Modulators of GABA_A receptor-mediated inhibition in the treatment of neuropsychiatric disorders: past, present, and future

Abstract

The predominant inhibitory neurotransmitter in the brain, γ-aminobutyric acid (GABA), acts at ionotropic GABA_A receptors to counterbalance excitation and regulate neuronal firing. GABA_A receptors are heteropentameric channels comprised from subunits derived from 19 different genes. GABA_A receptors have one of the richest and well-developed pharmacologies of any therapeutic drug target, including agonists, antagonists, and positive and negative allosteric modulators (PAMs, NAMs). Currently used PAMs include benzodiazepine sedatives and anxiolytics, barbiturates, endogenous and synthetic neurosteroids, and general anesthetics. In this article, I will review evidence that these drugs act at several distinct binding sites and how they can be used to alter the balance between excitation and inhibition. I will also summarize existing literature regarding (1) evidence that changes in GABAergic inhibition play a causative role in major depression, anxiety, postpartum depression, premenstrual dysphoric disorder, and schizophrenia and (2) whether and how GABAergic drugs exert beneficial effects in these conditions, focusing on human studies where possible. Where these classical therapeutics have failed to exert benefits, I will describe recent advances in clinical and preclinical drug development. I will also highlight opportunities to advance a generation of GABAergic therapeutics, such as development of subunit-selective PAMs and NAMs, that are engendering hope for novel tools to treat these devastating conditions.

Introduction

GABA (γ-aminobutyric acid) is the dominant inhibitory neurotransmitter in the brain. Released from a population of predominantly local interneurons, representing only ∼10% of all neurons in the brain, activation of GABA_A receptors by GABA increases membrane Cl⁻ conductance and hyperpolarizes the membrane potential of most neurons, thereby reducing the likelihood that postsynaptic neurons will generate action potentials. Although outnumbered ten-to-one, these inhibitory neurons serve as a powerful check on the ability of excitatory synapses to depolarize neurons. The critical balance between the strength of excitatory and inhibitory transmission is thus largely responsible for setting the level of excitability throughout the brain. Maintaining this balance is essential for controlling the ability of local circuits within all brain regions to perform their computational input-output functions. Notably, too little inhibition (or too much excitation) underlies the paroxysmal synchronization of neuronal discharge that characterizes epileptic discharge [1]. Conversely, it has been suggested that too much inhibition (too little excitation) may contribute to cognitive deficits in aging and dementia [2–4].

This article will review the evidence that changes in the balance of excitation and inhibition contribute to the genesis of several neuropsychiatric conditions. In addition, a new generation of medicines for neuropsychiatric diseases is being developed that targets GABAergic inhibition to provide relief of symptoms. Their psychiatric indications, targets, and mechanisms of action will be reviewed here, along with a look at opportunities for further development of novel therapeutics targeting GABA_A receptors.

GABA_A receptor heterogeneity

The actions of GABA are mediated by two classes of receptor: ionotropic GABA_A receptors and metabotropic GABA_B receptors. GABA_B receptors will not be discussed in this review. GABA_A receptors gate Cl⁻ permeable channels and contain several well-characterized allosteric binding sites capable of increasing and decreasing the efficacy with which GABA can activate the channel and inhibit neurons. These allosteric binding sites have been exploited to yield some of the most effective and widely used medicines targeting the brain, such as benzodiazepine anxiolytics, barbiturate sedatives, antiepileptic drugs, and many general anesthetics. Advances in the molecular pharmacology of GABA_A receptors, a better understanding of the pathophysiology of disease conditions, and successful development of novel compounds like brexanolone are now providing an impetus for the development of new GABA_A receptor targeting drugs to treat neuropsychiatric disorders, particularly various forms of depression and schizophrenia.

Brain GABA_A receptors are heteropentameric channels, assembled from subunits that are encoded by 19 genes in eight families: six α, three β, three γ, one δ, one ε, three ρ, one π, and one ϕ. The majority of brain GABA_A receptors are comprised of two α subunits, two β subunits, and one of the other subunits. The precise subunit composition of the receptor determines many aspects of its function, including the efficacy of GABA in opening the channel, channel opening and closing times, cellular localization, and sensitivity to allosteric modulators [5]. Some subunit compositions are ubiquitous, particularly those containing α1, β2, and γ2 subunits, whereas others exhibit highly specific localization. α6 subunits, for example, are expressed at high levels in cerebellum, α4 subunits are concentrated in the thalamus, and α5 subunits are enriched in the hippocampus and deep layers of the frontal cortex [6–8] (Fig. 1). There is considerable evidence of posttranslational modification of GABA_A receptors, particularly phosphorylation, which endows them with subunit-specific mechanisms for regulating their trafficking, their function, and the efficacy of allosteric modulators [9, 10].

Fig. 1. Images of adult human brain sections showing in situ hybridization for various GABA_A receptor subunits.

Upper row: adjacent sections through the superior surface of the orbital inferior frontal gyrus, corresponding to Brodmann’s area 46 or the dorsolateral prefrontal cortex. Lower row: adjacent sections through the hippocampal formation. Images taken from the Allen Brain Map (https://human.brain-map.org/ish).

GABA_A receptors have been localized to conventional subsynaptic sites, opposed to GABA releasing presynaptic nerve terminals, and to extrasynaptic sites along the dendritic membrane. These two receptor populations mediate different effects on the postsynaptic cell. Synaptic GABA_A receptors mediate fast phasic inhibition, i.e., canonical inhibitory postsynaptic potentials (IPSPs), in response to action potential-evoked GABA release from interneurons. Extrasynaptic GABA_A receptors, in contrast, respond to ambient levels of GABA in the extracellular space and mediate tonic inhibition [11].

Consistent with their distinct functional roles, synaptic and extrasynaptic GABA_A receptors have distinct molecular compositions. Synaptic GABA release causes short-lasting elevations of GABA to high concentrations (millimolar) within the confines of the synaptic cleft. IPSPs in many forebrain regions are prolonged by benzodiazepines and barbiturates, consistent with the ubiquitous expression of γ2 subunits [12]. Extrasynaptic GABA receptors can be activated by much lower concentrations of GABA (nanomolar to micromolar) because they contain high affinity α4 and α6 subunits, as well as δ or γ2 subunits [13–15].

In rodents, δ subunits are expressed at high levels in the dentate gyrus, thalamus, cortex, and cerebellum and at lower levels in the amygdala, hypothalamus, and basal ganglia [6, 16]. The presence of δ subunits in extrasynaptic GABA_A receptors appears to contribute to their low rate of desensitization, compared to synaptic receptors, allowing them to remain tonically activated [17]. In some cell types, such as hippocampal and layer 5 cortical pyramidal cells, α5 subunit-containing GABA receptors comprise a large portion of extrasynaptic GABA receptors [18]. Tonic activation of extrasynaptic receptors has significant impact on the resting membrane conductance of neurons and can thus significantly regulate cellular excitability and network function [19]. Lastly, GABAergic interneurons are also endowed with GABA receptors that are often of a different subunit composition from nearby principal cells. It should be noted that δ and α5 subunit-containing GABA receptors can also exist in subsynaptic locations, particularly in dendrites [20], and that extrasynaptic GABA_A receptors with α1–3 or α5, but not α4 or α6, are sensitive to benzodiazepines [18, 21].

Positive allosteric modulators (PAMs) of GABA_A receptors

Benzodiazepines, a class of compounds defined by their chemical structure, potentiate GABA_A receptors containing γ2 subunits, provided they express α1–3 or 5 subunits. Mutational strategies and the relatively recent determination of the crystal structure of the GABA_A receptor [22] have revealed that GABA binds at the interface between the α and β subunits. Typical benzodiazepines require a γ2 subunit for activity because they bind at the interface between it and an α subunit. This intersubunit interface provides the possibility of α subtype specificity [23].

Zolpidem and other so-called Z-drugs are chemically distinct from benzodiazepines and have achieved widespread use as a fast-acting, short-lasting hypnotic sleep aid, largely lacking in anxiolytic actions. Zolpidem competes with classic benzodiazepines for binding, and is displaced by the neutral antagonist flumazenil, like classic benzodiazepines. Consistent with its sedative effects, it displays higher affinity for α1 subunits than α2 or α3, and no affinity for α5 [24]. Mutagenesis studies and structural analysis reveal that zolpidem also binds to a second, pharmacologically distinct site from classic benzodiazepines, but still at the α/γ interface [25]. Unlike benzodiazepines, it has little or no anxiolytic activity.

Long-term use of benzodiazepines leads to both tolerance and physical addiction [26], a significant limitation on their use for treating chronic anxiety. Recent preclinical work has identified GABAergic interneurons within the ventral tegmental area (VTA) as critical to the development of addiction to benzodiazepines [27]. This brain region contains dopaminergic cells that project throughout the forebrain and is central to the development of addiction to many substances of abuse. Within the VTA, local circuit interneurons, but not dopaminergic neurons, express α1-containing GABA_A receptors. Benzodiazepines amplify the inhibition of GABAergic interneurons within the VTA, thereby paradoxically disinhibiting dopaminergic projection neurons and strengthening glutamatergic inputs onto them. This plasticity was shown to depend on activation of α1-containing GABA_A receptors with use of mice expressing benzodiazepine-insensitive α1 subunits [28]. Most promising for the development of non-addictive benzodiazepine therapeutics, a benzodiazepine site ligand that does not potentiate receptors containing α1 subunits failed to evoke this plasticity [28]. Furthermore, the slow escalation of benzodiazepine self-administration, a preclinical sign of liability for addiction, did not occur in the mice with benzodiazepine-insensitive α1 subunits [28].

Neuroactive steroids include both negative allosteric modulators (NAMs) of GABA_A receptors (e.g., dehydroepiandrosterone) and PAMs, such as pregnenolone and allopregnanolone, which are metabolites of progesterone, and tetrahydrodedoxycorticosterone, which is a metabolite of corticosteroids. Neurosteroid NAMs will not be considered further here because they lack psychiatric therapeutic indications at present. Neurosteroid PAMs bind at two distinct sites within the transmembrane domains of α and β subunits [29]. Low concentrations of steroids bind to a site within the transmembrane domains of α subunits that mediates their positive allosteric actions, whereas a second site at the α/β interface mediates direct activation of GABA_A receptors at high steroid concentrations. The involvement of the δ subunit in mediating the ability of neuroactive steroids to modulate GABA_A receptors is controversial, but there is evidence that neurosteroids are capable of potentiating both δ and γ2 containing receptors [30, 31]. Both synaptic and extrasynaptic GABA_A receptors are potentiated by neuroactive steroids [32]. Nevertheless, there may be explanations for a selective action at δ-containing receptors [31]. In particular, whereas neuroactive steroids are able to directly activate γ and δ containing GABA_A receptors, their ability to enhance GABA responses is greater for receptors containing δ subunits [33]. Mice lacking δ subunits, for example, display an attenuated sleep-inducing response and no anxiolytic response to neuroactive steroids [34]. It should also be noted that neuroactive steroids affect many targets in addition to GABA_A receptors [31].

Brexanolone, a formulation of the neuroactive steroid allopregnanolone for intravenous delivery, received federal approval for the treatment of postpartum depression in 2019 (Zulresso®)(discussed below). Brexanolone is active at both synaptic and extrasynaptic receptors and at both γ- and δ-containing receptors [35]. An orally active neuroactive steroid, zuranolone, has been developed by Sage Therapeutics. Zuranolone is active at synaptic and extrasynaptic GABA_A receptor compositions, displays no striking subunit preferences, and potentiates both tonic and phasic inhibition in brain slices [36]. Zuranolone has completed phase III clinical trials in major depressive disorder [37, 38] and was approved by the FDA for treating postpartum depression in 2023 (discussed below).

Barbiturates bind near the transmembrane interface of β subunits with α or γ subunits [39]. These sites may overlap with those for neuroactive steroids [22, 40]. Like neurosteroids, barbiturates can directly open GABA_A receptor channels in the absence of GABA. The ability of barbiturates to potentiate GABA responses is largely independent of the type of α subunit they express [41]. Consistent with unique binding sites, barbiturates increase the time channels remain open in response to a fixed concentration of GABA, whereas benzodiazepines increase channel opening frequency [42]. Neither affects single channel conductance.

Mechanisms of PAM action

How do nonspecific PAMs change brain function in a manner that promotes sleep and relieves anxiety? Obviously, by potentiating the synaptic and extrasynaptic actions of GABA, PAMs make almost all brain cells less excitable and less likely to be activated by afferent excitatory synapses. This decreased excitability will alter the balance of excitation and inhibition throughout the brain, with consequences for the behavior of the circuits those cells and synapses are embedded in. Indeed, electroencephalogram (EEG) recordings reveal that the sedative and anxiolytic behavioral actions of nonspecific PAMs are accompanied by changes in oscillatory activity [43, 44].

Sleep

Sleep, in particular, is characterized by various phases, each with its own unique EEG signature. The transition from wakefulness to sleep is accompanied by a transition of the EEG from a desynchronized state to one with robust synchronous oscillatory activity, characterized at first as slow wave sleep (4–7 Hz), followed by sleep spindles, consisting of regularly repeated bursts of 12–14 Hz activity [45].

Synaptic inhibition, particularly that mediated by GABAergic interneurons in the thalamic reticular nucleus, is critical to the timing of sleep spindles by modulating cortico-thalamic feedback circuits. Specifically, preclinical studies have revealed that spindles arise from the cyclic interaction of low-threshold voltage-dependent Ca²⁺ channels, which underlie the depolarizing phase, and the recruitment of GABAergic inhibition via the glutamatergic excitation of GABAergic interneurons in the thalamic reticular nucleus, whose inhibition of thalamocortical relay cells underlies the hyperpolarizing phase of the cycle [46]. By enhancing the hyperpolarization of relay cells, GABA PAMs increase the rebound activation of low-threshold Ca²⁺ channels, amplifying and slightly slowing sleep spindles [47].

Benzodiazepines, barbiturates, and neuroactive steroids all decrease latency to fall asleep and this is accompanied by a more rapid transition to non-rapid eye movement (NREM) sleep. In addition, PAMs suppress low frequency EEG activity and enhance sleep spindle activity (12–15 Hz), accompanying an increase in NREM sleep duration and a decrease in rapid eye movement (REM) sleep duration [48].

With regard to the use of brexanolone for postpartum depression, young and middle-aged women display sleep-related EEG signatures with greater power than males [49]. The frequency and power of sleep spindles vary across the menstrual cycle, with decreases during the follicular phase and increases during the luteal phases, corresponding to a decrease in sleep around ovulation and increase with onset of the luteal phase [45]. Administration of allopregnanolone enhances sleep through mechanisms shared by other GABA_A PAMs [50]. The normal increased production of endogenous neurosteroids during the luteal phase may thus mediate the increase in sleep spindles. In parallel to the changes in endogenous neurosteroids, the expression of the α4 GABA_A receptor subunit, a predominant thalamic subunit, is also regulated across the human menstrual cycle [51], suggesting that extrasynaptic GABA_A receptors may be particularly important for the effects of endogenous neurosteroids. Enhancement of sleep may contribute to the therapeutic actions of brexanolone and zuranolone.

Anxiety

There is less evidence regarding the mechanisms underlying the anxiolytic actions of GABA_A receptor PAMs, perhaps because no clearly accepted circuit that selectively controls anxiety levels has been identified due, in part, to difficulties in disentangling the partially overlapping experiences of fear, stress, and anxiety [52]. Chronic stress and early life trauma are often associated with human depression, and anxiety-like behavioral changes are often observed in preclinical chronic stress models. Acute stress elevates neurosteroid levels [53], whereas chronic stress leads to decreased neurosteroid levels [54]. Walton et al. [55] have recently observed that chronic stress in mice leads to anxiety-like behavioral effects that are accompanied by decreased GABA_A receptor δ subunit expression in the amygdala, a region often implicated in fear, as well as decreased allopregnanolone levels in the amygdala. Importantly, experimental recapitulation of these behavioral effects with manipulations of δ subunit expression or of the expression of neurosteroid synthetic enzymes in the amygdala of unstressed animals demonstrates a causative role for these stress-induced changes. Further evidence comes from the ability of experimentally enhanced neurosteroid production to ameliorate the effects of chronic stress [55]. Stress elevates endogenous neurosteroid levels in humans and administration of exogenous neuroactive steroids produces a sedating, calming effect, highlighting the capacity of this system to regulate stress [56].

In preclinical models, withdrawal of a range of chronically administered PAMs is accompanied by behavioral evidence of increased anxiety-like behavior. Furthermore, both the anxiolytic behavioral effects of benzodiazepines and the potentiating actions of benzodiazepines on electrophysiological GABA responses are decreased after six days of progesterone treatment [57]. These effects are likely to be mediated by changes in GABA_A receptor subunit expression because they are accompanied by increased expression of α4 subunits, typically expressed at extrasynaptic sites. These receptors are also likely to contain neuroactive steroid-sensitive δ subunits in some brain regions.

In summary, preclinical findings support the development and use of neuroactive steroids for anxiolytic actions in generalized anxiety disorder.

Premenstrual dysphoric disorder (PMDD)

PMDD is characterized by elevated anxiety, irritability, and lowered mood. Because its etiology is clearer, PMDD may offer some indications of how GABA_A receptor PAMs treat anxiety. PMDD symptoms typically manifest during the late luteal phase, when progesterone levels are falling. In a critical preclinical study, Maguire et al. [58] observed that hippocampal γ and δ subunits display altered patterns of expression across the estrous cycle, with δ subunits elevated during diestrus and γ subunits elevated during estrous. Corresponding to the increase in δ subunit expression, tonic inhibition in the dentate gyrus was increased. Mice in the estrous phase, when δ receptor expression is high, showed an increase in anxiety-like behaviors. It has thus been suggested that dysregulation of either δ receptor expression cycles or endogenous neurosteroids levels could contribute to the occurrence of PMDD [56, 58].

There is no consistent evidence, at present, that differences in endogenous neurosteroid levels are associated with the occurrence of PMDD in women [59]. Administration of allopregnanolone had no effect on anxiety in women with and without PMDD in one study, although anxiety in the two groups were not different at baseline [60]. On the other hand, a randomized, placebo-controlled clinical trial revealed that administration of the NAM sepranolone, an antagonist of neuroactive steroid activity, during the luteal phase reduced self-reported symptoms and negative mood in women with PMDD [61]. Better understanding of the interaction between GABA_A receptor expression and endogenous neurosteroid levels across the menstrual cycle is required for the development of rational therapeutics for PMDD.

Women with PMDD are also more sensitive to stress during the luteal phase, when progesterone levels are falling, both in self-reported stress perception and in physiological measures of hypothalamic-pituitary axis (HPA) activation [56]. There is also evidence of altered HPA axis regulation and endogenous steroid production in response to stress in women with PMDD [56].

Major depressive disorder

Major depressive disorder (MDD) is defined by the two cardinal symptoms of prolonged depressed mood and anhedonia, or the reduced ability to experience pleasure. These may be accompanied by a number of other wide-ranging symptoms, including altered sleep and appetite, lack of motivation, and impaired cognition. This broad symptomatology suggests wide-spread changes in brain function. Recent clinical and preclinical data suggests that genetic susceptibility and psychological and physical environmental factors combine to alter the balance of excitation and inhibition, leading to dysfunction of circuits responsible for mood, reward, motivation, executive function, and memory. While most work has been focused on changes in excitation [62–65], changes in GABAergic inhibition have also been observed which have been suggested to contribute to the pathogenesis of MDD [4, 66, 67].

Decreases in GABAergic inhibition could result from lower concentration of synthetic enzymes or GABA in presynaptic vesicles, loss of GABAergic interneurons and their synapses, or decreased expression of postsynaptic GABA receptors. Evidence favoring a GABA deficit in MDD includes reduced levels of GABA levels in cerebrospinal fluid and brain tissues from patients with depression, reduced expression of GABA-synthesizing enzymes in brain tissue from postmortem suicide completers, a reduced number of GABAergic interneurons in the brain tissue of patients with depression, and reduced expression of GABA_A receptor subunits [68–70]. These findings have not been replicated in all studies, however (e.g., ref. [71]). For example, lower CSF GABA levels were found not to be correlated with depression severity [72]. Similarly, low GABA was not correlated with suicidal behavior in men, but perhaps in women [73], although other studies have identified male only decreases in GABA [74]. There is also little evidence regarding how generalized these changes are across brain regions since most studies have examined only one cortical region. For example, Hasler et al. [75] observed a 10% decrease in GABA within the dorsomedial prefrontal cortex in MDD, but no change in the ventromedial prefrontal cortex.

If low GABA levels are causative in depression, then it may be expected that they would recover upon successful remission of symptoms. Bhagwagar et al. [76] observed that GABA levels remain lower in remitted patients than in healthy controls, however. Similarly, another study found that GABA levels were lower in MDD patients than healthy controls, but improvement of symptoms following electroconvulsive therapy was not accompanied by restoration of GABA levels [77]. Taken together, published differences in GABA levels are small and magnetic resonance spectroscopy is unable to distinguish the small, but critical, synaptic vesicle GABA pool from the much larger cytosolic GABA pool.

Changes in GABA concentrations could arise from changes in GABA synthesis. Differences in the expression of GABA synthetic enzymes have been observed in some, but not all, studies and generally not for the synaptic isoform, GAD-65, presumed to underlie inhibitory synaptic transmission [78, 79]. Decreased GAD-65 expression might reflect a decrease in the number of GABAergic interneurons and there is evidence to support this in some, but not all studies [79].

Postsynaptically, changes in GABA subunit expression in MDD are small, may not be widespread across depression-relevant brain regions [80], and have not been seen in all studies (e.g., ref. [81]). Indeed, radioligand binding studies have failed to detect differences in the density of benzodiazepine binding sites MDD in some studies [82] or only in some brain regions [83], but increases were observed in others [84, 85]. Synchronous paroxysmal epileptiform discharge is the consequence of experimentally impairing GABAergic function, but there is no evidence of hyperexcitability in EEG recordings from MDD patients, no consistent changes in power across different bands, and no evidence of asymmetries associated with depression [86, 87], as might be expected for a decrease in GABAergic inhibition.

Ultimately it is the functional strength of GABAergic inhibition that matters most. Non-invasive transcranial magnetic stimulation has been used to assess functional inhibition in awake human subjects by delivering pairs of stimuli at short intervals and quantifying the inhibition of the second response relative to the first at short latencies. Bajbouj et al. [88] reported lower levels of intracortical inhibition in unmedicated MDD patients compared to healthy controls, whereas others observed decreases in inhibition only in treatment resistant depression patients and not in unmedicated MDD patients [89, 90]. The magnitude of the decrease in inhibition was not correlated with symptom severity. Other studies have observed no significant differences in MDD [91]. Lastly, successful antidepressant therapy was not accompanied by changes in intracortical inhibition [92]. Unfortunately, because inhibition of muscle contraction was used to quantify the response to the stimuli in these studies, such experiments have only been performed over the motor cortex, not in cortical regions relevant to depressive symptoms.

The evidence of a correlation between the function of GABAergic synaptic transmission and the pathology of human MDD is thus inconsistent, at best, or largely negative (see also: ref. [79]), despite strong preclinical evidence [4, 9, 70]. A clear prediction of a causative role for impaired inhibition in the genesis of MDD is that treatments that potentiate GABAergic function should relieve depressive symptoms. The evidence of antidepressant actions of classic nonselective GABA PAMs, such as benzodiazepines and barbiturates, is largely negative [93–95]. Even when some positive effects were observed, it appears that the benefits may reflect improved sleep and lower anxiety, rather than improvement of core symptoms [96, 97]. Similarly, inhibiting GABA reuptake with tiagabine produced no benefits in bipolar depression [98] but did improve scores in MDD [99], although perhaps primarily by lowering anxiety [100]. Lastly, reducing GABA breakdown with vigabatrin elicits depression as a side effect [101, 102].

Zuranolone (also known as SAGE-217), an orally available synthetic neuroactive steroid created by Sage Therapeutics. The FDA approved Zuranolone for postpartum depression, but not MDD, in 2023. In a phase 2 placebo-controlled trial, zuranolone was administered to MDD patients for 14 days and symptoms were followed for six weeks [103]. Patients responded within the first few days of treatment. Immediately after cessation of treatment, patients treated with zuranolone showed a significantly greater improvement that those given placebo. Doubt has subsequently been cast on how well the effects persisted beyond this early time [104]. Furthermore, a subsequent phase 3 trial failed to detect a significant improvement with 14 days of dosing at 20 or 30 mg/kg [37]. Patients in a second phase 3 trial at 50 mg/kg did show a response slightly better than placebo during the dosing period, however, but the response was no better than placebo for the 1–4 weeks after the therapy was discontinued [38]. The lower dose trials reported no significant numbers of adverse events, however, 60% of patients in the higher dose study reported treatment-emergent adverse events that were not considered serious. In particular, 15% of treated patients reported somnolence and 7.5% reported sedation as side effects.

In summary, the results with nonspecific GABA_A PAMs in MDD have been disappointing. A potential explanation is the lack of receptor subtype or brain region specificity of the compounds tested to date. GABA_A receptors are expressed in every cell in the brain, hence these compounds do not target specific areas believed to be directly involved in the pathology of MDD. Alternatively, it is possible that the balance between GABAergic inhibition and glutamatergic excitation in MDD may be shifted towards more inhibition on average, not less, so that PAMs do not produce therapeutic benefits.

Postpartum depression

Episodes of major depression during pregnancy or within 4 weeks of delivery are clinically classified as postpartum depression (PPD). Occurring in 12% of new mothers, particularly those with a history of depression, anxiety, and severe stressors (childhood or adult trauma, poor social support, economic distress), PPD can have numerous adverse consequences for both the mother and the infant [105].

Clues to the genesis of PPD originated in preclinical research. Late stages of pregnancy are associated with high levels of endogenous neuroactive steroids in mother and fetus, increasing 3–10-fold over pregnancy [106]. Upon birth, these levels then plummet back to baseline levels within weeks. Knowing that neurosteroids preferentially act at δ subunit-containing GABA_A receptors, Maguire and Mody [107] discovered that the high levels of endogenous neurosteroids during pregnancy in mice were accompanied by downregulation of both γ2 and δ subunits and a corresponding decrease in the amplitude of synaptic and extrasynaptic GABA currents in dentate granule cells. Both subunit expression and electrophysiological responses recovered within days of parturition. They also showed that mice with genetic deletion of δ subunits displayed several anxiety- and depression-like behavioral abnormalities in the postpartum period that were not seen in wildtype mice. Furthermore, δ subunit-lacking mice displayed abnormal maternal behaviors. They concluded that proper downregulation of δ subunit expression during pregnancy and the rapid recovery of δ subunit expression upon birth are essential for coping with the elevated neurosteroid levels during pregnancy and their precipitous postpartum decrease. They further suggested that dysregulation of this plasticity could contribute to the etiology of PPD. In human PPD patients, postpartum allopregnanolone levels are positively correlated with altered functional connectivity of the prefrontal cortex, a region implicated in the pathophysiology of MDD [65], supporting the connection between behavioral changes in PPD and neuroactive steroid levels [108].

These findings led to the successful development of brexanolone (Zulresso®) for PPD by Sage Therapeutics. Administered intravenously over 60 h to women with PPD in an in-patient setting, brexanolone produces a rapid amelioration of symptoms [105, 109, 110]. Oral administration of zuranolone in an outpatient setting has also shown efficacy in a phase 3 placebo-controlled trial in PPD [111], perhaps providing a more efficient therapeutic option than intravenous brexanolone.

Although an unquestioned clinical advance, there remain considerable uncertainties about the mechanism of brexanolone actions in PPD (see ref. [31]). Does it correct for inadequate signaling at δ-subunit-expressing GABA_A receptors due to a pathological failure of women with PPD to properly upregulate their δ subunit expression after birth? Surprisingly, brexanolone restores normal postpartum behavior in mice lacking δ subunits [112], suggesting that it may act at multiple therapeutic targets. Does it relieve symptoms of depression by improving sleep? Altered sleep and insomnia is commonly associated with MDD and PPD, and sleep alterations are positively correlated with depressive symptoms [109]. In MDD, co-treatment with classical antidepressants and sedatives to eliminate insomnia improves outcomes better than treatment with antidepressants alone [113], testifying to the importance of proper sleep for mental health. Poor sleep quality during and immediately after pregnancy is a predictor for PPD. Indeed, cognitive behavioral therapy for insomnia alone during pregnancy can reduce the incidence of postpartum depressive symptoms [114]. Brexanolone was found to induce sedation and somnolence in 5% of treated PPD patients to a degree requiring discontinuation of treatment [105] It also induced a persistent increase in the remission of insomnia symptoms in treated patients [109].

Schizophrenia

Schizophrenia is characterized by continuous or recurring episodes of psychotic symptoms, such as delusions, hallucinations (often hearing ‘voices’), and other forms of disorganized thought, as well as social withdrawal, apathy, emotional blunting, impaired executive function, and impaired working memory. The latter processes are strongly associated with the frontal cortices (e.g., ref. [115]) and there is a correspondence between the late maturation of the frontal cortices and the developmental onset of schizophrenia [116]. Much like SSRIs and the formulation of the serotonin hypothesis of depression, the development of dopaminergic antagonists as antipsychotics led to the dominance of dopaminergic hypotheses for the origin of schizophrenia. Unlike serotonin and depression, however, there is abundant evidence for localized excess dopaminergic transmission in schizophrenia, particularly within regions of the frontal cortex and areas of the striatum to which they project [117]. More recently, the ability of the N-methyl-D-aspartate (NMDA) receptor blocker ketamine to induce dissociative perceptual dissociations and mild hallucinations in healthy subjects, resembling the symptoms of schizophrenia, and exacerbate symptoms in patients with schizophrenia [118], attracted interest in potential imbalances between excitation and inhibition in the genesis of the disease [119]. Preclinical studies have established that a prominent effect of ketamine is to reduce the excitation of GABAergic interneurons [120] and thus increase excitability and promote synchronous γ oscillations in rodents and humans [121]. Ketamine-induced gamma EEG oscillations thus parallel the increase in gamma power seen in the EEG of patients with schizophrenia in some studies [122–124]. Furthermore, ketamine-induced oscillations are diminished by antipsychotics [125]. In preclinical studies and healthy human subjects, administration of ketamine elevates striatal dopamine levels [126, 127]. Alterations of the balance between excitation that favor excitation can thus account for many of the classical observations of dopaminergic involvement in schizophrenia [128].

So how might excitation and inhibition become imbalanced? There is a large body of evidence supporting the dysfunction of GABAergic inhibition in a manner that is synergistic with, and potentially causative of, the above observations in schizophrenia. Radiolabeling studies in patients and in postmortem tissue have largely failed to detect substantial changes in the number of GABA_A and benzodiazepine binding sites [129, 130]. Similarly, spectroscopic evidence of changes in levels of GABA in human schizophrenia is inconsistent [130]. A reduction in the expression of the GABA synthetic enzyme GAD67, although not GAD-65, in postmortem brain tissue from patients with schizophrenia has been repeatedly observed, however [131–133].

Genome wide association studies of schizophrenia have identified genes encoding for GABA_A receptor subunits as potential risk factors. In particular, a risk locus on chromosome 5q contains the genes encoding α1, α6, β2, and γ2 subunits [134]. Schizophrenia associated haplotypes were associated with decreases in the expression of gene sets encoding a number of pre- and postsynaptic proteins in leukocytes, confirming some previous microarray results from brain tissue [135]. Microarray analysis identified decreased expression of α1, α4, β3, γ2, and δ subunits in the PFC [136], whereas real-time quantitative polymerase chain reaction analysis revealed changes in α1 and δ subunits [137]. These results are difficult to reconcile with the inconsistent changes observed in binding studies, however, changes in the relative expression of different subunits, their subcellular localization [138], or expression across cortical layers [139] have been reported and may be beyond the sensitivity of classic binding studies using nonspecific ligands to detect.

Functional studies using paired-pulse transcranial magnetic stimulation and EEG readouts has suggested decreased GABA_A receptor-mediated inhibition in the frontal cortex of patients with schizophrenia [140–142]. Despite the suggestion of pathologically impaired inhibition in schizophrenia, there is no evidence that nonselective GABA_A PAMS, chiefly benzodiazepines, exert beneficial therapeutic benefits, as might be expected if global impairment of GABAergic transmission was a cause of schizophrenia. They are used to reduce agitation during acute psychosis and sometimes as adjuncts to conventional antipsychotics, however [143].

One explanation for the inconsistent results is that, rather than a global decrease in GABAergic function, specific classes of interneurons become dysfunctional. GABAergic interneurons, particularly parvalbumin-expressing basket cells, which are largely responsible for fast, phasic, perisomatic inhibition, appear to be selectively dysfunctional in the prefrontal cortex [144, 145]. It remains uncertain whether there is a true loss of these neurons, or whether their expression of various markers is changed [146]. There is also evidence that the synaptic excitation of basket cells in the prefrontal cortex may be diminished in brain tissue from patients with schizophrenia [147]. This class of interneuron is particularly important for the proper generation of γ EEG activity in preclinical studies [148]. Similarly, changes in GABA_A receptor subunit expression are not equally distributed across all cortical layers [139] (Fig. 1). It is important to remember that such changes could be a cause of schizophrenia or a compensatory response of the brain to other pathological alterations.

Using a radioligand with high affinity for α5 subunit-containing GABA_A receptors, imaging in human patients with schizophrenia has revealed a negative association between symptoms and binding [149] and a lower overall binding patients with schizophrenia [150], consistent with a downregulation of α5 subunits in the hippocampus and prefrontal cortex, where they are uniquely expressed at high levels. Decreased α5 subunit expression was also observed with in situ hybridization [139]. Preclinical studies using methylazoxymethanol acetate (MAM) administration to mimic several behavioral abnormalities in schizophrenia revealed increased VTA dopamine release and reduced basket cell numbers. Gill et al. [151] used systemic administration of a novel α5 subunit-selective PAM in the MAM model and demonstrated that it reduced dopaminergic hyperactivity and abnormalities in amphetamine-induced hyperlocomotion. These effects were mimicked by intra-hippocampal infusion, consistent with its role in the cortico-mesolimbic system, and have subsequently been confirmed with two other novel α5-selective PAMs [152]. Similar effects were seen in an immune-challenge model [153].

Negative allosteric modulators (NAMs) of GABA_A receptors

Part of the rich pharmacology of GABA_A receptors is the availability of negative allosteric modulators. Known NAMs bind to the same site at the interface of α and γ subunits as benzodiazepine PAMs and can be displaced by the functionally neutral benzodiazepine antagonist flumazenil. Nonspecific NAMs include various β-carbolines, such as methyl-6,7-dimethoxy-4-ethyl-β-carboline-3-carboxylate, which inhibit GABA induced Cl⁻ currents in a manner that requires γ2 subunits and is prevented by flumazenil [154]. As such, nonspecific NAMs are proconvulsive and, because they exert an action opposite to anxiolytic benzodiazepine PAMs, anxiogenic [155], precluding their clinical utility.

α5 subunit-selective NAMS

Because known NAMs bind to the α-γ interface, it has been possible to design α subunit-selective NAMs. In particular, there has been considerable clinical interest in developing α5 subunit-selective NAMs because of the relatively selective expression of receptors containing these subunits in the hippocampus and prefrontal cortices in humans [156–159] and rodents [160, 161] (Fig. 2). They comprise ca. 5% of all brain GABA_A receptors, but in the hippocampus, they form roughly 20% of all GABA_A receptors [162] and comprise a large portion of the extrasynaptic GABA_A receptors [18].

Fig. 2. Localization of α5 subunit-containing GABA_A receptors.

Left image: Immunostaining for α5 subunits in a parasagittal section of the mouse brain (adapted from ref. [161]). Right images: PET labeling of α5 subunit-containing GABA_A receptors in healthy human subjects in parasagittal (left) and coronal (right) sections using [¹¹C]Ro15-4513 and spectral analysis (adapted from ref. [159]). In both species, α5 subunit expression is particularly concentrated in the hippocampus and frontal cortex.

Synaptic and extrasynaptic α5 subunit-containing GABA_A receptors are particularly important for limiting the ability of synaptic glutamate release to activate NMDA receptors [11, 163]. NMDA receptor channel activation requires the coincident binding of glutamate to the receptor and membrane depolarization to relieve the block of the ion channel by Mg²⁺ ions; relatively slow processes. α5 receptor-mediated hyperpolarization impairs that process, whereas α5-selective NAMs diminish the ability of synaptic inhibition to hyperpolarize the membrane potential and they thus promote NMDAR activation.

Synchronous oscillatory activity in the γ frequency range (30–120 Hz) is considered to play a role in memory and cognition. Hippocampal brain slices taken from mice with a genetic deletion of the α5 subunit are hyperexcitable [19] and display a change in the patterns of gamma oscillations they can produce [164]. Similarly, administration of α5-selective NAMs to mice results in an increase in the power of gamma oscillations in EEG recordings from the frontal cortex [165], an effect that is absent in α5 knockout mice [166].

The effects of NAMs on NMDA receptor activation and γ oscillations have led to efforts to develop them for their ability to enhance cognition and memory. Indeed, hippocampal-dependent learning is enhanced in mice with partial or complete deletion of α5-containing GABA_A receptors in fear and appetitive conditioning tasks [167, 168]. Similarly, spatial learning in the Morris water maze is selectively enhanced by both α5 subunit deletion and α5-selective NAMs [169, 170]. α5-selective NAMs also improve memory and learning defects in a mouse model of Downs Syndrome [171]. Importantly, α5-selective NAMs, unlike nonspecific NAMs, have not shown evidence of being anxiogenic or proconvulsant in mice or humans.

Basmisanil

Numerous α5-selective NAMs have been developed for their cognitive enhancing ability [172]. These compounds can display >10-fold selectivity for α5 over other α subunits in terms of binding affinity or functional efficacy and inhibit GABA responses at α5-containing GABA_A receptors by 50–75%. Only one, basmisanil, is approved for clinical testing in humans. In healthy human subjects, basmisanil is safe and well tolerated [173]. EEG recordings revealed an increase in relative power of theta frequency (ca. 4 Hz) activity and decrease in beta (ca. 20 Hz) frequencies. Unfortunately, gamma frequency activity was not adequately captured due to technical limitations in this study. Based on the preclinical findings, basmisanil was tested in a clinical trial for Downs Syndrome in adolescents and adults, but participants showed no improvement in cognition, executive function, language, or quality of life after 6 months of treatment, although it was well tolerated throughout the dosing [174]. Results from a second trial of 6 months of basmisanil treatment for cognitive impairment in schizophrenia were also negative (https://clinicaltrials.gov/ct2/show/NCT02953639). A third trial is currently recruiting for intellectual disability in children with Dup15q syndrome (https://clinicaltrials.gov/ct2/show/NCT05307679). This syndrome results from the duplication of a region of the chromosome containing genes encoding several GABA_A receptor subunits, including α5.

α5-selective NAMs in preclinical models of depression

Building on evidence that a single administration of ketamine exerts antidepressant actions by producing a brief period of partial disinhibition, Fischell et al. [175] hypothesized that a single administration of α5 subunit-selective NAMs would also shift the balance of excitation and inhibition in favor of excitation and exert a ketamine-like antidepressant action. Because the expression of α5 subunits is largely confined to regions of interest for cognition and reward, such as the PFC and hippocampus, unlike the ubiquitous NMDA receptors targeted by ketamine, they also hypothesized that α5-selective NAMs would exert fewer side effects than ketamine. Using chronic stress to produce an anhedonic phenotype in rats, two distinct α5 NAMs restored normal reward behaviors within 24 h, with benefits persisting for up to 7 days. Like ketamine, α5-selective NAM injection resulted in a prolonged period of enhanced gamma power in the EEG [165]. Unlike ketamine, α5-selective NAMs did not impair motor function or impair pre-pulse inhibition and did not induce a conditioned place preference [165]. Fischell et al. [175] also showed that restoration of excitatory synaptic strength at stress-sensitive synapses in the hippocampus accompanied the restoration of reward behaviors, providing a plausible biological mechanism for the behavioral response. The behavioral, synaptic, and EEG responses to α5-selective NAMs in mice are all prevented by flumazenil and are absent in mice lacking α5 subunits [166].

Where do we go from here?

Subunit-specific compounds

For many of the indications discussed above, nonselective modulators of GABA_A receptors have not proven effective. One approach that might advance GABA_A receptors as targets for novel therapeutics is to develop more compounds with selectivity for specific receptor subunits in order to either selectively affect specific behaviors and disease symptoms or to selectively eliminate unwanted drug actions. Benzodiazepines offer a clear example. The clinical efficacy of benzodiazepines for treating anxiety disorders is excellent, however, their use is severely limited by side effects, notably sedation and risk for addiction. This has led to a decades-long – and so far unsuccessful - search for the holy grail [21]: benzodiazepine site agonists with specificity for α subunits promoting anxiolytic responses, but not α subunits promoting sedation or abuse. Studies in mice using point mutations in genes for individual α subunits that eliminate benzodiazepine binding have revealed that the α1 subunit is required for the sedating, but not anxiolytic actions of benzodiazepines [176], whereas α2 and α3 subunits are more responsible for their anxiolytic actions [21]. Therefore, drugs that do not act at α1 subunit-containing receptors ought to be devoid of sedative side effects. Subsequent testing in genetically modified mice and the development of subunit-specific compounds and, along with early-stage clinical failures, such as ocinaplon [21], has led to doubts about this dichotomy. Furthermore, compounds with ‘ideal’ binding properties and promising behavioral profiles in preclinical experiments nevertheless had sedating actions in humans [21, 177].

The search for non-sedative anxiolytics continues, however, with α2/3 subunit-containing receptors the primary target [177–181]. One α2/3 preferring compound, TPA023, was shown to be anxiolytic but not sedating in preclinical species but exerted anxiolytic and sedative effects in humans [182]. It was ultimately withdrawn due to unrelated clinical complications. Even more exciting, in contrast to the general lack of effects of nonspecific PAMs in schizophrenia, four weeks of treatment with TPA023 improved cognition and improved task-related γ frequency responses in the EEG in patients with chronic schizophrenia in one study [183], although this result was not confirmed in a follow up trial [184]. The failure of this compound may be due to its low efficacy as a PAM [185]. AZD7325 in another α2/3 preferring PAM, but it did not display an anxiolytic action in a phase 2 clinical trial (https://clinicaltrials.gov/ct2/show/NCT00808249). AZD7325 was also tested for its effects on intracortical inhibition in human subjects, but no results were posted (https://clinicaltrials.gov/ct2/show/NCT02135198). It has now been licensed to Baergic Bio, who renamed it BAER-101, and are pursuing multiple indications.

The δ subunit is also promising target for drug development, given its implication in PPD and neurosteroid actions. One PAM that preferentially targets δ containing GABA_A receptors over γ containing receptors has been developed, DS2 [186]. In thalamic brain slices DS2 selectively enhances extrasynaptic GABA responses. Interestingly, it prefers α4 and α6 containing GABA_A receptors over α1-containing receptors and appears to bind to a novel site not shared by other PAMs. Preclinical studies have suggested that δ-selective PAMs may have benefits in opioid tolerance and withdrawal [187].

It is an exciting and hopeful sign of progress that there are many novel GABA_A receptor modulators advancing through the drug development pipeline for a range of neuropsychiatric and neurological disorders, many of them with subunit specificity (Table 1).

Table 1.

GABA_A receptor targeting therapeutics in commercial development.

Target	Compound	Action	Developer	Ind?	Indication
alpha2/3	KRM-II-81	PAM	RespireRX Pharmaceuticals	N	Epilepsy, pain
	GT-002	PAM	Gabather	Y	Cognition, schizophrenia, dementia
	ENX-101	PAMa	Engrail Therapeutics	Y	Epilepsy
	ENX-102	PAMa	Engrail Therapeutics	Y	Generalized anxiety disorder
	ENX-106	PAMa	Engrail Therapeutics	N	Spasticity, pain
	SAN711	PAM	Saniona	Y	Neuropathic pain, epilepsy
	SAN2219	PAM	Saniona	N	Epilepsy
	Baer-101 (AZD7325)	PAM	Baergic Bio	Y	Anxiety, autism, fragile X syndrome
	Darigabat	PAMb	Cereval Therapeutics	Y	Epilepsy
alpha5	basmisanil	NAM	Roche	Y	Cognition
	PNV-001	NAM	ProNovo Therapeutics	N	Depression, anhedonia
	alogabat	PAM	Genentech	Y	Autism
	Afizagabar (S44819)	antagonist	Servier	Y	Stroke
delta/	LYT-300	PAMc	Puretech	Y	Anxiety, PPD
nonspecific	brexanolone (Zulresso®)d	PAM	Sage Therapeutics	Y	PPD
	zuranolone	PAM	Sage Therapeutics	Y	Anxiety, depression, PPD
	Sage-319	PAM	Sage Therapeutics	?	GABA hypofunction
	Sage-689	PAM	Sage Therapeutics	?	Epilepsy
	ganaxolone (Ztalmy®)d	PAMe	Marinus Pharmaceuticals	Y	CDKL5 deficiency disorder
	PI-301	PAMf	Pantheric	?	Asthma
	GRX-917	PAMg	GABA Therapeutics	?	Anxiety, depression, PPD
	ETX-155	PAMh	Eliem Therapeutics	?	Depression, epilepsy

^aAlso alpha5 PAM, alpha1 NAM.

^bAlso alpha5 PAM.

^cAllopregnanolone prodrug.

^dFDA approved.

^eAllopregnanolone variant.

^fNot centrally active.

^gPromotes neurosteroid synthesis.

^hNeuroactive steroid.

There are several challenges to the development of subunit-specific compounds beyond chemistry. First, we lack good, brain-wide, high-resolution information about where GABA_A subunits are expressed and localized, and on which cell types. Although the data are good for mice, the human data are very incomplete. This is important because, for example, the extent to which α1- and α2-containing receptors are expressed on the same cell type, will make it harder for subunit-targeted compounds to produce differential effects, since both will tend to hyperpolarize the cell. This information must then be combined with knowledge of which circuits and brain regions to target. Preclinical advances and task-specific functional imaging in humans are progressing rapidly on this front. It should be acknowledged, however, that the ubiquitous and overlapping expression of so many receptor subunits, with the possible exception of the α5 subunit, makes the targeting of specific circuits and brain regions a difficult challenge.

Fortunately, the molecular pharmacology of GABA_A receptor subunits has now been very well characterized which should aid in the rational search for subunit-selective lead compounds. There is a vast and largely unexplored world of other binding sites at which novel subunit-selective compounds might be directed. For example, barbiturates and neurosteroids bind to sites that involve β subunits. Few compounds with selectivity for specific β subunits have been developed. Preclinical evidence suggests that PAMs with low activity at β1 subunits have more anxiolytic and less sedative activity [188].

Tool vs cure?

It should be obvious, but it bears restating: in drug development, as in all of science, a good mechanistic hypothesis with explanatory power is highly beneficial. We are used to thinking about how success in bringing compounds from preclinical stages to the market depends on good pharmacological and pharmacokinetic/pharmacodynamic properties and a lack of off-target actions, but we should be thinking equally hard about why and how we expect that a particular drug or drug target is going to be therapeutic in a given indication. There are two scenarios that apply to many of the conditions discussed above. Some therapeutics aim to provide a direct cure, i.e., to restore something missing or dysfunctional. An example is the development of brexanolone for PPD. There is evidence that the combination of falling neurosteroid levels and δ receptor downregulation in the postpartum period causes PPD, making provision of neurosteroids as a ‘cure’ a mechanistically sound hypothesis. The use of α5 subunit-selective PAMs for schizophrenia may also fall in this category.

Another potential therapeutic mechanism is that a compound provides an acute tool with which to engage endogenous processes that trigger beneficial responses. Single administration of psychedelics and ketamine, for example, seem to exert their persistent antidepressant actions by triggering endogenous neuroplastic responses that normally underlie learning and memory. α5-selective NAMs may share this ability [189]. Development of any hypotheses should take into account the anatomical distribution of the target receptor with the circuits that are likely dysfunctional in the indication. The development of non-sedating, non-habit-forming anxiolytics, for example, would benefit not only from subunit-selective PAMs, but also greater knowledge of the circuitry of anxiety and addiction.

Conclusion

The GABA_A receptor is one of the most accessible and powerful pharmacological targets in the brain and provides a powerful means to regulate the balance of excitation and inhibition in a directed manner to tune the function of brain circuits. The successful development of Zulresso, combined with new advances in the molecular pharmacology of the GABA_A receptor and new hypotheses about precision targeting of receptors containing specific subunits, offer great promise for a new generation of therapeutics for neuropsychiatric disease. Moving beyond our long established and powerful existing GABA therapeutics, we now stand, in the words of the pioneering neuropharmacologist Hans Möhler, on the edge of an era of a ‘new GABA pharmacology.’

Competing interests

The University of Maryland Baltimore has patents, on which I am listed as an inventor, covering the use of α5-selective NAMs to treat psychiatric disease.