The role of GABA_A receptors in hippocampus-dependent cognitive functions

1.1. Hippocampal Subregions and Memory

Located in the medial temporal lobe, the hippocampal formation comprises distinct subregions: the dentate gyrus (DG) and Cornu Ammonis areas: CA1, CA2, and CA3 (Amaral & Witter, 1989; Van Strien et al., 2009). As the primary input region of the hippocampus, DG receives excitatory projections from layer II of the entorhinal cortex though the perforant path and sends mossy fiber projections to CA3. CA3, in turn, sends excitatory projections to CA1 through Schaffer collaterals, completing what is known as the hippocampal tri-synaptic circuit. Monosynaptic projections from layer III of the entorhinal cortex terminate in CA1, comprising the hippocampal monosynaptic circuit. CA1, the primary hippocampal output region, sends projections, via the subiculum, back to deep layers of the entorhinal cortex (see Fig. 1). While later findings added further complexity to our understanding of intrahippocampal circuits (Amaral et al., 1990; Scharfman & Myers, 2013), this simplified model still represents the strongest informational flow streams within the hippocampus, serving as a descriptive framework for understanding how hippocampal anatomy supports function.

Due to the above-described flow of information within the hippocampus and the specific structures of each hippocampal subregion, it has been suggested that the subregions support different memory processes. Due to its large number of cells and sparse activation patterns, DG has been linked to pattern separation—transforming similar input patterns into orthogonal output representations, which helps reduce confounding of overlapping memories during encoding and retrieval. Lesion and genetic experiments show that impairing DG disrupts fine spatial discrimination (Gilbert et al., 2001; Leutgeb et al., 2007; McHugh et al., 2007; Yassa & Stark, 2011), although others have criticized the use of context discrimination as a behavioral proxy for neuronal pattern separation (Aimone et al., 2011). Other work suggested that only a subpopulation of DG cells might participate in discriminative encoding and the recruitment of DG cells to discrimination is dynamically regulated by reward-related and goal-directed information (Hainmueller & Bartos, 2020).

Located downstream of DG, CA3 is characterized as a dense network of recurrent collaterals supporting auto-associative neural coding. This feature has been theoretically linked to rapid, one-trial learning, temporary memory storage, and pattern completion, i.e., retrieval of entire memories from incomplete or degraded cues (Marr, 1971; Rolls & Treves, 1998; Rolls & Kesner, 2006). Empirical studies have supported CA3 involvement in these processes through introducing lesions or gene-targeted impairments of CA3 function and through electrophysiological recordings from CA3 (Cravens et al., 2006; Gold & Kesner, 2005; Leutgeb et al., 2007; Nakazawa et al., 2002).

As part of both the tri-synaptic and monosynaptic pathways, CA1 has been thought to integrate inputs regarding recent history from CA3 and novel sensory input from entorhinal cortex. As such, it is commonly attributed roles in temporal coding, integration, and transfer of processed information to neocortex for long-term consolidation and memory retrieval (Rolls & Kesner, 2006). Lesion, anatomically-specific gene knockout, and immediate-early gene studies show impairments in tasks involving temporal associations, sequence recall, and contextual memory upon CA1 disruption (Hall et al., 2001; Lee & Kesner, 2004; McHugh et al., 1996; Tsien et al., 1996).

1.2. Hippocampal Oscillations

Hippocampal oscillations, reflected in local field potentials (LFP), represent coordinated network activity of large populations of neurons that underlies key aspects of hippocampus-dependent memory. The most prominent rhythms of the hippocampus are theta (4–12 Hz), gamma (30–100 Hz), and sharp wave–ripples (SWRs; 100–250 Hz). Each of these rhythms occur during specific behavioral and brain states, contributing to different aspects of hippocampus-mediated memory. Theta oscillations are robust during active exploration, REM sleep, and during behaviors linked to encoding and retrieval of episodic memory, supporting temporal organization of ensemble firing and synaptic plasticity (Buzsáki, 2002; Colgin, 2016). Gamma rhythms are nested within theta cycles and have been implicated in information routing and memory encoding (Buzsáki & Wang, 2012; Colgin et al., 2009). The recruitment of physically scattered cell assemblies to the gamma rhythm and the resulting temporally aligned firing of these assemblies is thought to integrate sensory input and memory processes through the binding of these assemblies. SWRs emerge during times when theta activity is minimal (e.g., slow-wave sleep and quiet wakefulness). SWRs are characterized by the synchronous reactivation, i.e., “replay”, of neural sequences associated with recent experiences at fast speed and are thought to support memory consolidation and memory-guided planning (Buzsáki, 2015; Girardeau et al., 2009; Jadhav et al., 2012).

GABAergic interneurons are pivotal in orchestrating hippocampal oscillatory activity. Diverse subpopulations of interneurons—including parvalbumin-expressing basket cells and oriens-lacunosum moleculare cells—target distinct subcellular domains of pyramidal neurons, offering both spatial and temporal precision in inhibition. Fast phasic inhibition via synaptic GABA_ARs is central to the generation and maintenance of gamma (30–100 Hz) and theta (4–12 Hz) oscillations, as the rapid kinetics of these receptors enable synchronization across local and distributed neuronal ensembles (Buzsáki & Wang, 2012; Klausberger & Somogyi, 2008; Pelkey et al., 2017; Towers et al., 2004). Moreover, the molecular diversity and subunit composition of GABA_ARs (see below) modulate both the kinetics and localization of inhibitory currents, influencing the balance between phasic (synaptic) and tonic (extrasynaptic) inhibition and, consequently, the overall dynamics of oscillatory patterns (Allen & Monyer, 2015; Farrant & Nusser, 2005). Disruption of specific interneuron populations or GABA_ARs subtypes perturbs the temporal precision and coherence of hippocampal oscillations, with direct implications for cognitive processes and vulnerability to neuropsychiatric disorders (Klausberger & Somogyi, 2008; Pelkey et al., 2017).

1.3. GABA_A Receptors of the Hippocampus

The GABA_AR is a hetero-pentemeric ligand-gated ion channel (Engin et al., 2018; Jacob, 2019; Rabow et al., 1995; Tallman et al., 1978) permeable to chloride, with some permeability as well for bicarbonate. While the mechanics of channel opening and (in most mature neurons) chloride influx leading to hyperpolarization appear simple, the overall effect of GABA binding on the cell and on networks of cells depends heavily on receptor kinetics, subcellular location, and receptor motility, all of which are determined by receptor composition. The GABA_AR complex is composed of subunits belonging to various subunit families: α1–6, β1–3, γ1–3, δ, ε, θ, π, and ρ1–3 (Everington et al., 2018; Rudolph & Knoflach, 2011). The majority of GABA_ARs follow a pattern of organization consisting of two α subunits, two β subunits, and a γ or δ subunit. GABA_ARs are frequently classified by their α subunits and are referred to as αx-containing GABA_ARs or simply αxGABA_ARs (Fig. 2).

As noted above, the specific configuration of GABA_ARs determines several receptor characteristics. First among these is receptor kinetics. For instance, the activation time for α3GABA_ARs is three times longer than that for α1GABA_ARs, with the order of all subunits as follows: α1 < α2 < α4 < α6 < α5 < α3 (Sallard et al., 2021). Upon continued exposure to GABA, the receptors desensitize, with receptors containing α1, α2, or α4 subunits desensitizing faster than those containing α6, α3, or α5 subunits.

Secondly, the receptor composition determines its location along the membrane in relation to the synapse with α1, α2, and α3GABA_ARs more likely to form postsynaptic clusters, while α4, α5, and α6GABA_ARs are often found in persisynaptic and extrasynaptic locations (Rudolph & Knoflach, 2011). Synaptic receptors mediate fast and short-lasting phasic inhibition (Farrant & Nusser, 2005). With their fast activation and quick desensitization characteristics and synaptic location, α1 and α2GABA_ARs are perfectly suited to mediate fast hippocampal oscillations such as gamma and SWRs (Hasselmo, Bodelón, et al., 2002; Hasselmo, Hay, et al., 2002; Nyhus & Curran, 2010). Extrasynaptic receptors mediate long-lasting tonic inhibition (Fig. 3). The activation of extrasynaptic receptors decreases the ability of the neuron to reach an action potential by changing the cell’s response to voltage input, shifting the threshold for cell activation, and therefore decreasing the likelihood of an action potential (Fig. 4) (Farrant & Nusser, 2005). In networks where sparse activation patterns are essential to support complex computations, tonic inhibition participates in the maintenance of a high firing threshold, preserving sparsity (Engin et al., 2015).

In addition to membrane location in relation to the synapse, receptor composition affects subcellular location to different neuronal compartments (Nathanson et al., 2019). For instance, α2GABA_ARs are enriched at the axon initial segment, regulating signal transduction. Studies targeting the expression of α2GABA_ARs at this specific neuronal compartment have revealed the functional relevance of this specific subtype at the axon initial segment (Hines et al., 2018). α5GABA_ARs are located in dendrites rather than perisomatic compartments (Jacob, 2019; Schulz et al., 2018), while α1 GABA_ARs are found predominantly in the soma, perisomatic regions, and proximal dendrites (Kasugai et al., 2010; Nusser et al., 1996). Receptors expressed in distal dendritic regions are likely to play important roles in signal selection by eliminating weak signals in favor of strong signals, while receptors in areas closer to the soma are likely to play a role in signal integration (Fig. 5).

Finally, as reviewed elsewhere and emphasized below where necessary, different GABA_AR subtypes differ in abundance, with α1GABA_ARs as the predominant subtype in the adult brain, which also differs in each anatomical region, e.g., while α5GABA_ARs are expressed at low levels (<5%) across the brain, they constitute more than 20% of all GABA_ARs in the hippocampus) (see also Fig. 6) (Engin et al., 2018), suggesting a potential match between above-reviewed characteristics of each subtype and the physiological activity required to support the functions of each brain area. In this review, we would like to highlight the role of GABA_AR subtypes in the hippocampus to cognitive functions, with a particular focus on α5GABA_ARs. We also cover what is known regarding the roles of α1GABA_ARs, α2GABA_ARs, and α4GABA_ARs. α3GABA_ARs have so far not been convincingly linked to hippocampal-dependent cognitive functions, and α6GABA_ARs are not expressed in the hippocampus (Wisden et al., 1992).

1.4. Inhibitory Plasticity in the Hippocampus

While a large body of research has focused on the role of excitatory plasticity in memory formation and retrieval, more recent work points to the integral role of inhibitory plasticity in memory processes. Rather than merely serving as a counterweight to excitatory plasticity, inhibitory plasticity actively sculpts the temporal structure and spatial distribution of neural activity during learning. In this section, we provide a brief overview of emerging work elucidating how experience-dependent modifications at GABAergic synapses influence principal neuron participation in memory traces, integration of synaptic inputs, and systems-level consolidation through network oscillations.

A substantial body of work demonstrates that inhibitory plasticity shapes the recruitment, refinement, and stabilization of neuronal ensembles representing memory traces (i.e., engrams). Changes in the strength and distribution of GABAergic synapses dynamically tune the gain and timing of inhibition, thereby controlling which principal cells cross the firing threshold and join an emerging ensemble. These changes in GABAergic synapses are in turn implemented through presynaptic changes in GABA release and postsynaptic regulation of GABA_AR number, trafficking, ubiquitination, and lateral mobility (Luscher et al., 2011; Saliba et al., 2007).

During memory encoding, feedforward and feedback circuits built from parvalbumin-expressing (PV⁺) and somatostatin-expressing (SST⁺) interneurons can undergo long-term potentiation or depression of inhibitory synapses, thereby sharpening excitation–inhibition balance so that active granule cells or pyramidal cells suppress recruitment of their neighbors (lateral inhibition), which limits engram size and enforces sparse, pattern-separated representations in structures such as dentate gyrus (DG) and CA1 (Barron et al., 2017; Stefanelli et al., 2016; Vogels et al., 2011). In the DG–CA3 circuit, learning-driven strengthening of dentate granule cell (DGC) connections onto CA3 interneurons (particularly PV⁺ stratum lucidum cells) enhances feedforward inhibition onto CA3, which helps maintain DGC-defined engrams over time while reducing remote memory generalization(Guo et al., 2018). In addition to inhibitory plasticity of inputs onto neighboring neurons which results in the refinement and stability of the engram, inhibitory inputs onto the engram-participating neurons also show changes, influencing the temporal dynamics of activity in engram neurons. Yap et al. (2021) report strengthening of PV⁺ and weakening of CCK⁺ perisomatic inhibition onto engram-participating neurons in CA1, potentially promoting the coordinated participation of the ensemble in fast network oscillations and supporting systems level consolidation (Yap et al., 2021). Indeed, disruption of the Fos pathway abolishes the changes in PV/CCK balance as well as disrupting fast network oscillations. In line with these findings, Brito et al. (2024) reports a biphasic increase in NPAS4 expression following contextual fear-conditioning with a strong conditioning paradigm(Brito et al., 2024). Suppression of the late-phase NPAS4 expression leads to increased freezing response 24 hours later, suggesting enhanced contextual memory consolidation, as well as strengthened CCK⁺-mediated inhibition. Combined with the findings of Yap et al., these studies suggest that NPAS4 expression might mediate suppression of non-engram principal neurons via plasticity of CCK⁺ interneuron synapses(Yap et al., 2021).

Another key molecular handle on the formation and stabilization of engrams is tonic inhibition mediated by high-affinity, largely extrasynaptic GABA_ARs, notably α5GABA_ARs. In CA1 and DG, α5GABA_ARs set the threshold for LTP induction, regulate gamma oscillations, and provide a homeostatic brake that prevents runaway excitation of engram cells, while their developmental or activity-dependent redistribution between extrasynaptic and synaptic sites can “freeze” further potentiation and stabilize excitatory weights once learning has occurred(Brickley & Mody, 2012; Davenport et al., 2021; Engin et al., 2015; Martin et al., 2010). Finally, computational and experimental work converge on the idea that inhibitory synapses themselves form “inhibitory engrams” that mirror excitatory ensembles. These inhibitory engrams, also termed “negative images“, are thought to play important roles in behavioral habituation and latent memory and are likely formed and maintained through inhibitory and excitatory plasticity of the synapses on GABAergic neurons(Barron et al., 2017; Koolschijn et al., 2019)

To summarize, synapse-specific inhibitory plasticity during consolidation gradually enhances inhibition onto non-ensembled cells and sculpts inhibitory input onto engram cells, allowing memory representations to become more selective and resistant to interference without changing overall firing rates (Ujfalussy & Makara, 2020; Vogels et al., 2011). Thus, inhibitory plasticity at multiple spatial scales — from tonic conductances and dendritic/perisomatic synapses up to circuit-level feedforward and feedback loops — is essential not only for initial ensemble recruitment but also for the refinement, stabilization, and long-term specificity of memory traces.

At the molecular level, inhibitory plasticity relies on mechanisms that parallel, complement, or interact with excitatory plasticity. Activity-dependent trafficking of GABA_ARs, including phosphorylation-dependent insertion or removal from synapses, is a major determinant of inhibitory long-term potentiation (iLTP) and long-term depression (iLTD). These processes are regulated by intracellular signaling cascades involving CaMKII, PKA, PKC, and neuromodulatory pathways, and they contribute directly to learning-related changes in inhibitory strength(Luscher et al., 2011).

Recent work has also highlighted the role of extracellular proteolysis in the induction of iLTP and for iLTP-induced plastic changes at GABAergic synapses(Wiera et al., 2021). Other findings have expanded the molecular repertoire of inhibitory plasticity through the discovery that glutamatergic GluD1 receptors can bind GABA and modulate inhibitory synapse function, revealing unexpected cross-modal regulation of inhibitory transmission(Piot et al., 2023). Such molecular diversity provides multiple routes through which learning can reshape inhibitory circuits.

At the network level, activity-dependent inhibitory plasticity reshapes hippocampal interneuron networks that generate sharp wave–ripples (SWRs), directly linking changes in GABAergic synapses to ripple-mediated replay events that support memory consolidation(Buzsáki, 2015). In vivo studies demonstrate that contextual learning is followed by sleep-dependent strengthening and reorganization of PV⁺ interneuron networks in CA1, which stabilizes SWR patterns and is necessary for successful consolidation(Ognjanovski et al., 2017). These findings are consistent with the broader view that interneuron activity and connectivity are themselves plastic memory substrates rather than fixed modulators. Further emphasizing the role of different interneurons in SWR-mediated memory consolidation is the finding that SWR events are systematically preceded by reduced activity of CCK⁺ interneurons and followed by enhanced PV⁺ basket-cell firing(Vancura et al., 2023). The magnitude of these changes predicts ripple duration and performance in hippocampal-dependent tasks. At the synaptic level, GABAergic synapses onto CA1 pyramidal neurons undergo state-dependent LTP, with wake and sleep states differentially regulating the strength of PV⁺ cell inputs via α5GABA_AR insertion, providing a mechanism by which inhibitory gain around SWRs can be persistently up- or downregulated to bias which assemblies are most likely to be reactivated(Wu et al., 2022). Finally, intracellular and unit-recording studies reveal that heterogeneous, plastic inhibitory “ramps“ before SWRs and dynamic changes in interneuron firing rates determine the precise timing of pyramidal-cell spikes within a ripple, so that adjustments in inhibitory drive can retime ensemble replay without necessarily altering excitatory input, offering a direct route through which inhibitory plasticity refines the temporal structure of ripple-mediated memory consolidation(Noguchi et al., 2022).

Together, these studies establish inhibitory plasticity as a fundamental mechanism through which learning shapes neural circuits. By regulating the recruitment of neurons into memory ensembles, controlling dendritic and somatic integration, and orchestrating the oscillatory dynamics that support consolidation, inhibitory plasticity ensures that memories are encoded with specificity, stabilized over time, and retrieved with fidelity.

2.1. Properties of α5GABA_ARs

As noted, α5GABA_AR expression is concentrated in the hippocampus, with similarly strong expression in the olfactory bulb, some expression in deep layers of cortex, and low levels of expression in the subiculum, hypothalamus, and amygdala. Within the hippocampus, α5GABA_ARs are expressed heavily in CA1, CA2, and CA3, with lower but still significant expression in DG (see Fig. 6). Early work investigating the subcellular location of α5GABA_ARs suggested that these receptors are primarily expressed in extrasynaptic locations and mediate tonic inhibition, as well as a slow form of phasic inhibition with a decay constant >30ms (Brünig et al., 2002; Capogna & Pearce, 2011; Caraiscos et al., 2004; Crestani et al., 2002; Glykys & Mody, 2006; Prenosil et al., 2006; Zarnowska et al., 2009). Interestingly, rather than showing diffuse staining throughout extrasynaptic space, α5GABA_ARs were observed to form extrasynaptic clusters like those formed at inhibitory synapses by synaptic receptors. α5GABA_AR clustering within the extrasynaptic compartment is mediated by their interaction with radixin, a protein anchored to the actin cytoskeleton (Hausrat et al., 2015). Phosphorylation of radixin via the RhoA/ROCK signaling pathway activates radixin, promoting its binding to α5GABA_AR and thereby maintaining their localization at extrasynaptic sites.

Dephosphorylated radixin is inactive, resulting in a reduction of extrasynaptic and in an increase of intrasynaptic localization of α5GABA_ARs. The phosphorylation status of radixin is activity-dependent, with suppressed neuronal activity promoting radixin phosphorylation and increased activity suppressing phosphorylation and dissolving extrasynaptic α5GABA_AR clusters. In addition to the above interaction with radixin in extrasynaptic compartments, α5GABA_ARs interact directly with the inhibitory synaptic scaffolding protein gephyrin, forming synaptic clusters (Brady & Jacob, 2015). In line with above findings demonstrating reduced extracellular clustering of α5GABA_ARs with increased neuronal activity, induction of an excitatory LTP was shown to relocate α5GABA_ARs from extrasynaptic to synaptic locations, most likely promoting interactions with gephyrin (Davenport et al., 2021) (Fig. 3).

At least in CA1, α5GABA_ARs are expressed primarily on the dendrites of pyramidal neurons, where they receive GABA input from somatostatinergic interneurons, as well as from neurogliaform cells (Schulz et al., 2018). As noted above, α5GABA_ARs are often expressed at the base of dendritic spines, where α5 is expressed in proximity to the post-synaptic marker PSD-95 (Brünig et al., 2002). Further adding to the idea of α5GABA_ARs as an inhibitory functional counterpart of NMDA receptors is the finding that α5GABA_ARs have a greater hyperpolarizing effect when membrane potential is depolarized(Lodge et al., 2021; Schulz et al., 2018, 2019)

While α5GABA_ARs are expressed predominantly on principal neurons in all subregions of the hippocampus, expression has also been observed on the dendrites of somatostatin-positive interneurons, where they are targeted by vasoactive intestinal peptide (VIP)-positive and calretinin-positive interneurons (Magnin et al., 2019). Despite low-level expression, α5GABA_ARs on interneurons clearly have behaviorally relevant effects (Zhu et al., 2023), and more studies in this area are needed.

2.2. Involvement of hippocampal α5GABA_ARs in cognitive processes

Early behavioral work providing direct evidence for the involvement of hippocampal α5GABA_ARs in memory processes came from studies that involved global manipulations of α5GABA_AR expression or activity but demonstrated effects in tasks that are known to be hippocampus-dependent. For instance, in a mouse line carrying a point mutation in the α5 subunit that rendered α5GABA_ARs insensitive to modulation by diazepam, significant functional alterations were observed (Crestani et al., 2002). The mutation unexpectedly led to a 25% reduction in the expression of the α5 subunit in the hippocampus and these mice were thus studied as partial α5 knockdowns. The α5 knockdown mice displayed enhanced hippocampus-dependent trace fear conditioning, a learning process where the association between a neutral stimulus and an aversive event is formed across a temporal gap (Crestani et al., 2002; Yee et al., 2004). In contrast, hippocampus-independent delay fear conditioning, where the shock was delivered immediately at the end of the tone presentation, was unaltered in the α5 knockdown mice (Crestani et al., 2002), confirming that only hippocampus-dependent memory was influenced. Similarly, it has been reported that young adult mice with targeted disruption of the α5 gene (global α5KO mice) exhibit significantly improved performance in the Morris Water Maze (MWM), a hippocampus-dependent spatial learning task (Collinson et al., 2002). Global α5KO mice performed similarly to controls in non-hippocampal tasks, such as in a two-way active avoidance test.

Further evidence for the involvement of α5GABA_ARs of the hippocampus in learning and memory processes came from studies using pharmacological approaches or a combination of mouse genetics and pharmacology. For instance, studies in global α5 KO mice showed that while α5GABA_ARs are not required for the sedative action of the intravenous general anesthetic etomidate, they are required for its amnestic effects, as shown in hippocampus-dependent spatial (Morris Water Maze) and contextual (contextual fear conditioning) tasks (Cheng et al., 2006). Etomidate also reduced LTP in wild type mice but not in global α5 KO (Cheng et al., 2006; Orfila et al., 2019; Zurek et al., 2014). Further supporting the role of α5GABA_ARs in mediating the amnestic effects of general anesthetics, it was observed that exposure to the injectable anesthetic etomidate induced a sustained increase in tonic inhibitory currents mediated by α5GABA_ARs and enhanced their surface expression for up to one week, leading to memory impairments in the novel object recognition task during this post-exposure period (Zurek et al., 2014). However, pre-treatment with the α5-selective negative allosteric modulator (α5-NAM) L-655,708, administered 30 minutes before training, completely reversed the memory deficits induced by etomidate. Using α5-NAMs, a similar α5-mediated mechanism was also uncovered for the acute and lasting amnestic effects of inhalant anesthetics isoflurane and sevoflurane (Zhao et al., 2019; Zurek et al., 2012a), diazepam (Ballard et al., 2009), and alcohol (Nutt et al., 2007). Others have reported that the adverse effects of processes such as chronic stress and neuroinflammation on memory might be at least partially mediated by α5GABA_ARs (Li & Zhang, 2021; Prevot et al., 2019; Wang et al., 2012). Interestingly, in aged mice beneficial effects of α5GABA_ARs have been suggested, i.e., prevention or reversal of surgery-induced neuroinflammation in the hippocampus and of cognitive deficits (Nagarajan et al., 2025).

Complementary insights have been obtained through systemic administration of α5-selective negative and positive allosteric modulators of GABA_ARs, α5-NAMs and α5-PAMs, respectively. In an early example, it was shown that the α5-NAM named α5IA has a benzodiazepine site occupancy of 25%, and enhanced cognition in the delayed matching-to-position (DMTP) task, which measures working memory. RO4938581, an α5-NAM that exhibits dose-dependent binding to α5GABA_ARs with approximately 30% receptor binding, also improved rat choice accuracy in a DMTP task. Additionally, the compound reversed diazepam-induced impairments in spatial learning, as measured in a water maze task, in Lister Hooded rats and improved executive function in either the object retrieval task or the detour reaching task in monkeys (Ballard et al., 2009). The α5-NAM TB21007 enhanced learning and memory in the radial arm maze task (Koh et al., 2013). Interestingly, the compound had no effect on aged rats (24-26 months). In contrast, the α5-PAM Compound 44 significantly improved long-term memory in aged rats in the water maze test. The α5-PAM Compound 6 improved memory performance of aged rats in the radial arm maze, while not influencing young adult rats (Koh et al., 2013). Similarly, chronic treatment of mice with the α5-PAM GL-II-73 reversed the cognitive and neuronal degeneration associated with aging by enhancing dendritic branching and spine density in cortical neurons and reversed the age-related decline in memory performance in the Y-Maze test (Prevot et al., 2021). Thus, while reducing the activity of α5GABA_ARs in young animals leads to improvements in at least certain types of hippocampal memory, this effect seems to be reversed with age, where improving α5GABA_AR activity leads to memory improvement (see also Fig. 7). These findings are further corroborated by the observation that isoflurane, which has amnestic effects in both young and aged animals, has opposing effects on Gabra5 mRNA expression in the hippocampus, increasing expression in young rats and decreasing expression in aged rats (Zhao et al., 2019). This change in effect might be caused by changes in α5GABA_AR abundance, trafficking, and subcellular location (i.e., synaptic/extrasynaptic) that occur as a result of aging.

While these early reports suggested that α5GABA_ARs mostly serve as a brake on memory processes in young adult animals, with reduced α5 levels leading to improved hippocampal memory, later work employing α5 knockdown mice demonstrated a more complex picture in young adult animals as well. α5 knockdown mice showed significant impairments in object-in-place memory retrieval, while object, position, and context memories were mostly preserved (Prut et al., 2010). Notably, α5 knockdown mice preferred object-based recognition strategies over spatial processing, further supporting the idea of impairments in specific hippocampal functions. Interestingly, performance of α5 knockdown mice in T- and eight-arm mazes was indistinguishable from wild type. Thus, α5GABA_ARs appear to support specific types of hippocampus-dependent memory, suppress others, and have no effect on yet other tasks in young adult animals, while mostly supporting learning and memory processes in aged animals. A more in-depth understanding of the specific hippocampal processes involved in each of these memory tasks and how α5GABA_ARs are involved in these processes would elucidate the mechanisms underlying these differential effects.

2.3. Distinct roles of α5GABA_A receptors in different hippocampal subregions

The studies mentioned above have mostly been performed in animals carrying global genetic mutations or that were treated with α5-selective agents systemically. While the outcome measures involved memory tasks that are known to rely heavily on hippocampal function, as all memory tasks involve complex coordination between many brain regions and as α5GABA_AR expression is not strictly limited to the hippocampus, it is not possible to conclude unequivocally that the observed effects were mediated by hippocampal α5GABA_ARs. To address this limitation, α5-NAM TB-21007 was infused into the dorsal hippocampus during the acquisition trial of the shock-probe burying task (McEown & Treit, 2013). This manipulation resulted in impaired encoding of fear memory, as demonstrated by increased number of shock-probe contacts in a test conducted 24 hours later. Infusions of the same α5-NAM into the ventral hippocampus had no effect on fear memory, suggesting a specialized role for α5GABA_ARs subunits in the dorsal hippocampus in the regulation of conditioned fear memory.

As noted in earlier sections, different roles in learning and memory have been proposed and empirically demonstrated for hippocampal subregions (Fig. 1). To study the role of α5GABA_ARs in defined neuronal subpopulations, conditional knockdowns of the α5 subunit in CA1 pyramidal neurons, CA3 pyramidal neurons, and dentate gyrus (DG) granule cells have been generated (Engin et al., 2015, 2020) (Table 1).

Table. 1. Role of α5-containing GABA_A receptors in hippocampal subregions.

This table is based on conditional knockouts of the GABAA receptor α5 subunit in CA1 and CA3 pyramidal neurons and dentate gyrus (DG) granule cells. Arrow: upward (↑) refers to improved function; downward (↓) refers to impaired function; - refers to largely no effects on respective function. Sources: 1) Engin et al., 2015; 2) Engin et al., 2020; 3) this review (Fig. 8); 4) unpublished. ND = not done.

Open in a new tab

Mice with selective deletion of the α5 subunit in CA1 pyramidal neurons (α5CA1KO mice) exhibited enhanced performance in spatial memory tasks, such as the Morris water maze, and in trace fear conditioning (Fig. 8), indicating that α5GABA_ARs in CA1 pyramidal neurons attenuate spatial and temporal memory (Engin et al., 2020). The same mice exhibited deficits in context discrimination during fear conditioning, a task that is sometimes conceptualized as a behavioral counterpart to pattern separation. This finding was particularly interesting as discrimination between highly overlapping stimuli is often attributed to DG function (A. M. Morris et al., 2012), with the encoding of highly similar stimuli being accomplished by distinct, nonoverlapping DG neuronal ensembles (Schmidt et al., 2012). The reactivation of DG neuronal ensembles linked to a specific context in turn biases CA1 activity towards neural states specific to this context (Coelho et al., 2024), eventually leading to the retrieval of the context-specific memory at the expense of competing with similar memories. It is possible that the critical role of α5GABA_ARs on CA1 pyramidal neurons lies in this “biasing” of CA1 activity based on DG input through inhibition of competing neuronal states rather than at the discrimination stage itself. These observations highlight that α5GABA_ARs in the same neuronal population can dampen some memory processes while improving others (Engin et al., 2020).

Unlike α5CA1KO mice, mice with a selective knockout of α5GABA_ARs in CA3 hippocampal pyramidal neurons (α5CA3KO) exhibited enhanced contextual fear conditioning but were similar to controls in context discrimination (Fig. 8). In the learning phase of the MWM, these mice displayed enhanced spatial learning, but they did not show significant differences in the MWM reversal task or in a latent inhibition to conditioned freezing paradigm (Engin et al., 2015).

As noted above, DG has long been linked to pattern separation (Myers & Scharfman, 2009; Schmidt et al., 2012), based on its anatomical and connectivity properties. Specifically, the large number of cells and high levels of baseline inhibition, which maintain sparse activation patterns in DG, were proposed to underly its ability to disambiguate overlapping inputs. To directly evaluate the role of high-threshold inhibition in the dentate gyrus (DG), mice with a selective knockout of the α5 subunit in DG granule cells (α5DGKO) were generated (Engin et al., 2015). α5DGKO mice had reduced tonic (but not fast phasic) inhibition in DG, as well as increased c-Fos activation in DG following exposure to a novel environment, confirming the role of α5-mediated tonic inhibition in maintaining sparse activation patterns in DG. Behaviorally, α5DGKO mice showed impairments in context discrimination, in reversal learning in the MWM, and in fear extinction (Fig. 8), i.e., in tasks involving high memory interference. These findings are in line with the proposed role of DG in disambiguation of overlapping inputs and particularly for the role of α5GABA_AR-mediated tonic inhibition of the DG. Further supporting the importance of the specific role of α5-mediated tonic inhibition in the disambiguation of high-interference memories was the finding that α5CA3KO mice had a distinct behavioral phenotype with no specific deficiencies in high-interference memory tasks. Similarly, a DG granule cell-specific knockout of the synaptic α2GABA_ARs, which reduced fast phasic inhibition in DG without affecting tonic inhibition (Engin et al., 2016) also did not lead to specific memory impairment in high-interference tasks (Engin et al., 2015).

While the above findings emphasize the specific contribution of DG α5GABA_ARs to the control of memory interference, it should be noted that α5GABA_ARs mediate less than half of baseline tonic inhibition in DG, with the remainder mediated by α4GABA_ARs (Brickley & Mody, 2012; Glykys et al., 2008). Thus, one important question is whether any reduction of DG tonic inhibition, for instance, through a similar DG granule cell selective knockout of α4GABA_ARs would lead to similar specific memory impairments. While α5GABA_ARs mediate tonic inhibition at baseline, as noted above, unlike α4GABA_ARs, α5GABA_ARs tend to shift to synaptic spaces and form clusters here following intense neuronal activity, which can be induced by a complex memory task. When α5GABA_ARs are removed from the DG, it affects not only baseline tonic inhibition but also this form of neuronal plasticity. Thus, it is possible that the above noted role of DG α5GABA_ARs in managing memory interference relies not only on baseline tonic inhibition in this region, but also on inhibitory neuronal plasticity made possible by dynamic shift of α5GABA_ARs between synaptic and extrasynaptic compartments. In this case, the effects would be specific to α5GABA_ARs, and a similar behavioral phenotype would not be observed following DG-selective knockout of α4GABA_ARs.

α5-selective NAMs have been suggested to be potentially useful for the treatment of cognitive deficits based on the earlier observations that a lack or reduction of the activity of α5GABA_ARs improves spatial and temporal memory (Ballard et al., 2009; Collinson et al., 2002; Crestani et al., 2002). Along with above noted findings that an α5GABA_AR knockdown may lead to impairments in object-in-place memory (Prut et al., 2010), the findings in α5DGKO mice might call into question whether the strategy to develop NAMs as cognitive enhancers is viable. However, both insufficient and excessive α5GABA_AR-mediated tonic and slow phasic activity may lead to cognitive deficits. For instance, in the Ts65Dn mouse model of Down Syndrome, α5GABA_AR-mediated inhibition in the DG and CA1 pyramidal cells is excessive (Fernandez et al., 2007; Schulz et al., 2018, 2019).

Treatment with the α5-NAM, RO4938581 or deletion of one or two copies of the Gabra5 gene significantly ameliorated learning and memory deficits, improved LTP, and normalized density of GABAergic synaptic markers (Martínez-Cué et al., 2013; Schulz et al., 2018, 2019; Vidal et al., 2018). In the Dup15q syndrome mouse, a model for autism with increased brain α5-GABA_A receptor expression, chronic treatment with the α5-NAM RO4938581 normalized inflexibility in water-maze reversal learning phases along with other behavioral abnormalities, supporting the idea that reducing α5-mediated inhibition can improve hippocampus-dependent learning flexibility when α5 signaling is excessive (Nakagawa et al., 2026). This work strengthens the “α5-NAMs as pro-cognitive agents” framework in a context with known increased expression and thus likely increased activity of α5GABA_ARs. On the other hand, studies have suggested that α5GABA_AR expression may be reduced in the brains of individuals with autism spectrum disorders (ASD)(Fatemi et al., 2010) and schizophrenia(Marques et al., 2021). MP-III-022, an α5-selective positive allosteric modulator, was shown to ameliorate reversal learning deficits as well as social and repetitive behavioral abnormalities in the valproate model of autism spectrum disorder (ASD) (Santrač et al., 2022). Similarly, α5-PAMs have been effective in normalizing hippocampal activity and cognitive flexibility deficits in a MAM model of schizophrenia (Donegan et al., 2019; Gill et al., 2011). A recent preclinical study of the α5-PAM alogabat (RO7017773/RG7816) reported beneficial effects of α5-positive allosteric modulation in mouse models of neurodevelopmental disorders with impaired GABA_A signaling, showing reduced repetitive self-grooming in autism-relevant BTBR and Cntnap2^−/− mice at >~50% α5GABA_AR occupancy. Importantly, learning and memory in the contextual fear conditioning test and in the Morris water maze probe test were impaired only at higher α5GABA_AR occupancy (~75–81% or ~79%, respectively), whereas lower-to-moderate engagement did not significantly impair contextual or spatial memory(Cecere et al., 2025). These findings suggest that both α5-NAMs and α5-PAMs may have value as cognitive enhancers, depending on the specific pathology underlying the observed cognitive deficits.

2.4. Potential mechanisms of hippocampal α5GABA_AR involvement in learning and memory

Long-term potentiation (LTP), a specific form of synaptic plasticity, has long been viewed as a neuronal substrate for learning and has been known to be attenuated by α5GABA_AR-mediated inhibition (Martin et al., 2010). Thus, early studies sought evidence that reductions in α5GABA_AR expression that led to improvements in specific forms of hippocampus-dependent memory would also improve LTP in specific hippocampal synapses. While initial studies employing α5 knockdown and global α5 KO mice failed to find direct evidence for large scale changes in LTP (Collinson et al., 2002; Crestani et al., 2002), later work employing α5-NAMs demonstrated that these drugs to promote LTP in CA1 (Dawson et al., 2006; Kawaharada et al., 2018). The effect of α5GABA_ARs on LTP seems to be a fine-tuned and modulatory one, where α5GABA_AR activity increases the threshold for induction of LTP, but specifically for stimulation frequencies of 10-20 Hz; the slopes of the field excitatory postsynaptic potentials (fEPSPs) for these frequencies were increased in global α5 KO mice(Martin et al., 2010). This specificity in the activity range where α5GABA_ARs seem to be relevant for modulating synaptic plasticity perhaps explains the task-specificity of the involvement in α5GABA_ARs in learning and memory tasks. Moreover, increased clustering of α5 at synapses following increased neuronal activity or a learning prevents further excitatory LTP in the activated synapses, serving as a way of “freezing” the synapses following initial encoding, assuring the permanence of this neuronal substrate for memory (Davenport et al., 2021). These behavioral and physiological findings all point to the conclusion that the involvement of α5GABA_ARs in hippocampal memory is far more complicated than the initially proposed “α5 up, memory down; α5 down, memory up” scenario.

In addition to the synaptic level, there is some evidence of α5 involvement in the regulation of population-level activity that underlies cognitive processes. Sharp wave ripples (SWRs), which are high-frequency oscillations generated in the hippocampus, are associated with the consolidation of memory and the coordination of network activity during sleep. The α5-NAM L-655,708 reduced the rate of occurrence of SWRs and enhanced the probability of their generation in clusters in the ventral hippocampus, but not in the dorsal hippocampus, suggesting subregion-specific contributions to network oscillations and memory processing Click or tap here to enter text.(Papatheodoropoulos & Koniaris, 2011; Trompoukis et al., 2021). These findings indicate that α5GABA_ARs play an important role in the modulation of SWR activity which may affect the consolidation of memory.