Age-dependent alterations in anterior cingulate excitation-inhibition balance in major depressive disorder revealed by 7-Tesla magnetic resonance spectroscopy Authors: Yujiro Yoshihara, Taro Suwa, Manabu Kubota, Momoko Hatakoshi, Yusuke Kyuragi, Jamie Near, Masoumeh Dehghani, Shin-ichi Urayama, Tomohisa Okada, Toshiya Murai Published: 2026 ABSTRACT Highlights•7T MRS reveals age-dependent alterations in excitation–inhibition balance in treated MDD.•A significant group × age interaction is observed in dACC Glu/GABA ratio.•The normative age-related increase in Glu/GABA ratio is absent in MDD.•Preserved glutamate–GABA coupling suggests altered inhibitory trajectory rather than uniform deficits.Keywords: Major depressive disorder, 7T magnetic resonance spectroscopy, Anterior cingulate cortex, Glutamate, Excitation/inhibition balance Introduction The balance between neuronal excitation and inhibition (E/I balance) in the brain is critical for the proper functioning of neural circuits involved in information processing and social behaviour (Shu et al., 2003, Yizhar et al., 2011). Findings from postmortem, animal, and human neurophysiological studies converge to indicate disrupted inhibitory regulation within cortical circuits in depression, consistent with an altered excitation–inhibition balance (Hu et al., 2023, Page and Coutellier, 2019, Duman et al., 2019, Fuchs et al., 2017, Krystal et al., 2023, Fee et al., 2017, Fitzgerald and Watson, 2018). Furthermore, modifying the E/I balance has become a key target for therapeutic interventions for depression, including pharmacotherapy, electroconvulsive therapy (ECT), and transcranial magnetic stimulation (TMS). (Sanacora et al., 2003, Salvadore et al., 2012, Croarkin et al., 2011, Ermis et al., 2021, Merkl et al., 2011, Narayan et al., 2022). Magnetic resonance spectroscopy (MRS) enables the in vivo quantification of glutamate (Glu), a major excitatory neurotransmitter, and gamma-aminobutyric acid (GABA), the principal inhibitory neurotransmitter. Previous 3T MRS studies have struggled to separate glutamate (Glu) and glutamine (Gln) from the combined Glx signal (Glu + Gln). Utilizing high magnetic field strength, 7-tesla (7T) MRS allows improved spectral separation of Glu and Glu (Tkac et al., 2001) and facilitates GABA quantification (Terpstra et al., 2016). The separation of Glu and Gln has been shown to be important for understanding neuropsychiatric disorders (Kubota et al., 2025). Previous studies using 3T MRS in patients with depression have evaluated the excitation/inhibition (E/I) balance based on Glu, Glx (Glu + Gln), and GABA (Dai et al., 2025, Narayan et al., 2022, Tran et al., 2024, Persson et al., 2021). The 3T MRS findings have been inconsistent across studies, showing both increases and decreases in Glu, Glx, and GABA, likely reflecting differences in the illness stage of depression, treatment status, and brain region examined, as well as the inherent heterogeneity of depression (Tao et al., 2025, Godfrey et al., 2018, Truong et al., 2021, Schur et al., 2016). In addition to clinical heterogeneity related to illness stage, treatment status, and brain region examined, methodological limitations associated with 3T MRS, including relatively low signal-to-noise ratio (SNR) and limited spectral resolution, may also contribute to inconsistent findings across studies. Studies employing 7T MRS in unmedicated patients with major depressive disorder (MDD) have reported no significant differences in Glu levels compared with healthy controls (Godlewska et al., 2018, Evans et al., 2018, Smith et al., 2021, Song et al., 2024). Following pharmacological treatment, 7T MRS studies have reported conflicting changes in Glu levels, with both increases (Evans et al., 2018) and decreases observed (Smith et al., 2021). Similarly, findings for GABA have been variable, with studies showing either no change (Smith et al., 2021) or reductions in treated patients with MDD (Song et al., 2021). Thus, even with the use of 7T MRS, results of Glu and GABA in patients with MDD remain discordant. Previous studies in both healthy individuals and psychiatric populations using 7T MRS have suggested a close relationship between Glu and GABA levels (Rideaux et al., 2022, Biria et al., 2023). However, 7T MRS studies on E/I balance in treated patients with MDD remain limited. Therefore, it is necessary to further investigate whether the Glu-GABA relationship is altered in treated patients with MDD. Beyond examining the presence of an association, comparing the strength of Glu-GABA coupling between groups may further clarify whether excitatory-inhibitory coordination is altered in depression. The anterior cingulate cortex (ACC) is a central hub within large-scale brain networks involved in emotional and cognitive regulation (Seeley et al., 2007, Menon and Uddin, 2010). Among ACC subregions, the dACC plays a key role in stress regulation, emotional processing, and higher-order cognitive functions (Bush et al., 2000, Shackman et al., 2011, Palomero-Gallagher et al., 2009). Structural MRI studies have reported reduced gray matter volume in the ACC in patients with depression (Bora et al., 2012, Wise et al., 2017), while task- and resting-state functional imaging studies have revealed aberrant functional connectivity in the ACC (Vasic et al., 2009, Carballedo et al., 2011, Zeng et al., 2020, Gong et al., 2020). These findings suggest that structural and functional abnormalities in the ACC may contribute to core depressive symptoms, such as anhedonia, avolition, and rumination (Horn et al., 2010, Northoff and Sibille, 2014, Levar et al., 2017, Demenescu et al., 2017, Park et al., 2019). Moreover, systematic reviews and meta-analyses have reported decreased concentrations of Glx, Glu, and GABA in the ACC in depression; however, some studies have reported elevated levels, resulting in inconsistencies in the literature (Godfrey et al., 2018, Draganov et al., 2020, Simmonite et al., 2023). Thus, neurochemical alterations in the ACC in treated patients with MDD remain incompletely understood. Importantly, most previous findings have been derived from conventional 3T MRS studies, which have limited spectral resolution and often quantify glutamate and glutamine together as Glx. This methodological limitation may partly account for the inconsistent findings reported in the literature. Region-of-interest approaches remain useful for characterizing neurochemical alterations in cortical regions implicated in depression. In this study, we focused on patients with MDD following therapeutic intervention and targeted the dorsal anterior cingulate cortex (dACC) as an a priori region of interest. The dACC plays an important role in cognitive control, conflict monitoring, and stress regulation. Converging evidence from structural and functional neuroimaging studies has implicated the dACC in the pathophysiology of major depressive disorder. In addition, the relatively homogeneous magnetic field properties of the dACC facilitate reliable spectroscopic measurements of glutamatergic and GABAergic metabolites. Using 7T MRS in treated patients with MDD, we assessed the Glu/GABA ratio in the dACC to investigate (1) whether abnormalities in excitatory and inhibitory balance exist, (2) how E/I balance relates to depressive symptoms and cognitive function, and (3) whether the relationship between Glu and GABA is altered in treated patients with MDD. Given prior evidence that glutamatergic and GABAergic measures may vary according to sex (Seney et al., 2018, Theriault and Perreault, 2019, Kantrowitz et al., 2021) and age (Porges et al., 2021, Kaiser et al., 2005, Lissemore et al., 2018), group effects on the Glu/GABA ratio were examined using a full-factorial ANCOVA including group × sex and group × age interactions, while symptoms- and cognition-related associations and Glu-GABA coupling were evaluated using correlation-based approaches. Methods Participants Participants were recruited from the Longitudinal MRI Study Identifying the Neural Substrates of Remission/Recovery in Mood Disorders (L/R Study), an observational study, as part of the Brain MINDS/Beyond project. Patients with MDD and healthy controls (HCs) were enrolled from the Department of Psychiatry at Kyoto University. All patients received treatment as part of routine clinical care, with therapeutic decisions made by the attending psychiatrists at the time of recruitment. Eligibility required a clinical decision to modify antidepressant treatment (addition or switch) and/or to initiate an additional treatment modality, such as cognitive behavioral therapy (CBT) or ECT. Patients receiving CBT participated in a standardized individual CBT program consisting of weekly 50-minute sessions for a total of 16 sessions. All therapists were experienced in CBT for depression and received supervision throughout the study period. Patients receiving ECT were treated according to the revised Japanese ECT guidelines based on the American Psychiatric Association Task Force recommendations. ECT was generally administered under general anesthesia twice weekly, with a mean of 11.0 ± 2.5 sessions. In most cases, non-pharmacological treatments were provided as add-on approaches to ongoing pharmacotherapy. All patients underwent clinical assessments and MRS scans either within two weeks after the final CBT or ECT session, or approximately six weeks after medication initiation or modification (eight patients in total), or six months after treatment (33 patients). The inclusion criteria for MDD were a diagnosis based on the Mini-International Neuropsychiatric Interview (Sheehan et al., 1998), age between 20 and 85 years, and a Hamilton Depression Rating Scale (HAMD-17) score of 8 or higher at the time of participation (Ishikawa et al., 2025). The inclusion criteria for healthy participants were HAMD-17 score ≤ 7 (Zimmerman et al., 2013), Montgomery-Asberg Depression Rating Scale (MADRS) score ≤ 6 (Thase et al., 2021), and Beck Depression Inventory – Second Edition (BDI-2) score ≤ 13 (Beck, 1996). The exclusion criteria were a history of substance use disorder or alcohol abuse, suicidal ideation or attempts, dementia, MRI contraindications, and other serious neuropsychiatric or physical conditions. Clinical assessment The severity of depressive symptoms in all patients with MDD was evaluated using the MADRS, BDI-2, and HAMD-17. The dosages of medications (antidepressants, anxiolytics, antipsychotics, hypnotics/sedatives, and lithium) were standardized using the Defined Daily Dose (DDD), a unit recommended by the World Health Organization (WHO) (http://www.whocc.no/atc_ddd_index/). The DDD represents the average maintenance dose per day for a drug used for its main indication in adults. Cognitive function was assessed in both groups using the Brief Assessment of Cognition in Schizophrenia Japanese Version (BACS-J), which covers six cognitive domains: verbal memory, working memory, motor speed, attention, verbal fluency, and executive function. MRI acquisition All participants underwent MRI and MRS scans on a 7-T MRI scanner (MAGNETOM 7T, Siemens Healthcare, Erlangen, Germany) equipped with a whole-body gradient system (70 mT/m maximum amplitude, 200 mT/m/ms maximum slew rate). A single-channel transmit and 32-channel receiver head coil (Nova Medical, MA, USA) were used. Structural T1-weighted images were acquired using magnetization-prepared 2 rapid acquisition gradient echo (MP2RAGE, research prototype sequence) with the following parameters: repetition time (TR) = 6000 ms; echo time (TE) = 2.9 ms; inversion times (TI) = 800 and 2700 ms; flip angle = 4 and 5 degrees; number of slices = 256; slice thickness = 0.7 mm; field of view = 195 × 198 mm; matrix size = 280 × 284; isotropic spatial resolution = 0.7 × 0.7 × 0.7 mm. MRS scans were conducted using the semi-adiabatic short TE spin-echo full-intensity-acquired localized single voxel spectroscopy (sSPECIAL) sequence (Near et al., 2013, Oz, G., Deelchand, D.K., Wijnen, J.P., Mlynarik, V., Xin, L., Mekle, R., Noeske, R., Scheenen, T.W.J., Tkac, I., Experts' Working Group on Advanced Single Voxel, H.M., 2020). The volume of interest (VOI) was positioned in the dorsal ACC, and the voxel size was 20 × 30 × 20 mm3 (in right-left, anterior-posterior, and superior-inferior directions). The parameters for the MRS scans were TE = 16 ms, TR = 6500 ms, spectral bandwidth = 6 kHz, and 48 averages. After performing automated shimming using the Fast, Automatic Shim Technique using Echo-planer Signal readout for Mapping Along Projections (FASTESTMAP, Siemens prototype sequence), manual shimming was conducted when the full-width at half maximum (FWHM) value of the water spectrum was more than 25 using the online monitoring. MRS data processing Spectral processing was performed using the FID-A toolkit (Simpson et al., 2017) running on MATLAB2024a (MathWorks, Inc., Natick, MA, USA). Processing steps included: weighted combination of the receiver channels, removal of motion-corrupted averages, spectral registration for frequency and phase drift correction and alignment of inversion on/off sub-spectra prior to subtraction of sub-spectra, and signal averaging. MRS spectral modelling was conducted using LCModel (ver. 6.3-1N, Stephen Provencher Inc., Oakville, Ontario, Canada) using a basis set of simulated model spectra for the Siemens 7T sSPECIAL sequence. The basis set included the following metabolites: GABA, Gln, Glu, glutathione (GSH), myo-inositol (mI), N-acetylasparate (NAA), taurine (Tau), total choline (tCho: glycerophosphocholine + phosphocoline), and total creatine (tCr: creatine + phosphocreatine). Simulated macromolecules at 0.9, 1.2, 1.4, 1.7, and 2.0 ppm were also included. The macromolecule at 2.0 ppm consists of peaks at 1.95, 2.08, 2.25, and 3.00 ppm. In the current study, Glu, Gln, and GABA levels were of primary interest, but additional analyses considered all other metabolites in the basis set. The tissue composition inside the VOIs was estimated based on T1-weighted image segmentation using the gannetCoRegister and gannetSegment functions in Gannet 3.0 (Edden et al., 2014) to calculate the volume fractions of white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF). Metabolite concentrations were corrected for segmental water concentration and partial volume of CSF inside the VOI and presented as institutional units, assuming water concentrations in WM, GM and CSF as 35,880, 43,300, and 55,556 mmol/L, respectively using the formula below (Provencher, 2021): Water concentration = (fwm*35880 + fgm*43300 + fcsf * 55556)/(1 - fcsf), where fwm, fgm, and fcsf are fractions of WM, GM, and CSF, respectively. The signal-to-noise ratio (SNR) was calculated using the NAA peak height at 2.01 ppm divided by noise calculated as twice of the root mean square of residuals from the 0.2 to 4.0 ppm. MRS Spectra quality was ensured by including only data with SNR > 10, full-width at half maximum (FWHM) < 0.05ppm, and Cramér–Rao lower bound (CRLB) ≤ 20%. Group-wise outliers of metabolite concentrations were defined as observations lying more than three times the interquartile range (IQR) below the first quartile or above the third quartile. Three participants were excluded from the analysis for the following reasons: one due to excessive head motion that precluded data acquisition, one due to spectrum abnormality, specifically a markedly distorted baseline above 3.4 ppm, and one due to failure to meet the criterion of FWHM < 0.05 ppm. In the case of GSH quantification, one healthy control was excluded because the value exceeded three IQRs, whereas no outliers were identified for the other metabolites. The number of exclusions for each metabolite was compared between patients and controls using the chi-square test to check for potential data selection bias. Voxel overlap maps after spatial normalization and representative spectra for each group are shown in Fig. 1A–D. Representative LCModel spectral fitting and basis set components are shown in Supplementary Fig. S1. Percent overlap of the volume of interest (VOI) at the dorsal anterior cingulate cortex (dACC) after spatial normalization and placed over the averaged anatomical images for (A) HC and (B) MDD. Individual (black) and their averaged (red) spectra are plotted after baseline removal for (C) HC and (D) MDD. Voxel localization and spectral spectral quality were comparable between groups. Statical Analysis All statistical analyses were conducted using SPSS 27.0 (SPSS Inc.) and GraphPad Prism version 9 (Graph-Pad Software, La Jolla, CA, USA, www. graph pad. com). Demographic and clinical differences between groups were examined using Mann-Whitney U or chi-square tests. Analysis of covariance (ANCOVA) with age as a covariate was performed to examine the differences in each domain score of the BACS-J between MDD and HC groups. Group differences in ANCOVA were considered significant at p < 0.05/6 (Bonferroni-corrected). The dependent variable was the Glu/GABA ratio in the dorsal ACC, which was used as an MRS-based proxy of excitation/inhibition (E/I) balance. A univariate analysis of covariance (ANCOVA) was conducted with group (MDD vs. HC) and sex as fixed factors, and age (centered on the sample mean) as a covariate. A full factorial model ANCOVA was specified to examine the main effects of group, sex, and age, as well as their two-way interactions (group × sex, group × age). Type III sum of squares was applied, and partial eta-squared (ηp2) values were reported as measures of effect size. To follow up on a significant group × age interaction, simple effect analyses were performed to clarify the interaction pattern. Specifically, the simple slope of age within each group was calculated, and conditional group differences were tested at the representative ages (mean ± 1 standard deviation) using the estimated marginal means (EMMs) with 95% confidence intervals (CIs). The primary analysis focused on the Glu/GABA ratio as the main E/I balance index, for which no multiple comparison correction was applied (p < 0.05). Secondary analyses were conducted for Glu and GABA separately using the same ANCOVA model to supplement the primary E/I findings. For these additional analyses, a Bonferroni-adjusted threshold of p < 0.025 (0.05/2) was applied. The simple slopes of Glu and GABA levels with age within each group were calculated. Other brain metabolite (Gln, GSH, mI, NAA, Tau, tCho, and tCr) levels were analyzed using ANCOVA, with age and sex as covariates (p < 0.05/7). Furthermore, considering the medication treatment effects, we examined ANCOVA with sex and age as covariates, comparing patients receiving versus not receiving each of the five medication classes (antidepressants, anxiolytics, antipsychotics, hypnotic/sedatives, and lithium). For the analysis, a Bonferroni-adjusted threshold of p < 0.01 (0.05/5) was applied. Correlational analysis between neurometabolite levels and clinical measures We first performed correlational analyses between Glu/GABA and HAMD-17, MADRS, and BDI-2 scores within the MDD group. We computed Spearman’s rank correlation coefficients and partial rank correlation coefficients, controlling for age. The statistical significance threshold was set at p < 0.05/3. Additionally, the same correlational analyses were applied to Glu and GABA levels individually. Furthermore, the relationship between Glu and GABA levels was assessed using Spearman’s rank correlation coefficient in each group. To compare the strength of this coupling between groups, z-transformed correlation coefficients (using Fisher’s transformation) were compared using two-tailed t-tests, with a statistical significance of p < 0.05. Results Demographic and clinical data Detailed demographic data are presented in Table 1. A total of 41 patients with MDD following treatment and 35 HCs were included in the analysis. No significant differences in age or sex were observed between the groups. Among the patients, 30 (73.2%) were in remission, defined as a HAMD-17 score of ≤ 7. Antidepressant medication was administered to 85.4% of the patients. Compared with healthy controls, treated patients with MDD exhibited significantly lower performance in the motor speed, attention, and verbal fluency domains of the BACS-J (all p < 0.05/6). Table 1Demographic and clinical characteristics of patients with major depressive disorder (MDD), and healthy controls (HC)MDD (n = 41)HC (n = 35)StatisticsVariablesMedian (IQR)Median (IQR)PAge (years)43.0 (29.0, 51.0)45.0 (40.5, 67.5)0.169aSex (female/male), n21/2021/140.443bTreatment (CBT/Med/ECT), n30/7/4-HDRS-175.0 (3.0, 9.0)0.0 (0.0, 0.0)< 0.001a**MADRS7.5 (4.0, 13.0)0.0 (0.0, 0.0)< 0.001a**BDI-211.0 (7.0, 19.0)2.0 (0.0, 4.0)< 0.001a**Duration of illness (months)29.0 (19.0, 53.0)-Current Smoking (Yes/No/NA), n8/29/42/33/00.024b*Medication% use [DDD (SD)]Antidepressants85.4 [1.3 (0.9)]-Anxiolytics12.2 [0.0 (0.1)]-Antipsychotics14.6 [0.0 (0.1)]-Hypnotics and Sedatives31.7 [0.4 (0.6)]-Lithium7.3 [0.2 (0.7)]-BACS-J(n = 37)(n = 35)Median (IQR)Median (IQR)Verbal Memory57.0 (42.0, 61.0)54.0 (48.0, 59.0)0.079cWorking Memory24.0 (18.0, 26.0)24.0 (21.5, 26.0)0.025cMotor Speed69.8 (62.0, 82.0)76.0 (66.0, 84.0)0.003c***Attention64.0 (59.0, 74.0)68.0 (61.0, 81.5)< 0.001c***Verbal Fluency47.7 (41.0, 56.0)53.0 (45.5, 64.5)0.002c***Executive Function19.0 (18.0, 19.0)19.0 (18.0, 21.0)0.152cNote: MDD, major depressive disorder; HC, healthy controls; IQR, interquartile range; CBT, cognitive behavior therapy; Med, medication therapy; ECT, electroconvulsive therapy; HDRS-17, Hamilton-17 item Depression Rating Scale for depression; MADRS, Montgomery-Asberg Depression Rating Scale; BDI-2, Beck Depression Inventory - Second Edition; NA, not applicable; DDD, defined daily dose; BACS-J, The Brief Assessment of Cognition in Schizophrenia Japanese Version; %, proportion receiving medication. DDD values are presented as mean (SD).* p <0.05** p <0.017 (Bonferroni corrected for three comparisons)*** p <0.008 (Bonferroni corrected for six comparisons)aMann-Whitney U testbChi-square testcMain effects from analysis of covariance (ANCOVA) with a covariate as age Demographic and clinical characteristics of patients with major depressive disorder (MDD), and healthy controls (HC) Note: MDD, major depressive disorder; HC, healthy controls; IQR, interquartile range; CBT, cognitive behavior therapy; Med, medication therapy; ECT, electroconvulsive therapy; HDRS-17, Hamilton-17 item Depression Rating Scale for depression; MADRS, Montgomery-Asberg Depression Rating Scale; BDI-2, Beck Depression Inventory - Second Edition; NA, not applicable; DDD, defined daily dose; BACS-J, The Brief Assessment of Cognition in Schizophrenia Japanese Version; %, proportion receiving medication. DDD values are presented as mean (SD). * p <0.05 ** p <0.017 (Bonferroni corrected for three comparisons) *** p <0.008 (Bonferroni corrected for six comparisons) Mann-Whitney U test Chi-square test Main effects from analysis of covariance (ANCOVA) with a covariate as age MRS quantification There were no significant differences between the two groups in terms of spectral SNR, FWHM, or CRLB for Glu and GABA, as defined by our quality criteria (Supplementary Table S1). Regarding GABA, the CRLB exceeded 20 in seven patients and six HCs; however, no significant differences were observed in the rate of data rejection between treated patients with MDD and HC (p = 0.994). As shown in Supplementary Table S2, there were no significant group differences in the volume proportions of GM, WM, and CSF within the VOIs. Effects of diagnostic group, age, and sex on Glu/GABA ratio As shown in Table 2 and Fig. 2A, significant main effects of group and age but not sex on Glu/GABA ratios were observed (p = 0.019, 0.013, and 0.059, respectively). The age × group interaction was also significant (p = 0.005), indicating that the group difference varied as a function of age. Subsequent representative-age effects analysis revealed that while the group difference was not significant at the mean − 1SD (ΔMDD−HC = 0.15, p = 0.681), it was significant at the mean + 1SD (ΔMDD−HC= −1.45, p = 0.001) (Table 3). To clarify this interaction, we examined the simple slopes of age within each group (Fig. 2B). In treated patients with MDD, age was not significantly associated with Glu/GABA ratio (β = −0.004, 95% confidence intervals (CI) [−0.025, 0.017], p = 0.690), whereas in the HC group, the slope was significant (β = 0.041, 95% CI [0.015, 0.066], p = 0.003). To further address the potential influence of partially unbalanced age distributions between groups, we conducted an additional sensitivity analysis restricted to participants aged 25–75 years. The group × age interaction in the Glu/GABA ratio remained significant (p = 0.003), supporting the robustness of the observed findings (Supplementary Table S3). Table 2Effects of group, age, and sex on Glutamate/GABA ratio in the dACC (ANCOVA)MetaboliteSourceF (df1, df2)Ppartialη2Number (MDD, HC)Glutamate/GABAGroup5.82 (1, 57)0.0190.093(34, 29)Age6.55 (1, 57)0.0130.103Sex3.71 (1, 57)0.0590.061Age × Group8.40 (1, 57)0.0050.128Sex × Group0.00 (1, 57)0.9980.000Note: Age was mean-centered prior to analysis. MDD, major depressive disorder; HC, healthy controls; GABA, γ-aminobutyric acid. Interaction terms are denoted by “×”. Effects of group, age, and sex on Glutamate/GABA ratio in the dACC (ANCOVA) Note: Age was mean-centered prior to analysis. MDD, major depressive disorder; HC, healthy controls; GABA, γ-aminobutyric acid. Interaction terms are denoted by “×”. A. Group differences in Glu/GABA ratio between patients with major depressive disorder (MDD; green) and healthy controls (HC; orange), adjusted for age and sex (ANCOVA). *p = 0.019 for the main effect group. B. Age-related changes in Glu/GABA ratio in MDD and HC. Solid and dashed lines represent regression slopes for MDD and HC, respectively. Shaded areas indicate 95% confidence intervals. The group × age interaction was significant (p = 0.005). Table 3Conditional group differences in Glu/GABA ratio at representative agesRepresentative ageGroupEMM (95% CI)ΔGlu/GABA (MDD – HC)P-valueMean –1 SD (29.3 years)MDD7.18 (6.75, 7.62)+0.150.681HC7.03 (6.46, 7.61)Mean (46.6 years)MDD7.14 (6.77, 7.50)–0.650.019HC7.79 (7.39, 8.18)Mean +1 SD (63.9 years)MDD7.09 (6.28, 7.70)–1.450.001HC8.54 (8.00, 9.08)Note: EMM, estimated marginal means; CI, confidence interval; SD, standard deviation; MDD, major depressive disorder; HC, healthy controlsEMMs were adjusted for sex and age (centered) in the ANCOVA model. P-values are Bonferroni-adjusted. ΔGlu/GABA represents the mean difference (MDD – HC). Conditional group differences in Glu/GABA ratio at representative ages Note: EMM, estimated marginal means; CI, confidence interval; SD, standard deviation; MDD, major depressive disorder; HC, healthy controls EMMs were adjusted for sex and age (centered) in the ANCOVA model. P-values are Bonferroni-adjusted. ΔGlu/GABA represents the mean difference (MDD – HC). In the secondary analysis, ANCOVAs for Glu and GABA showed no significant diagnostic effects or interactions (Supplementary Table S4). Simple slopes indicated that in the HC group, age was significantly correlated with GABA levels (β = −0.007, 95% CI [−0.013, −0.001], p = 0.023), while no other slopes reached significance (Fig. 3A, B). Age-related associations of glutamate (Glu; A) and γ-aminobutyric acid (GABA; B) in patients with major depressive disorder (MDD; green) and healthy controls (HC; orange). Measured metabolites values are presented in institutional units. Regression lines are shown separately for patients with MDD (green) and HC (orange). Solid and dashed lines represent regression slopes for MDD and HC, respectively. Shaded areas indicate 95% confidence intervals. Age was significantly correlated with GABA levels in HC (β = −0.007, 95% CI [−0.013, −0.001], p = 0.023), while no other slopes reached significance. Within treated patients with MDD, no significant differences in Glu/GABA, Glu, or GABA levels were found between patients receiving and not receiving specific medications (Supplementary Table S5). To examine the potential influence of treatment heterogeneity, we conducted an additional sensitivity analysis excluding the single patient who underwent ECT, while applying the same ANCOVA model used in the primary analysis. Importantly, the group × age interaction in the Glu/GABA ratio remained significant after exclusion of the ECT-treated patient (p = 0.013), indicating that the main findings were not driven by ECT-related effects (Supplementary Table S6). In addition, no significant main effect of treatment category was observed within treated patients with MDD (Supplementary Table S7). A supplementary ANCOVA also showed no significant effect of scan timing on the Glu/GABA ratio within treated patients with MDD (Supplementary Table S8). Correlation between Glu/GABA ratio and clinical features No significant correlations or partial correlation analyses controlling for age were observed between the Glu/GABA ratio and clinical symptoms or cognitive function (Supplementary Table S9). In addition, Glu and GABA levels were not significantly associated with symptoms or cognition (Supplementary Table S10). Correlation between Glu and GABA in treated patients with MDD and HC In both MDD and HC groups, a significant positive correlation was observed between glutamate and GABA (MDD: r = 0.669, p < 0.001; HC: r = 0.476, p = 0.009) (Fig. 4). Comparisons of Fisher’s z-transformed correlation coefficients between MDD and HC groups revealed no significant differences in the strength of the relationship between Glu and GABA (z = 1.08, p = 0.282). Positive association between glutamate (Glu) and γ-aminobutyric acid (GABA) in patients with major depressive disorder (MDD; green) and healthy controls (HC; orange). Measured metabolite values are presented in instituional units. Solid lines represent Spearman’s rank correlations within each group. MDD: r = 0.669, p < 0.001; HC: r = 0.476, p = 0.009. The difference between correlation coefficient was not significant (z = 1.08, p = 0.282). Diagnostic Group Differences in the other metabolites The levels of other brain metabolites did not differ significantly between treated patients with MDD and HC groups (Supplementary Table S11). Discussion Using 7T MRS, we identified a significant interaction between the Glu/GABA ratio and age in the dorsal anterior cingulate cortex (dACC) of treated patients with MDD and HCs. In the HC group, the Glu/GABA ratio showed a significant positive correlation with age, whereas this age-related increase in the Glu/GABA ratio was absent in treated MDD group. Furthermore, the Glu/GABA ratio in treated patients with MDD was not significantly correlated with symptom severity, and no significant associations were found between the Glu/GABA ratio and cognitive performance in either group. Consistent with previous reports, Glu and GABA levels were significantly correlated within each group; however, the strength of this coupling did not differ significantly between groups. Because all patients were scanned after therapeutic intervention, the present findings might partially reflect state-related neurochemical alterations associated with treated or remitted phases of MDD rather than stable trait-related abnormalities. A previous MRS study of the dACC in depression at 3T reported no significant differences in the GABA/Glu ratios between patients with MDD and HCs (Persson et al., 2021). Although that study expressed the ratio inversely to ours, their findings similarly indicated no significant overall alteration in the E/I balance. However, in this study, Glu was estimated from spectra acquired during GABA editing, which may not have provided sufficient precision for accurate Glu quantification. In contrast, our use of 7T MRS provided a higher SNR and enhanced spectral resolution, enabling more accurate simultaneous quantification of Glu and GABA. Furthermore, the participants were younger (average age ∼30 years) and had higher symptom severity (MADRS ∼30) compared with our cohort. Our results thus suggest a reduced Glu/GABA ratio specifically in treated, older treated patients with MDD exhibiting relatively stable clinical status. While brain Glu and GABA levels are generally known to decline with age in healthy subjects (Sailasuta et al., 2008, Gao et al., 2013, Maghsudi et al., 2020, Cassady et al., 2019), no previous studies have reported an age-related increase in the Glu/GABA ratio across a broad adult age range. A prior 7T MRSI study in healthy adolescents (aged 10–30 years) demonstrated no age-related E/I changes in the ACC (Perica et al., 2022). Our findings in adults indicate a significant group difference in the age-dependence of the Glu/GABA ratio, with a positive age-related slope in healthy controls but not in treated MDD. When examining these neurometabolites separately, Glu was not significantly associated with age in either group. In contrast, GABA showed a significant age-related decline only in the HC group, which drove the observed group-by-age interaction in the Glu/GABA ratio. This age-related decline in GABA in HCs is consistent with a meta-analysis (Porges et al., 2021) and 3T MRS studies of the medial frontal lobe in adulthood (Gao et al., 2013, Rowland et al., 2016, Porges et al., 2017). Because the Glu/GABA ratio reflects the relative balance between excitatory and inhibitory metabolites, the observed increase in the ratio with age in healthy controls is likely driven mainly by the age-related decline in GABA. One possible interpretation is that age-related alterations in GABAergic regulation may differ between healthy individuals and treated patients with MDD, potentially reflecting adaptive or compensatory neurobiological processes. However, the precise mechanisms underlying these differences remain unclear. Given that the treated patients with MDD in this study were undergoing various treatments, including medication, CBT, and ECT, our findings should be interpreted within this clinical context. We observed no significant group differences in Glu levels, while GABA levels exhibited a marginally significant increase in the treated MDD group (p = 0.084, see Table S3). Regarding pharmacological effects, a previous 7T MRS study in elderly patients with MDD found no significant changes in dACC Glu or GABA levels following treatment with the antidepressant citalopram (Smith et al., 2021). Similarly, previous 3T MRS studies have reported no significant changes by antidepressant in Glu in the mid-cingulate cortex (Grimm et al., 2012), nor differences in Glu/Cr or GABA/Cr levels between patients taking antidepressants and controls in the dACC (Persson et al., 2021). Conversely, one study reported a significant increase in GABA levels in the prefrontal cortex, including the ACC, following antidepressant treatment (Draganov et al., 2020). While benzodiazepines have been associated with reduced in some reports (Ferland et al., 2021, Goddard et al., 2004), several studies found no effect on Glu or Glx levels (Brambilla et al., 2002, Yildiz et al., 2010, Henry et al., 2010). In our cohort, 85.4% and 12.2% of the patients were receiving antidepressants and anxiolytics (benzodiazepines), respectively. Although no significant differences in GABA levels were found between patients with and without antidepressants and anxiolytics in our study, their use may have contributed to the observed trend toward increased GABA in the MDD group. In addition, supplementary analyses revealed no significant effects of treatment category on Glu/GABA ratio, supporting the robustness of the observed findings. Non-pharmacological treatments are also unlikely to have majorly confounded our results. A considerable portion of our patient sample (73.2%) received CBT; however, prior research showed no change in occipital GABA levels following CBT (Sanacora et al., 2006), and no studies to date have reported CBT-induced alterations in Glu. Similarly, 3T MRS studies on ECT found no changes in Glx or GABA levels in the dACC (Njau et al., 2017, Erchinger et al., 2020) or other regions (Knudsen et al., 2019). Additional sensitivity analyses excluding the ECT-treated patient yielded similar results, supporting the robustness of the observed group-by-age interaction. From a biological perspective, several postmortem gene expression studies have reported altered expression of GABAergic and glutamatergic system-related genes in the dorsal ACC (Hu et al., 2023). This provides a plausible biological bias for the trend toward increased GABA levels in our patient cohort and suggests that underlying molecular abnormalities may contribute to altered neurotransmitter dynamics observed in treated patients with MDD. The dorsal anterior cingulate cortex (dACC), the focus of the present study, is implicated in emotional regulation and anxiety (Comte et al., 2016, Tolomeo et al., 2016, Ganesh et al., 2019). However, we found no significant correlation between the Glu/GABA ratio and the severity of depressive symptoms. A similar observation was reported in the pregenual ACC, where changes in the Glx/GABA ratio over an 8-week antidepressant treatment period were not associated with changes in HAMD scores (Narayan et al., 2022). In contrast, neurochemical correlates of symptom severity have been reported in treatment-resistant depression (TRD); for instance, reduced occipital GABA concentrations were observed in patients with TRD (Price et al., 2009). Additionally, an EEG-based study assessing E/I balance before and after ECT reported increased frontal E/I balance in responders, that is, patients who showed a reduction in HAM-D scores (Stuiver et al., 2024). Unlike EEG, which measures neural activity, MRS reflects total tissue concentrations (both intracellular and extracellular) of Glu and GABA rather than direct synaptic E/I flux. Accordingly, the Glu/GABA ratio should be interpreted as an indirect MRS-based proxy of excitation/inhibition balance rather than a direct measure of synaptic excitatory or inhibitory activity. Our findings may thus suggest that the dACC Glu/GABA ratio as measured by MRS does not directly mediate depressive symptoms. Given the dACC is involved in cognitive processes (Shackman et al., 2011), we further examined the relationship between Glu/GABA ratio and cognitive performance. However, no significant correlations were found between the Glu/GABA ratio and BACS-J scores in either group. While a study on schizophrenia reported positive correlations between dACC Glx levels and working memory and attention performance (Bojesen et al., 2021), we did not find that the impaired performance in motor speed, attention, and verbal fluency in treated patients with MDD was significantly associated with the Glu/GABA ratio. Several factors may account for this lack of association. First, MDD is a heterogeneous condition with diverse symptom profiles and pathophysiological subtypes (Winokur, 1997). Second, depressive symptoms and cognitive deficits may be mediated by complex neural networks rather than the dACC in isolation. Third, as noted above, the MRS-derived signal reflects total metabolite pools, which may limit its specificity as a proxy for excitatory synaptic activity. The significant relationship between Glu and GABA has been previously demonstrated in the frontal cortex of healthy individuals using 7T MRS (Rideaux et al., 2022). Notably, this correlation is specific to Glu and GABA rather than Glx and GABA, suggesting that the superior resolution of 7T spectroscopy enables a more precise characterization of the E/I balance. A similar association was recently reported in the dACC, where a significant correlation between Glu and GABA was observed in both HCs and patients with obsessive-compulsive disorder (OCD) using 7T MRS (Biria et al., 2023). In OCD, elevated dACC Glu levels were associated with an increased Glu/GABA ratio. In contrast, our findings demonstrated a trend toward increased GABA levels in the treated MDD group, contributing to a diminished Glu/GABA ratio. Despite these divergent directional changes in the Glu/GABA ratio, both studies suggest that the fundamental relationship between Glu and GABA remains preserved across clinical and non-clinical populations. This preservation of the Glu–GABA coupling is consistent with EEG evidence from animal models demonstrating stability in the E/I balance at the network level (Sukenik et al., 2021). Moreover, a 3T MRS study showed that the Glx/GABA relationship remains consistent across different scanner vendors and participant demographics (Steel et al., 2020). Taken together, our results extend these findings by showing that the dACC Glu-GABA correlation is preserved in treated patients with MDD. This suggests that while individual metabolite levels or the Glu/GABA ratio itself may be altered, the intrinsic homeostatic mechanisms governing the balance between excitatory and inhibitory neurotransmission may remain fundamentally intact in these cortical regions. Limitations The present study has several limitations. First, MRS measurements were restricted to a single voxel in the dACC, which limits the ability to assess regional heterogeneity in E/I balance across the brain. Second, as we assessed patients following treatment, 28 of the 41 participants (approximately 68%) had achieved remission, with HAMD-17 scores below 8. This high proportion of remitted patients suggests that the observed attenuation of the E/I balance in the dACC may reflect neurochemical characteristics associated with treated or remitted states rather than the acute pathophysiology of depression. Another limitation is the absence of repeated longitudinal measurements in both patients and healthy controls. Because the present study employed a cross-sectional design following treatment, within-subject changes in Glu/GABA dynamics over time could not be assessed. Therefore, individual variability in treatment response or aging trajectories may have contributed to the observed heterogeneity. Although only one patient underwent ECT, sensitivity analyses excluding this participant yielded similar results. Future longitudinal studies using repeated 7T MRS measurements will be important to clarify state-related and trait-related neurochemical alterations in MDD. Third, the majority of patients (85.4%) were receiving antidepressant medications at the time of scanning; therefore, the E/I balance reported here reflects a state potentially influenced by pharmacological interventions. Finally, the inclusion of patients who had undergone diverse treatment modalities—including CBT, medication adjustments, and ECT—introduces clinical heterogeneity. While our sub-group analyses suggested no significant differences based on medication class, the potential for these varied interventions to confound neurochemical outcomes cannot be entirely ruled out. Conclusion Utilizing the high sensitivity and resolution of 7T MRS, we simultaneously quantified Glu and GABA levels, revealing a reduced Glu/GABA ratio, used here as an MRS-based proxy of excitation/inhibition balance, within the dACC of treated patients with MDD. We identified a significant interaction between the Glu/GABA ratio and age, characterized by the absence of the age-related increase observed in the HC group. While this altered E/I balance was not directly associated with clinical symptom severity or cognitive performance, our findings suggest that a diminished Glu/GABA ratio and its divergent age-dependent trajectory may reflect neurochemical characteristics associated with treated or remitted states of MDD. Declaration of Generative AI and AI-assisted technologies in the writing process During the preparation of this manuscript, the authors used generative artificial intelligence tools (ChatGPT and Google Gemini) to assist with English language editing and refinement of grammatical expression. All content was subsequently reviewed and revised by the authors, who take full responsibility for the accuracy and integrity of the final manuscript.Uncited reference Beck et al. (1996). Funding This study was supported in part by Grant in-Aid for the Strategic International Brain Science Research Promotion Program (Brain/MINDS Beyond) [JP19dm0307102 to T.M.] from the Japan Agency for Medical Research and Development (AMED), Grants-in-Aid for Scientific Research Grants-in-Aid for Scientific Research (B) [25K02591 to T.M.] and [21H03806 to T.O.] from the Japan Society for the Promotion of Science, a research grant from SRF Foundation, and a research grant [MRA-IPA3] from Siemens Healthcare K.K., Japan. The funders had no role in the study design, data collection, data analysis, data interpretation, or writing of the manuscript. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements We would like to thank all the participants in this study and all research collaborators, Japan Agency for Medical Research and Development (AMED) Kyoto team including clinical psychologists (Yuri Ito, Makiko Nakamoto, Sayuri Murakami, and Manami Tanaka), assistant nurses (Etsuko Shinoda, and Natsuko Sugita), and clinical doctors of psychiatry department (Sugita Naoko and Hazama Masaaki), working at Kyoto University Hospital for their invaluable work in data collection and analysis. Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.nicl.2026.104018. Contributor Information Yujiro Yoshihara, Email: yujiroyo@kuhp.kyoto-u.ac.jp. Manabu Kubota, Email: m_kubota@kuhp.kyoto-u.ac.jp. Momoko Hatakoshi, Email: momokoh@kuhp.kyoto-u.ac.jp. Yusuke Kyuragi, Email: kyuragi@kuhp.kyoto-u.ac.jp. Jamie Near, Email: jamie.near@utoronto.ca. Masoumeh Dehghani, Email: dm.dehghani@gmail.com. Shin-ichi Urayama, Email: urayama@kuhp.kyoto-u.ac.jp. Tomohisa Okada, Email: tomohisa.okada@gmail.com. Toshiya Murai, Email: murai@kuhp.kyoto-u.ac.jp. Supplementary data The following are the Supplementary data to this article: Supplementary Data 1 Data availability The authors do not have permission to share data. References Associated Data Supplementary Materials Supplementary Data 1 Data Availability Statement The authors do not have permission to share data.