Presynaptic Trafficking of Glutamate Decarboxylase Isoforms Is Dispensable for Basal GABAergic Neurotransmission

Abstract

Two major glutamate decarboxylase isoforms (i.e., GAD65 and GAD67) together synthesize the majority of γ-aminobutyric acid (GABA) in our nervous system. However, the subcellular distribution of these enzymes and their relative impacts on synaptic GABA release remain unclear. To address this important question, here we monitored their synaptic trafficking in male and female mouse brains and dissociated neuronal cultures. We noticed that, unlike some major glutamate-biosynthesizing enzymes, e.g., glutaminase and glutamate dehydrogenase, which were primarily associated with perisomatic mitochondria, both GADs together were highly enriched at GABAergic presynapses. Nevertheless, when expressed separately in GAD-deficient human neurons derived from a male stem cell line, GAD65 exhibited preferential distribution at presynapses over GAD67. Despite these differences in subcellular localization, both GADs produced equivalent levels of intracellular GABA, which adequately diffused to axon terminals, and triggered robust GABAergic activities. These findings raised the question of whether the presynaptic recruitment of GADs is, after all, necessary for reliable GABAergic transmission. To examine this hypothesis, we further swapped or removed the trafficking signals from both GAD isoforms and even artificially restricted them at nonsynaptic compartments, including the cell nucleus. Despite our attempts, the chimeric and mutant GAD variants continued to produce sufficient amount of intracellular GABA for vesicular loading and presynaptic release. These results indicate that GAD65 and GAD67 are functionally redundant in GABA production, if expressed equitably in neurons, and irrespective of GADs’ subcellular trafficking profile, diffusion of GABA molecules from distant sources can effectively supply and replenish the presynaptic terminals for functional activities.

Significance Statement

The inhibitory transmitter γ-aminobutyric acid (GABA) plays a vital role in modulating neuronal excitability. GABA is metabolized from glutamate by two enzymes, GAD65 and GAD67. These proteins contain distinct trafficking signals, exist as soluble versus membrane-bound forms, and exhibit contrasting subcellular distribution. It is unclear if these differences between GAD isoforms indirectly influence GABAergic neurotransmission. Here we demonstrate that, when expressed equitably, both GADs synthesized comparable amounts of GABA that diffused throughout the intracellular space and were adequately packaged into synaptic vesicles for action potential (AP)-dependent and AP-independent release. Mutating GADs’ N-terminal residues impaired their trafficking without affecting GABA production by C-terminal domains, which was sufficient for effective synaptic activities. Hence, neurotransmitter diffusion can support synapse function over a distance.

Introduction

Neurotransmitters are generally derived from precursor molecules abundantly present in neurons. Depending on the transmitter type, they are manufactured from a variety of substances including amino acids (glutamine, glycine, phenylalanine, tryptophan, tyrosine), small metabolites (choline, α-ketoglutarate), and even modified from other neurotransmitters [glutamate to γ-aminobutyric acid (GABA), dopamine to epinephrine or norepinephrine; Bloom, 1984; Hyman, 2005]. Since these substrates are not physically confined within a specific subcellular region, it raises questions if their enzymatic conversion into neurotransmitters occurs locally or remotely from presynaptic sites. This fundamental issue remains unresolved even for glutamate and GABA, the major excitatory and inhibitory neurotransmitters. The glutamate and GABA vesicular transporters, vGLUTs and vGAT, typically accumulate at glutamatergic versus GABAergic presynapses to optimize vesicular packaging (McIntire et al., 1997; Bellocchio et al., 2000). Conversely, intracellular distributions of glutamate versus GABA synthesis enzymes, glutaminase (GLS) versus GADs, and their functional relevance on presynaptic neurotransmitter availability and release are not well understood.

There are two major GLS isozymes, kidney and liver types (Curthoys and Watford, 1995). The kidney-type GLS is selectively expressed in the brain, operating as phosphate-activated amidohydrolase to catalyze glutamate production from glutamine (Bradford et al., 1978; Masson et al., 2006). GLS appears to be highly enriched in mitochondria but also exists as a diffused pool in cytoplasm (Aoki et al., 1991; Kvamme et al., 2000). Previous studies with hippocampal and cortical slices revealed that GLS is primarily present at neuronal cell bodies and dendritic branches but also partially localizes to a fraction of synaptic terminals (Dienel et al., 1977; Altschuler et al., 1985; Aoki et al., 1991). Although synaptic GLS recruitment can ideally enhance the local glutamate level for release (Marx et al., 2015), its abundance relative to nonsynaptic population is unknown for most brain areas. In addition to GLS, glutamate dehydrogenase (GDH) can operate as a bidirectional enzyme which reversibly catalyzes glutamate synthesis from α-ketoglutarate and vice versa, in sync with tricarboxylic acid (TCA) cycle (Andersen, 2025). Yet, the distribution of GDH in neurons and its enrichment at synaptic versus nonsynaptic compartments remain unclear.

The relationship between synaptic trafficking and enzymatic action is also puzzling for GADs. Both GAD67 and GAD65 isoforms require pyridoxal 5′-phosphate cofactor and, together, synthesize the majority of GABA (Kaufman et al., 1986; Baekkeskov et al., 1990; Erlander et al., 1991). Due to subtle structural change, GAD67 behaves like holoenzyme, whereas GAD65 acts as apoenzyme, leading to strikingly different kinetic properties (Battaglioli et al., 2003; Fenalti et al., 2007). In the rodent brain, GABAergic neurons usually contain both GADs, but GAD67 is expressed at a higher degree than GAD65 (Esclapez et al., 1994), which might explain its more severe phenotypes in knock-out animals (Asada et al., 1996, 1997; Kash et al., 1997). Two GADs also differ in their trafficking pattern (Esclapez et al., 1994). GAD67 is considered to be mostly soluble and dispersed evenly within cells, whereas GAD65 is concentrated at presynapses (Kaufman et al., 1991), potentially by membrane anchoring via N-terminal domain (Kanaani et al., 2002), and interactions with heat shock cognate 70 (HSC70) or vesicle-associated cysteine string protein (Hsu et al., 2000). A minor fraction of GAD67 is recruited at presynapses because of axonal targeting through N-terminus and heterodimerization with GAD65 (Dirkx et al., 1995; Kanaani et al., 1999, 2010). These findings led to a notion that GAD65 and GAD67 might differentially contribute to basal intracellular GABA levels for general metabolic and tonic inhibitory function versus its vesicular packaging and presynaptic release during neuronal activities. Nevertheless, these predictions have not been systematically tested yet, which require comprehensive analyses of GADs' subcellular localizations and their direct impact on GABAergic neurotransmission.

Here, we have assessed (1) the relative distributions of major glutamatergic and GABAergic enzymes at their corresponding presynaptic compartments, in mouse and human neurons, both in vivo and in vitro, and (2) asked if synaptic association can influence their capacity to supply neurotransmitters for vesicular loading. Our findings emphasize that neurotransmitter production can be physically uncoupled from their synaptic release in neurons.

Materials and Methods

Institutional approvals

All cell culture methods and lentivirus production procedures were authorized by the Institutional Biosafety Committee (IBC protocol number 19-059B) of Colorado State University. All experiments with mice were formally approved by the Institutional Animal Care and Use Committee (IACUC protocol number 5237).

Cell lines and animals

The human-induced pluripotent stem (iPS) cells (WTC-11, male line) were donated by Dr. Michael E. Ward (National Institute of Neurological Disorders and Stroke) to Dr. James Bamburg (Colorado State University) and were acquired by us for this study. For all experiments with mice, both male and female sexes (strain C57BL/6) were used to dissect out brain slices (3-month-old animals), as well as prepare primary hippocampal neurons or astrocytes (postnatal day P0-P1) for coculture experiment. Human embryonic kidney 293 (HEK 293T, containing SV40T-antigen) cells were available commercially (catalog #632180, Takara Bio).

Brain slice preparation

Eight mice were first perfused with phosphate-buffered saline (PBS), followed by 4% paraformaldehyde (PFA) dissolved in PBS. The brains were removed and further postfixed overnight in PFA. Coronal sections were cut across the medial region at ≈30 μm thickness on a vibratome (Leica VT1200S).

Primary neurons

The mouse primary neuronal cultures were prepared as described previously (Benner et al., 2023). Hippocampal sections were surgically isolated from newly born wild-type (WT) pups, digested inside tissue-culture incubator for ≈15 min at 37°C, with 10 unit/ml papain (catalog #LS003126, Worthington Biochemical) in Hanks' balanced salt solution (HBSS) buffer containing 0.5 mM EGTA. The brain tissues were thoroughly washed with HBSS only, and subsequently dissociated in Neurobasal Plus medium (already contains 1.8 mM CaCl₂ and 0.5 mM GlutaMAX; Thermo Fisher Scientific) + B27 supplement (Thermo Fisher Scientific) + 10% fetal bovine serum (FBS; Atlas Biologicals) + 1% penicillin–streptomycin mix (Thermo Fisher Scientific), using the 1 ml plastic pipette tips.

Triturated cells were seeded onto Matrigel (Corning, Sigma-Aldrich)-precoated glass coverslips, placed inside individual wells of 24-well dishes. The day of neuronal plating was considered as 0 d in vitro (DIV 0). During DIV 2–3, FBS concentration was gradually reduced to ≈5% and then to 2.5% by adding almost equal volumes of Neurobasal media but without any serum and also included 5-fluorodeoxyuridine (FdU; 10 μM) to prevent astrocyte proliferation after reaching 70–80% confluency. Fresh media of 50% volume + 2.5% FBS + FdU were added at DIV 7 and DIV 12. The hippocampal cultures were subsequently analyzed at DIV 16–18.

Stem cell culture

WTC-11 iPS cells were plated on six-well dishes precoated with Matrigel (catalog #354277, Corning) and cultured under feeder-free conditions in mTeSR 1 or mTeSR Plus (STEMCELL Technologies) media, which was exchanged every day. At 60–70% confluency, the cells were dissociated with PBS + 0.5 mM EDTA and split at a 1:6 dilution until ready for neuronal differentiation. During all passages, the mTeSR media were also supplemented with a ROCK-inhibitor Y-27632 (2.5 μM, MedChemExpress) but were excluded thereafter.

Neurogenesis from stem cells

The Neurogenin-2 (Ngn2)-inducible WTC-11 iPS cells were rapidly converted into neurons as reported earlier (Burlingham et al., 2022; Carricaburu et al., 2024). They were plated at 1:12 dilution in N3 media [composition, DMEM/F12 (Thermo Fisher Scientific) + N2 (Thermo Fisher Scientific) + B27 (Thermo Fisher Scientific) + insulin (20 μg/ml, Sigma-Aldrich) + penicillin–streptomycin mix (1%, Thermo Fisher Scientific)]. The next day, doxycycline (2 μg/ml, Sigma-Aldrich) was added to the media to induce Ngn2 expression and trigger neurogenesis. After 5–7 d, the cells were dissociated with EDTA, mixed with passage 1–2 mouse primary glia, and replated onto Matrigel-coated glass coverslips placed inside 24-well dishes. During the first 24 h, N3 media contained 10% FBS, which was serially diluted down to a final concentration of 2.5% by consecutive half-replacements over the next 3–4 d. It also included 10 µM FdU to inhibit any glial proliferation after reaching a ≈70–80% confluency. Around 15 d of coculture, these cultures were gradually switched to the Neurobasal Plus media, additionally supplemented with B27, 2.5% FBS, 10 µM FdU, and 1% penicillin–streptomycin, by half-exchanges every 3–4 d. These human cultures were analyzed on Day 7 or Days 50–56 after differentiation.

DNA constructs

The WT constructs for vGAT, GAD65, and GAD67 proteins were originally cloned by us from human cDNA preparations (Burlingham et al., 2022). Both GADs were fused at their C-termini with either a hemagglutinin (HA) tag or class I monopartite nuclear localization signals (NLSs) borrowed from SV40 (PKKKRKV) or cMyc (PAAKRVKLD). For GAD65 mutant constructs, site-directed mutagenesis was performed to substitute for the two critical cysteine residues (i.e., amino acid 30 and 45) with alanine, and/or its N-terminal 1–23 residues were deleted. For the GAD67 ΔN construct, the first 1–89 amino acids were removed from its N-terminus, and an HA tag was added at the C-terminal. For GAD65 and GAD67 chimeric constructs, N-terminal 1–93 residues of GAD65 were fused with C-terminal 99–594 residues of GAD67, and N-terminal 1–99 residues of GAD67 were fused with C-terminal 93–585 residues of GAD65. Expression cassettes encoding the WT and recombinant proteins were inserted into lentiviral vector under human Synapsin (hSyn1) promoter, followed by Woodchuck regulatory element, flanked by 5′ to 3′ long terminal repeats. An empty vector was used as control.

Lentivirus production

Three helper plasmids (i.e., pRSV-REV, pMDLg/pRRE, and VSV-G; 7 µg each) and one expression vector (15–20 µg) were cotransfected using polyethylenimine (dissolved in Opti-Mem) into 70–80% confluent HEK 293T cells growing in DMEM media [Genesee Scientific; premixed with sodium pyruvate, glucose, and l-glutamine + 1% nonessential amino acids + 1% penicillin–streptomycin + 10% FBS + 0.002% β-mercaptoethanol (BME)], on 10 cm dishes. At ≈10–12 h post-transfection, the culture medium was exchanged completely, and supernatants containing the lentiviral particles were collected after another ≈24, 48, and 72 h. The supernatants were then pooled and spun at ≈800 × g for 6–8 min to remove all HEK cell debris. The supernatants were then spun again at ≈120,000 × g for another 2 h at 4°C (Beckman L8-70 M ultracentrifuge, equipped with SW41Ti rotor). The supernatants were removed by aspiration, and viral pellets were resuspended overnight in ≈100 µl of DMEM media, subsequently aliquoted, and frozen at −80°C before use.

Immunoblotting

Neurons were lifted, dissociated with PBS + EDTA, and then pelleted by spinning at ≈800 × g for 5–10 mins. The cell pellets were lysed with RIPA buffer (Thermo Fisher Scientific) supplemented with Halt protease inhibitor cocktail (Thermo Fisher Scientific) for 1 h. The cell lysates were centrifuged at ≈800 × g for 15 min, and supernatants containing the protein extracts were collected, aliquoted, and stored at −80°C for future use.

Western blots were performed similarly as described before (Benner et al., 2023). Lysates were mixed with 4× Laemmli Sample Buffer (Bio-Rad Laboratories) supplemented with sodium dodecyl sulfate (SDS) and BME. The protein samples were run on 4% stacking and 7.5% resolving polyacrylamide gel electrophoresis (SDS-PAGE) for ≈2.5 h and transferred onto nitrocellulose membranes for 1 h. The membranes were blocked overnight while rocking at 4°C in Tris-buffered saline (TBS) containing 5% bovine serum albumin and 5% nonfat milk, supplemented with 0.1% Tween 20 detergent (TBST). The blots were first incubated with primary antibodies dissolved in blocking buffer for 2–3 h at room temperature, washed four times with TBST, incubated with secondary antibodies (≈1:3,000; DyLight 680/800, Invitrogen) for another 1–2 h at room temperature, and washed again four times with TBST. Membranes were imaged immediately using LI-COR Odyssey CLx system; signal brightness and/or contrast were uniformly adjusted by the Image Studio software (version 5.2). Protein contents were estimated by drawing regions of interest (ROIs) and calculating band intensities.

Immunocytochemistry and histochemistry

Primary neuron cultures were rinsed once with PBS only and then immediately fixed with 4% PFA for 30–45 min at room temperature. They were washed three times with PBS and blocked by 10% cosmic calf serum + 0.1% Triton X-100 detergent dissolved in PBS while rocking for ≈ 1 h at 37°C. Next, the primary antibodies were mixed in the blocking buffer at specific concentrations, added to the samples, and incubated for 2–3 h while rocking at 37°C. For brain slices, the primary antibody incubation was performed overnight at 4°C. After this, the samples were washed four times and then exposed to blocking buffer containing secondary antibodies conjugated with Alexa Fluor dye (405/488/546/647, Invitrogen) for 1–2 h at 37°C. Following this, samples were washed four more times with blocking buffer and once with PBS only. For nuclear staining, cells were further incubated with DAPI diluted in blocking buffer for 10 min at 37°C and then washed twice with PBS. The culture coverslips were mounted upside down on glass slides using Fluoromount-G (SouthernBiotech). The brain slices were processed very similarly, except occasionally incubated with primary antibodies overnight at 4°C and additionally shielded by cover glasses after mounting on the glass slides. All samples were left at room temperature to solidify overnight, prior to confocal imaging.

Confocal microscopy

All images were acquired using an inverted STELLARIS 5 (Leica Microsystems) laser scanning microscope with 405/488/561/638 nm wavelengths containing Power HyD detectors. Samples were imaged through a series of 0.5-µm-thick optical z-sections under 40× (1.3 NA) and 63× (1.4 NA) oil immersion, plan-apochromatic objectives. The images were processed through Leica Application Suite version X (LasX) and analyzed using the ImageJ (FIJI, NIH) software. Images from ROIs were presented as a single optical section or superimposed as maximum-intensity z-projections (10–30 slices). The colocalization parameters between two signals were assessed by first thresholding individual channels appropriately to eliminate any background noise and then measuring pixel-based percent coincidence between projected areas. For the cluster size of synaptic proteins, appropriately thresholded signals from one channel were used as a mask to detect inside versus outside signals from another channel. Three-dimensional reconstruction of cell bodies was rendered from confocal images using the Reconstruct software (SynapseWeb; Fiala, 2005). Pearson's and Mander's colocalization coefficients were calculated using the JACoP plugin in FIJI, and the intensity histograms were plotted using EzColocalization.

List of antibodies

The antibodies selected for Western blot and/or immunostaining assays included appropriate combinations of chicken anti-MAP2 (1:1,000; catalog #Ab5392 or Ab92434, Abcam), mouse anti-Synapsin1 (1:500; catalog #106011; clone, 46.1, Synaptic Systems), rabbit anti-Synapsin1/2 (1:500; catalog #106002, Synaptic Systems), guinea pig anti-Synapsin1/2 (1:500; catalog #106004, Synaptic Systems), guinea pig anti-vGLUT1 (1:500; catalog #135304, Synaptic Systems), rabbit anti-vGLUT1 (1:500; catalog #135303, Synaptic Systems), mouse anti-vGLUT1 (1:500; catalog #135011, Synaptic Systems), rabbit anti-vGLUT2 (1:500; catalog #135403, Synaptic Systems), rabbit anti-vGAT (1:500; catalog #131003, Synaptic Systems), mouse anti-vGAT (1:500; catalog #131011, Synaptic Systems), mouse anti-GAD65 (1:500; catalog #198111, Synaptic Systems), rabbit anti-GAD65 (1:500; catalog #198103, Synaptic Systems), mouse anti-GAD67 (1:500; catalog #198211, Synaptic Systems), rabbit anti-GDH (1:500; catalog #A5176, ABclonal), rabbit anti-GAD67 (1:500; catalog #198208, Synaptic Systems), rabbit anti-GLS (1:500; catalog #A3885, ABclonal), mouse anti-GLS (1:500; catalog #66265-1-Ig, Proteintech), rabbit anti-GABA_AR (GABA_A receptor) α3 (1:500; catalog #A11636, ABclonal), mouse anti-gephyrin (1:500; catalog #147111; clone, 3B11, Synaptic Systems), rabbit anti-GABA (1:1,000; catalog #A2052, Sigma-Aldrich), mouse anti-HA (1:1,000; catalog #h3663, Sigma-Aldrich), rabbit anti-HA (1:1,000; catalog #C29F4, Cell Signaling Technology), mouse anti-Cytochrome-C (1:500; catalog #66264-1-Ig, Proteintech), mouse anti-GAPDH (1:5,000; catalog #60004-1-Ig, Proteintech), mouse anti-SV2A (1:500; catalog #SV2, DSHB), mouse anti-Syt1 (1:500; catalog #105011, Synaptic Systems), rabbit anti-parvalbumin (PV; 1:500; catalog #A2791, ABclonal), rabbit anti-Ankyrin-G (1:500; catalog #PA5-143596, Thermo Fisher Scientific), and rabbit anti-Neurofilament (1:500; a gift from Dr. Peter Hollenbeck, Purdue University, to Dr. James Bamburg, Colorado State University). DAPI (1:50,000; catalog #D1306, Thermo Fisher Scientific) was included in the secondary antibody solution, for all nuclear staining purposes.

Electrophysiology

Whole-cell patch–clamp recordings were performed similarly to those described previously (Cast et al., 2021; Burlingham et al., 2022). In brief, the reprogrammed human neurons were patched using internal solutions consisting of (in mM) ≈120 KCl (for current-clamp configuration) or CsCl (for voltage-clamp configuration), plus 5 NaCl, 10 EGTA, 1 MgCl₂, 10 HEPES, 3 Mg-ATP, and 0.3 Na-GTP, 310 mOsm, and pH adjusted at 7.3–7.4 using either KOH (for current clamp) or CsOH (for voltage clamp). For voltage clamp, the internal solution additionally included 2.5 mM QX-314. The extracellular bath solution contained the following compositions (in mM) ≈140 NaCl, 5 KCl, 2 CaCl₂, 2 MgCl₂, 10 glucose, and 10 HEPES, 300 mOsm, and pH adjusted at 7.4 with NaOH. Electrophysiological recordings were conducted using an integrated patch-clamp amplifier (Sutter Instrument) with a customized Igor Pro (WaveMetrics) data acquisition and analysis system.

Current-induced AP firings were recorded approximately at a holding potential (V_hold) = −60 mV, by injecting small currents to adjust the membrane potential (V_m) accordingly. The voltage-clamp recordings for GABA_AR-mediated inhibitory postsynaptic currents (IPSCs) were conducted at a V_hold = −70 mV. Evoked IPSCs were triggered by field stimulation using a matrix electrode (catalog #MX21AEW-RT2, FHC) connected to an A365RC isolated pulse stimulator (World Precision Instruments). GABA_AR-mediated synaptic currents (both spontaneous and evoked) were isolated in the presence of both CNQX (AMPA receptor blocker) and CPP (NMDA receptor blocker), 50 μM each (Tocris Bioscience). Acute 5 min bath applications of TPMPA (250 μM; Tocris Bioscience) or GABAzine (25 μM; SR 95531, Tocris Bioscience) were used, respectively, to block GABA_CR (ρ subunit) or GABA_AR-mediated IPSCs.

Computational model

GABA diffusion was modeled using MCell4 through the CellBlender interface in Blender (Husar et al., 2024). The model neuron consisted of a spherical soma with 10 µm diameter and an axon with 100 µm length and 0.25 µm diameter. Including the soma in this model proved impractical because it imposed an enormous computational burden with ≈3.1 × 10⁸ GABA molecules, without providing additional insights into their diffusion down the axon. Instead, the axon entry point was modeled as a concentration-clamped boundary with GABA levels maintained at 1 mM (Kerr et al., 2008), i.e., within range of its typical cytosolic concentration in neurons (Apostolides and Trussell, 2013). This boundary condition is valid if we assume that somatic GABA concentration remains relatively constant through a homeostatic regulation and that this mechanism can meet the GABA demands imposed by the axon terminals downstream. At the synaptic end of this axon, the activity of vGAT was mimicked as a single molecule of absorber with binding (i.e. absorption) rate constants of 2.18 × 10⁷ M^–1 s^–1, 2.18 × 10⁸ M^–1 s^–1, or 2.18 × 10⁹ M^–1 s^–1, to represent presynaptic terminals of different activity levels. These three absorption rate constants were explicitly chosen based on the biophysical principles of molecular collisions and reactions in diffusion theory (i.e., from the Einstein–Smoluchowski equation for the rate of molecular encounter) to assign a binding probability per discrete collision event of p = 0.01, p = 0.1, and p = 1 occurring during each 1 µs time step in the simulations [for detailed derivations, see Stiles and Bartol (2001) and Kerr et al. (2008)]. The diffusion constant of GABA in saline is from 7.7 to 10 × 10⁻⁶ cm² s⁻¹ (Herz et al., 1969; Rodrigo et al., 2017); we conservatively used the lower value and divided that by 3 (i.e., 2.56 × 10⁻⁶ cm² s^–1) to correct for the lower rate of diffusion in a viscous cytosol (Verkman, 2002; Ando and Skolnick, 2010). The simulation time step was 1 µs, so GABA in the model had an average radial diffusion step length of 36 nm. The positions of GABA molecules entering the axon were tracked for 25 s (for the p = 1 case) to 40 s (for the p = 0.1 and p = 0.01 cases), at which point the number of GABA molecules in the axon had reached equilibrium.

Data presentation

For all figure panels, the average values (bar graphs, filled squares, and pie charts) reflect means ± SEM (i.e., standard deviation of a given parameter divided by the square root of sample numbers) and are presented with total number of replicates, e.g., cultures examined (immunoblots) or field-of-views imaged (immunostainings) or neurons patched (electrophysiology) from independent experimental batches. Individual data points are included as color-matched symbols and connected with lines for paired comparison. The type and strength of all statistical evaluations between experimental groups are mentioned in corresponding figure legends. These included either paired (batchwise assessments between experimental conditions) or unpaired (multiple measurements from each batch), two-tailed, Student’s t test (single averages), or one-way analysis of variance (ANOVA; multiple averages), with ***p < 0.001; **p < 0.01; *p < 0.05; and ns, not significant, p > 0.05.

Results

GAD isoforms but not GLS are preferentially recruited by presynaptic terminals

To determine the primary subcellular regions for glutamate versus GABA biosynthesis, we sought to monitor their catalytic enzymes, i.e., GLS versus GADs, and inquired whether they can associate with their respective vesicular transporters, i.e., vGLUT1 versus vGAT, located at glutamatergic versus GABAergic presynaptic terminals. At first, we examined six different broadly defined brain regions, i.e., somatosensory cortex layer IV and V, stratum oriens (SO) of the CA1 hippocampus, molecular layers (MLs) of the dentate gyrus, the basolateral amygdala (BLA), posterior complex and ventral posteromedial nucleus (VPM) of the thalamus, ventromedial nucleus (VMH), and arcuate nucleus (ARC) of the hypothalamus, from adult mouse brain slices (Fig. 1A). Intriguingly, we observed that the majority of GLS signals did not associate with vGLUT1 puncta, regardless of the brain areas analyzed (Fig. 1B,C). Instead, they commonly aggregated into large clusters around neuronal cell bodies labeled with DAPI (Fig. 1D,E), even including those of GABAergic identities (Fig. S1A). This was remarkably different for the GAD isoforms (i.e., GAD65 and GAD67 combined), which manifested obvious colocalization with vGAT puncta at every brain region explored (Fig. 1B,C). Therefore, the enzymes producing GABA but not glutamate are spatially coupled with corresponding vesicular packaging sites at their respective presynaptic terminals.

Figure 1.

Contrasting presynaptic association of GLS versus GADs at multiple brain regions. A, Coronal view of an adult mouse brain section adopted from the Allen Brain Atlas (http://atlas.brain-map.org/), with several ROIs (somatosensory cortex layer IV and V, SO region of the CA1 hippocampus, ML of the dentate gyrus, posterior complex and VPM of thalamus, BLA region of the amygdala, both VMH and ARC nucleus of the hypothalamus) specified using boxes (dotted squares), that were subsequently analyzed in B and C. B,C, Sample images (B) of designated brain regions (top to bottom) immunolabeled in pairs for (i) vGLUT1 + GLS or (ii) vGAT + GADs (65 and 67), together with a nuclear DAPI stain; boxed areas from superimposed merged views (left panels) are further expanded, separated into individual channels for a single optical plane (right panels); yellow arrowheads, colocalized signals; cyan arrowheads, limited overlap between the two channels. Pearson's coefficients of correlation (C), measured between vGLUT1 and GLS versus vGAT and GADs. D, Representative images (i, single optical sections) of different areas from mouse brain sections (as labeled), immunostained for GLS, vesicular glutamate transporter vGLUT1, and nuclear DAPI; asterisks, neuronal cell bodies surrounded by prominent GLS signals, exhibiting minimal colocalization with vGLUT1. Three-dimensional reconstructions of cell bodies (ii) illustrate extensive GLS clusters around DAPI-positive nuclei. E, The average size of GLS clusters formed inside versus outside cell soma, from cortical and hippocampal sections. F, Example image (left; a boxed area magnified with single channels split) and average intensity (right; arbitrary unit, AU) of GLS signals colocalized (yellow arrow) versus noncolocalized (cyan arrow) with vGLUT1 or vGLUT2 or both vGLUT1- and vGLUT2-labeled glutamatergic presynapses in the Layer IV and V of the somatosensory cortex. G, Same as F, except for GDH signals located around DAPI-positive soma, inside versus outside vGLUT1 puncta. H–I, Same as G, except for GAD65 (H) and GAD67 (I) signals present inside (+) versus outside (−) the GABAergic presynapses labeled by vGAT, and their respective cluster sizes; arrowheads, vGAT puncta with (yellow) or without (cyan) the GADs. Notice a greater degree of presynaptic enrichment for GAD65 over GAD67, as the latter was also detected at the DAPI-positive cell bodies of GABAergic neurons (asterisk in I, example image). Quantifications reflect means ± SEM, with values on the bar graphs indicating the total number of brain slices from two animals (C) or ROIs/brain slices (E–I). Single data points are plotted as color-matched symbols, adjoined by lines when calculated from same images. The statistical comparison between conditions was conducted by two-tailed, unpaired (panel C) or paired (panels E–I), Student's t test, with ***p < 0.001; **p < 0.01; and *p < 0.05.

In cortical sections, we continued to detect minimal colocalization between GLS and vGLUT2 or vGLUT1/2 combined, confirming its negligible recruitment at presynapse (Fig. 1F). A similar assessment of GDH, another major enzyme that produces glutamate from α-ketoglutarate, also revealed its weak association with vGLUT1-labeled glutamatergic presynapses (Fig. 1G). The GABAergic presynaptic compartments did not contain GDH either (Fig. S1B) but recruited a considerable amount of individual GADs (Fig. 1H,I). Notably, between the two GADs, GAD65 demonstrated preferential enrichment at synapses over its GAD67 counterpart (Fig. 1H,I), as the latter in particular was also detected in the cell bodies of PV-positive neurons (Fig. S1C,D). Therefore, although both GADs were present at presynaptic sites at levels much higher than GLS or GDH, the GAD65 versus GAD67 isoforms exhibited noticeable differences between each other.

GAD isoforms are associated with presynapses at distinct subregions in the brain

To further assess the relative enrichment of GAD isoforms at different subsections within a given brain region and their relative association with local GABAergic presynapses, we analyzed them at several sublayers of the hippocampus and dentate gyrus. We observed that, although GAD65 is widely dispersed in the hippocampal CA1 subregions including the SO, stratum pyramidale (SP), radiatum (SR), and lacunosum moleculare (SLM), it is particularly enriched at the SLM area, where it highly correlated with vGAT puncta presumably formed on the distal dendrites of pyramidal neurons (Fig. 2A,C). In contrast, GAD67 showed abundant expression and major colocalization with vGAT signals in the SP area containing the cell bodies of pyramidal neurons (Fig. 2B,C). Hence, GAD65 versus GAD67 are selectively and differentially enriched at specific hippocampal strata, where they respectively associate with the dendritic versus somatic GABAergic presynaptic structures formed on CA1 pyramidal neurons.

Figure 2.

An isoform-specific distribution of GAD65 versus GAD67 at hippocampal sublayers. A, Example images of the hippocampal CA1 region, immunolabeled for GAD65 and vGAT and stained with DAPI; boxed regions from SO, SP, SR, and SLM sublayers are further magnified below and split into single channels. B, Same as panel A, except for the GAD67 isoform. C, The average values (means ± SEM, filled squares) indicate (left to right) relative intensities of vGAT, GAD65, and GAD67 signals, as well as the total fraction of vGAT puncta cocontaining GAD65 or GAD67 signals, at SO, SP, SR, and SLM sublayers of the CA1 hippocampus. Individual data points from each field-of-view are provided as color-matched open circles joined by straight lines, with numbers on the bar graphs indicating ROIs/brain slices. The degrees of colocalization between vGAT, GAD65, and GAD67 were quantified using the percentage values of Mander’s coefficient calculated from single optical sections, thresholded uniformly across different subregions. D, Example images of hippocampal CA3 region, immunolabeled for GAD65 and vGAT, and stained with DAPI; boxed regions from SO, SP, SL, and SR sublayers are further magnified below and split into single channels. E, Same as panel D, except for the GAD67 isoform. F, Same as panel C, except for SO, SP, SL, and SR sublayers within the CA3 region of mouse hippocampus.

The two GAD isoforms were also differentially distributed in the CA3 hippocampal region. GAD65 displayed a limited presence and vGAT association at the stratum lucidum (SL) areas and was mostly restricted to SO and SR layers, with intermediate SP signals (Fig. 2D,F), whereas high GAD67 levels were observed particularly at the SP regions with pronounced vGAT colocalization (Fig. 2E,F). In the dentate gyrus, GAD65 was predominantly localized to the outer ML, with fairly modest appearances at the inner ML, granule cell (GC) layer, or hilus (Fig. S2A,C), whereas both the GC layer and hilus contained prominent GAD67 signals (Fig. S2B,C). Despite their differences in layer-specific distributions, a vast majority of GAD65 and a large fraction of GAD67 signals continued to be associated with vGAT-positive terminals in the dentate gyrus (Fig. S2A–C). These were highly contrasting to vGLUT1-positive glutamatergic presynapses, which rarely contained either GLS or GDH signals (Fig. S2D,E).

GLS is associated with mitochondrial network present in the perisomatic region

To further validate the divergent subcellular distributions of GLS versus GADs, we next generated mouse primary hippocampal cultures, which provide a better optical resolution for synaptic structures formed in monolayered neurons. Once again, consistent with brain slices, we found that the GLS clusters were predominantly enriched at the cell bodies and proximal dendrites of most neurons (Fig. 3A), regardless of their non-GABAergic or likely glutamatergic (e.g., pyramidal cells) versus GABAergic interneuron identities (Fig. S3A). Moreover, the vast majority of GLS signals present in cell soma colocalized extensively with local Cytochrome-C clusters, implying their direct association with perisomatic mitochondria (Fig. 3A). In contrast, both GAD65 and GAD67 proteins were coexpressed in most GABAergic neurons by a variable degree, processed at their cell bodies, but also trafficked elaborately to distant subcellular regions, often appearing as colocalized spots (Fig. 3B).

Figure 3.

Contrasting presynaptic association of GLS versus GADs also in cultured neurons. A, Example images of neurons (cyan arrows) from mouse hippocampal culture coimmunolabeled for dendritic MAP2, GLS, and a mitochondrial marker Cytochrome-C (Cyto-C), either as separate channels or their merged view; the boxed areas from a single soma are magnified below as series of single optical sections, with yellow arrowheads pointing at large somatic clusters of colocalized GLS and Cyto-C signals. Bar graphs, average intensities of GLS signals calculated from cell bodies versus nonsomatic areas and with (+) or without (−) Cyto-C. B, Sample images of mouse primary hippocampal neurons immunolabeled for MAP2; the GABAergic neurons (yellow arrowheads) were readily distinguished from neighboring non-GABAergic neurons (cyan arrowheads), as they expressed both GAD65 and GAD67 with similar distribution patterns. The boxed area enlarged below with split channels display both colocalized (yellow arrows) and noncolocalized (cyan arrows) GAD isoforms. Bar graphs, average fractions of GAD65 and GAD67 puncta distributed at somatic versus nonsomatic regions. C,D, Sample images (C) from mouse primary hippocampal cultures, immunolabeled for MAP2 and Synapsin, in combination with either GLS, GAD65, or GAD67 (left to right); arrowheads point at colocalized (yellow) or independent (cyan) signals. Average values (D) indicate percentages of the total GLS, GAD65, and GAD67 area colocalized with Synapsin (left) or their relative cluster sizes when located inside versus outside synapses (right). E, F, Example images (E) and average parameters (F; Pearson's values and synaptic vs nonsynaptic cluster size) present relative degrees of colocalization between vGLUT1 and GLS versus vGAT and GAD65 or GAD67. G, H, Dendritic segments coimmunolabeled (G) for MAP2, Synapsin and both GAD65 and GAD67; visualized as z-projected merged images (left) or the boxed region enlarged and channels split into series of single optical sections (right); arrows, Synapsin puncta with (yellow; GABAergic) or without (cyan; non-GABAergic) GADs. Colocalization between the two GAD isoforms, GAD65 and GAD67 (H), displayed as signal intensities from a single synapse (left; measured along the dotted line in panel G), multiple synapses (middle; raster-plot fit with a straight line), or their total fractions overlapped with each other (right; pie charts with respective percentages). All average data reflect means ± SEM, along with total number of ROIs analyzed/independent experimental batches (provided with bar graphs) and individual data points (color-coded symbols, connected with lines for pairwise comparisons). Statistical significance was estimated by two-tailed, either paired (panels A, B, D, and F) or unpaired (panels D and F), Student's t test, with ***p < 0.001; *p < 0.05; ns, not significant, p > 0.05.

Our immunostaining experiments revealed that a considerable fraction of both GAD65 and GAD67, but not GLS, preferentially coincides with pan-presynaptic marker Synapsin, with enlarged cluster volume (Fig. 3C,D). When immunolabeled in pairs with their corresponding vesicular transporters at presynapse, GLS continued to demonstrate minimal correlations with vGLUT1, whereas both GADs displayed substantial colocalization with vGAT (Fig. 3E,F). We also noted a consistent lack of correlation between GLS and vGLUT2 or vGAT signals, corroborating its limited presence at either glutamatergic or GABAergic synapses, as evidenced in brain slices (Fig. S3C,D). Similar to GLS, we also observed minimal trafficking of GDH at the presynaptic sites labeled with Synapsin (Fig. S3E), which, instead, also colocalized with somatic clusters of mitochondria labeled with Cytochrome-C (Fig. S3F). Therefore, our results imply that glutamate versus GABA, the two essential neurotransmitters, are likely biosynthesized at distinct subcellular compartments, although both of them could be efficiently loaded into presynaptic vesicles and released from axon terminals.

To determine if GAD65 and GAD67 are recruited by the same or separate GABAergic terminals of primary hippocampal neurons, we next probed for both GADs together with Synapsin (Fig. 3G). We noticed that a vast majority of Synapsin puncta that included GAD65 also accommodated GAD67 in a highly reciprocal fashion, as the bulk of their presynaptic signals overlapped extensively with each other (Fig. 3G,H). Hence, a GABAergic presynapse can co-contain both GAD isoforms, especially when coexpressed by the same neurons. However, we frequently observed that, relative to GAD65, the GAD67 isoform appeared to manifest a particularly strong signal inside the cell bodies of GABAergic neurons irrespective of their PV versus non-PV subtype identities, once again hinting at possible trafficking differences between the two GADs (Fig. 3B; Fig. S3B).

GAD65 exhibits a higher degree of presynaptic localization than GAD67 isoform

Since GAD65 and GAD67 are known to form both homo- and heterodimers (Dirkx et al., 1995; Kanaani et al., 1999; Fenalti et al., 2007), we wondered whether they can also traffic to presynapses independently of each other. To investigate that, we reprogrammed human iPS cells (WTC-11 line; Wang et al., 2017; Carricaburu et al., 2024) into purely glutamatergic neurons that are deficient of any endogenous vGAT and GAD expression by forced induction of a single transcription factor Ngn2 (Zhang et al., 2013; Chanda et al., 2019). We infected these neurons with lentiviruses encoding human vGAT in combinations with either GAD65 or GAD67 or both GADs mixed at equal ratios (1/2:1/2), cocultured them with mouse primary astrocytes for further maturation, and analyzed them at multiple time-points (Fig. 4A). We first ran an immunoblot with cell extracts 7 d postdifferentiation and already detected a prominent degree of transgene induction with high specificities (Fig. 4B). This approach allowed us to avoid any potential contamination from endogenous GADs or heterodimerization between the two GAD isoforms and characterize each of them separately in a human cellular context.

Figure 4.

Differential localization of GAD65 versus GAD67 isoforms at GABAergic presynapse. A, Ngn2-inducible iPSCs (WTC-11 line) were reprogrammed into human neurons by doxycycline, infected with lentiviruses encoding vGAT + GAD65 or GAD67, cocultured with mouse glia, analyzed at indicated time-points. Images depict iPS cells at Day 0 and differentiated neurons with elaborate neurite outgrowth already by Day 5. B, Example Western blot (left) and average expression levels (right) of exogenous vGAT and HA-tagged GAD65 or GAD67, at Day 7 of neuronal differentiation, either in control (Ctrl) condition or when transduced with lentiviruses encoding vGAT+ either GAD65 or GAD67; GAPDH was used as loading control. C, D, Representative images (C) of Day 50 neurons transduced with a Ctrl virus versus lentiviruses encoding vGAT + GAD65, vGAT + GAD67, or vGAT + GAD65 and GAD67 mixed at equal ratios, when labeled with MAP2 and both GAD65 and GAD67 antibodies; insets, boxed regions are further expanded below; arrowheads, GAD67 at soma (cyan), GAD65 and GAD67 puncta colocalized (yellow). Average intensities (D) of GAD65 and GAD67 signals from the entire field-of-views (left; bar graphs) or soma versus nonsomatic axonal areas (raster plot; right). E, Representative images (left) of neurons infected with vGAT + GAD65 versus vGAT + GAD67 lentiviruses, when immunostained for both GAD isoforms, together with Synapsin and MAP2; arrowheads, synaptic enrichment of GADs; asterisks, GAD67 labeled neuronal cell bodies. Relative fraction of GAD65 versus GAD67 puncta area or their average cluster sizes (right), when situated outside (negative) versus inside (positive) of Synapsin signals. F–H, Same as E, except for GAD colabeling with endogenous SV2 (F) and Syt1 (G) or exogenous vGAT (H). All summary plots (bar graphs and filled squares) indicate means ± SEM and report the number of independent replicates (B) or field-of-views analyzed/individual batches (D–H). The single data points were included as color-coded open circles (adjoined by lines for pairwise comparison). Statistical powers between experimental groups were evaluated using two-tailed, Student's t test, with ***p < 0.001; *p < 0.05; ns, not significant, p > 0.05.

At Days 50–56, we observed a robust endogenous GLS expression in these neurons, which was consistently present at nonsynaptic regions including the cell bodies, and in alignment with glutamate production required as the GABA precursor (Fig. S4A,B). When transduced with exogenous GAD65 versus GAD67, we confirmed a sustained expression of both GAD proteins in their respective cultures without cross-reactivity (Fig. 4C,D). Once again, GAD65-alone clusters were almost exclusively located at axon terminals developed along the MAP2-positive dendritic arbors, whereas a large pool of GAD67-alone signals were apparently retained in cell soma (Fig. 4C,D). Human GADs were transduced at an equivalent degree to their endogenous versions in mouse neurons, arguing against any overexpression artifact beyond physiological ranges (Fig. S4C). When expressed together, GAD65 and GAD67 often manifested spatial overlaps as they coassociated with Synapsin, possibly because of their presynaptic cotrafficking as heterodimers (Fig. S4D).

To evaluate the synaptic versus nonsynaptic distribution of GAD65 versus GAD67, we inspected distantly located nonsomatic GAD clusters and measured their relative association with presynaptic markers. We found that the size of both GAD65 and GAD67 puncta was consistently potentiated if present at synaptic terminals labeled by endogenous Synapsin, SV2, Syt1, or exogenous vGAT (Fig. 4E–H). Nevertheless, when compared with GAD67 counterparts, the GAD65 clusters depicted a greater tendency for presynaptic accumulation, in terms of the total signal area shared with each of those presynaptic proteins individually (Fig. 4E–H). These results suggest that the homodimers of two major GAD isoforms are not equitably recruited by human presynaptic compartments.

GAD65 and GAD67 can individually trigger efficient GABAergic synaptic activity

We inquired if the distinct subcellular distribution of GAD65 versus GAD67 could differentially impact their relative abilities to biosynthesize GABA from ambient glutamate, supply it to axon terminals for vesicular transmission, and activate postsynaptic GABA_ARs. To this end, we monitored Day 50–56 human neurons transduced with either GAD65- or GAD67-only or both GAD65 + GAD67 in combination with vGAT. Lentiviral transduction of exogenous GADs did not change the intrinsic membrane properties, action potential (AP) firing, or voltage-gated Na⁺ and K⁺ channel currents (Fig. S5A–D). Notably, the presynaptic terminals of Ngn2-induced human neurons were already apposed by gephyrin scaffolds at their postsynaptic interfaces, which recruited and organized abundant GABA_ARs (Fig. S6A; Carricaburu et al., 2024). This allowed us to perform electrophysiology recordings to directly estimate and compare the GABAergic activities.

In the control condition, we detected virtually no spontaneous IPSCs (sIPSCs), in agreement with negligible content of endogenous vGAT and GADs in Ngn2 neurons (Fig. 5A). However, in the presence of lentivirally delivered exogenous vGAT, either GAD65- or GAD67-only or both GAD65 and GAD67 combined triggered robust sIPSCs with high frequency and amplitude, corroborating efficient GABA synthesis and its adequate vesicular supply by individual GAD isoforms (Fig. 5A,B). Interestingly, all sIPSC parameters including their event kinetics were found to be very similar between the GAD65-only or GAD67-only and GAD65 + GAD67 conditions (Fig. S6B). Furthermore, when presynaptic outputs were activated globally by electrical stimulation, each GAD isoform illustrated reproducible GABAergic evoked IPSCs with comparable amplitude and coefficient of variation, confirming an equivalent probability of vesicular GABA release (Fig. 5C).

Figure 5.

Comparable GABA synthesis and its presynaptic release by GAD65 and GAD67. A, B, Representative traces of GABAergic sIPSCs (A), their cumulative probability plot (B, left), or average event frequency and amplitude (B, right), as recorded from Ctrl versus vGAT + GAD65, vGAT + GAD67, or vGAT + GAD65 + GAD67 condition, in the presence of 50 μM CNQX + 50 μM CPP; insets in A, boxed regions magnified below. C, Representative traces of GABAergic IPSCs evoked by five successive trials of presynaptic stimulation (left), their average amplitudes, and coefficients of variation (right), as recorded from Ctrl versus vGAT + GAD65, vGAT + GAD67, and vGAT + GAD65 + GAD67 conditions; the stimulus artifacts (arrow) are mostly erased for clarity. D, Example traces (left) of GABAergic IPSCs evoked at various time intervals (arrows, Δ time), and their PPRs calculated as IPSC₂/IPSC₁ (right), from neurons expressing vGAT + GADs; inset, first three Δ time magnified. E, Sample traces (left) of GABAergic IPSCs evoked with a 10 Hz train stimulation, IPSC total charge transfers, or all IPSC amplitudes normalized to IPSC₁ (right), recorded from cells expressing vGAT + GAD65 versus GAD67. F–G, Example images (F) and average levels of GABA production or vGAT expression (G), in Day 50 human neuronal cultures from Ctrl versus vGAT + GAD65, vGAT + GAD67, or vGAT + GAD65 and GAD67 (equal ratios) conditions, labeled with nuclear DAPI, immunostained for MAP2, vGAT, and GABA, depicted as all channels merged; boxed areas are magnified below; yellow arrows, a limited amount of GABA and vGAT in Ctrl cells. The brightness and contrast of all channels were calibrated equally to eliminate any background signal across experimental conditions (i.e., vGAT + GAD65 and/or GAD67), using thresholds for Ctrl condition as baselines. H, Neurons from vGAT+ (i) GAD65, (ii) GAD67, or (iii) GAD65 + GAD67 conditions. Left, Sample images of axonal branches containing diffused GABA and clustered vGAT signals or their mutual coincidence (arrows) in overlayed views, from superimposed z-stacks (merge) or series of optical sections. Right, Signals from a single synapse (intensity profiles along the dotted lines) and fraction of the vGAT area with or without GABA (pie charts). All average data (bar graphs and filled symbols) are presented as means ± SEM, along with the total number of neurons patched (panels A–E) or images acquired (panel G)/independent experimental batches. The pie charts (panel H) represent relative fractions of the vGAT cluster area coenriched with intracellular GABA signals. All individual data points were provided as color-coded open circles. Statistical assessments were performed using either two-tailed, unpaired, Student's t test (the bar graphs in panels B, C, and G; box plot in E), or one-way ANOVA (for multipoint comparisons; panels D and E), with ***p < 0.001; ns, not significant, p > 0.05.

Presynaptic stimulation at variable time intervals also triggered robust evoked IPSCs for both GAD65 and GAD67 conditions with similar paired-pulse ratios (Fig. 5D). Moreover, when stimulated with a train of pulses, both GAD isoforms produced a series of IPSCs with equivalent charge transfer and demonstrated short-term plasticity with obvious synaptic depression (Fig. 5E). Acute treatment of selective GABA_C receptor (GABA_CR) antagonist TPMPA had a marginal effect on evoked IPSCs, whereas bath application of GABAzine effectively blocked these synaptic responses, suggesting that they are primarily mediated by GABA_ARs (Fig. S6C). Of note, TPMPA also acts as a low-affinity–competitive antagonist for GABA_ARs and, thus, partially protects a fraction of them from postsynaptic receptor saturation (Markwardt et al., 2009; Ammer et al., 2015). Nonetheless, although the bath application of TPMPA in GAD65-transduced human neurons caused a minor but significant decrease in IPSC amplitude, it failed to effectively alter IPSC paired-pulse ratios, suggesting that the synaptic depression observed in this system likely had a presynaptic origin and presumably reflected high probability of transmitter release (Fig. S6C). In sum, these results suggested that GAD65 and GAD67 can successfully produce and deliver a substantial amount of GABA for both vesicular packaging and activity-dependent release, irrespective of their disproportionate trafficking to the human presynaptic terminals.

Both GAD65 and GAD67 generate a considerable amount of intracellular GABA

We subsequently explored how the GABA molecules generated by each individual GAD isoform were spatially distributed within a cell. We immunostained GAD65- and GAD67-expressing human neurons for MAP2, vGAT, and nuclear DAPI, paired with anti-GABA antibody (Fig. 5F). We noticed that, relative to the control condition, GAD65 and GAD67 alone as well as both GADs together produced abundant amounts of GABA that diffused throughout the intracellular space, including cell bodies and neuronal processes (Fig. 5G). This was matched by a substantial quantity of exogenous vGAT expression, which showed punctate axonal distribution along the MAP2-labeled dendrites (Fig. 5F,G). In all conditions, most of the vGAT clusters were commonly occupied by a soluble GABA signal, inferring its presynaptic availability regardless of either GAD isoforms used (Fig. 5H).

To further gauge this diffusion phenomenon, we probed for intracellular GABA signals in rodent GABAergic neurons that generally coexpress both GAD isoforms (Fig. 3B,G–H). We immunostained mouse cortical slices and primary hippocampal cultures for GABA. In alignment with our observations in human neurons transduced with exogenous GADs, mouse neurons containing endogenous GADs also elicited intense GABA signal which dispersed globally throughout their soma and neurite projections (Fig. S7A). The GABA signals were also adequately detected at presynaptic spots coimmunolabeled with endogenous vGAT, GAD65, and to some extent GAD67 (Fig. S7B–E). Collectively, our results predict that GABA molecules, once manufactured by either GAD isoform, might sufficiently diffuse to synapse terminals to drive vesicular release.

Diffusion of GABA from distal source can account for its presynaptic availability

We naturally wondered why localization of GAD65 versus GAD67 at distinct subcellular regions did not appreciably affect GABA levels at the presynapse. For individual neurons, the GABA signals appeared to be most intense at larger compartment volumes, always high in the cell bodies, some in dendrites, with thinner axons often depicting punctate labels at the presynaptic sites cocontaining vGAT (Fig. 6A). Since GABA is a small (≈0.1 kDa), polar, and soluble molecule, we asked how quickly it diffuses away from the point of synthesis and distributes to distant parts of a neuron in a time- and distance-dependent manner. To understand this, we next developed a 3D particle or agent-based reaction–diffusion model using the MCell simulator (see Materials and Methods). We assumed a cell soma as the sole source for GABA supply with a fixed concentration, implying continuous production (physiological ranges ≈ 5–7 mM; Apostolides and Trussell, 2013), and monitored GABA diffusion along a single axon originating from it (thickness in Ngn2-induced neurons ≈0.42–0.78 µm; Fig. 6B). For this current model, we conservatively used lower values for all parameters, i.e., a constant concentration of 1 mM GABA at its source and an axonal diameter of 0.25 µm (Fig. 6C). The positions of individual GABA molecules were tracked throughout the simulation period, as they entered and propagated through the model axon for 25–40 s, during a stochastic diffusion (Fig. 6C).

Figure 6.

Rapid diffusion and global distribution of GABA throughout intracellular space. A, Representative images from Ngn2 neurons cotransduced with vGAT + GAD67, immunolabeled for MAP2, vGAT, GABA, and stained with DAPI; asterisks indicate high GABA levels in the MAP2-positive neuron soma (yellow) adjacent to MAP2-negative astrocyte (cyan), with arrows pointing at GABA signals in dendrite (cyan), axon (white), or presynapses (yellow); the boxed area from merged image is enlarged and split as dual channels. B, Left images, Ngn2-induced human neurons were coimmunostained for dendritic MAP2 (white arrowheads) and axonal Ankyrin-G (Ank-G), indicating the axon initial segment (AIS, cyan arrowheads) or distal branches (yellow arrowheads), with the boxed region further enlarged below. Right images, Ngn2 neurons immunolabeled for MAP2, paired with axonal Neurofilament (NF). Box plot, Average axon widths (mean ± SEM, filled squares; n = 30 per condition, open circles), using the indicated markers (Ank-G and NF), at both AIS and distal processes. C, A 3D reaction–diffusion model of GABA diffusion and absorption: implementation of the axon diffusion model in MCell4-CellBlender. GABA molecules are represented as oversized yellow spheres for visualization purposes. GABA concentration is maintained in the cell body at 1 mM (mimicking continuous enzymatic production), as its propagation along the axon is simulated (diffusion constant, D); distant GABA molecules are removed from the presynaptic terminal by absorption at variable rates (binding constants, K_r), imitating vesicle loading and release. D, Average time (mean ± standard deviation; open circles) taken for the first three GABA molecules to arrive at the synaptic terminal and be removed by absorption, calculated from 10 independent simulations (filled circles). E, F, A time-dependent distribution of GABA molecules (panel E) inside the model axon (visualized in C), as they gradually approach steady states (gray-dotted lines) and their subsequent removal after arriving at the terminal (panel F), for three different K_r values of absorption with different binding probabilities (color shades, listed in C). G, Concentration gradient of GABA molecules at 0.5, 2, 5, 10, 25, and 40 s (color shades) after tracking them entering the axon from soma, as they stochastically diffuse to the presynaptic site subject to different K_r values (panel i, 2.18 × 10⁹M^–1s^–1; panel ii, 2.18 × 10⁸M^–1s^–1; and panel iii, 2.18 × 10⁷ M^–1 s^–1).

The first GABA molecules to reach a model presynaptic terminal 100 µm away from cell body arrived within ≈1.5 s (in diffusion theory, this statistic is referred to as the “first passage time”; Fig. 6D), and a steady-state gradient was achieved by ≈15 s at the highest rate of absorption (Fig. 6E,F). The nature of this equilibrium depended on the rates of GABA consumption at the terminal. We first examined the presynaptic terminal as a perfect absorber (i.e., the diffusion-limited case with an absorption rate constant = 2.18 × 10⁹ M^–1 s^–1; Fig. 6C), consuming every GABA molecule upon contact that diffused as far as the terminal. In this case, at equilibrium, the GABA concentration declined linearly from 1 mM at the axon entry point to 0 mM at the terminal (Fig. 6Gi). A substantial pool of GABA molecules accumulated just upstream of this terminal, implying the presence of a local reservoir to support their presynaptic deliveries. This scenario represented highly active terminals, where GABA is immediately consumed for vesicular release (Fig. 6Gi). We also considered absorption efficacies 10 and 1% of the one described above, to model conditions with lower rates of GABA consumption, i.e., moderate to low activity levels (rate constant = 2.18 × 10⁸ and 2.18 × 10⁷ M^–1 s^–1, respectively; Fig. 6C). In these cases, GABA also quickly attained equilibrium state with even higher concentrations (≈0.2–0.7 mM) at the terminal (Fig. 6Gii,iii). Such terminals would develop even larger reservoirs of local GABA molecules for vesicle loading as activity increases during periods of peak demand, in addition to more GABA arriving rapidly from the soma.

We compared these computational models with estimates of GABA utilization in functional human neurons (Fig. 5A–E). In the absence of significant AP-driven synaptic release, the average sIPSC rate in these cultures was ≈ 2 Hz (Fig. 5A,B). This is equivalent to each neuron releasing, on average, ≈2 vesicles/s across all of its synaptic terminals. Since each synaptic vesicle contains ∼3,000 GABA molecules (Farsi et al., 2016), a single neuron would consume ≈6,000 GABA molecules/s to maintain that basal sIPSC activity. In our model neuron, we found that individual GABA molecules entered the axonal segment at a rate of ∼64,000 s^–1, which would be more than enough to support spontaneous mIPSC release at desired frequency and sufficient for up to 10 terminals downstream of the soma. GABA consumption during AP-driven release depends on the rate of neuronal activity, which is highly context-dependent. For an initial estimate, the average evoked IPSC amplitude was ≈1 nA (Fig. 5C), and the average sIPSC size was ≈100 pA (Fig. 5A,B). Therefore, each evoked IPSC utilized ≈10 vesicles, i.e., ∼30,000 GABA molecules. During the paired or train stimulations (Fig. 5D,E), GABA utilization would increase; however, the total number of GABA molecules reserved inside the axon and the GABA-filled vesicles stored at presynaptic terminals are substantial. The total GABA content would be even greater for axons of higher diameter, which could support the IPSC amplitude for a period of time. Thus, a GABA gradient established through rapid diffusion from distal compartments, e.g., cell bodies, may support the restocking of presynaptic terminals during neurotransmission, although it would be important to evaluate constraints that arise at higher activity levels and more release sites.

Impeding the synaptic trafficking of GADs does not affect global GABA diffusion

Intrigued by these in silico modeling results, we next aimed to diminish and perhaps eliminate the presynaptic enrichment of GAD isoforms and directly measure their impacts on GABAergic neurotransmission. To further test the GABA diffusion hypothesis, we created a number of recombinant GAD65 and GAD67 constructs. We systematically (1) swapped the N-terminal domains between GAD65 versus GAD67 that variably determine their membrane association, Golgi targeting, and synaptic localization (Dirkx et al., 1995; Kanaani et al., 2002) and fused them with corresponding C-terminal parts essential for their cofactor binding and/or catalytic functions (i.e., 65N-67C and 67N-65C; Fig. 7A; Battaglioli et al., 2003; Fenalti et al., 2007); (2) substituted two critical cysteine residues (amino acids 30 and 45) of GAD65 with alanine that are necessary for its palmitoylation and membrane association (i.e., Ala^30/45; Fig. 7A); (3) deleted the N-terminal 1–23 residues of GAD65 (i.e., ΔN^1-23; Fig. 7A) required for its Golgi trafficking (Kanaani et al., 2002), even (4) created a version containing both of those modifications (i.e., Ala^30/45/ΔN^1-23; Fig. 7A), and (v) deleted the entire N-terminus (amino acids 1–89) of GAD67 (i.e., ΔN^term; Fig. 7A) to completely eliminate its synaptic targeting (Kanaani et al., 1999, 2010).

Figure 7.

Impeding presynaptic GAD trafficking does not prevent vesicular GABA release. A, Schematic diagrams of chimeric (65N-67C and 67N-65C) or mutated (amino acid substitutions, 65-Ala^30/45, or domain-deleted, 65-ΔN^1-23 and 67-ΔN^term, or both combined, 65-Ala^30/45/ΔN^1-23) versions of the two GAD proteins. B, Sample images of Day 50–56 human neurons coimmunostained for MAP2 and GABA, from Ctrl condition versus those expressing GAD65 and GAD67 with their N and C-terminal domains exchanged; asterisks, cell soma. C, Representative images (i) of neurons expressing HA-tagged GAD65, WT versus mutant versions (Ala^30/45, ΔN^1-23, or Ala^30/45/ΔN^1-23), immunolabeled for MAP2, Synapsin, and vGAT; yellow arrows point at cell soma; boxed regions are enlarged to the right demonstrating colocalization between two or more channels (as labeled). The bar graphs (ii) represent (left to right) GAD65 signal at cell soma, and its colocalization with Synapsin or vGAT. D, Example images of neurons expressing GAD65 variants (as indicated), immunostained for DAPI and GABA. E, Example waveforms of GABAergic sIPSCs (left; inset, boxed area expanded below) and their average parameters (right; amplitude and frequency), recorded from Ctrl versus vGAT + GAD65-Ala^30/45/ΔN^1-23 conditions. F, Representative traces (left; three consecutive pulses overlayed) of GABAergic IPSC evoked by presynaptic stimulations, their average peak amplitudes and total charge transfers (right), when recorded from Day 50 to 56 human neurons in Ctrl versus vGAT + GAD65-Ala^30/45/ΔN^1-23 conditions; black arrow points at stimulus artifacts. G, Example images (i) and average colocalization between HA and Synapsin signals (ii), in Ctrl condition versus neurons expressing HA-tagged GAD67, either WT or ΔN^term versions; boxed areas are enlarged to the right; arrows point at human neurons (yellow) or mouse glia (white) identified by their distinct heterochromatinization patterns in DAPI channel. Notice a near-complete loss of presynaptic accumulation for ΔN^term compared with WT. H, Human neurons from Ctrl versus vGAT + GAD67 ΔN^term conditions coimmunolabeled for MAP2 and GABA; asterisks, cell bodies; the boxed area from GAD67 ΔN^term magnified to the right with channels split and shown together with coimaged vGAT signal; arrowheads, GABA signals with (yellow) or without (magenta) vGAT. I, Same as panel E, except for Ctrl versus vGAT + GAD67 ΔN^term conditions. J, Same as panel F, except for Ctrl versus vGAT + GAD67 ΔN^term conditions. Summary graphs are displayed as means ± SEM, containing total number of images analyzed or cells patched/independent replicates. The data points are provided as color-matched open circles. All statistical evaluations were performed by two-tailed, unpaired, Student's t test, with *** p < 0.001 and ns, not significant, p > 0.05.

We analyzed these mutant GAD constructs in GAD-deficient Ngn2–induced human neurons, which display elaborate morphology, containing highly branched axonal arbors extending up to ≈10 mm in cumulative length (Chanda et al., 2019). When transduced for 50–56 d (Fig. 4A), both chimeric GADs 65N-67C and 67N-65C continued to synthesize adequate amounts of GABA that diffused throughout the intracellular space, despite their swapped domains (Fig. 7B). The ΔN^1-23 mutation caused a significant reduction in GAD65 levels, as less proteins were detected at cell bodies and Synapsin-labeled presynapses, including vGAT terminals; the Ala^30/45 mutation in GAD65 triggered its retention at the soma and, thus, decreased its association with vGAT, whereas the GAD65-Ala^30/45/ΔN^1-23 mutation exhibited a combination of both phenotypes (Fig. 7C). However, despite the impaired synaptic trafficking, all mutant constructs produced highly diffused GABA signals (Fig. 7D). Furthermore, when compared with GAD-deficient control neurons, the GAD65-Ala^30/45/ΔN^1-23 condition successfully elicited robust synaptic response, characterized by high-frequency sIPSCs and evoked IPSCs of large amplitude (Fig. 7E,F).

The loss of entire N-terminus in GAD67 also led to complete disruption of its presynaptic targeting relative to the WT version and resulted in its uniform distribution throughout the cytosol (Fig. 7G). Yet again, the GAD67-ΔN^term mutant continued to maintain soluble GABA production, which often colocalized with vGAT-positive terminals (Fig. 7H), and similarly triggered both spontaneous and evoked IPSCs, corroborating highly reliable GABAergic neurotransmission (Fig. 7I,J). Hence, the presynaptic availability of GABA molecules for vesicular loading and release appeared to be virtually insensitive to the subcellular localization of GAD isoforms.

A nuclear production of GABA can sufficiently activate its presynaptic terminals

To convincingly restrict the GADs at a subcellular compartment distant from all axonal terminals, we attached two distinct NLS, either from simian virus 40 (SV40) or c-Myc transcription factor, to the C-termini of both GAD65 and GAD67 (i.e., NLS^SV40 or NLS^cMyc, respectively; Fig. 8A). When transfected into HEK 293 T-cells for a brief period, both WT and ΔN^term GADs commonly remained in cytoplasm and were completely excluded from DAPI-positive nuclear boundaries, whereas all NLS-tagged GAD variants favorably resided within the nucleus, thus confirming a major disruption of their regular trafficking patterns (Fig. 8B,C).

Figure 8.

Normal GABAergic synaptic activities induced by nuclear-localized GAD variants. A, Schematic diagrams of chimeric NLS-tagged GAD67 constructs, generated and analyzed in this current study. B,C, Experimental strategy (B) for validating the distribution of NLS (SV40 and cMyc)-tagged GAD mutants in HEK 293T cells; example images (C, top) and signal intensities (C, bottom) across the dotted lines, when cells were labeled for both GADs and DAPI; arrowheads, transfected (yellow) versus nontransfected (white) cells. D, Representative images of human neurons transduced with GAD65 (top panels) or GAD67 (bottom panels), WT versions versus when fused with NLS of SV40 or cMyc, stained for GADs, Synapsin, and DAPI. All channels were merged to the right, and boxed regions were expanded to visualize synaptic clusters. Arrowheads, nuclear signals from neurons (yellow) or cocultured astrocytes (white); somatic signals (cyan) particularly for GAD67 WT. E, Bar graphs illustrate cellular distribution of GAD65 (left) and GAD67 (right) variants (i.e., WT vs NLSSV40 or NLScMyc) at DAPI-labeled nuclei (top), or Synapsin-positive presynaptic terminals (bottom) of human neurons. F, Neurons expressing GAD65 or GAD67 variants (as indicated) and coimmunostained for MAP2 and GABA. Note the diffused GABA signal evenly distributed at MAP2-negative axons (arrowheads); asterisks, cell soma. G,H, Sample traces of GABAergic sIPSCs (panel G) and their average parameters (panel H, amplitude and frequency), recorded from neurons expressing vGAT + GAD65 or GAD67 variants (WT, NLSSV40, or NLScMyc). I, Example waveforms (left; three consecutive pulses overlayed) of GABAergic IPSCs evoked by presynaptic stimulations and their average peak amplitudes (right), recorded from Day 50–56 human neurons expressing vGAT + GAD65 (top) or GAD67 (bottom) variants (WT, NLSSV40, or NLScMyc); black arrows, stimulus artifacts. J, Representative waveforms (left) of GABAergic IPSCs evoked by high-frequency train stimulations (i.e., 20 Hz; black arrowheads) and normalized peak amplitudes of individual pulses (right), as recorded from human neurons transduced with WT versus NLS-tagged (SV40 and cMyc) GAD65 and GAD67 isoforms, when combined with vGAT. Summary plots are shown as means ± SEM, with the number of field-of-views analyzed (for imaging) or cells patched (for electrophysiology)/independent experimental batches. The individual data points from each group are provided as color-coded open circles. Statistical weights between experimental conditions were calculated by either two-tailed, unpaired, Student's t test (bar graphs; panels E, H, I) or one-way ANOVA (multipoint comparisons; panel J), with *** p < 0.001; ** p < 0.01; * p < 0.05; ns, not significant, p > 0.05.

When expressed in human neurons, we continued to notice that both NLS sequences confined the GAD variants primarily inside the DAPI-positive nuclei, with most prominent phenotypes observed for NLS^cMyc over NLS^SV40 and GAD67 over GAD65 (Fig. 8D,E). This artificial retention of GADs inside nucleus was inversely correlated with their reduced availability at all other subcellular compartments, which virtually eliminated the majority of synaptic signals, especially for GAD67 (Fig. 8D,E). Irrespective of that, every GAD-NLS construct produced substantial quantity of GABA that diffused out in the soma, dendrites, and axonal branches (Fig. 8F). Therefore, both GADs can operate remotely from nonsynaptic regions to maintain intracellular GABA levels.

To further check if this nonsynaptic GABA synthesis by recombinant GADs at distant sources can possibly allow its vesicular packaging by presynaptic vGAT and drive release, we next conducted electrophysiological recordings from these neurons. Once again, we detected robust and recurrent GABAergic sIPSCs for all GAD-NLS constructs with event frequency, amplitude, and kinetic properties similar to WT versions (Fig. 8G,H; Fig. S8A,B). Moreover, these nuclear-localized GADs effectively generated sizable GABAergic evoked IPSCs, with both amplitude and coefficients of variation equivalent to WT GADs, assuring very similar probability of synaptic vesicle release (Fig. 8I; Fig. S8C). Finally, when subjected to a high-frequency and repetitive electrical stimulation, all GAD-NLS mutants also achieved highly reproducible evoked IPSC trains with a noticeable short-term depression, comparable to their corresponding WT counterparts (Fig. 8J). Hence, our experimental results collectively provide strong evidence that global diffusion of GABA, even from distal sources, can play a key supporting role in maintaining the basal GABAergic synaptic transmission.

Discussion

A neuron can form hundreds of en passant synapses along the length of its axons, which specialize in vesicular release of neurotransmitters. Existing models suggest that neurotransmitter homeostasis at the presynapse relies on three major processes, i.e., local synthesis, direct reuptake, and axonal diffusion (Marx et al., 2015), and their relative involvements may differ depending on the synapse types analyzed. Importantly, while the local synthesis model predicts that most small-molecule transmitters are metabolized in situ at synaptic terminals by a local pool of enzymes, their cellular distribution does not always manifest a universal trend consistent with this expectation. Moreover, even though synaptic vesicle endocytosis and plasma-membrane transporters can retrieve a fraction of transmitters from the synaptic clefts, neurons are in constant need to biosynthesize these chemicals for reliable trans-synaptic communication. Under these constraints, diffusion of transmitters from distal sources may provide a vital supporting route for their presynaptic deliveries, a possibility that warrants further investigation. Here, we aimed to understand this critical issue by monitoring the synaptic association of glutamate and GABA producing enzymes and impacts of GABA diffusion on vesicular release. Our efforts yield several important conclusions.

We observed that the major glutamate-producing enzyme, GLS, does not efficiently localize at glutamatergic presynapses in multiple brain regions or neuronal cultures but rather associates with perisomatic mitochondria (Figs. 1 and 3; Figs. S1–S3). The advantages of maintaining a local pool of GLS have been linked with utilization of astrocytic glutamine at tripartite synapses, especially to sustain a repeated glutamate release (Tani et al., 2014; Cheung et al., 2022), i.e., during high-frequency activities (Bacci et al., 2002; Tani et al., 2010). Nevertheless, the influence of glutamine–glutamate cycles on basal neurotransmission has been reported to be rather slow, as acute potentiation or inhibition of this pathway does not immediately affect synaptic strength (Kam and Nicoll, 2007). Importantly, although it is estimated that ≈70% of synaptic glutamate pool is derived from the glutamate–glutamine cycles (Kvamme, 1998; Lieth et al., 2001; Sibson et al., 2001), astrocytic glutamine supply may not be strictly confined at synapses. Most of this enzymatic action can certainly occur at perisomatic regions because both the (1) GLS for glutamate synthesis (Figs. 1 and 3) and (2) neuronal System A transporter (SNAT1) for glutamine import are preferentially distributed at neuronal cell bodies and proximal dendrites (Mackenzie et al., 2003; Yamada et al., 2019). We evidenced a similarly nonsynaptic and predominantly perisomatic distribution also for GDH, another major enzyme involved in glutamate biosynthesis (Fig. 1; Figs. S1–S3). Thus, a bulk of glutamate production in neurons may not happen locally but at places away from axon terminals.

As opposed to glutamatergic presynapses, the GABAergic synaptic terminals contained high levels of GAD enzymes (Figs. 1–3; Fig. S2). Hence, the production of two principal neurotransmitters may not follow a general trend, since they are mostly manufactured at different cellular compartments. Notably, the GAD isoforms differed significantly in their endogenous expression and vGAT association at various strata and sublayers within mouse hippocampus and dentate gyrus, highlighting their enrichment at specific subregions of the brain (Fig. 2; Fig. S2). Even among the two GAD isoforms, GAD65 was more favorably recruited at presynapses over GAD67, as the latter also showed strong somatic signals, including when tested individually in human neurons (Figs. 1 and 4; Fig. S1). The trafficking profiles of human GADs generally phenocopied those reported by earlier studies in rodent systems (Kaufman et al., 1991; Dirkx et al., 1995; Kanaani et al., 1999, 2002, 2010; Hsu et al., 2000), validating that their molecular properties are shared across multiple species and operate similarly in human cellular environment. Synaptic abundance of GADs but not GLS also made us curious about the benefit of local transmitter sources, at least for GABA if not glutamate.

Surprisingly, the exogenously expressed GAD65 and GAD67 isoforms also triggered recurrent GABAergic sIPSCs with very similar event properties, as well as evoked IPSCs with comparable amplitude and short-term synaptic plasticity, suggesting an equivalent release probability for both cases (Fig. 5; Figs. S4–S6). Notably, the GAD67 isoform is known to exhibit a superior enzymatic function compared with GAD65, as the latter more frequently undergoes transamination reactions to produce an inactive apoGAD form and needs to reassociate with pyridoxal 5′-phosphate cofactor that restore a functionally active holoGAD (Battaglioli et al., 2003; Fenalti et al., 2007). Furthermore, the endogenous expression of GAD65 is reported to be much lower than GAD67 in rodent brain (Esclapez et al., 1994), making it difficult to normalize their relative contributions in GABA production and/or release. Our data implied that, in the absence of GAD67, the GAD65 isoform alone is catalytically active and fully capable of synthesizing sufficient GABA, especially when expressed to equivalent degrees for prolonged period of time (Fig. 5). These findings complement the results obtained from GAD65 null mice, which indeed develop defects in GABAergic synaptic transmission (Kash et al., 1997; Stork et al., 2000; Song et al., 2011; Lange et al., 2014), despite its moderate to low expression in the brain (Esclapez et al., 1994).

In agreement, we subsequently determined that a contrasting subcellular distribution of GAD65 versus GAD67 does not ultimately lead to any major functional differences between them, in terms of their abilities to produce GABA that diffuses freely throughout the cytosol, either in vivo or in vitro, for both murine and human neurons, if expressed individually or combined together (Fig. 5, Fig. S7). Of note, both anterograde and retrograde movements of large organelles, proteins, or mRNAs along the axonal length are relatively slow, highly constrained, and energetically demanding, as active transports on the microtubule tracks by Kinesin versus Dynein motors are often deemed necessary (Guedes-Dias and Holzbaur, 2019). These molecular machineries are also required to carry the large dense core vesicles containing neuropeptides from soma to axon terminals (van den Pol, 2012). However, such limits on the soluble and small-molecule neurotransmitters, e.g., glutamate or GABA, are mostly unclear. Our in silico modeling suggested that, with a substantial reserve pool at physiological concentrations, the intracellular space can effectively serve as a GABA reservoir, since GABA molecules can rapidly diffuse away from its sources and arrive at distances hundreds of micrometers away within only a few seconds to support presynaptic activity (Fig. 6). In accordance, we detected the presence of ample GABA signals at vGAT-labeled presynaptic structures, for both GADs, irrespective of their local enrichment (Fig. 5).

It has been proposed that heterodimerization with GAD65 might facilitate the synaptic trafficking of GAD67 (Dirkx et al., 1995). However, we noticed that GAD67 can partially localize to synapses even in the absence of GAD65, via its N-terminal residues (Figs. 4 and 7), in agreement with other studies (Kanaani et al., 1999, 2010). Nevertheless, a complete loss of its N-terminal domain did not alter GABA production, intracellular diffusion, or vesicular release from the presynapse (Fig. 7). An identical result was obtained when the N-terminal trafficking signals were deleted from GAD65 or the catalytic domains were swapped between the two GADs (Fig. 7). In addition, we also noticed that the GAD variants fused to NLSs, even though artificially retained inside nucleus, continued to synthesize a considerable amount of GABA that successfully diffused out in the soma and axonal processes through nuclear pores (molecular-weight limit ≈60 kDa for nongated passive diffusion; Nigg, 1997; Görlich, 1998), reached axon terminals, was packaged into presynaptic vesicles, and was released both spontaneously and in an AP-dependent manner to stimulate the postsynaptic GABA_ARs, generating sizable GABAergic IPSCs (Fig. 8, Fig. S8). Finally, an earlier study has demonstrated that, instead of a GAD-dependent GABA synthesis, even an inclusion of GABA in the intracellular recording pipette placed at the soma can partially maintain presynaptic vesicle recycling (Wang et al., 2013). Therefore, a local, presynaptic production of small-molecule neurotransmitters, e.g., glutamate or GABA, may not be compulsory for their basal vesicular release.

Since Ngn2-induced human neurons express negligible amounts of GAD65 or GAD67, this system enabled us to execute a systematic structure-function analysis with the exogenous GADs, without any interference from their endogenous versions. In the future, it will be important to reexamine whether GABA diffusion can act as a sustainable model for presynaptic GABA supply, also for in vivo environment and in a cell-subtype independent manner. Furthermore, in parallel with these GAD isoforms, GABA reuptake via nonvesicular GABA transporter GAT1 can function as a secondary source for intracellular GABA storage and release, as illustrated for several neuronal subtypes from multiple brain regions (Minelli et al., 1995; Conti et al., 2011; Scimemi, 2014; Melani and Tritsch, 2022). Because the Ngn2 neurons reportedly express low amount of GAT1 (Yang et al., 2017), it allowed us to focus on GAD mechanisms, without significant contributions from GABA uptake pathways. Thus, presynaptic GABA homeostasis can depend on additional factors that were not directly addressed in this study. While GABA diffuses efficiently throughout the extensive processes of Ngn2 neurons (Chanda et al., 2019), the time to establish an equilibrium gradient along axons will depend on their lengths, which can differ considerably both between and within various GABAergic interneurons or projection neuron subtypes (Buzsáki et al., 2007; Tamamaki and Tomioka, 2010; Booker and Vida, 2018). Moreover, the overall contributions of GABA diffusion to GABAergic neurotransmission may vary substantially depending on the number and distribution of functional synapses, their spontaneous activity, AP firing rate, and release parameters that influence GABA consumption.