the three Ovid databases identified 2 057 records (Fig. 1). After deduplication, 1 154 records remained, and 532 of them were deemed ineligible based on the publication type alone. These records were excluded, and the remaining 622 underwent screening. A total of 35 publications were eligible for the systematic review. A supplementary search of reference lists identified 7 additional eligible articles, bringing the final number of included studies to 42 (Supplementary Fig. S1).

Fig. 1. PRISMA flow diagram of study selection process.

PRISMA flow chart for systematic review synthesizing evidence from magnetic resonance imaging studies examining baseline sgACC structure and function among participants with MDD vs. HC.

Basic study characteristics

Among the reviewed studies, 11 studied the anatomy of the sgACC, while 33 studied its function (Supplementary Fig. S2 and S3a). Among anatomical studies, 10 examined GM volume in sgACC voxels using morphometric techniques [22, 53–61], 1 examined GM density [59], and 1 examined CT and surface area [62]. Among functional studies, 15 articles studied the activity of the sgACC, and 20 articles studied its network coherence with other brain regions. Twenty functional studies acquired scans during the resting state, whereas 14 were task-based.

Quality assessment

QA results for each study are provided in Supplementary Tables S1-S3. The quality of the research design was very high: all 42 studies (100%) clearly defined their inclusion criteria, reliably assessed symptoms, and identified confounders and respective mitigation strategies; 97.6% described the study setting and participants in detail and used appropriate statistical analyses; and 92.9% measured outcomes in a valid and reliable way. In terms of MRI methods and reporting, the QA score for all studies was ≥14 (out of 20), with the exception of one study published in 1997 [22]. The average QA score was 16.9 ± 1.6.

MRI methodology

Acquisition parameters and metrics of interest for anatomical and functional studies are presented in Supplementary Tables S4-S8 and Supplementary Fig. S3b-d. Among anatomical studies, T1-weighted scans were acquired from 1 T head-only scanner in 1 study [58], 1.5 T scanners in 6 studies [22, 53–57], and 3 T scanners in 4 studies (Supplementary Fig. S4a,d and S5a) [59–62]. Among resting-state studies using advanced fMRI techniques, 1 study acquired a perfusion-weighted pulsed ASL scan from a 1.5 T scanner [63], and the rest acquired T2*-weighted EPI scans from 1.5 T scanners in 2 studies [64, 65] and from 3 T scanners in 17 studies (Supplementary Fig. S4b,e and S5b) [61, 66–81]. Among task-based functional studies, 4 studies acquired T2*-weighted EPI scans from 1.5 T scanners [53, 82–84], 8 from 3 T scanners [70, 85–91], and 2 from 4 T scanners (Supplementary Fig. S4c,f and S5c) [92, 93]. One study [70] used a multi-band multi-echo sequence for both the resting-state and task-based acquisition, whereas all other studies were single-echo. Task-based studies used paradigms probing emotion processing [82, 84–86, 88, 89, 91, 93], cognitive control [53, 70, 84], attention [88, 90], perceptual processing [85, 90], reward selection, anticipation, and feedback [70, 92], memory [84, 89], pain [83], attachment [87], and personal relevance [87, 89] (Supplementary Fig. S5d).

Participant characteristics

Collectively, all 42 studies enrolled 1 485 patients with MDD and 1 479 HC, and data from 1 369 MDD and 1 442 HC were included in their respective MRI analyses (Supplementary Fig. S6a and Table S9). Thirty-seven studies enrolled participants of both sexes, 4 studies enrolled females only [53, 57, 58, 82], and 1 study did not specify the sex of participants [85]. Among patients, the severity of MDD ranged from moderate to very severe (Supplementary Fig. S6b,c). MDD and HC did not differ in mean age (MDD, M_G = 35.2 ± 5.7, range: 22.75-44.6; HC, M_G = 34.1 ± 5.1, range: 23.0-47.2), which was supported by the random-effects model (z = 1.77, p = .08, Cohen’s d_avg = 0.08, 95% CI -0.01 to 0.16, n = 41), as well as the absence of heterogeneity (Q[40] = 47.21, p = .20, τ² = 0.01, I² = 16.04) and funnel plot asymmetry as indicated by rank correlation test (p = .27) and Egger’s test (p = .13) (Supplementary Fig. S6d-h, S7-S8 and Table S10). The average standardized depression score was 45.9% ± 9.6% (n = 38 studies), which is equivalent to a total score of 23.9 ± 5.0 on the 17-Item Hamilton Depression Rating Scale (HAM-D-17), indicating very severe depression (Supplementary Fig. S6i-m and Table S10).

Anatomy of the sgACC

Among 11 anatomical studies, 5 reported decreased sgACC GM volume in MDD compared to HC [22, 53, 55, 57, 60], and 5 reported no significant difference (Fig. 2a and Supplementary Table S11) [54, 56, 58, 59, 61]. One study by Meier et al. (2016) reported a decreased surface area of the sgACC but no difference in cortical thickness [62]. Qualitatively, anatomical studies using both lower-field ( ≤ 1.5 T) and higher-field (3 T) scanners reported mixed sgACC GM volume findings (decreased and null), with a modest tendency toward null results in 3 T studies. Five out of 10 studies provided numerical values for GM volume metrics in MDD and HC groups and were included in the classical meta-analysis. The random-effects model did not support a significant difference in GM volume between MDD and HC, neither for the left (z = -0.81, p = .42, Cohen’s d_avg = -0.17, 95% CI -0.60 to 0.25) nor for the right sgACC (z = -1.56, p = .12, Cohen’s d_avg = -0.27, 95% CI -0.61 to 0.07) (Supplementary Fig. S9-S12). There was no significant heterogeneity (left sgACC, Q[4] = 9.34, p = .05, τ² = 0.13, I² = 57.29; right sgACC, Q[4] = 5.89, p = .21, τ² = 0.05, I² = 33.45) or funnel plot asymmetry as revealed by the rank correlation (left sgACC, p = .48; right sgACC, p = 1.00) and Egger’s tests (left sgACC, p = .39; right sgACC, p = .50). For the total sgACC volume, the random-effects model also did not support a significance difference between MDD and HC (z = -1.09, p = .28, Cohen’s d_avg = -0.29, 95% CI -0.82 to 0.24), although heterogeneity was significant (Q[4] = 14.30, p = .006, τ² = 0.26, I² = 72.20) (Supplementary Fig. S13-14). To assess whether this heterogeneity could be attributed to methodological differences across studies, exploratory random-effects meta-regressions were conducted. Neither publication year (β = −0.045, SE = 0.064, z = −0.71, p = .48) nor scanner field strength ( ≤ 1.5 T vs ≥3 T; Q_M = 1.32, p = .25) significantly moderated sgACC GM volume effect sizes, and substantial residual heterogeneity remained in both models (I² = 75.1% and 69.6%, respectively). These findings suggest that variability in sgACC volumetric findings is unlikely to be explained solely by broad differences in MRI acquisition era or field strength. There was no funnel plot asymmetry, as revealed by the rank correlation (p = .82) and Egger’s tests (p = .40).

Fig. 2. Structural, functional, and connectivity alterations of the sgACC in major depressive disorder.

(A-C) Distribution of main results for the sgACC (A) GM volume (n = 10 studies), (B) resting-state fMRI BOLD signal (n = 3 studies), and (C) task-based fMRI BOLD signal (n = 12 studies) in participants with MDD compared to HC. (D-E) Directional change in (D) resting-state (n = 17 studies) and (E) task-based (n = 3 studies) network coherence of the sgACC with other brain regions. (F) Brain regions displaying decreased (left, blue nodes) and increased (right, orange nodes) resting-state network coherence with the sgACC (n = 17 studies). CBMp = Cerebellum, posterior lobe; dlPFC = Dorsolateral Prefrontal Cortex; dmOFC = Dorsomedial Orbitofrontal Cortex; dmPFC = Dorsomedial Prefrontal Cortex; HPC = Hippocampus; Ins = Insula; L = Left; lOFC = Lateral Orbitofrontal Cortex; MTG = Middle Temporal Gyrus; PCC = Posterior Cingulate Cortex; PCu = Precuneus; PHG = Parahippocampal Gyrus; pgACC = Perigenual Anterior Cingulate Cortex; pIPL = Inferior Parietal Lobule, posterior; R = Right; sgACC = Subgenual Anterior Cingulate Cortex; vlPFC = Ventrolateral Prefrontal Cortex; vmPFC = Ventromedial Prefrontal Cortex. Created with BioRender.com (Panel F), RRID:SCR_018361.

CBMA identified one significant cluster, with peak coordinates close to the midline but in the right sgACC, with decreased GM volume in MDD compared to HC based on 2 studies (58 participants, 2 foci) (Fig. 3a and Table 1) [53, 57]. Similarly, CBMA revealed one significant, more inferior cluster spanning bilateral sgACC, with a peak in the right sgACC, with no between-group differences in GM volume based on 1 study (42 participants, 2 foci) (Fig. 3b) [61].

Fig. 3. Convergent anatomical alterations of the sgACC in major depressive disorder identified using ALE meta-analysis.

ALE results for anatomical MRI studies reporting convergence of sgACC coordinates with (A) decreased GM volume (x = 2, y = 30, z = -2) and (B) no difference in GM volume (x = 0, y = 26, z = -10) in MDD vs. HC.

Table 1.

ALE-derived clusters of sgACC structural, functional, and connectivity differences in major depressive disorder.

	Cluster #, Label	Cluster Volume (mm3)	Laterality	BA	MNI Coordinates			ALE max.	p-value	Z-score	FSN, k
					X	Y	Z
GM Volume
Decreased	1, sgACC	1624	R	BA24	2	30	-2	0.009	.000004	4.48	5
No difference	1, sgACC	1816	R	BA11	0	26	-10	0.004	.0005	3.28	2
Activity
Resting-state studies
Decreased	1, sgACC	1888	R	-	20	9	-13	0.01	.00002	4.17	NAa
Task-based studies
Decreased	1, sgACC	664	L	-	-2	12	2	0.009	.000008	4.32	1b
Increased	1, sgACC	1056	R	BA24	2	26	-2	0.009	.00001	4.23	3
	2, sgACC	1024	R	BA25	4	18	-18	0.009	.000004	4.47	3
No difference	1, sgACC	2056	R	-	2	14	-10	0.02	< .000001	6.34	12
	2, sgACC	384	R	-	8	-14	-24	0.009	.00001	4.21	1b
Network Coherence
Resting-state studies
Decreased	1, sgACC	4808	R	BA32	6	34	-4	0.04	< .000001	7.94	>90c
			L	BA24	-6	34	-4	0.03	< .000001	7.75	>90c
Decreased sgACC-vmPFC	1, sgACC	1096	L	BA11	-10	28	-12	0.01	.000004	4.46	1b
	1, vmPFC		L	BA11	-12	33	-21	0.01	.00004	3.96	2
Decreased sgACC-Ins	1, Ins	768	R	BA44	36	9	9	0.01	.00002	4.12	1b
	2, sgACC	656	R	BA24	6	28	-4	0.009	.00003	3.98	1b
Increased	1, sgACC	2296	L	BA11	-10	28	-12	0.01	< .000001	5.19	6
			R	BA11	0	26	-12	0.01	.000004	4.47	6
No difference	1, sgACC	2032	L	BA32	-4	39	-2	0.01	.000008	4.33	3
			R	BA32	5	41	6	0.009	.00002	4.08	1b
Task-based studies
No difference	1, sgACC	1032	L	BA32	-6	30	-8	0.009	.00001	4.23	1b

Clusters and peak coordinates identified by the activation likelihood estimation meta-analysis of anatomical andfunctional MRI studies examining the sgACC gray matter volume, activity, and network coherence differencesbetween MDD and HC.

ALE activation likelihood estimation, BA brodmann area, FSN fail-safe n, Ins insula, k = number of studies in the FSN approach, L left, NA not applied, R right, sgACC subgenual anterior cingulate cortex, vmPFC ventromedial prefrontal cortex.

Only ALE subanalyses with significant clusters are reported.

^aFSN not applied because ALE results are based on the contribution of one study.

^bFSN ≤ k/2, indicating that results may not be robust to bias due to missing null (non-significant) studies in the meta-analysis.

^cFSN > 10k.

Activity of the sgACC

Among 15 activity studies, 3 were resting-state [63, 71, 72] and 12 were task-based [53, 70, 82–84, 86–90, 92, 93]. At rest, increased perfusion [63] but decreased fractional amplitude of low-frequency fluctuations (fALFF) [72] and brain entropy (BEN) [71] were identified in the sgACC of MDD participants compared to HC (Fig. 2b and Supplementary Table S12). Because perfusion-weighted ASL, fALFF, and BEN index distinct physiological properties, these resting-state metrics were interpreted descriptively rather than pooled quantitatively. The direction of effects appeared to track the physiological metric and experimental context rather than scanner era. Across functional tasks, 4 studies identified overactive sgACC in MDD during emotional face viewing [82], painful heat stimulation [83], attachment-related picture viewing [87], and reward-based incentive flanker task [70] in MDD (Fig. 2c and Supplementary Table S13). CBMA revealed the presence of task-based overactivity in 2 clusters with a peak in the right sgACC based on 2 studies (66 participants, 2 foci) (Fig. 4a-d and Table 1) [83, 87]. Two studies identified underactive sgACC in MDD during the reward-based wheel of Fortune task [92] and perception-based detection task during rapid serial visual presentation [90], whereas 6 studies reported no difference in the BOLD response during the Stroop [53], emotional word stimulus [86], selective attention [88], emotional autobiographical memory [89], incidental emotion processing and n-back working memory [84], and emotional face viewing tasks [93]. Two clusters with no significant difference between MDD and HC were revealed among task-based studies with a peak in the right sgACC based on 5 studies (190 participants, 6 foci) [53, 86, 88, 89, 93]. Variability in sgACC activation was not confined to early low-field acquisitions and may relate more to task demands and analytic choices than scanner era alone. As an exploratory sensitivity analysis, we stratified task-based activation studies by task family. Affective paradigms showed a mixed pattern with both hyperactivation (3/6) and null effects (3/6), whereas cognitive paradigms more commonly reported no group differences (2/3), and reward paradigms showed bidirectional findings (1 increased, 1 decreased).

Fig. 4. Convergent functional alterations of the sgACC in major depressive disorder identified using ALE meta-analysis.

ALE results for functional MRI studies reporting convergence of sgACC coordinates with (A) decreased task-based activity in the sgACC, (B) increased task-based activity in the sgACC, (C) no difference in task-based activity within the sgACC, (D) decreased resting-state activity in the sgACC, (E) decreased resting-state network coherence of the sgACC with other brain regions, (F) increased resting-state network coherence of the sgACC with other brain regions, (G) no difference in resting-state network coherence of the sgACC with other brain regions, and (H) no difference in task-based network coherence of the sgACC with other brain regions in MDD vs. HC.

Network coherence of the sgACC

Among 20 network coherence studies that examined FC, 17 were resting-state [61, 64–70, 73–81] and 3 were task-based [85, 90, 91]. Abnormally decreased resting-state network coherence of the sgACC in MDD was reported by 11 studies [61, 64, 66–69, 74, 75, 78, 80, 81], notably with the dlPFC and vmPFC bilaterally, insula in the right hemisphere, perigenual and posterior divisions of the cingulate gyrus, right parahippocampal gyrus, left middle temporal gyrus, posterior nodes of the default mode network (DMN), occipital cortex, and regions of the posterior cerebellum (Fig. 2d,f and Supplementary Table S12). CBMA confirmed decreased resting-state network coherence in one cluster with two local maxima in bilateral sgACC based on 9 studies (755 participants, 15 foci) (Fig. 4e and Table 1) [61, 64, 66–68, 74, 75, 80, 81]. Decreased network coherence of the sgACC with bilateral vmPFC and right insula in MDD was reported by several of these studies [75, 80, 81], and CBMA results support the involvement of the left vmPFC (170 participants, 4 foci) and right insula (151 participant, 3 foci), respectively (Fig. 5). This global pattern of dysconnectivity was further highlighted in a graph theory study by Yun et al. (2023), which revealed decreased centrality values for sgACC nodes in young adults with MDD [69].

Fig. 5. Convergent decreases in sgACC resting-state functional connectivity in major depressive disorder identified using ALE meta-analysis.

ALE results for studies reporting decreased resting-state network coherence between (A) left sgACC (x = -10, y = 28, z = -12) and left vmPFC (x = -12, y = 33, z = -21) and (B) right sgACC (x = 6, y = 28, z = -4) and right insula (x = 36, y = 9, z = 9). ALE = Activation Likelihood Estimation; sgACC = Subgenual Anterior Cingulate Cortex; vmPFC = Ventromedial Prefrontal Cortex.

Four studies identified abnormally increased resting-state network coherence of the sgACC with dorsolateral, dorsomedial, and ventrolateral regions of the left prefrontal cortex, orbitofrontal cortex, bilateral hippocampi, and regions of the right parietal lobe, such as precuneus and temporoparietal junction (Fig. 2f) [70, 73, 77, 80]. CBMA confirmed increased resting-state network coherence in one cluster with two peaks in bilateral sgACC (182 participants, 2 foci) (Fig. 4f and Table 1) [73, 80]. Notably, Cheng et al. (2021) [80] reported both increased and decreased resting-state network coherence of the sgACC in the same sample of MDD participants but with different brain regions (Supplementary Table S12). One study that examined machine learning model performance reported abnormal FC of the sgACC with 11 different nodes, which was discriminative of MDD [65]. Two studies reported no difference in resting-state FC of the sgACC between MDD and HC [76, 79], which was also revealed by CBMA in one cluster with two peaks in bilateral sgACC based only on 1 study (60 participants, 2 foci) (Fig. 4g and Table 1) [79]. Resting-state connectivity findings were largely derived from 3 T studies, and heterogeneous results persisted within the high-field era, indicating that discrepancies were not uniquely attributable to early low-field imaging.

Across functional tasks, 1 study reported increased FC between the sgACC and bilateral amygdalae during the emotional face matching paradigm [91], and 2 studies reported no differences in FC of the sgACC between MDD and HC during the song listening task [85] and detection task during rapid serial visual presentation (Fig. 2e and Supplementary Table S13) [90]. Only studies with no between-group difference provided the stereotactic data, and CBMA showed one converging cluster with a peak in the left sgACC based on 2 studies (97 participants, 2 foci) (Fig. 4h and Table 1) [85, 90]. Task-based connectivity evidence was limited to high-field studies.

Discussion

This systematic review offers a snapshot of anatomical and functional results reported for the sgACC between medication-free participants with MDD and HC prior to any experimental manipulation. Studies were subgrouped into three categories based on the direction of reported between-group difference (i.e., decreased, increased, no difference). Classical random-effects meta-analysis was performed for studies reporting values for GM volume, and CBMA was run to identify convergent clusters within the sgACC across studies that provided stereotactic data. The key findings of the meta-analysis highlight the differential involvement of sgACC subregions in MDD pathophysiology, with different clusters exhibiting distinct directional between-group differences. Specifically, loci reported as sgACC in the primary literature most consistently implicated ventral/caudal cingulate territories (BA24/25), with decreased GM volume and elevated task-evoked activity in MDD compared to HC, whereas more posteroinferior ventromedial prefrontal loci, often labelled as orbitofrontal (BA11) in the original reports, showed no consistent between-group differences. Moreover, reduced resting-state network coherence of sgACC subregions proximate to BA24 and BA32, particularly with the vmPFC and right insula, was replicated across several studies, indicating disrupted FC of the sgACC-associated network in MDD.

In the present review, we use the term subgenual anterior cingulate cortex (sgACC) to refer primarily to BA25 and adjacent ventral/caudal portions of BA24 (and ventral BA32 where reported as subgenual/ventral ACC in the source literature). We acknowledge, however, that anatomical labelling of ventromedial cingulate and adjacent vmPFC/OFC territories is not uniform across studies, and that coordinate-based synthesis inherits this heterogeneity from original ROI definitions, spatial normalization, and smoothing (particularly in older studies acquired at lower field strengths). Accordingly, loci reported as “sgACC” in primary studies may extend into adjacent perigenual or ventromedial/orbitofrontal regions, and such findings should be interpreted as reflecting ventromedial cingulate/vmPFC heterogeneity rather than definitive subregional boundaries of sgACC.

Hemispheric lateralization has long been used as a framework for understanding depression-related circuitry, with influential models proposing relatively reduced left-hemisphere/approach-related activity and relatively greater right-hemisphere/withdrawal-related processing [94–96]. Within mood disorders, an altered balance between left and right sgACC function has also been proposed as a contributor to affective and autonomic/neuroendocrine features of depression [15]. This framework provides one interpretive lens for lateralized patterns observed across neuroimaging modalities, while acknowledging that the asymmetry effects are not uniform across studies and methods [96, 97].

Gray matter volume alterations show anterior localization with a right-sided peak in the sgACC

GM volume of the sgACC is frequently reported to be decreased in participants with MDD [16, 22, 98]. For instance, a meta-analysis [99] of volumetric MRI studies with patients with MDD and bipolar disorder showed the existence of a volumetric decrease in the GM of left sgACC, whereas anatomical changes in the right sgACC were less robust. This meta-analysis did not exclude studies based on medication status, and it is well-known that treatment with antidepressants has neuroprotective and neuroplastic effects that can potentially drive a GM volume increase in certain brain structures, including those of the prefrontal cortex, such as the sgACC [100]. Our CBMA identified two clusters: one cluster in the right sgACC (BA24) with decreased GM volume in MDD and another bilateral cluster, peaking in the right sgACC (BA11), showing no significant between-group differences. The right-sided peak of the most spatially convergent GM locus can be interpreted within broader hemispheric asymmetry frameworks in depression. Classic lateralization models propose an imbalance between left-sided approach/regulatory processes and right-sided withdrawal/negative affect and autonomic responding, and this lens has historically informed neuromodulation rationales in MDD [15, 95]. Applied to ventromedial cingulate circuitry, this suggests that a right-leaning sgACC pattern may plausibly relate to the salience of negative affective and autonomic/neuroendocrine features of depression. At the same time, structural asymmetry effects are not uniform across neuroimaging studies and analytic pipelines, and we therefore treat this hemispheric emphasis as a contextual, hypothesis-generating interpretive frame rather than definitive evidence of intrinsic right-versus-left specialization [96, 97].

These results suggest potential lateralized structural changes in MDD, which align with findings from a sgACC-DBS study [101], where the right sgACC was shown to play a unique role in predicting treatment response. Moreover, a reduction in GM volume appears to be constrained to more anterior subregions of the sgACC corresponding to BA24, whereas posteroinferior subregions in the orbitofrontal areas (BA11) likely display no such differences. The existence of sgACC subregions with no volumetric differences between MDD and HC is also supported by the literature – for example, a retrospective analysis of the GM volume from a TRD cohort that underwent sgACC-DBS [102] indicated no significant volumetric differences between responders and non-responders to DBS, highlighting the limitations of GM volume alone in predicting clinical outcomes. Indeed, while the right sgACC may show anatomical differences in MDD, volumetric data alone is insufficient to explain variability in DBS outcomes, and our findings need to be corroborated through multi-modal neuroimaging with higher field strengths (of at least 3 T) and larger sample sizes, as well as investigation of the FC for the identified subregions.

Overactivity of the sgACC in MDD is lateralized and task-dependent

At rest, MDD participants were found to have increased perfusion in the sgACC but decreased fALFF and BEN relative to HC [63, 71, 72]. These findings may be explained by our limited sample size, as only 3 of the 15 included activity studies were resting-state. Moreover, perfusion (measured via ASL), fALFF, and BEN capture different processes associated with brain activity – namely, vascular activity [103], neural oscillatory behavior [104], and irregularity of signal patterns [105, 106], respectively. Since these metrics assess distinct processes, it is expected that they might show divergent patterns when comparing MDD and HC.

For task-based analyses, there was marked heterogeneity in sgACC activity, with both over- and underactive responses observed across different experimental contexts. This variability appeared to be strongly task-dependent rather than contradictory across studies. In total, 4 studies reported overactive sgACC during tasks with a prominent affective component. Two studies reported sgACC underactivity; notably, these paradigms did not involve affective processing but instead probed reward-related and perceptual processes. CBMA confirmed sgACC overactivity during painful heat stimulation and attachment-related picture viewing [83, 87], with two significant clusters peaking in the right sgACC and corresponding to BA24 and 25. Both paradigms place substantial demands on autonomic and interoceptive processing, consistent with the role of the sgACC in context-dependent modulation of bodily arousal states [107].

For tasks that displayed no difference in sgACC activity between MDD and HC [53, 86, 88, 89, 93], CBMA identified clusters spanning both hemispheres and peaking in the right sgACC, located posteroinferiorly relative to the overactive clusters. Most of these studies employed paradigms with a significant cognitive component, although some passive affective tasks, such as emotional face viewing, were also included [93]. Together, these findings suggest that sgACC task-evoked activity in MDD is highly sensitive to task demands and motivational or autonomic salience, rather than reflecting a uniform direction of abnormality across paradigms. However, the limited number of studies within each task category constrains formal task-specific comparisons and underscores the need for more systematic task-based investigations targeting the sgACC.

Overall, these findings suggest strong context-dependence in sgACC activity patterns across paradigms, alongside a recurrent right-leaning emphasis in reported peak loci across the available studies. As noted above, hemispheric asymmetry frameworks provide one lens for interpreting this pattern (i.e., potential links to negative affect and autonomic/interoceptive responding), but task-evoked results are especially sensitive to paradigm design, preprocessing choices, and analytic strategy, and therefore should not be taken to imply a uniform direction of sgACC abnormality across contexts. Within this interpretive framework, while both the right and left sgACC contribute to mood regulation, the right sgACC is more often linked to negative affect and autonomic responding [13, 108], whereas the left sgACC is more commonly discussed in relation to regulatory processes such as cognitive reappraisal [14, 109]. More broadly, these findings support a circuit-level view in which sgACC abnormalities reflect network dysregulation rather than a strictly regional effect. For example, in a lesion network mapping analysis of five independent datasets with different lesion etiologies [110], lesion locations associated with MDD did not correspond to a single brain region, such as the sgACC, but instead mapped onto a specific brain circuit. Accordingly, identifying paradigms that reliably and meaningfully engage sgACC-linked circuits, and demonstrating adequate test–retest reliability while capturing quantifiable behavioral indices, remains an important priority for future work.

sgACC displays a global dysconnectivity pattern with resting-state and task-dependent functional networks in MDD

Resting-state network coherence of the sgACC is a core component of MDD pathophysiology. This has been consistently highlighted by large-scale systematic reviews and meta-analyses at the whole-brain level [111, 112]. Studies have repeatedly shown abnormal FC of this region in patients with MDD compared with HC [113, 114], correlating symptom severity [115, 116] and duration [117]. Further, the sgACC has been emphasized as a hub with decreased whole-brain FC in MDD [118]. This is supported by our findings, which demonstrate a widespread pattern of reduced resting-state network coherence with multiple anterior and posterior regions across the brain, based on the results of 11 studies. Some of these regions identified by our review, notably the nodes of the DMN such as the posterior cingulate cortex, precuneus, and inferior parietal lobule, exhibit an immediate reduction in their activity when sgACC is stimulated with DBS in the MRI scanner (i.e., DBS-ON state), and these acute activity changes have been show to predict individual long-term antidepressant response [119]. Via CBMA, the cluster identified for a subset of studies reporting reduced resting-state network coherence of the sgACC in MDD was also the largest (4808 mm³, with two peaks in the right and left sgACC) and most robust (FSN > 90). Reduced resting-state network coherence of the sgACC with vmPFC and right insula is particularly noteworthy and may be more relevant to MDD symptoms than coherence with other brain regions, as highlighted by several studies included in our systematic review [75, 80, 81].

The vmPFC is a large and heterogeneous region that forms part of the ventromedial affect and reward networks (VMN) implicated in MDD, and its subregions activate in response to negative stimuli, as well as stimuli associated with reward and loss [19, 21]. The sgACC itself is part of the vmPFC and VMN, which serves as a central node involved in affect [120]. Studies suggest that its decreased network coherence with other VMN regions, including other parts of the vmPFC, is likely associated with anhedonia [121–123], although some studies specifically establish this association with context-dependent task-based network coherence involving pleasurable stimuli rather than intrinsic resting-state network coherence [121]. The sgACC also receives decreased top-down inhibition from the vmPFC in MDD [1], which drives inappropriately elevated negative affect, emotional arousal, and threat generalization.

The right insula is also involved in the sgACC dysconnectivity pattern during the MDD resting state. Interestingly, lower resting-state FC between the sgACC and insula in MDD participants predicts worse clinical outcomes and a lower reduction in depressive symptoms [124], suggesting that the baseline synchrony between the two regions is an important prognostic marker of treatment response. The insula is a region known to link interoception with emotional feedback, facilitating interactions between cortical and limbic brain networks to integrate signals from both internal and external environments [125]. The FC between the sgACC and insula may integrate the affiliative value with interoceptive information, possibly refining emotional states and assessing social threat salience [124, 126]. Consequently, the reduced FC may result in suboptimal interoceptive and emotional integration, rendering individuals more vulnerable to overgeneralizing social feelings [127].

Increased resting-state network coherence of the sgACC, on the other hand, potentially underpins rumination in MDD [128]. This pattern predominated in 4 studies reporting increased sgACC coherence with regions of the frontal and parietal cortex, as well as bilateral hippocampi [70, 73, 77, 80]. With supporting CBMA, these findings indicate a tendency towards resting-state hyperconnectivity in individuals with MDD with higher-order nodes of the CEN and DMN, throughout both hemispheres. One study also reported increased FC between the sgACC and bilateral amygdalae [91] during the emotional face matching paradigm, which reflects inappropriately elevated network coherence between regions involved in emotional processes in MDD. This paradigm is known to reliably activate the amygdala [129, 130], and it can be leveraged further to study the network coherence of the VMN and limbic regions.

Identified sgACC subregions and associated networks as candidate targets for invasive and non-invasive neuromodulation

Given that the sgACC displays reduced network coherence with the bilateral vmPFC and right insula in MDD, as evidenced by our meta-analysis, these regions may represent candidate nodes within depression-relevant circuits that warrant consideration in future neuromodulation research. Importantly, the present findings are cross-sectional and associative, and therefore should be interpreted as hypothesis-generating rather than establishing causal treatment targets. Moreover, reduced sgACC-vmPFC or sgACC-insula coherence should not be interpreted as implying that increasing connectivity is universally optimal; rather, these findings motivate a baseline-dependent framework in which neuromodulation may aim to normalize aberrant coupling or restore network flexibility, depending on whether an individual shows hypo- vs. hyper-coupling at baseline. In invasive neuromodulation such as DBS, electrodes can be implanted in either sgACC-adjacent regions or functionally connected network nodes. While direct stimulation of the sgACC remains a primary neurosurgical strategy, the sgACC subregions and peak coordinates identified here may help inform target refinement or network-level targeting strategies, rather than serving as definitive stimulation sites. For non-invasive neuromodulation, the vmFPC and insula are theoretically accessible targets; however, stimulation of these regions is technically challenging due to their depth and anatomical location within the longitudinal and lateral fissures, respectively. Standard figure-eight coils preferentially stimulate superficial cortical regions such as the dlPFC, the most common and clinically approved rTMS target for MDD. Targeting deeper regions, including the vmPFC or insula, typically requires specialized coils (e.g., H-coils) [131]. Moreover, both the vmPFC and insula are anatomically extensive, underscoring the importance of fMRI-guided neuronavigation and individualized targeting strategies to identify functionally relevant subregions [132].

Emerging rTMS protocols for MDD [133–135] already leverage resting-state FC between the sgACC and dlPFC to guide stimulation, with dlPFC sites showing stronger negative coupling to the sgACC identified as optimal targets. A similar connectivity-informed framework could be explored for the vmPFC and insula, using their resting-state FC with the sgACC to identify candidate stimulation sites. However, whether reduced sgACC-vmPFC or sgACC-insula coherence represents a modifiable mechanism underlying symptom improvement remains unknown. Prospective interventional and longitudinal studies will be required to determine whether baseline connectivity patterns predict treatment response or change as a consequence of neuromodulation. Finally, emerging non-invasive approaches designed to modulate deeper brain structures, such as transcranial focused ultrasound [136, 137] and temporal interference stimulation [138, 139], may offer future opportunities to experimentally test the causal relevance of these circuits in MDD. At present, however, these findings should be viewed as providing a circuit-level framework for hypothesis-driven neuromodulation research, rather than definitive guidance for clinical targeting.

Limitations

This review is constrained by the specificity of our search strategy, as it only allowed us to retrieve papers mentioning sgACC in the title and abstract. Hence, mainly hypothesis-driven studies were reviewed, and potential sgACC-related findings identified through data-driven methods as part of unspecified whole-brain analyses may have been inadvertently omitted. As a result, large open-source neuroimaging datasets were not consistently captured, as many such studies rely on whole-brain or atlas-level analyses and do not report sgACC-specific results or peak coordinates suitable for CBMA. Furthermore, due to the failure of studies to report coordinates, sample sizes for some of our ALE meta-analyses were small. The file drawer problem should be acknowledged, as currently, there are no clear cut-offs for publication bias in CBMA, and disease studies, in general, tend to be less robust against noise experiments [140]. Some of the clusters presented in this meta-analysis are subject to publication bias due to low FSN.

The present synthesis focused on MRI modalities that report voxel-wise sgACC findings compatible with CBMA. Diffusion MRI studies were not included, as tract-based and network-level diffusion metrics are not readily compatible with CBMA approaches. Pooling was guided by construct and signal comparability: structural measures (e.g., GM volume and CT) were synthesized separately from functional measures (e.g., BOLD activation and FC), and physiologically distinct signals such as perfusion-weighted ASL were not pooled with BOLD/FC because they index different underlying processes and have different acquisition and processing characteristics. Although CT and surface area are valid morphometric measures, estimates in ventromedial regions such as the sgACC can be sensitive to thin cortex, complex folding, and surface reconstruction choices, and should therefore be interpreted cautiously [62]. In addition, molecular imaging modalities such as PET, including FDG-PET studies of glucose metabolism that historically motivated hypotheses regarding sgACC involvement in MDD, were not included because of differences in physiological signal, preprocessing pipelines, spatial resolution, and reporting conventions that limit compatibility with MRI-focused CBMA. These PET studies provide complementary insights and warrant dedicated PET-specific or multimodal meta-analytic investigation in future work. Although ASL was included as an MRI-based perfusion technique compatible with CBMA, ASL studies remain relatively sparse compared to BOLD-based fMRI, limiting the power to draw modality-specific conclusions regarding sgACC CBF alterations in MDD. In addition, baseline treatment status was variably reported across studies. Although all participants were unmedicated at the time of imaging, medication-naïve status was not uniformly reported, and heterogeneity in prior medication exposure or washout procedure may contribute to residual variability in sgACC findings.

Inclusion of task-based fMRI studies introduces additional heterogeneity related to task design, cognitive and affective demands, and analytical approaches. Although resting-state and task-based studies were analyzed and reported separately, and task-related findings were interpreted as context-dependent, variability across paradigms may still contribute to between-study heterogeneity and limit the generalizability of task-evoked sgACC effects. Moreover, this review was intentionally restricted to human neuroimaging studies; although animal models provide important causal insight into sgACC-related circuitry, they fall outside the scope of a CBMA and represent an important direction for future integrative work. Finally, different MRI techniques and acquisition paradigms can introduce stereotactic heterogeneity in reported GM volume, activity, and network coherence patterns, and this variability can affect the convergence of activation across studies and, consequently, the robustness of the ALE results. While the ALE meta-analysis aims to account for such heterogeneity, the inherent methodological diversity of the included studies should be considered when interpreting the results and the peak coordinates reported here.

Future directions

Heterogeneity in MRI acquisition across studies should also be mentioned. The literature covered by this review has been published over the span of >25 years, and this time period has witnessed considerable advances in hardware and the development of new MRI sequences. Replicating some of the older studies today with higher field strengths and better spatial resolution may lead to different results, and we actively encourage the neuroimaging community to run simple cross-sectional experiments comparing the anatomy, activity, and network coherence of the sgACC between patients with MDD and HC using cutting-edge MRI approaches (e.g., high field strength scanners, multi-band multi-echo acquisition sequences). The significant clusters presented here may also be used to perform meta-analytic connectivity modeling or as regions of interest in future neuroimaging analyses of the sgACC activity and network coherence patterns in MDD. Performing similar meta-analyses comparing medicated patients with MDD and HC may also be informative with regard to MDD etiology and precision medicine (e.g., brain stimulation targeting), as well as understanding the effects of common antidepressant medications on the sgACC structure and function relative to the baseline depressed state. Sex-specific effects could not be examined robustly, as sex-stratified results and diagnosis-by-sex interaction tests were not consistently reported in a manner compatible with CBMA. Future large-scale studies and meta-analytic efforts would benefit from standardized reporting of sex-stratified sgACC metrics and interaction models, given the sex-dependent nature of MDD. Lastly, selecting the right behavioral paradigm that reliably engages the sgACC remains a challenge, so basic science and test-retest reliability neuroimaging experiments tackling this particular problem are encouraged.

Conclusions

In MDD, the sgACC displays distinct anatomical and functional changes across its subregions. This systematic review and meta-analysis present evidence of reduced GM volume and task-based overactivity of the sgACC subregions corresponding to BA24/25 in unmedicated individuals with MDD compared to HC, whereas more posteroinferior ventromedial/orbitofrontal loci (often labelled BA11 in original reports) did not show consistent differences. Furthermore, the sgACC displays a wide dysconnectivity pattern with multiple brain regions in MDD, particularly with vmPFC and right insula. The behavioral and clinical relevance of these anatomical and functional changes in MDD should continue to be explored.

Author contributions

ID conceptualized the systematic review and meta-analysis, developed the study methodology, led data curation and analysis, interpreted the results, prepared figures and tables, and drafted the initial manuscript. RLS conducted the literature search, performed study screening and selection, extracted data from included studies, and contributed to the quality assessment of the included studies. VAB conducted the literature search, performed study screening and selection, extracted data from included studies, and contributed to the quality assessment of the included studies. IT conducted the literature search, performed study screening and selection, extracted data from included studies, and contributed to the quality assessment of the included studies. NR contributed to the quality assessment of included studies and assisted with verification of extracted data. TAS contributed to interpretation of the results, provided methodological input, and supervised the project. SHK contributed to interpretation of the findings and provided clinical expertise and critical feedback on the manuscript. AB contributed to interpretation of the results and provided expertise on neuromodulation and clinical translation. AML contributed to interpretation of the results and provided expertise on neurosurgical neuromodulation and clinical implications. KD contributed to interpretation of the results and provided expertise on neuromodulation and clinical applications. VB conceptualized the study, provided supervision and strategic guidance throughout the project, contributed to interpretation of the findings, and critically revised the manuscript. All authors reviewed the manuscript critically for important intellectual content and approved the final version of the manuscript.

Data availability

Data supporting the findings of this systematic review and meta-analysis are available within the article and its supplementary information file.

Competing interests

RLS, VAB, IT, NR, TAS, and AB have no conflicts of interest. ID is a Vanier Scholar supported by the Canadian Institutes of Health Research (#513715). SHK has received research funding or honoraria from Abbott, Alkermes, Allergan, Boehringer Ingelheim, Brain Canada, Canadian Institutes of Health Research, Janssen, Lundbeck, Lundbeck Institute, Ontario Brain Institute, Ontario Research Fund, Otsuka, Pfizer, Servier, Sunovion, and Xian-Janssen, and has received stock options from Field Trip Health. AML holds the Hudson Chair in Neurosurgery and is a consultant for Abbott, Boston Scientific, and Functional Neuromodulation Inc. KD is supported by an Academic Scholar Award from the Department of Psychiatry, University of Toronto, and has received research support from the Canadian Institutes of Health Research, Brain & Behavior Foundation, St. Michael’s Foundation, and the American Foundation for Suicide Prevention. VB is supported by an Academic Scholar Award from the Department of Psychiatry, University of Toronto, and has received research support from the Canadian Institutes of Health Research, Brain & Behavior Foundation, Ministry of Health Innovation Funds, Royal College of Physicians and Surgeons of Canada, Department of National Defence (Government of Canada), New Frontiers Research Fund, American Foundation for Suicide Prevention, University of Toronto Connaught Funds, University of Toronto EMH Seed Fund, and investigator-initiated trials from Roche Canada, Eisai Canada, Novartis, Associated Medical Services Inc Healthcare, and the National Research Council of Canada.

Supplementary information

The online version contains supplementary material available at 10.1038/s41398-026-04006-5.