Excitation-inhibition balance plays a major role in determining neural activity (1). At a microscopic level, the influence of the brain's major excitatory and inhibitory neurotransmitters [glutamate and γ-aminobutyric acid (GABA)], on neural activity has been studied in detail [see Isaacson and Scanziani (1), Lauritzen et al. (2)]. At a macroscopic level, this investigation has been achieved through the development and optimisation of neuroimaging techniques enabling quantification of GABA and glutamate concentrations via proton magnetic resonance spectroscopy (¹H-MRS), and of neural activation via measurement of blood oxygen level dependent (BOLD) signal with functional magnetic resonance imaging (fMRI). BOLD signal has been linked to local field potentials of dendritic origins, implying the signal is likely to reflect incoming input (3), thereby being closely linked to, and affected by, neurotransmission. However, the relationship between neurotransmitter levels and task-related neural activity remains unclear.

A previous narrative review by Duncan et al. (4) sought to address this question with a comprehensive overview of studies using imaging techniques such as magnetoencephalography, electroencephalography, fMRI, positron emission tomography and ¹H-MRS, as well as behavioural measures as proxies for neural activity and pharmaceutical manipulations of neurotransmitter levels. The authors reported that GABA levels were related to reduced positive BOLD response or reduced negative BOLD response (NBR) within the same brain region, while glutamate levels were more commonly related to inter-regional neural responses. The number of publications using multimodal fMRI and ¹H-MRS techniques has more than doubled since this review was published, and studies using this type of multimodal combination are expected to keep increasing in the future due to the ease of collecting both types of data within a single scanning session. More recent reports outside of the previous review include studies on regions of interest that had previously been examined too infrequently to permit meta-analysis, providing important insights into the nature of neurochemistry-neurophysiology associations in a variety of brain regions and task domains.

The present study aims to address these issues by conducting a systematic review and meta-analysis of ¹H-MRS and task-based fMRI studies in the healthy brain. We then meta-analyse homogeneous studies based on the location of ¹H-MRS voxel placement and fMRI paradigm. Finally, we discuss the findings in light of abnormalities in these relationships in psychiatric populations, such as schizophrenia patients, and their potential implications on brain processes in psychiatric disorders.

Systematic searches were performed on Pubmed and Ovid (Medline, Embase and PsycINFO) databases from database inception to 4 December 2019. Search strategies for the individual databases included the terms “functional magnetic resonance imaging” AND “magnetic resonance spectroscopy” AND (“GABA” OR “glutamate”) (for full search syntax see Supplementary Table 1). We included studies that used ¹H-MRS to measure GABA, glutamate or Glx (combined glutamate and glutamine) metabolite levels to test associations with task-related fMRI activation. Eligible study designs included observational studies, cohort studies, case-control studies, cross-sectional studies, or experimental studies (for full eligibility criteria see Supplementary Table 2). The population of interest were healthy human participants, or healthy human control groups from case-control studies. Interventional studies were excluded, unless those studies conducted and reported the outcome of interest analysis with baseline measures. Studies assessing only changes in metabolite levels before, during and after a task together with BOLD activation were not included as they do not investigate the relationship between resting-state neurotransmitter levels and neural activity modulated by stimuli. Two reviewers (AK and CB) independently screened search result titles and abstracts of papers for eligibility according to the inclusion and exclusion criteria. All potentially eligible studies were screened full-text for inclusion. Reference lists of eligible studies were hand-searched for further relevant articles.

Two independent reviewers (AK and PBL) extracted data using a standardised template datasheet. The primary outcomes extracted were the metabolite type, ¹H-MRS metabolite sampling region, fMRI task paradigm, neural activation region, and relationship between metabolite and neural activity. Studies investigating the relationship between ¹H-MRS GABA or glutamate and task-related BOLD signal performed analyses in one or both of two different ways: by including metabolite levels as a covariate in a regression, or by extracting the percentage of BOLD signal change (%SC) from a region of interest (ROI) and using correlation analysis with the metabolite levels. Both types of analysis methods were included in this review. Regressions either covered the whole-brain or a predetermined ROI such as the ¹H-MRS voxel, an anatomical ROI or a mask derived from an activation cluster. Correlation analyses used %SC extracted from sources such as the peak voxel within or mean activation across an activated cluster, an anatomically defined ROI, or the ¹H-MRS voxel. Additional extracted data for the systematic review included the statistical significance of the relationship, covariates tested in the analysis, sample size, age and sex characteristics of the sample, scanner strength, ¹H-MRS sequence, and reference metabolite. In cases where no statistical significance value was reported for the relationship itself and only Pearson's r correlation coefficients were reported [e.g., (5)], p-values were calculated through conversion of r values to test statistic t with the formula

The p-value was then derived using the sampling distribution of Student's t. This calculation of the p-value was used to ascertain the significance of the correlation coefficient.

Some studies verified their main findings (¹H-MRS x fMRI response) in the following ways: (1) substituting the metabolite levels from their primary region of interest with those from a different, control sampling region; (2) substituting the fMRI response from their primary task paradigm with that from a control task; (3) substituting the fMRI response from their primary region of interest with that from a control region during the same task. Such control analyses may be found in Supplementary Tables 7, 8. A formal quality assessment was not conducted due to a lack of standard quality assessment for such types of studies. Imaging and study design parameters were, however, collected (Supplementary Tables 3–6, 9). Review findings [see Figure 1 for PRISMA flowchart (6)] are presented according to the brain region from which metabolite levels were sampled.

The main finding of this meta-analysis was that GABA levels were negatively correlated with co-localised neural activity within the ACC during emotion processing and within the occipital lobe during visual stimulation. Further meta-analyses showed that glutamate levels within the ACC were not correlated with co-localised neural activity during either cognitive control or emotion processing, however, the latter meta-analysis narrowly missed significance. These findings were further supported by a systematic review of all fMRI-¹H-MRS association studies to date, where increasing resting GABA levels were repeatedly associated with either weaker neural activation or stronger neural deactivation. On the other hand, glutamate levels were largely not correlated with intra-regional neural activity, although a handful of studies did find a positive correlation. An additional finding was that glutamate concentrations were largely positively correlated with regions outside of the ¹H-MRS sampling area; however, negative inter-regional associations were also identified in multiple studies. GABA concentrations were less likely to be associated with regions outside of the ¹H-MRS sampling area with only a few studies providing evidence of a positive or negative relationship with inter-regional neural activation depending on the metabolite sampling region and the neural activation region.

These findings support and extend prior reported trends. Previously, in a narrative review Duncan et al. (4) highlighted that GABA tended to be negatively correlated with regional neural activity in the visual, auditory, sensorimotor and anterior cingulate cortices, while glutamate was positively correlated with neural activity in regions outside of the sampling region of the metabolite. In the present study, by using a systematic rather than narrative review method and capitalising on the large increase in available studies specifically using ¹H-MRS and fMRI, a more region-dependent relationship between these specific measures emerged. The negative relationship between GABA levels and regional neural activation was confirmed within the occipital cortex, ACC/mPFC, temporal lobe, and PCC. However, regions such as the sensorimotor cortex, auditory cortex and dorsolateral prefrontal cortex either did not exhibit this relationship clearly or in a few studies the relationship was in the opposite direction (30, 50). A negative relationship between GABA levels and regional neural activation appears intuitive: increased inhibition as measured through resting state ¹H-MRS would be associated with decreased net activation of a region through regulation of resting excitatory activity (4, 65). Prior work in animals found that pharmacological manipulation increasing endogenous GABA concentrations produced lower positive BOLD signal changes (66). In the context of NBR, previous work in rats found hyperpolarisation of neurons through neuronal inhibition to correspond to a decrease in blood oxygenation (67). Findings of a positive relationship between GABA levels and neural activity may be a result of more complex neural processes that require suppression of neural response to irrelevant stimuli to promote concurrent activation to target stimuli (68). A relationship between GABA levels and neural activity in distal regions had not previously been identified in the literature (4). While initial studies suggested that the relationship between GABA levels and neural activity was limited to the same region (4, 68), a handful of recent studies reported both negative and positive relationships with activation outside of the ¹H-MRS voxel (32–34, 58–60). Negative relationships with inter-regional neural activity may be the result of local inhibition of excitatory projections that regulate activity in other regions (32–34).

In terms of glutamate, a few studies observed a positive relationship between glutamate and local neural activity in the ACC and PCC (29, 44, 46, 60). Additionally, both positive and negative relationships with distal neural activity were found. These findings add to the simplified positive relationship with inter-regional neural activity reported in Duncan et al. (4). A negative relationship between BOLD activity in DMN regions, such as the PCC/PCu, were speculated to be due to higher ACC glutamate levels supporting facilitation of salience processing, thus resulting in stronger deactivation of DMN regions (39, 45). However, this relationship seems to be region-dependant as glutamate levels measured within the DMN, specifically the PCC, were positively related with DMN neural activation, where higher glutamate levels predicted lower local deactivation (60). The positive relationship between glutamate levels and intra-regional neural activity may stem from an excitatory drive stimulating neural activation within the same region through recurring feedback (46, 69). However, the relationship between glutamate levels and intra-regional neural activity may be difficult to replicate: one study failed to replicate a positive correlation between ACC glutamate and ACC neural activity at a 6-week follow-up scan within the same cohort (46). In fact, Enzi et al. (47) suggested resting glutamate levels may be more closely related to resting-state activation, thus being less intimately linked with task-related activity. As previously observed by Duncan et al. (4), glutamate levels were positively associated with neural activity outside of the sampling regions for glutamate. The long-range projections of glutamatergic neurons (70) are thought to influence neural activity in distal regions, thus giving rise to this relationship. However, a number of studies have shown that these projections may not simply reflect positive relationships with neural activity, but may also indicate that glutamate levels can negatively predict neural activity inter-regionally.

The ACC has been commonly investigated across studies, for which effects have been inconsistent. The discrepancies in ¹H-MRS x fMRI relationships may arise from functional segregation within the ACC (71), its involvement in different behaviours (72), and that it hosts both task-positive and task-negative subdivisions (e.g., pgACC and supragenual ACC). The ability of ¹H-MRS to discriminate between ACC subregions is still limited due to the ¹H-MRS voxel size, especially in the case of GABA. This may be improved by the use of high-field MRI scanners that will allow for smaller voxel sizes (73). An additional consideration is that relationships between metabolite levels and neural activity may also be task-specific. Studies that used an active control task, rather than a low-level baseline condition such as a fixation cross [e.g., (44, 53, 54)], indicate that baseline metabolite levels are not simply correlated with any type of neural response. When neural response was confirmed for both the task of interest and the control task, different relationships between GABA levels and neural response were found indicating that metabolite levels may not be simply associated with all types of neural activity within a region (54).

Despite some incongruent findings, a pattern of neurochemical-neurofunctional relationships has emerged, which give valuable insight into psychiatric disorders where this relationship is aberrant. Measurement of metabolite levels at rest with ¹H-MRS is thought to reflect individual differences in number of glutamatergic and GABAergic neurons, number of glutamatergic and GABAergic synapses, and glutamate and GABA turnover (68). It is conceivable, then, that abnormalities in metabolite levels would be reflected in differential neurochemical-neurofunctional relationships. In psychiatric disorders such as schizophrenia, case-control studies have suggested an aberrant relationship between ACC glutamate and local neural activity (46) and in regions receiving inputs from the ACC, such as the insula and DMN (34, 46, 74). Differences were found in the relationship between hippocampal Glx and inferior frontal gyrus activation, with medicated schizophrenia patients showing a less robust relationship (56) potentially reflecting differential coupling between the two regions (75). Activation related to prediction error within the substantia nigra was not correlated with regional glutamate levels in schizophrenia patients, which was interpreted as a glutamatergic dysfunction leading to abnormal prediction error coding (38). Furthermore, a relationship between glutamate and working memory activation, both within the dlPFC, was found in unmedicated schizophrenia patients where no such relationship existed in healthy controls—potentially due to compensatory action to glutamatergic insensitivity (52). Given the known and increasingly well-characterised abnormalities in glutamatergic and GABAergic function associated with this psychiatric disorder (76, 77) and the close relationship between metabolites and neurofunction, abnormalities in brain function should arise. The neurochemical-neurofunctional relationship has been investigated less frequently in other psychiatric disorders such as obsessive-compulsive disorder and major depressive disorder. While this line of investigation showed no differences in the relationship in patients with obsessive compulsive disorder (40, 51), patients with major depressive disorder showed stronger coupling between glutamate levels and NBR compared with healthy controls (31). Walter et al. (31) concluded that reduced resting state activity was driven by glutamatergic metabolism in patients with major depressive disorder. Differences in ¹H-MRS x fMRI relationships may be the result of variations in neurochemistry, differential abnormal functions regionally and inter-regionally, as well as structural abnormalities. Despite the correlational nature of all the studies identified here, understanding healthy relationships between metabolite and neural activity allows some insight into compensatory action within mental illness and can provide potential pharmaceutical targets for intervention. Targeting GABAergic or glutamatergic neurotransmission may be a promising avenue to treating symptoms of schizophrenia (78) and major depressive disorder (79).

The significant expansion of studies combining multi-modal ¹H-MRS and fMRI in recent years has enabled more detailed examination of the interactions between neurochemistry and neurophysiology in the healthy brain. These studies have shown robust negative correlations between GABA levels and local activation in regions such as the occipital cortex and ACC, and positive correlations between glutamate levels and activation in distal regions. These findings have implications for our understanding of pathophysiology of psychiatric disorders such as schizophrenia, a field in which studies combining ¹H-MRS and fMRI are burgeoning in recent years. In this context, findings from multimodal studies may provide key insights into the neurobiology of schizophrenia not only by informing how the interplay between neurochemistry and neurophysiology may be altered prior and across disease stages, but also by their translational potential to elucidate whether pharmacological modulation of glutamate and/or GABA may be a new target for addressing abnormal neurophysiology in this psychiatric disorder.

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.

GM, AK, and MJK: concept and design. CB and AK: systematic search. PBL and AK: data extraction. AK: statistical analysis and drafting of the manuscript. MJK and AK: statistical support. CD, JMS, PBL, GM, and AK: critical revisions of the manuscript. CD, JMS, and GM: supervision. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.