Approximately 35% of patients with Major Depressive Disorder (MDD), which ranked among the top 25 leading causes of disease burden worldwide in 2019, do not respond to two or more different antidepressant medications and meet the criteria for Treatment-Resistant Depression (TRD) (1). TRD is associated with a high proportion of physical and psychiatric comorbidities, long-lasting functional impairment, and increased suicide risk (2–4). To find effective therapeutic approaches, MDD has been investigated as a disease of monoamine deficiency over the last century, which coincided with an explosion of pharmaceutical agents targeting the system (5, 6). Under the monoamine hypothesis, the antidepressant response was thought to be achieved by increasing the levels of monoamine neurotransmitters serotonin, norepinephrine, and dopamine in the synaptic cleft via the targeting of the molecular mechanisms of reuptake inhibition and enzymatic blockade. However, in the following years, several studies have cast doubt on the validity of low monoamine levels as an underlying all-or-nothing principle behind the etiology of MDD (7–10). Furthermore, the heterogeneity of response to monoaminergic antidepressants, low remission rates, and increased treatment refractoriness remain unmet challenges that have warranted the search for alternative treatment options with a rapid onset of action (11, 12).

In the early 1990s, the momentous discovery that N-methyl-D-aspartate receptor (NMDA-R) antagonists, notably ketamine, led to rapid improvements in depressive symptoms in rodents (13) and humans (14) served as an impetus for proposing the glutamate hypothesis of depression. Supplemented with early evidence of significantly higher plasma glutamate levels in patients with mood disorders (15), this etiological framework postulates that alterations in glutamate release, clearance, and metabolism lead to sustained accumulation of glutamate in cortical and limbic brain areas that regulate emotions, cognition, and behavior, thereby promoting a depressive state (16, 17). According to this theory, altered levels of synaptic and extrasynaptic glutamate transmission would result in impaired synaptic connectivity in these regions, manifested by decreased synaptogenesis, excitation-inhibition imbalance, neuronal loss and atrophy, and deficits in the inhibitory fine-tuning (18). This hypothesis has shifted drug discovery efforts toward identifying and investigating the properties of novel pharmacological agents that target the glutamatergic system, establishing a new paradigm in the research and treatment of MDD (Table 1).

In the clinical setting, significant efforts have been undertaken toward establishing the efficacy, feasibility, and safety of glutamatergic interventions for MDD and TRD, with intravenous ketamine and intranasal s-ketamine researched the most extensively (19). However, the attempts to identify biological predictors of rapid antidepressant response to glutamate-mediating interventions have not achieved a clinically meaningful predictive value at an individual level. To date, glutamatergic clinical research is mainly hypothesis-driven, as opposed to being data-driven, and the field is missing translational predictive preclinical models. The findings concerning the effects of glutamate-mediating compounds on the human brain connectome remain unconsolidated. This makes the identification of robust and reproducible biomarkers of treatment effects and rapid antidepressant response challenging, thus limiting the application of these findings to the real-world psychiatric setting, where the heterogeneity of symptoms and treatment response as well as the presence of physical and psychiatric comorbidities are significant factors.

Motivated by these limitations, recent studies have sought to employ new methodologies to advance our understanding of individual differences that predict and characterize antidepressant response. Research has largely focused on techniques that provide quantifiable metrics of biological structure and function, such as multi-omics (transcriptomics, proteomics, metabolomics), neuroimaging, and network pathway analysis (19). Neuroimaging, in particular, provides robust and reproducible models of the functional neuroanatomy and the brain's network architecture, which can serve as predictive correlates of clinical and functional outcomes (20). One of the major advances in the field of neuroimaging is the demonstration that the activity of brain regions is organized into coherent networks that are functionally distinct (21). These networks, termed intrinsic connectivity networks (ICNs), represent coupled brain regions correlated over time in spontaneous or task-evoked activity fluctuations (22). ICNs are associated with human cognition and behavior that can be studied at rest or using neuropsychological paradigms (23). They are highly replicable (24–26) and are thought to be constrained by brain anatomy, sufficiently reflecting the structural topology of the brain (27–29). Studying functional alterations in ICNs specific to MDD and TRD can elucidate the heterogeneity of symptom manifestation and treatment response, serving as predictive biomarkers of resistance to treatment and impacting clinical and functional outcomes (30–32).

In this article, we aim to summarize the key advances in our understanding of how glutamate-mediating interventions modulate ICNs of the human brain. First, we provide a brief overview of the functional architecture of human glutamatergic networks and how alterations in their activity and connectivity features could overlap with those previously described in the context of ICNs pertinent to MDD. Next, we review brain imaging studies across modalities that report ICN activity and connectivity alterations predicting or characterizing rapid antidepressant response to glutamate-mediating interventions. Finally, we discuss how these ICN biomarkers contrast with those of slow-active monoamine-based treatments for MDD and provide commentary on the direction for the next generation of neuroimaging biomarker studies for glutamatergic treatments.

Glutamate is a major excitatory neurotransmitter in the brain, and ~80% of all neocortical synapses are glutamatergic (33). Upon release in the synaptic cleft, glutamate triggers changes in the conduction of action potential, neurotrophic function, and apoptosis pathways by binding to ionotropic NMDA-Rs, α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPA-Rs), and kainate receptors on the postsynaptic membrane (34). In parallel, glutamate binds to G-protein-coupled metabotropic glutamate receptors (mGluRs), mediating changes in cellular processes that are regulated by second messenger molecular cascades (34). At the systems level, these molecular alterations translate into structural and functional changes within glutamatergic neurocircuitry, which could contribute to psychopathology either directly, via the recruitment of the core brain areas and ICNs receiving glutamatergic innervation, or indirectly, via the modulation of monoamine, gamma-aminobutyric acid (GABA), and other neurotransmitter circuits.

In the human brain, several regions relevant to the core cognitive, behavioral, and affective functions are deeply interconnected via glutamatergic neurons. These include the prefrontal cortex (PFC), anterior cingulate cortex (ACC), nucleus accumbens (NAc), hippocampus (HPC), amygdala (AMYG), thalamus (TH), hypothalamus (HPT), and brainstem neurotransmitter centers regulating the release of norepinephrine (i.e., locus coeruleus), serotonin (i.e., raphe nuclei), dopamine [i.e., ventral tegmental area (VTA), substantia nigra pars compacta (SNc)], and GABA (i.e., substantia nigra pars reticulata). Broadly, brain glutamatergic networks can be categorized into cortical and subcorticolimbic. For the purposes of this review, we define cortical pathways as those with at least one node situated in the cerebral cortex, and subcorticolimbic pathways as those with no nodes in the cerebral cortex. The cortical glutamatergic pathways can be further subdivided into five major arcs (35, 36), illustrated in Figure 1. Similarly, the subcorticolimbic glutamatergic pathways can be categorized into hippocampal and amygdalar, where HPC and AMYG are the central limbic nodes (37), illustrated in Figure 2. Functionally, the subcorticolimbic pathways interconnect brain regions responsible for the generation of lower-order emotional and behavioral outputs in response to environmental inputs. Some notable examples include reward and motivation (NAc-VTA/SNc), memory formation and emotional engagement (HPC), detection of threat and activation of fear response (AMYG), and autonomic and homeostatic regulation (HPT). Cortical pathways, on the other hand, regulate lower-order outputs (“top-down” processing) and the subsequent fine-tuning of already existing higher-order representations (“bottom-up” processing). Taken together, due to the abundance of glutamatergic neurons, the transmission of glutamate through cortical and subcorticolimbic pathways accounts for the greatest proportion of the brain's functional connectivity profile, making it a principal mediator in the broad management and manifestation of complex cognitive and emotional processes.

A growing literature supports the notion that a widespread network dysconnectivity, as opposed to aberrant responses within select brain regions, is what characterizes and possibly drives pathophysiological changes associated with MDD (35, 38). Behind these changes, a variety of molecular processes related to glutamate cycling and release may be implicated, including the decreased expression NMDA-Rs (39–41), AMPA-Rs (41, 42), or mGluRs (43), which have been reported both in humans and animal models of depression, as well as in postmortem brains of suicide victims (39, 44). Ultimately, impaired glutamate neurotransmission would compromise synaptic connectivity, leading to reduced synaptogenesis and impaired cell signaling (45). However, these impairments are most likely non-generalized but rather involve structural and functional aberrations within specific nodes of major networks, for example, the PFC, HPC, or AMYG. Due to the highly interconnected nature of the brain, these regional alterations in node synaptic properties would likely further drive downstream changes in more distant nodes, causing widespread dysconnectivity of glutamatergic neurocircuitry both cortically and subcortically. Nevertheless, a definitive integrative pathway-focused framework that would describe how molecular changes in glutamate transmission contribute to brain connectivity impairments is yet to be proposed and experimentally validated.

However, in a similar fashion to how the monoamine deficiency hypothesis was proposed, the glutamate hypothesis initially attempted to associate the pathophysiology of depression with directional changes in absolute levels of glutamate within the synapse. For instance, existing in vivo proton magnetic resonance spectroscopy (¹H-MRS) literature largely supports the hypoglutamatergic view of depression, supplemented with the robust evidence of the reduction in glutamate and GLX (glutamate + glutamine) levels in those with MDD (46). A review of 16 ¹H-MRS studies, with a total of 281 MDD patients and 301 healthy controls (HC), reported significantly lower levels of glutamate and GLX in MDD participants, primarily in the ACC (47). Subsequently, the hypoglutamatergic hypothesis has been confirmed by a more recent meta-analysis on 1,180 MDD patients and 1,066 HC, which concluded that lower levels of GLX, primarily in the medial PFC (mPFC), were inherently linked with the etiology of MDD (48). The reports of successful restoration of GLX levels to normal that were observed in the ACC (49) and dorsolateral PFC (DLPFC) (50) of MDD patients post-electroconvulsive therapy (ECT) further support this notion. Several ¹H-MRS studies, on the other hand, have reported increased (51–53) and unchanged (54, 55) glutamate/GLX levels in MDD participants, suggesting that the direction and magnitude of glutamate-specific alterations may differ depending on the brain region or the network of interest (56). Therefore, studying the whole-brain connectome as if comprised of a set of major ICNs, each having their corresponding functional and behavioral significance, may provide a more specific direction toward characterizing regional changes in absolute glutamate levels and how those could translate into specific symptoms of depression.

While the exact number of ICNs and the functional role of each are not yet fully known, the consensus of neuroimaging studies has revealed the existence of 7–17 distinct functional ICNs based on stable network parcellations (57, 58). Multi-modal neuroimaging literature, including studies of functional magnetic resonance imaging (fMRI), electroencephalography (EEG), magnetoencephalography (MEG), and positron emission tomography (PET), have identified several candidate ICNs that are functionally relevant to the symptomatology and pathophysiology of MDD. In this review, we focus on seven candidate ICNs that have been consistently replicated in MDD neuroimaging studies, illustrated in Figure 3: the default mode network (DMN), ventromedial affective network (AN), ventral frontostriatal reward network (RN), frontoparietal central executive network (CEN), anterior cinguloinsular salience network (SN), frontocerebellar sensorimotor network (SMN), and frontovagal central autonomic network (CAN) (23, 38, 59–61).

Although the definition and characterization of these ICNs primarily emerge from the functional brain imaging data, each of these networks has a structural connectivity substrate that overlaps with anatomically defined glutamatergic pathways, proposed here based on the core nodes forming each ICN. Glutamatergic circuits existing in the human brain may interconnect brain areas within a single ICN, determining within-network connectivity patterns (Table 2A), as well as brain areas distinctly involved in two or more ICNs, determining between-network connectivity patterns (Table 2B). For instance, the DMN, the best known and the most studied ICN, comprises a reproducible set of regions active during stimulus-free control conditions of task-based neuroimaging (“at rest”) that become deactivated during a loaded cognitive task (62, 63). While it is known that central DMN nodes are mainly localized within the cerebral cortex, given the abundance of glutamatergic neurons in the brain (i.e., ~80%) and our understanding of neuroanatomy, cortical-cortical glutamatergic connections would constitute the majority of synapses within the DMN (edges), thus forming a substrate for its within- and between-network functional connectivity patterns. A similar theoretical approach could be applied to other ICNs, where our knowledge of the core network nodes and existing white matter tracts might shed light on the specific glutamatergic connections that possibly represent the edges of a network or a subnetwork (Tables 2A,B). While these assumptions are hypothetical and should be interpreted as such, they nevertheless rely on the well-described patterns of human white matter neuroanatomy and could greatly inform the neuroimaging community as to which particular glutamatergic connections within the brain may warrant attention of future investigations.

When considering cognitive impairment and prominent affective dysregulation characterizing MDD, a network model of depression views these deficits, as well as other MDD symptoms, as the manifestation of altered connectivity within and between major ICNs (Figure 3) (38). To summarize, enhanced DMN connectivity is associated with excessive self-referential processes and maladaptive rumination (64–67). Similarly, elevated connectivity of the AN, a crucial network linked with the processing and regulation of emotions, may underlie excessive negative feelings, or dysphoria (68–70). Conversely, attenuated RN connectivity has been suggested to account for the symptoms of anhedonia, characterized by the loss of pleasure, reward, and motivation, as well as reduced magnitude and duration of positive affect upon exposure to positive stimuli (71–74). The SN, responsible for task-switching and goal-directed behaviors, also appears to be hypoactive in MDD, which translates into poor incentive salience, impaired response inhibition and selection, and diminished ability to initiate a behavior through premotor and motor regions, thus further contributing to anhedonia (64, 75–77). In turn, depression is also characterized by cognitive and executive deficits, marked by inefficient top-down regulation of emotions and negative thinking. Impaired top-down control is thought to be attributed to diminished CEN connectivity, which has been seen in depressed individuals at rest (64, 78–80) and during tasks involving working memory (81), executive control (82), and affective interference (83). Hypoconnectivity within the SMN has been linked with the symptoms of psychomotor retardation (84, 85), and several studies have shown this association among patients with MDD (30, 86–88). A relatively “new” ICN incorporated into the network model is the CAN, which involves the HPT, AMYG, insula, and brainstem nuclei responsible for the control of bodily autonomic responses (61, 89). The HPT appears to be the central node of the CAN that integrates autonomic, endocrine, and sleep functions, and CAN hyperconnectivity possibly accounts for the strong association between depression and physiological mechanisms inappropriately regulated by the autonomic nervous system, including the distinct markers of elevated heart rate, low heart rate variability, increased arousal, and responsivity to stress. All these subsequently translate into changes in the sleep cycle, neuroendocrine response, sexual function, and appetite that are frequently observed in MDD patients.

While within-network connectivity within each of the seven core ICNs could potentially represent a neural substrate for specific affective, cognitive, behavioral, and physiological constructs of depression, the interaction among their respective nodes is far more complex. Functional communication between the ICNs, or between-network connectivity, also plausibly contributes to the expression of depressive phenotypes. For instance, the state of negative self-referential thinking has been attributed to disrupted between-network connectivity of the DMN and AN with the SN and CEN, as well as the hyperactive subgenual anterior cingulate cortex (sgACC)—a central node of the AN (23, 90). Specifically, the posterior DMN displays elevated connectivity with the SN and CEN, while the anterior DMN displays diminished connectivity with the AN and SN. The connectivity between the SN and AN also appears to be reduced in depression. Hence, it is not the elevated connectivity within the DMN alone but also the interaction between the specific DMN nodes and nodes of other ICNs that account for the overall manifestation of the functional profile of negative self-referential thinking. Therefore, when considering ICNs as biomarkers and biologically plausible therapeutic targets, one should not treat ICNs as distinct units but should view them as integrated whole-brain functional circuits. In fact, the basal ganglia (91) and cerebellum (92) have their own ICNs forming their respective organization and topography, which further supports this notion.

Since ICNs, at large, constitute functional rather than structural units, determining the neurochemical source driving ICN changes is challenging. ¹H-MRS is a specialized neuroimaging technique that enables in vivo quantification of metabolites within pre-specified regions of interest, and among them, the concentrations of glutamate, glutamine, and GLX (93). Although this technique is subject to confounds, such as functional status of recruited samples, magnet strength (3T vs. 7T), and exposure to medication (94), pairing ¹H-MRS with functional neuroimaging modalities may partially mitigate the challenge of determining the neurochemical source of signal, providing an insight into the role of glutamate in the circuits involved in the pathophysiology of depression (95). While the interpretation of neuroimaging findings is generally constrained by the inverse problem, where the source of observed signals can be inferred only indirectly, the abundance of glutamatergic neurons (~80%) (33) and the role of glutamate in neuroenergetics (96–98) may in part account for ICN-related changes.

Mechanistically, glutamatergic signaling contributes to the excitation-inhibition balance responsible for the generation of neural oscillations that underlie slow fluctuations of neuronal activity visualized by means of brain imaging (99, 100). In other words, glutamate is a neurotransmitter that orchestrates synchronized neuronal activity across the entire brain, and the temporal correlation between these patterns of activation is what comprises the ICNs. The emerging literature has actively examined the relationship between glutamate levels and functional network small-world and global properties. As demonstrated by in vitro experiments, acute glutamate treatment leads to increases in similarity and connectivity weight between cultured neuronal networks, implicating enriched communication between neurons and stronger functional connection (101, 102). Animal studies have demonstrated an association between glutamatergic neurotransmission, neuronal firing rate, and blood-oxygen-level-dependent (BOLD) signal in the rat brain (103, 104), while in vivo human multi-imaging studies have shown that glutamate plays a vital role in modulating BOLD response both at rest and during functional tasks (98, 105, 106), as well as the functional connectivity between brain regions (105, 107–114). Additionally, a recent systematic review and meta-analysis of ¹H-MRS-fMRI studies combining the two modalities reported converging evidence supporting a positive association between glutamate levels and local (activity inside the spectroscopy voxel) as well as distal (activity outside the spectroscopy voxel) task-related brain activity (115). This evidence indicates that glutamate exerts a global effect on the BOLD response via abundant long-range glutamatergic projections to other cortical areas rather than modulating a local BOLD response within the acquired ¹H-MRS voxel (111, 116).

From this standpoint, changes in synchronized neuronal oscillations caused by disrupted glutamate neurotransmission may indeed be associated with the aberrant dynamics of ICNs, and the relationship between glutamate levels and ICN activity and connectivity has been confirmed in psychiatric populations. ¹H-MRS studies have demonstrated an association between altered metrics corresponding to glutamate levels and functional changes in the nodes of the DMN (108, 109, 112, 117), AN (118–123), RN (124, 125), CEN (126, 127), SN (98, 110, 116, 128), SMN (129), and CAN (130). Of note, a study by Levar et al. (126), which examined the relationship between glutamate levels in the dorsal ACC (dACC) (i.e., the core SN node), as well as the within-network connectivity of the DMN, SN, and CEN, reported no significant association between resting-state connectivity patterns in any of the ICNs and absolute glutamate levels (126). However, low GABA/glutamate ratios were linked with elevated connectivity within the DMN and SN, suggesting that the interplay between GABA and glutamate levels, rather than absolute glutamate levels, might drive network functional connectivity.

Given the plausibility that functional connectivity of ICNs might be positively correlated with absolute or relative glutamate levels between the synapses, the brain network model fails to support the absolute state of cortical hypo- or hyperglutamatergia in depression. Instead, in an MDD patient, glutamate levels would be dynamically altered as a function of ICN involvement during a particular mental state or task engagement, representing a “functional pathology” as opposed to a static neurochemical one. For instance, glutamate levels might remain inappropriately elevated at rest during the DMN involvement but would be pathologically diminished when the engagement of the SN or CEN is warranted (i.e., switching from the resting state to the state of information processing). Furthermore, another hypothetical assumption is that baseline glutamate levels in depression may vary across the cortex based on the neuroanatomy of specific ICN nodes and edges. For instance, cortico-cortical tracts between mPFC/rostral ACC (rACC) and posterior cingulate cortex (PCC) (i.e., the core DMN nodes) might have increased glutamatergic neurotransmission, while cortico-cortical tracts interconnecting dACC and anterior insula (AI) (i.e., the core SN nodes) might display diminished glutamatergic neurotransmission. This is despite the fact that both tracts pass in anatomical proximity to each other through the cingulum bundle, given that the nodes of DMN and SN are localized within rACC and dACC, respectively. From the mechanistic point of view, targeting the networks selectively makes treatment particularly challenging since currently available glutamatergic compounds lack this specificity and precision, and their basic pharmacodynamic properties do not get modulated by specific mental states.

In the following section of the article, we review the present state of evidence from studies that utilized glutamatergic treatments to investigate changes in ICN functional dynamics in MDD participants. The ICN changes discussed in this review emerge from studies that administered glutamate-modulating compounds as treatment for MDD/TRD, and most studies did not directly measure glutamate, glutamine, or GLX levels (Table 3). Therefore, these findings should be interpreted with caution, as the association between ICNs and glutamate signaling is presumed based on understanding the pharmacodynamics of administered compounds and not the direct association between neuroimaging and neurochemical metrics. We present the results of the reviewed studies grouped into the effects of glutamatergic interventions on ICN functional metrics (Table 4A) and the ICN biomarkers that predict antidepressant response (Table 4B). Observed changes in specific ICNs would provide mechanistic insights into the etiology of brain network abnormalities in depression. This is the first step toward designing prospective glutamate-focused ¹H-MRS-fMRI whole-brain connectivity studies, furthering the development of future glutamate-mediating treatments to selectively target these functional abnormalities within the brain connectome.

The discovery of rapid antidepressant effects of ketamine, an arylcyclohexylamine derivative and a non-selective NMDA-R antagonist, has revolutionized the field of investigational treatments for depression. The first data demonstrating the antidepressant properties of racemic ketamine were published by Berman et al. (14). Since then, several meta-analyses have confirmed its efficacy for depression, including TRD (170–172). Ketamine is believed to antagonize the NR2B subunit of an NMDA-R situated on GABAergic interneurons. This results in local circuit disinhibition and a consequent increase in the release of glutamate and brain-derived neurotrophic factor (BDNF), triggering the upregulation of AMPA-R expression and the stimulation of non-NMDA-R-mediated glutamatergic neurotransmission (19, 173). Molecularly, these processes modulate the second messenger intracellular signaling pathways involving mammalian target of rapamycin (mTOR), Calcium/Calmodulin-Dependent Protein Kinase II (CaMKII), the Eukaryotic Elongation Factor 2 Kinase (eEF2K) pathway, methyl-CpG-binding protein 2 (MeCP2) phosphorylation, and brain glycogen synthase kinase-3 (GSK3) (174–179). At a systems level, these molecular cascades stimulate synaptic plasticity and synaptogenesis, leading to the reorganization of neural networks in adaptation to the inputs from the environment (19, 180). Behaviorally, while the rapid antidepressant effects appear to be evident from clinical observations, increased synaptogenesis could contribute to long-term changes in ICNs and, as a corollary, a sustained antidepressant response.

The efficacy, feasibility, safety, and tolerability of alternative glutamate-mediating interventions have been assessed in numerous MDD clinical trials. Despite these efforts, the ICN biomarker signatures of rapid antidepressant response to these compounds have not yet been identified. At present, there is a minimal number of published peer-reviewed neuroimaging studies that have examined connectivity markers among MDD participants treated with these glutamatergic compounds, although several research groups have incorporated neuroimaging protocols into their active trials and have presented preliminary results at scientific meetings and conferences. The widespread interest in the glutamate hypothesis of depression and the existing evidence showcasing that ketamine successfully targets disrupted brain connectivity in depression makes it incredibly likely that the clinical and academic community can expect novel exciting developments in this area in the next decade.

Slow-onset monoaminergic pharmacotherapies and their relationship to brain connectivity changes is a major research topic of interest, given that, in the clinical setting, monoamines are still prescribed as the first-line treatment for MDD. As of now, the need for robust and reproducible biomarkers of monoamine antidepressant response is apparent since those potentially have a greater prospective clinical utility. Conventional antidepressants have, in fact, demonstrated promise in terms of mitigating brain network connectivity abnormalities among MDD participants. For instance, fMRI studies have reported changes in the DMN, AN, RN, and CEN functional connectivity following antidepressant treatment with SSRIs or serotonin-norepinephrine reuptake inhibitors (SNRIs) (23, 30, 32, 38, 314–318). Further, research has suggested that baseline structural and functional connectivity of the DMN, SN, AN (sgACC, HPC, AMYG), and CEN (DLPFC) may predict antidepressant treatment response (23, 38). In particular, stronger posterior-anterior DMN connectivity prior to treatment, as well as increased connectivity within this network generally, have been found to correlate with the treatment response following monoaminergic pharmacotherapy (32, 314, 318). Further, greater RN-DMN pre-treatment functional connectivity, specifically between the NAc and rostral ACC, has been found to be associated with bupropion treatment response (319). With respect to the AN and SMN, decreased AMYG functional connectivity with the left SMA and right precentral gyrus, as well as increased AMYG functional connectivity with the right central opercular cortex, have been found to reflect a favorable response to SSRI treatment (320). Previous research has also found that areas of the CEN with low functional connectivity, such as the DLPFC, are associated with a greater response to sertraline (321). In contrast, decreased functional connectivity of the right insula, a node of the SN, has been linked with insufficient antidepressant response (322). Between-network connectivity has also been suggested to be a predictor of response to pharmacotherapy. A large study on patterns of functional connectivity and sertraline treatment outcomes in MDD found that, in general, greater between-network connectivity predicted a better response to sertraline (314). In addition, greater hippocampal connectivity to various networks (e.g., HPC-DAN) and decreased connectivity with the SN was associated with a better placebo response (314). Taken together, these findings suggest that connectivity biomarkers in MDD are complementary in predicting response to slow-onset monoaminergic pharmacotherapies.

Despite these advances, the therapeutic time lag and inadequate response rates that are associated with monoamine therapies make them a suboptimal treatment option for MDD (323), and novel glutamatergic treatment options might be a better alternative. The hypothesis of monoamine deficiency lacks a solid evidence base, but since a variety of treatment interventions of different forms and styles are efficacious for depression, including conventional and non-conventional pharmacotherapy, brain stimulation, psychotherapy, it is reasonable to infer that these modalities would act on a converging final common pathway that underlies antidepressant response (7). Neuroplasticity, or structural and functional processes related to growth, maturation, apoptosis, and communication of neurons, as well as their ability to change in response to the environment, is thought to be the “final common pathway” of antidepressant response that can be targeted by multiple treatment interventions.

Both monoaminergic and glutamatergic antidepressants improve morphological and functional neuroplasticity. In fact, mechanisms of action of monoaminergic and glutamatergic antidepressants are thought to converge at the synaptic connectivity level, where both modalities target neurotrophic factors leading to neuroplasticity changes in synaptic connectivity and neural wiring (45, 212, 324). The key molecular pathways where the two types of antidepressants converge include the reduction of the depolarization-evoked release of presynaptic glutamate, which mitigates the effect of excitotoxicity and stimulates neurogenesis and synaptic strength, the enhancement of AMPA-R promoted by the inactivation of NMDA-R, the induction of long-term potentiation processes, and the release of growth factors, such as BDNF [for review, see (7)]. These molecular mechanisms subsequently result in changes in neurocircuitry and network properties, leading to behavioral manifestations of antidepressant response (45). This is supported by recent advances in neuroimaging, where connectivity studies have shown a high association between glutamatergic and monoaminergic CFPs in the MDD population and where a unique antidepressant CFP identified for ketamine predicted treatment response to monoaminergic pharmacotherapy (151, 212).

However, monoamines mainly act on a serotonin transporter, which only indirectly regulates the activity of glutamatergic receptors. While it plays an important role in modulating neuroplasticity, this pathway is much weaker, slower, and less efficient in comparison to the direct modulation of ionotropic glutamate receptors exerted by ketamine and other interventions (7). Glutamate is the brain's major excitatory neurotransmitter, and all brain functions related to cognition and emotions are ultimately mediated by the interplay between excitatory glutamatergic transmission and inhibitory GABAergic transmission (16). The ICNs discussed in this review comprise glutamatergic connections by more than 80%, while monoaminergic connections account for a significantly lower proportion. Ketamine and other related interventions would take a faster and shorter route to promote neuroplasticity and induce network-level remodeling, leading not only to a faster antidepressant response at the behavioral level but also to more widespread and long-term neurobiological changes.

Given that MDD is marked by limited responsivity to conventional pharmacotherapy and psychotherapy (1, 11), biomarkers of response to newly emerging alternative treatments are in high demand. Advances in functional neuroimaging have enabled the identification of neural correlates of rapid antidepressant response with robustness and specificity, and, as evident from this review, they will continue serving as a valuable tool for the future generation of biomarker studies. While the field of glutamatergic treatments for mood disorders greatly benefits from emerging functional neuroimaging, genetic, and neurophysiological studies, structural brain imaging research remains limited. Likewise, the functional neuroplasticity hypothesis that is currently the leading model in the field needs to be accompanied by specific evidence of structural plasticity, which provides opportunities for reverse translation to animal models. In clinical research, conversely, supplementing the field with diffusion-based neuroimaging and longitudinal volumetric data can be of major benefit to the future generation of biomarker studies that will strive to elucidate where and how glutamatergic interventions act upon distinct brain systems. Integrating structural and functional data will also enhance the accuracy and power of predictive models since models constructed through multiple integrated modalities tend to perform better than those trained using a single modality (325).

This review proposes a neuroanatomical framework for studying glutamatergic ICNs in depression. While this approach may reflect a reliance on a limited number of nodes and edges that are selected a priori, it can improve the precision and specificity of glutamate-based therapeutics and put research findings into medical perspective, bringing them closer to clinical practice and prospective use for diagnostic and treatment purposes. The ICN framework, in particular, is a valid and reproducible approach inspired by the principles of graph theory that mirrors the hierarchical organization of the brain's intrinsic architecture (182, 183, 188). Neuroimaging studies striving to describe the systems-level effects of glutamatergic treatments may greatly benefit from the characterization of ICNs provided here, which will aid in enabling the exploration of connectomic signatures of rapid antidepressant effects in a harmonized fashion. Correspondingly, the ICN theoretical model can be applied across brain imaging studies for the purposes of repeatability and reproducibility, as well as systematized interpretation of findings. This may ensure faster and more efficient progress in the development of effective glutamate-based therapies for mood disorders.

Existing literature reviews in the field of glutamatergic ICNs support the notion that glutamate-based therapies act on several brain networks, leading to a rapid antidepressant response (38, 326, 327). As discussed by Alario and Niciu (326), further research exploring how glutamate-based therapies for depression, such as ketamine, have long-lasting effects with short-term connectivity changes is needed. The promising findings that exist to date suggest the importance of coordinating the timing of study and neuroimaging procedures in a way that is physiologically appropriate to capture acute vs. delayed effects of glutamate-based therapies within affected ICNs (326). Future research may also employ long-term follow-ups to generate an understanding of how such delayed effects occur at a physiological level and provide insight into sustainable long-term changes in brain connectivity following therapeutics that target the glutamate system.

Furthermore, the biomarker studies reviewed here are characterized by a notable variability in employed neuroimaging procedures and analysis methods (Table 3). Other authors have also raised this issue. For instance, Kotoula et al. (327) points to a discrepancy between ways of measuring brain connectivity across healthy and depressed samples, as well as between ways of measuring acute and delayed effects of treatment interventions, including ketamine. This makes the interpretation of ICN findings challenging, and the use of different analysis pipelines within a single study may act as a potential confound. Further, the tendency to use small neuroimaging sample sizes with such novel therapeutics may also prevent a true understanding of how glutamate-based therapies act on various brain networks. Future research should use harmonized neuroimaging procedures and analyses and larger MDD and HC sample sizes across studies and sites.

Nevertheless, it seems evident that ICN changes induced by ketamine and other glutamatergic treatments comprise both the elevation and the reduction in functional connectivity and nodal strength. The prevailing theoretical model behind these mechanisms is the modulation of functional neuroplasticity, where glutamatergic compounds lead to rapid alterations in synaptic connectivity. As noted by Abdallah et al. (212), it seems likely that both increases and decreases in synaptic connectivity are needed to induce antidepressant effects. However, how changes in neuroplasticity at a molecular and cellular level translate into system-level changes in ICNs remains largely unknown, both in the context of glutamatergic modulation and in the context of depression. Marrying the modalities of genetics, cellular electrophysiology, neuroimaging, and behavioral markers has the potential to bridge this translational gap, and trials with this aim that explore serotonergic antidepressant response are already underway (328, 329). For glutamate-based therapeutics, this should be the focus of prospective research in the next decade, where neuroimaging would be used as a “tool in the toolkit” (23) and not as the only method for drawing conclusions about complex biological processes that drive network changes.

Through this review, we have summarized advances in the understanding of how glutamate-modulating interventions impact ICNs of the human brain. The reviewed evidence suggests that there are favorable prospects in using glutamate-modulating interventions to target human ICNs that are implicated in depression. Further, we have identified relevant biomarkers of rapid antidepressant response, such as altered functional connectivity of limbic, cognitive, and executive nodes, to aid in understanding and rationalizing the use of glutamate-based therapeutics as a neurobiologically sound alternative to existing slow-acting treatment options for depression. As research in this field continues to emerge, it may become evident how structural and functional changes within implicated ICNs translate into improvements in clinical and behavioral manifestations of depression. Finally, we have contrasted the connectomic theoretical principles underlying slow-onset monoaminergic pharmacotherapies with those of glutamatergic interventions, highlighting the convergence of their mechanisms of action on processes that regulate and drive neuroplasticity. Such preliminary neurobiological evidence supports the use of glutamate-based pharmacotherapies for disorders of brain dysconnectivity, such as depression.

ID, KD, and VB co-designed the theoretical framework for this article. ID and VB conceptualized the manuscript. ID performed the literature search and analysis, drafted the initial version of the manuscript, created figures and tables, and participated in subsequent review and editing. VT contributed to manuscript writing, performed literature investigation, assisted with the preparation of tables, and participated in review and editing. SK contributed valuable ideas to the conception of the article and assisted with review and editing. KD contributed valuable ideas to the conception of the article, provided writing supervision, and assisted with review and editing. VB provided primary supervision for creating this review and critically revised the manuscript. All authors approved the final version of this article.

SK has received honoraria or research funds 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, Sun Pharmaceuticals, and holds stock in Field Trip Health. KD was supported by an Academic Scholar Award from the University of Toronto, Department of Psychiatry. VB is supported by an Academic Scholar Award from the University of Toronto, Department of Psychiatry, 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 Defense (Government of Canada), New Frontiers in Research Fund, Associated Medical Services Healthcare Inc, Novartis, and from Roche Canada for an investigator-initiated trial. The remaining 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.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.