Abstract
The hypothalamic paraventricular nucleus (PVN) is strongly inhibited by γ-aminobutyric acid (GABA) from the surrounding peri-nuclear zone (PNZ). Because glutamate mediates fast excitatory transmission and is substrate for GABA synthesis, we tested its capacity to dynamically strengthen GABA inhibition. In PVN slices from male mice, bath glutamate applied during ionotropic glutamate receptor blockade increased PNZ-evoked inhibitory postsynaptic currents (eIPSCs) without affecting GABA-A receptor agonist currents or single-channel conductance, implicating a presynaptic mechanism(s). Consistent with this interpretation, bath glutamate failed to strengthen IPSCs during pharmacological saturation of GABA-A receptors. Presynaptic analyses revealed that glutamate did not affect paired-pulse ratio, peak eIPSC variability, GABA vesicle recycling speed, or readily releasable pool (RRP) size. Notably, glutamate–GABA strengthening (GGS) was unaffected by metabotropic glutamate receptor blockade and graded external Ca2+ when normalized to baseline amplitude. GGS was prevented by pan- but not glial-specific inhibition of glutamate uptake and by inhibition of glutamic acid decarboxylase (GAD), indicating reliance on glutamate uptake by neuronal excitatory amino acid transporter 3 (EAAT3) and enzymatic conversion of glutamate to GABA. EAAT3 immunoreactivity was strongly localized to presumptive PVN GABA terminals. High bath K+ also induced GGS, which was prevented by glutamate vesicle depletion, indicating that synaptic glutamate release strengthens PVN GABA inhibition. GGS suppressed PVN cell firing, indicating its functional significance. In sum, PVN GGS buffers neuronal excitation by apparent “over-filling” of vesicles with GABA synthesized from synaptically released glutamate. We posit that GGS protects against sustained PVN excitation and excitotoxicity while potentially aiding stress adaptation and habituation.
Significance Statement
Here we found that PVN GABA inhibition, which dominates glutamatergic excitation in the unstressed state, is rapidly strengthened by increased synaptic glutamate release. This negative-feedback homeostatic response reflects an apparent quantal mechanism with pharmacological inhibition studies implicating neuronal glutamate uptake by EAAT3 and de novo enzymatic synthesis of GABA, culminating in apparent GABA vesicle “over-filling.” This glutamate–GABA strengthening response restrains PVN neuronal discharge responses with potential to mitigate behavioral, endocrine, and autonomic responses to glutamatergic excitation during stress.
Introduction
The hypothalamic paraventricular nucleus (PVN) regulates behavioral, endocrine, and autonomic responses to homeostatic (interoceptive) and perceived (exteroceptive) stresses (Jankord and Herman, 2008; Wamsteeker Cusulin and Bains, 2015; Lamotte et al., 2021). As elsewhere in the brain (Foster and Kemp, 2006), firing of PVN neurons reflects the dynamic balance of synaptic excitation and inhibition (Herman et al., 2004; Bardgett et al., 2014; Wamsteeker Cusulin and Bains, 2015; Li and Pan, 2017). Interestingly, PVN neuronal responses to glutamate in the unstressed animal are typically modest compared with those elicited by GABA-A receptor blockade (Cole and Sawchenko, 2002; Chen et al., 2003; Herman et al., 2004; Li et al., 2006; Bardgett et al., 2014), indicating that GABA inhibition is dominant at rest and strongly blunts glutamatergic excitation. Consistent with this, PVN neurons exhibit little or no resting discharge in vivo (Lovick and Coote, 1988; Chen and Toney, 2003, 2010), consistent with the dominance of GABA inhibition (Decavel and Van den Pol, 1990; Herman et al., 2002) and with evidence that GABAergic axo-somatic synapses are plentiful in the PVN (Miklos and Kovacs, 2002). Hence, driving PVN discharge appears to require that mechanisms establishing and defending GABA dominance be overpowered or otherwise subverted (Verkuyl et al., 2005; Wamsteeker and Bains, 2010). Despite this, local mechanisms governing the strength of GABA tone are incompletely understood. This is a critical knowledge gap because diminished GABAergic inhibition is a key feature of persistent pathogenic PVN activation (Li et al., 2006; Park et al., 2009; Wamsteeker and Bains, 2010; Qin et al., 2018; Lamotte et al., 2021).
Although studies of several brain regions have shown that ambient glutamate can strengthen phasic GABA inhibition through multiple mechanisms (Sepkuty et al., 2002; Mathews and Diamond, 2003; Erickson et al., 2006; Stafford et al., 2010; Ishibashi et al., 2013), studies in PVN have primarily emphasized the role of presynaptic mGluR (Inoue et al., 2013; Colmers and Bains, 2018). Hence, mechanisms governing the balance of glutamate and GABA signaling in the PVN remain relatively unexplored. Not surprisingly, glutamatergic inputs are a major driver of PVN activation (Herman et al., 2002, 2004; Bardgett et al., 2014; Li and Pan, 2017), which prompted us to investigate coupling of PVN glutamatergic excitation with increased GABAergic inhibition and to seek insight into local mechanisms that dynamically govern PVN output.
One possible coupling mechanism relies on ionotropic glutamate receptor activation triggering GABA-A receptor insertion into the postsynaptic membrane (for review see Chapman et al., 2022). Such a mechanism has not been described in the PVN. Alternatively, glutamate could increase GABA uptake and vesicle recycling through upregulation of presynaptic GABA transporters (Rowley et al., 2012; Pandit et al., 2015; Schitine et al., 2015; Ryan et al., 2021). However, elevated glutamate release is reported to decrease, not increase, GABA uptake (Schitine et al., 2015), and GABA uptake specifically in the PVN is reported to play a modest role in setting the strength of GABA inhibition (Park et al., 2009; Pandit et al., 2015).
The above findings support a model of presynaptic PVN GABA transmission that relies more on the rate of GABA synthesis (Lee et al., 2019) than recycling (Ryan et al., 2021). This resembles the hippocampus where synaptic glutamate release is reported to dynamically strengthen quantal GABA transmission (Mathews and Diamond, 2003; Stafford et al., 2010). Here, increasing ambient glutamate in the PVN, either by bath application or increased synaptic release, strengthened GABA transmission to an extent sufficient to restrain neuronal firing. We posit that the glutamate–GABA strengthening (GGS) mechanism autoregulates PVN output by coupling GABA inhibition to glutamate excitation, protecting against neuronal hyperactivation and exaggerated stress responses.
Materials and Methods
Animals
Ninety-five adult male mice (25–30 g) were used in this study. Of these, 90 were wild-type C57Bl/6J and five expressed enhanced green fluorescent protein (EGFP) driven by the vesicular GABA transporter promoter (VGAT-EGFP). The latter, generated by crossing hemizygous VGAT-Cre (#028862) and Ai6 (#007906) mice (Jackson Labs), were used in initial studies to assess the distribution of GABAergic neurons in the PNZ that surrounds the PVN, thereby facilitating accurate placement of stimulating electrodes. Regardless of strain, mice were group-housed (2–4/cage) in a temperature-controlled (22–24°C) vivarium on a 14/10 h light/dark cycle. Standard chow (Harlan Teklad) was available ad libitum. All experimental procedures conformed to the Guide for the Care and Use of Laboratory Animals and were approved by the Animal Care and Use Committee of the University of Texas Health San Antonio.
Brain slice preparations
Mice were anesthetized (3% isoflurane in O2), and coronal brain slices (300 µm) containing the PVN were cut with a vibratome (Leica VT1000S) while submerged in ice-cold oxygenated solution consisting of (in mM): 140 choline chloride, 2 KCl, 1 CaCl2, 1 MgCl2, 10 D-glucose, 10 HEPES, and 0.4 ascorbic acid, pH 7.4. Prior to (≥60 min) and during recording, slices were maintained at room temperature (RT = ∼22°C) in O2 equilibrated artificial cerebrospinal fluid (aCSF) consisting of the following (in mM): 140 NaCl, 2.5 KCl, 2 CaCl2, 1 MgCl2, 10 D-glucose, 10 HEPES, and 0.4 ascorbic acid, pH 7.4 (Cato and Toney, 2005; Chen and Toney, 2009; Chen et al., 2010). In some experiments, aCSF Ca2+ was lowered (1 mM) and raised (4 mM) by changing the amount of CaCl2 added.
Electrophysiology
Slices were transferred to a recording chamber perfused (1–2 ml/min) with RT aCSF, and whole-cell recordings were made with 4–6 MΩ borosilicate glass electrodes. Regardless of the experiment, only one neuron was recorded per brain slice (2–3 PVN slices per mouse). Except where noted, intracellular solution consisted of the following (in mM): 97 K-gluconate, 48 KCl, 1 MgCl2, 1 EGTA, 10 HEPES, 2 Mg-ATP, 0.5 Na-GTP, and 5 QX-314, pH 7.2. Recorded signals were acquired with a MultiClamp 700B amplifier and digitized with a Digidata 1550B (Molecular Devices) at 10–20 kHz and low-pass filtered at 1–2 kHz. Access resistance was continuously monitored, and data were discarded when a change of >10% occurred.
Voltage-clamp recording
Inhibitory postsynaptic currents (IPSCs) were isolated by blocking α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) and N-methyl-D-aspartate (NMDA) ionotropic glutamate receptors (iGluR) with 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX, 10 µM) and D-(-)-2-amino-5-phosphono-pentanoic acid (APV, 50 µM; Tocris), respectively, and were recorded from a holding potential of −70 mV (ECl = −28 mV) such that GABA-A receptor-mediated currents were directed inward. Evoked IPSCs (eIPSCs) were recorded every 15 s by delivering 0.5 ms single or paired (20 Hz) stimuli through a concentric (250 µm) bipolar electrode positioned in the GABA-enriched PNZ surrounding the PVN (Fig. 1A). Spontaneous excitatory postsynaptic currents (sEPSCs), also directed inward, were recorded (Vhold = −70 mV) with GABA-A receptors blocked with picrotoxin (PTX, 100 µM). In postsynaptic GABA-A receptor saturation experiments, the GABA-A/C receptor antagonist 3-(α-L-arabinopyranosyloxy)-2-(3,4-dihydroxyphenyl)-5,7-dihydroxy-4H-1-benzopyran-4-one (quercetin, 30 µM, Sigma-Aldrich; Goutman and Calvo, 2004; Fan et al., 2018) was bath applied starting 10 minutes prior to bath perfusion with aCSF containing high (4 mM) Ca2+ or L-glutamate (100 µM). Group I, II, and III metabotropic glutamate receptors were blocked with a cocktail of (RS)-α-cyclopropyl-4-phosphonophenylglycine (CPPG, 200 µM; Hello Bio) and (RS)-α-methyl-4-carboxyphenylglycine (MCPG, 500 µM; Hello Bio). Excitatory amino acid transporters (EAATs) were globally inhibited with DL-threo-b-benzyloxyaspartic acid (TBOA, 100 µM; Tocris). Glia-specific transporters EAAT1 (GLAST) and EAAT2 (GLT-1) were inhibited with 2-amino-5,6,7,8-tetrahydro-4-(4-methoxyphenyl)-7-(naphthalen-1-yl)-5-oxo-4H-chromene-3-carbonitrile (UCPH101, 10 µM; Tocris) and dihydrokainate (DHK, 250 µM; Tocris), respectively. Glutamic acid decarboxylase was inhibited with allylglycine (AG, 2 mM; Sigma-Aldrich). High potassium aCSF was made by replacing 10 mM NaCl with KCl. Synaptic glutamate release was inhibited by bath perfusion of the vesicular glutamate transporter (VGLUT) inhibitor Rose Bengal (RB, 30 µM; Tocris; Bole and Ueda, 2005).
Bath glutamate strengthens PVN GABA inhibition. A, Experimental protocol, (a) baseline, (b) post glutamate (top). A coronal PVN slice from a VGAT-EGFP mouse illustrates the distribution of GABA neurons (green) in the surrounding PNZ. Electrical stimuli were applied to PNZ every 15 s and evoked IPSCs (eIPSCs) were recorded from PVN neurons (bottom). f, fornix; PNZ, peri-nuclear zone; PVN, hypothalamic paraventricular nucleus; 3V, 3rd cerebral ventricle. B, Representative responses of a parvocellular type III neuron showing the average of eight consecutive eIPSCs at (a) baseline, (b) 10 min after bath glutamate (100 µM, 10 min), and (c) in the presence of the GABA-A receptor antagonist picrotoxin (PTX, 100 µM; top). Time course of the GABAergic eIPSC amplitude response of this neuron to bath glutamate (bottom). C–E, Traces illustrating the current-clamp protocol used to elicit low-threshold Ca2+ spikes that are absent from type 1 (C), present in type II (D), and augmenting in type III (E) PVN neurons (top). Bath glutamate (100 µM, 10 min) increased the amplitude of IPSCs evoked from the ipsilateral PNZ in each PVN cell type (bottom). F, Summary analysis of type I, II, and III neurons averaged in 2 min bins (mean ± SD) showing that normalized eIPSC amplitude increased within 10 min of glutamate washout (n = 19). *p = 0.0107, **p = 0.0020. G, Control experiments showing no effect of time (p = 0.2106; n = 8). H, Representative traces of mIPSCs recorded in the absence of TTX (Table 3) recorded during iGluR blockade (CNQX, 10 µM; APV, 50 µM), before (baseline, gray), and 10 min after bath glutamate (100 µM, 10 min; post Glu, green) at low (top) and high (bottom) time resolution. I–L, Summary data of mIPSC amplitude (I, *p = 0.0256), frequency (J, p = 0.8951), 10–90% rise time (K, p = 0.4197), and 90–10% decay time (L, p = 0.2810) before (baseline, gray) and after bath Glu (post Glu, green; two-tailed paired t tests; n = 8 cells).
Focal drug application
Using a pico-spritzer (Dagan), the GABA-A receptor agonist isoguvacine (ISO, 100 µM; Tocris) was delivered from a glass pipette (>4 MΩ) to the slice surface (Cato and Toney, 2005; Pakhomov et al., 2006) at 20 s intervals to examine effects of bath glutamate on postsynaptic GABA-A receptor currents. L-Glutamate (100 µM, Sigma-Aldrich) was puff-applied in duplicate before and after bath Rose Bengal to affirm no effect of VGLUT inhibition on postsynaptic membrane current responses (IGlu).
Nonstationary fluctuation analysis
Peak-scaled decay-phase nonstationary fluctuation analysis (De Koninck and Mody, 1994; Traynelis and Jaramillo, 1998; Nusser et al., 2001) was performed to estimate the single-channel conductance (γ) and the relative number (N) of GABA-A receptors active at the peak of IPSCs. Although analyzed IPSCs were recorded during iGluR blockade and the absence of tetrodotoxin (TTX), the amplitude distribution of these recorded events was unchanged by bath TTX (1 µM, Alomone Labs; Table 3), indicating that, as expected, action potential-dependent IPSC activity was absent from our slice preparations. Therefore, analyzed sIPSCs were, in fact, miniature IPSCs (mIPSCs), making them suitable for fluctuation analysis (Silver et al., 1998; Traynelis and Jaramillo, 1998; Scheuss and Neher, 2001; van Huijstee and Kessels, 2020). Additionally, IPSCs subjected to fluctuation analysis fulfilled established criteria for estimating values of γ and N, including small amplitude and minimal variance near the peak (De Koninck and Mody, 1994; Nusser et al., 2001).
To isolate current amplitude fluctuations reflecting stochastic channel closing at baseline and post glutamate, we scaled the peak of individual mIPSCs to that of the mean waveform. For each recorded neuron, decay-phase amplitude fluctuations around the mean current were separated into 30 bins of equal amplitude decrement from the peak to 10% of the mean current. The latter was performed to prevent/minimize data contamination with background noise. The average binned mIPSC variance, calculated at baseline and 10 min after bath glutamate washout, was plotted versus the corresponding mean amplitude of analyzed events. The peak-scaled variance–mean current relation is given by σ2 = iI − I2 / N, where σ2 = peak-scaled variance, I = mean current, i = weighted mean single-channel current, and N = number of active channels at the peak of mIPSCs. The slope of the line extrapolated to the zero current value (i = σ2 / Ī) of the resulting parabolic function yields the predicted single-channel current (i), while the vertex is proportional to the number of channels (N) active at the peak of mIPSCs. Single-channel conductance (γ) values were calculated as the chord conductance γ = i / (Ehold − EGABA) with Ehold = −70 mV and EGABA = −28 mV.
Coefficient of variation analysis
The contribution of increased GABA vesicle release probability and number of vesicle release sites (n) in modulating effects of glutamate on GABAergic IPSC amplitude was assessed by analyzing the coefficient of variation [CV, standard deviation (SD) mean−1] of IPSCs evoked during high-frequency (20 Hz) PNZ stimulation consisting of trains of 10 pulses (0.1 ms duration) repeated 10 times at a 30 s interval before and 5 min after bath glutamate (100 µM, 10 min) washout. Data were plotted as normalized values of 1/CV2 relative to normalized mean eIPSC peak amplitude. In accordance with the binomial release model of synaptic transmission, we assumed that n was unchanged over the time course of our experiments (Brock et al., 2020; van Huijstee and Kessels, 2020). Changes in the relationship between 1/CV2 and mean eIPSC amplitude were therefore interpreted to reflect proportional changes in vesicle release probability (Pr).
Estimating GABA vesicle pool size and dynamics
At baseline and again 5 min post-glutamate washout, high-frequency (20 Hz) stimulus trains consisting of 20 pulses (0.1 ms duration) were delivered to the PNZ and repeated for 10 trials at a 30 s interval to depress eIPSC amplitude. The latter primarily reflects presynaptic vesicle depletion (Gielen et al., 2020), though a contribution from calcium channel inactivation cannot be ruled out (Regehr, 2012). The time course of synaptic depression was quantified by separately averaging IPSCs from the 10 stimulus trains delivered before and after bath glutamate and normalizing each to the corresponding amplitude of eIPSC #1. To assess the effects of glutamate on the size of the RRP of GABA vesicles, we plotted the cumulative amplitude (pA) of normalized eIPSCs before and after bath glutamate washout and iteratively applied least squares regression to determine the linear portion of each curve (stimuli #7–20). Extrapolating regression lines to the y-axis indexed RRP size prior to synaptic depression under baseline and post-glutamate conditions. Immediately thereafter the effect of glutamate on recovery from depression/vesicular depletion was assessed by plotting, as a percent of the average pre-depression baseline, the amplitude of PNZ eIPSCs at progressively longer inter-stimulus intervals.
Current-clamp recording
Effects of bath glutamate on PVN neuronal firing were determined in slices perfused with aCSF using electrodes filled with the following (in mM): 135 K-gluconate, 10 KCl, 1 MgCl2, 1 EGTA, 10 HEPES, 2 Mg-ATP, and 0.5 Na-GTP, pH 7.2. Recording began ∼5 min after whole-cell access. Estimated liquid junction potential (−15.8 mV) was corrected offline. Discharge frequency (Hz), quantified once stable (∼2–3 min), was analyzed from rate meter records generated at baseline, during glutamate (10 min), and 10 min post-glutamate. Comparison was to time control recordings and recordings made with positive current injected to mimic glutamate excitation. Contributions of GABA-A receptors and glutamate uptake to effects of glutamate on cell firing were separately determined by exposing slices to glutamate in the continuous presence of the GABA-A receptor antagonist picrotoxin (PTX, 100 µM) and the pan-specific glutamate uptake inhibitor TBOA (100 µM).
Immunohistochemistry
Distribution of PVN EAAT3 was mapped to presumptive local GABA terminals identified by immunoreactivity (ir) to the 65 kD isoform of glutamic acid decarboxylase (GAD65) and the presynaptic terminal marker synaptophysin (SYP). Anesthetized (3% isoflurane) mice (n = 6) underwent transcardiac perfusion exsanguination with phosphate-buffered saline (PBS) followed by fixation with 4% paraformaldehyde (PFA) in 0.2 M phosphate buffer. Brains were removed, post fixed (4 h, RT), and cryoprotected in 30% sucrose–PBS (48 h, 4°C). PVN was cut in 30-μm-thick coronal sections (Leica SM 2000R) that were stored (−20°C) in polyvinyl-pyrrolidone (PVP) cryoprotectant. Brain tissue from EAAT3 (SLC1a1) gene knock-out mice was generously provided by Dr. Zhiyi Zuo, University of Virginia, and was processed identically. Prior to immunostaining, sections were rinsed (PBS), incubated (30 min) in 0.5% sodium borohydride–PBS to reduce autofluorescent aldehydes, and then underwent epitope retrieval (30 min, 80°C) in sodium citrate buffer (0.01 M, pH 8.6). Cooled sections were rinsed and incubated (30 min) in 5% normal donkey serum (NDS) blocking buffer containing 0.2% Tween 20. Rinsed sections incubated in blocking buffer without Tween 20 (1 h) were transferred to blocking buffer (4°C) with or without monoclonal mouse anti-synaptophysin 1° antibody (1:100; ab8049; Abcam) (18 h, RT) and then to donkey anti-mouse 2° antibody conjugated to Alexa Fluor (AF) 405 (1:250; ab175658; Abcam; 2 h, RT). Rinsed sections were blocked in 10% normal goat serum (NGS, 1 h, RT) and then incubated (18 h, 4°C) with or without a cocktail of monoclonal rabbit anti-EAAT3 (1:50; 14501; Cell Signaling Technology) and polyclonal guinea pig anti-GAD65 (1:500; 198104; Synaptic Systems) 1° antibodies. Rinsed sections were transferred (2 h, RT) to NGS blocking solution containing AF488-conjugated goat anti-guinea pig IgG (1:250; A11073; Thermo Fisher) to reveal GAD65-ir and biotinylated goat anti-rabbit IgG (1:250; AP132B; Thermo Fisher). EAAT3-ir was revealed in rinsed sections incubated in streptavidin–AF594 (1:250; S11227; Thermo Fisher; 15 min, RT). Rinsed sections were mounted on slides and coverslipped with ProLong Diamond anti-fade mounting medium (Thermo Fisher).
Imaging and analysis
Images of PVN immunofluorescence were captured with a 16 bit photomultiplier tube interfaced with a Zeiss LSM710 laser scanning confocal microscope equipped with appropriate laser lines. Images were captured at 20× (NA 0.8) and 100× (NA 1.4) with pinholes of 47.5 and 283.4 μm, respectively. From each 100× image (∼7.2 kµm2), ImageJ was used to generate separate pixel maps of synaptophysin (SYN, blue), GAD65 (green), and EAAT3 (red) immunofluorescence. Pixel threshold intensity and pixel overlap detection for dual and triple labeling was determined by comparison with the mean of tissue processed without primary antibodies as previously described (Mitchell et al., 2018; Maruyama et al., 2019).
Statistical analysis
Using Prism (v9.1) software (GraphPad), response data were tested for Gaussian distribution using Shapiro–Wilk or Kolmogorov–Smirnov tests, and all datasets were deemed compatible with parametric statistical testing. Statistical significance of means was compared by paired two-tailed Student’s t tests or by one- or two-way ANOVA with repeated-measures and mixed-effects design as appropriate. Post ANOVA pairwise comparisons were made using Sidak or Tukey’s tests corrected for multiplicity error. Parameter estimates from nonstationary fluctuation analyses (Fig. 2) were compared using the Kolmogorov–Smirnov (KS) test. Summary data are expressed as mean ± SD. Differences between means were considered significant at p < 0.05.
PVN GGS reflects activation of a greater number of postsynaptic GABA-A receptors but without postsynaptic plasticity. A, Representative Im responses to puff application of the GABA-A receptor agonist isoguvacine (ISO, 100 µM). Traces are the average of six consecutive ISO responses spanning a 2 min period at (a) baseline and (b) 10 min after bath glutamate (Glu, 100 µM, 10 min) washout. B, Summary time course of normalized IISO amplitude responses showing no effect of bath Glu (p = 0.5601; n = 5 cells). Each time point represents a 2 min average of peak Im responses to ISO applied every 20 s (mean ± SD). C, Representative (gray) and average mIPSCs at baseline (top, black) and 10 min after bath Glu (100 µM, 10 min) washout (bottom, green). D, Representative scaling of individual mIPSCs in C to the average of their peak amplitudes. Bold traces in each pair are the average mIPSCs (Ī) at baseline (black) and post Glu (green) with peak-scaled raw traces (I, gray) shown to illustrate decay variance of individual mIPSCs relative to the mean. E, Decay-phase mean–variance relationship of mIPSC data from panel C. Lines represent the linear fit of data points comprising the initial −5 pA of mean mIPSC amplitude. Slope values provide estimates of single-channel current (I, inset) from which single-channel conductances (γ, inset) were calculated [EIPSC = (−28 mV)-Vhold (−70 mV)]. F, Summary data showing that single-channel conductance values before (baseline, black circles) and 10 min after bath Glu (100 µM, 10 min) washout (post Glu, green squares) were unchanged (p = 0.7445; n = 9 cells). G, Summary data of the normalized decay-phase mean–variance relationship before (baseline, black circles) and after bath Glu (post Glu, green squares; n = 9 cells). Kolmogorov–Smirnov analysis revealed the number of GABA-A receptor channels active at the peak of mIPSCs at baseline was significantly increased post Glu (*p < 0.0001).
Results
Increased glutamate availability strengthens PVN GABAergic IPSCs
Inwardly directed IPSCs were recorded from PVN neurons in whole-cell configuration and in the continuous presence of iGluR antagonists (CNQX, 10 µM; APV, 50 µM; Fig. 1A, top) to isolate IPSCs and avoid membrane resistance changes induced by bath glutamate while holding Vm at −70 mV. To optimize placement of stimulating electrodes used for evoking IPSCs, we performed initial studies in slices from VGAT-EGFP reporter mice (n = 5), which revealed the localization of peri-nuclear zone (PNZ) GABAergic neurons surrounding the PVN (Fig. 1A, bottom). Initial recordings revealed that eIPSC amplitude at baseline was increased ∼5 min after bath glutamate (100 µM, 10 min) washout. Events were GABAergic eIPSCs as they were blocked by the GABA-A receptor antagonist picrotoxin (PTX, 100 µM; Fig. 1B). Extending the duration of bath glutamate exposure (100 µM, 20 min) resulted in a sustained increase in eIPSC amplitude without further strengthening (Table 1).
Effect of glutamate application time for PVN GGS
Recorded PVN neurons were classified as type I magnocellular neurosecretory cells (n = 67), type II parvocellular neuroendocrine cells (n = 40), and type III parvocellular pre-autonomic cells (n = 42) according to electrophysiological criteria encompassing their passive membrane properties (Table 2) and the absence (type I neurons) or presence (type II and III neurons) of low-threshold calcium spikes that underwent time-dependent augmentation selectively in type III neurons (Fig. 1C–E, top; Tasker and Dudek, 1991; Luther and Tasker, 2000; Stern, 2001). Initial studies testing effects of bath glutamate (100 µM, 10 min) revealed an effect in 10 of 12 type I, 5 of 5 type II, and 4 of 5 type III neurons. Among responsive neurons, eIPSC amplitude after bath glutamate washout increased to 131 ± 9%, 137 ± 12%, and 134 ± 14% of baseline, respectively (Fig. 1C–E, bottom). Effects of bath glutamate were not different among responsive PVN neuron subtypes (F(2,16) = 0.5205; p = 0.6040; ANOVA). Therefore, eIPSC amplitude data were combined for summary analysis, which revealed that eIPSC amplitude was significantly increased beginning ∼5 min after glutamate washout (F(12,226) = 69.74; *p = 0.0107; **p < 0.0020; n = 19 cells; Fig. 1F). This GGS response reflected ionotropic receptor-independent actions of glutamate as amplitude was unchanged in time control experiments (F(4.498,51.40) = 1.501; p = 0.2106; n = 8 cells; Fig. 1G).
Passive membrane properties of PVN neuron subtypes
Analysis of the amplitude distribution of spontaneous IPSCs recorded under the same experimental conditions, that is, ionotropic glutamate receptors blocked with CNQX (10 µM) and APV (50 µM), revealed that the mean, median, mode, and SD were not different and unchanged by the addition of bath TTX (1 µM), indicating events were mIPSCs whose amplitudes were Gaussian distributed (Table 3; n = 9 cells). In a separate group of eight cells, bath glutamate increased mIPSC amplitude (t(7) = 2.824; *p = 0.0256; Fig. 1H,I) without affecting frequency (t(7) = 0.1841; p = 0.8951; Fig. 1J), 10–90% rise time (t(7) = 0.2104; p = 0.4197; Fig. 1K), or 90–10% decay time (t(7) = 0.6085; p = 0.2810; Fig. 1L). Collectively, data in Figure 1 indicate that in the absence of ionotropic glutamate receptor signaling, elevated ambient glutamate dynamically and similarly strengthens GABAergic synaptic transmission among PVN neuron subtypes.
Amplitude distribution parameters of PVN spontaneous and miniature IPSCs
PVN GGS is not mediated at a postsynaptic locus
To assess postsynaptic contributions to PVN GGS, we first determined that the Im response to puff application of the short-acting GABA-A receptor agonist isoguvacine at baseline (−472 ± 106 pA) was unchanged after bath glutamate washout (−463 ± 136 pA; F(2.897,11.59) = 0.7108; p = 0.5601; one-way ANOVA; n = 5 cells; Fig. 2A,B). For a representative type II neuron, examination of an equal number of mIPSCs (n = 80) at baseline and 10 min after glutamate washout revealed expected peak amplitude variability (Fig. 2C), reflecting temporal variation in the number of released GABA vesicles (Scheuss and Neher, 2001). Normalizing mIPSC amplitudes to the average of their peak values at baseline and after glutamate washout revealed that rates of decay were also variable (Fig. 2D), reflecting a variable number of GABA-A receptor channels activated at the peak of individual mIPSCs (De Koninck and Mody, 1994; Nusser et al., 2001). As shown in Figure 2E, we calculated mIPSC variance across 30 equally sized time bins and plotted these relative to the corresponding mean current of each bin. Resulting variance–mean plots at baseline and after glutamate washout yielded the expected parabolic functions reflecting stochastic open/closing behavior of GABA-A receptor channels (De Koninck and Mody, 1994; Nusser et al., 1998, 2001; Silver et al., 1998; Scheuss and Neher, 2001). Established methods were then used to assess effects of glutamate on GABA-A receptor single-channel current (i ;Regehr, 2012). Linear regression was applied to the rising phase of parabolic functions, and regression lines were extrapolated to the zero current value (y-intercept). Slopes of these lines indicate that the single GABA-A receptor channel current was remarkably similar at baseline and after glutamate. For a group of nine neurons, the mean number of mIPSCs analyzed per neuron was 79 ± 46 at baseline and 78 ± 40 post-glutamate. Regression line slopes before (1.49 ± 0.78) and after glutamate (1.50 ± 0.75) were not different (t(8) = 1.543; p = 0.1614; paired t test), affirming that i was unaltered. Bath glutamate also did not affect single-channel conductance (γ; baseline, 30.1 ± 11.4 pS; post Glu, 30.8 ± 10.8 pS; t(8) = 0.3364; p = 0.7445; paired t test; Fig. 2F). Analysis of normalized variance relative to normalized mean current for pooled data (n = 9 cells) at baseline and after glutamate washout confirmed that peak variances at the vertex of corresponding parabolic functions were significantly different (D = 0.5667; p < 0.0001; Kolmogorov–Smirnov; Fig. 2G), indicating that bath glutamate increased the number of GABA-A receptor channels active at the onset of mIPSCs. Because bath glutamate increased m/eIPSC amplitude without increasing the postsynaptic response to GABA-A receptor agonist activation or GABA-A receptor single-channel conductance, findings collectively indicate that PVN GGS involves a presynaptic increase in synaptic GABA release and recruitment of spare postsynaptic GABA-A receptors.
PVN GGS is mediated presynaptically by a mechanism not dependent on increased GABA vesicle release probability
To assess presynaptic contributions to PVN GGS, we first assessed the effects of bath glutamate on the paired-pulse ratio (PPR) of PNZ eIPSCs while slices were perfused with aCSF containing low (1 mM), normal (2 mM), and high (4 mM) bath Ca2+ so that PPR varied from facilitating (PPF; n = 6 cells), to near unity (PPU; n = 7 cells), to depressing (PPD; n = 11 cells), respectively (Fig. 3A–C). Under each ambient Ca2+ condition, the prevailing PPR was unchanged by bath glutamate (F(2,21) = 2.085; p = 0.1493; two-way ANOVA), indicating no effect of glutamate on the initial probability of synaptic GABA release regardless of the level of terminal Ca2+ (Fig. 3D). As expected in the absence of postsynaptic receptor saturation (Foster et al., 2002, 2005), we observed a trend for the magnitude of PVN GGS to varying with bath Ca2+ (F(2,22) = 2.496; p = 0.1055; one-way ANOVA), an effect nullified when post-Glu eIPSC amplitudes were normalized to their respective baselines under each level of bath Ca2+ (Fig. 3E).
PVN GGS does not rely on presynaptic Ca2+ modulation or mGluR signaling but requires spare GABA-A receptors. Similar magnitude PVN GGS responses (arrows, top traces) were observed under low (A), normal (B), and high (C) bath Ca2+ to change PPR from facilitation (PPF; PPR > 1.1) to unity (PPU; 1.1 ≥ PPR ≥ 0.9) to depression (PPD; PPR < 0.9), respectively. Representative eIPSC responses (average of 8 sweeps) to paired-pulse PNZ stimulation at baseline (gray) and 10 min after bath glutamate (Glu, 100 µM, 10 min) washout (post Glu, PPF, blue; PPU, green; PPD, orange) under each bath Ca2+ concentration. D, E, Summary graphs of PPR (D) and eIPSC amplitude (E) values at baseline (gray bars) and after bath Glu (PPF, blue; PPU, green; PPD, orange; D: PPF, p = 0.3628, n = 6 cells; PPU, p = 0.4730, n = 7 cells; PPD, p = 0.1517, n = 11 cells; E: PPF, **p = 0.0090, n = 6 cells; PPU, ##p = 0.0093, n = 7 cells; PPD, ****p < 0.0001, n = 11 cells). F, Time course of glutamate effect on eIPSC amplitude in quercetin (30 µM) treated neurons (n = 7 cells). G, Summary data comparing eIPSC amplitude at baseline and during bath glutamate (p = 0.0730) in the presence of continuous quercetin. H, Summary data comparing eIPSC amplitude (**p = 0.0020) and PPR (p = 0.9105) at baseline and post quercetin. I, J, Time course of high bath Ca2+ effect on eIPSC amplitude in the absence or presence of quercetin. K, Summary data comparing eIPSC amplitude at baseline and during high bath Ca2+ in the absence (black vs blue; ***p = 0.0001; n = 8 cells) or presence (red vs blue; p = 0.8136; n = 6 cells) of continuous quercetin. L, Representative PVN GGS response during blockade of mGlu receptors. M, Summary data (n = 5 cells) comparing eIPSC amplitude (*p = 0.0390, left) and PPR (p = 0.3579, right) at baseline (black) and during bath glutamate (green) in the presence of continuous combined blockade of group I, II, and III mGlu receptors.
To further assess mechanisms of PVN GGS, we bath applied the GABA-A/C receptor negative allosteric modulator quercetin at low concentration (30 µM), thereby partially and insurmountably occupying postsynaptic GABA-A receptors (Fan et al., 2018). By ensuring that GABA released by PNZ eIPSCs at baseline was sufficient to fully saturate GABA-A receptors not occupied by quercetin, bath glutamate is expected to fail to increase eIPSC amplitude if PVN GGS is mediated by a presynaptic increase in GABA release above baseline. Consistent with this mechanism of action, bath glutamate failed to strengthen PNZ eIPSCs in the presence of quercetin (Fig. 3F,G; t(6) = 2.171; p = 0.0730; paired t test). Notably, quercetin reduced eIPSC amplitude prior to bath glutamate application (t(6) = 5.220; p = 0.0020; paired t test), confirming partial blockade of postsynaptic GABA-A receptors (Fig. 3H, left). Quercetin alone, however, did not change PPR (t(6) = 0.1172; p = 0.9105; paired t test), indicating that it did not act presynaptically to modulate the initial probability of synaptic GABA release (Fig. 3H, right). We further tested effects of quercetin to prevent eIPSC strengthening caused by bath application of high Ca2+ aCSF, which is expected to strengthen IPSCs by increasing terminal Ca2+ availability and thereby increasing vesicle release probability (Kirischuk et al., 2002; Kirischuk and Grantyn, 2002). As expected, slice perfusion with high Ca2+ aCSF alone increased PNZ eIPSC amplitude (Fig. 3I,K, left; t(7) = 7.669; p = 0.0001; paired t test). In the presence of quercetin, however, this effect was not observed (Fig. 3J,K, right; t(5) = 0.2485; p = 0.8136; paired t test), consistent with baseline eIPSCs saturating postsynaptic GABA-A receptors in the presence of quercetin, thereby preventing eIPSC strengthening mediated by increased GABA vesicle release probability.
Next, we tested for metabotropic glutamate receptor (mGluR) involvement in PVN GGS by blocking group I, II, and III mGluR with a cocktail of CPPG (200 µM) and MCPG (500 µM) before and during bath glutamate (Fig. 3L). PVN GGS was preserved (Fig. 3M, right; t(4) = 3.023; p = 0.0390; pair t test) and PPR remained unaffected (Fig. 3M, left; t(4) = 1.038; p = 0.3579; paired t test). Collectively, data in Figure 3 indicate that glutamate enhancement of GABA IPSC amplitude, though mediated presynaptically (Fig. 2), does not reflect a presynaptic increase in vesicle release probability or activation of mGluR.
PVN GGS does not change GABA vesicle release, pool size, or recycling dynamics
To further assess presynaptic actions of GGS, we evoked IPSCs by trains of 20 stimuli delivered at 20 Hz to the PNZ and calculated peak amplitude coefficients of variation (CV, SD mean−1) at baseline and 10 min after glutamate washout (Fig. 4A). With CV values not corrected for background noise (Korn and Faber, 1991) and with sweeps lacking eIPSCs excluded, data were expressed as 1/CV2, a quantity that varies in direct proportion to vesicle release probability and number of vesicle release sites (Traynelis and Jaramillo, 1998; Brock et al., 2020; van Huijstee and Kessels, 2020). For summary analysis (n = 6 cells), CV values were calculated from 1,125 eIPSCs at baseline (188 ± 19/cell) and 1,070 after Glu washout (178 ± 30/cell). Findings revealed (Fig. 4B) that 1/CV2 was unchanged by bath glutamate (t(5) = 0.03196; p = 0.9757; paired t test), suggesting that glutamate incubation for 10 min had no effect on GABA vesicle release probability or the number of vesicle release sites over a range of presynaptic Ca2+ levels and vesicle pool sizes.
Effect of PVN GGS on GABA vesicle release, pool size, and recycling dynamics. A, Left traces show 10 consecutive eIPSC responses (gray) at baseline (top) and 10 min after Glu (100 µM, 10 min) washout (bottom), overlaid with their respective averages (bold). Right traces (gray) illustrate response variability around the mean (bold) at baseline (top, black) and post Glu (bottom, green). Mean current amplitudes were subtracted from individual eIPSC amplitudes (I − Ī) at each time point (right). B, Summary plot of peak eIPSC amplitude 1/CV2 values at baseline (gray) and 10 min after bath Glu washout (green; p = 0.9757; n = 6 cells). C, Representative high-frequency PNZ stimulation-induced (20 pulses, 20 Hz) depression of eIPSC responses at baseline (top, gray) and post Glu (bottom, green). Each trace is the average of 10 consecutive trains. D, Summary eIPSC high-frequency depression curves to PNZ stimulation at baseline (black circles) and post Glu (green squares). Peak eIPSC amplitudes were normalized within each group to eIPSC #1. Depression time constants (τ, inset) were not different (p = 0.9437; n = 6 cells). E, Cumulative eIPSC amplitudes plotted as a function of stimulus number (#1–20) during high-frequency depression at baseline (black circles) and 10 min after bath Glu washout (green squares). Extrapolation of the linear portion of the curve (#7–20) to the y-intercept indexes the size of the RRP of vesicles, which was unaffected by bath Glu (p = 0.6078; n = 6 cells). F, Representative time-dependent recovery of eIPSC amplitude following high-frequency depression at baseline (gray, left) and 10 min after bath Glu washout (green, right). Each trace is the average response to 10 consecutive stimulus trains consisting of pulses delivered at the indicated post-depression time points. G, Summary plot of eIPSC recovery from high-frequency depression at baseline (black) and post Glu (green). Recovery time constants (τ, inset) were not different (p = 0.5041, n = 6 cells).
We next determined the impact of GGS on the size and use dynamics of the GABA vesicle RRP. Returning to high-frequency (20 stimuli at 20 Hz) PNZ stimulation, we first determined for six cells that the time constant (τ) of amplitude depression at baseline (123 ± 41 ms) and after bath glutamate washout (121 ± 21 ms) were not different (t(5) = 0.07422; p = 0.9437; paired t test), consistent with no effect on release probability (Fig. 4C,D). Likewise, the apparent size of the RRP was unchanged (baseline, −655 ± 384 pA; glutamate, −635 ± 348 pA; Fig. 4E) as determined by the similarity of y-intercept values of regression lines fitted to cumulative eIPSC amplitude plots at baseline and after glutamate (t(5) = 0.5457; p = 0.6087; paired t test). Recovery from eIPSC depression across multiple cell recordings (Fig. 4F), which primarily reflects the steady-state vesicle recycling speed (Regehr, 2012), followed the expected exponential time course with a time constant (τ) unaffected by bath glutamate (t(5) = 0.7193; p = 0.5041; paired t test; Fig. 4G). Collectively, data in Figure 4 indicate that glutamate enhancement of GABA IPSC amplitude does not reflect a presynaptic increase in vesicle release probability, number of vesicle release sites, or RRP size/dynamics.
PVN GGS relies on neuronal glutamate uptake and new GABA synthesis
Given the above findings, we reasoned that PVN GGS could involve increased synthesis and packaging of GABA into vesicles, whose release and recycling dynamics were unaffected by bath glutamate. Given that GABA is synthesized intracellularly from glutamate, we tested dependence of PVN GGS on increased glutamate uptake by membrane transporters. Using maximally effective concentrations of the glial EAAT1 inhibitor UCPH-101 (10 µM; n = 7 cells; Abrahamsen et al., 2013) and EAAT2 inhibitor DHK (250 µM; n = 6 cells; Arriza et al., 1994), we determined that bath glutamate retained its capacity to increase eIPSC amplitude (UCPH: 131 ± 15% baseline, t(6) = 5.402, p = 0.0017, Fig. 5A,D; DHK: 128 ± 14, t(5) = 4.700, p = 0.0053, Fig. 5B,D, paired t tests). In contrast, a maximally effective concentration of the pan-specific glutamate uptake inhibitor TBOA (100 µM; n = 5 cells; Shimamoto et al., 1998) effectively prevented glutamate strengthening of eIPSC amplitude (83 ± 17% baseline) (t(4) = 2.213; p = 0.0913; paired t test; Fig. 5C,D). Collectively, these findings indicate that PVN GGS depends on neuronal, not glial, glutamate uptake, arguing against the glial glutamine shuttle playing a significant role in supplying glutamate to PNZ GABA neurons as part of the strengthening mechanism. PPR was again unaffected by bath glutamate (Fig. 3A–D) whether in the presence of UCPH-101 (0.90 ± 0.18 to 0.90 ± 0.16; t(6) = 0.0036; p = 0.9972; paired t test) or DHK (1.11 ± 0.03 to 1.09 ± 0.06; t(5) = 0.2155; p = 0.8379; paired t test) or TBOA (0.91 ± 0.10 to 0.93 ± 0.13; t(4) = 0.2887; p = 0.7872; paired t test; Fig. 5E), affirming that PVN GGS does not involve an increase in GABA vesicle release probability.
PVN GGS relies on neuronal glutamate uptake by EAAT3 and de novo GABA synthesis. A–C, Experimental protocols (top) with representative average traces of eight consecutive eIPSC responses to paired-pulse electrical stimulation of the PNZ (bottom). Beginning 5 min before bath glutamate (Glu, 100 µM, 10 min), the glial specific inhibitor of EAAT1 (UCPH-101, 10 µM, left) or EAAT2 (DHK, 250 µM, center) or the pan-specific EAAT inhibitor TBOA (100 µM, right) was bath applied. D, Summary of eIPSC amplitude changes to bath Glu with concurrent EAAT inhibition with UCPH-101 (blue; ***p = 0.0017; n = 7 cells), DHK (purple; **p = 0.0053; n = 6 cells) or TBOA (orange; p = 0.0913; n = 5 cells). E, EAAT inhibition had no effect on eIPSC PPR (UCPH-101, p = 0.9972; DHK, p = 0.8379; TBOA, p = 0.7872). F, Experimental protocol (top) and representative eIPSC responses (average of 8 sweeps, bottom) in the continuous presence of the membrane permeable GAD inhibitor allylglycine (AG, 2 mM) at baseline and 10 min after exposure to bath glutamate (100 µM, 10 min). G, H, Summary data showing effects of AG on Glu strengthening of PVN (G) eIPSCs (n = 7 cells) and (H) mIPSCs (n = 5 cells). Paired t tests revealed that bath Glu failed to change eIPSC amplitude (p = 0.1329), mIPSC amplitude (p = 0.1904), or mIPSC frequency (p = 0.0772) in the presence of AG.
Dependence of PVN GGS on neuronal glutamate uptake by EAAT3 raised the possibility that glutamate cleared from the extracellular space could mediate GGS by undergoing enzymatic conversion to GABA by GAD65. To test this, we bath applied the membrane-permeable GAD inhibitor allylglycine (AG; 2 mM; n = 7 cells; Horton et al., 1978) prior to and during bath glutamate (Fig. 5F, top). In addition to reducing mIPSC amplitude by ∼50% (data not shown), AG pretreatment (1–3 h) effectively prevented glutamate strengthening of PNZ eIPSCs (t(6) = 1.738; p = 0.1329; paired t test; Fig. 5G). Analysis further revealed no effect of AG inhibition of GAD on the amplitude (t(4) = 1.575; p = 0.1904) or frequency (t(4) = 2.366; p = 0.0772) response of mIPSCs to bath glutamate (Fig. 5H). Collectively, data in Figure 5 support a model of GGS in which glutamate cleared by EAAT3 on GABA terminals strengthens GABA inhibition by increasing GABA synthesis, with subsequent packaging into vesicles at greater than normal density (i.e., increased GABA vesicle quantal size; Mathews and Diamond, 2003; Stafford et al., 2010).
Given that the above findings implicate direct neuronal EAAT3-mediated glutamate uptake and de novo synthesis of GABA in PVN GGS, we examined the potential for EAAT3-mediated glutamate uptake directly into PVN GABA terminals. Triple-label immunostaining was performed and revealed that EAAT3-ir was localized to putative PVN GABA terminals based on co-staining for the GABA terminal marker GAD65 (Petroff, 2002; Patel et al., 2006) and the presynaptic vesicle associated membrane protein synaptophysin (SYP; Elferink and Scheller, 1993). As expected, all three antigens were densely distributed throughout the PVN (Fig. 6Aa,b,c), with GAD65-ir and SYP-ir appearing to surround cell bodies (Fig. 6Ae,f). EAAT3 antibody staining specificity was assessed by lack of immunofluorescence in PVN (Fig. 6Ba,b) and hippocampal (Fig. 6Bc,d) sections from EAAT3 (SLC1a1) gene knock-out mice. EAAT3 and GAD65 double-labeling with SYP (Fig. 6Ca,b,d,e) revealed that each appeared strongly localized to presumptive synaptic terminals (EAAT3, 43.5 ± 6.4%; GAD65, 58.7 ± 8.7%; n = 6; Fig. 6Da,b). Triple labeling (Fig. 6Cc,f) revealed that >50% of EAAT3-ir was localized with elements double-labeled for GAD65 and SYP (Fig. 6Dc), consistent with PVN EAAT3 dominantly associating with local GABA terminals.
EAAT3 localizes to presumptive GABA terminals in PVN. A, Representative low magnification (20×) confocal images of EAAT3 (red, panel a), GAD65 (green, panel b), and SYP (blue, panel c) immunofluorescence. Dashed lines demark the PVN boundary. Boxed regions are shown at high magnification (100×) in panels (d–f). Bottom panels (g–i) show high magnification images of tissue processed without primary antibodies. B, Representative images of EAAT3 immunofluorescent staining of PVN (20×, left) and hippocampus (2×, right) from wild-type (top) and EAAT3 (SLC1a1) gene knock-out (bottom) mice. C, Merged low (top) and high (bottom) magnification images of double labeling of PVN EAAT3 and SYP (purple, a,d) as well as GAD65 and SYP (cyan, b,e) together with triple labeling of EAAT3, GAD65, and SYP (white, c,f). D, Summary graphs (n = 6 mice) of percent pixel overlap of merged high-magnification images.
Synaptic glutamate release is sufficient to elicit PVN GGS
We next examined whether endogenous synaptic glutamate release would strengthen PVN GABA transmission like bath glutamate. We first confirmed that bath perfusion with high-K+ (+10 mM) aCSF significantly increased synaptic glutamate release (t(5) = 4.778; p = 0.0050; n = 6 cells; paired t test; Fig. 7A,B) and that Rose Bengal (RB; 30 µM, 20 min), a potent noncompetitive VGLUT inhibitor (Bole and Ueda, 2005), effectively diminished synaptic glutamate release (t(5) = 3.325; p = 0.0209; n = 6 cells; paired t test; Fig. 7C,D) without affecting the Im response to puff-applied glutamate (t(3) = 0.9018; p = 0.4336; paired t test; Fig. 7E,F). We then investigated whether increased glutamate release from PVN synapses can enhance GABA IPSCs. With AMPA and NMDA iGluR blocked, we recorded IPSCs evoked from the PNZ while slices were perfused for 10 min with high-K+ aCSF in the absence and presence of RB (Fig. 7G,H). For a group of eight cells, high-K+ increased eIPSC amplitude 10 min after washout (F(3.399,23.79) = 11.97; p = 0.0047–0.0002; one-way ANOVA; Fig. 7I, blue), an effect not observed during vesicular glutamate depletion with RB (F(4.955,34.68) = 6.135; p = 0.0002; one-way ANOVA; Fig. 7I, red). Further analysis revealed that under control conditions, eIPSC amplitudes at specific time points during bath high-K+ washout were significantly greater compared with those recorded in the presence of RB (F(5.506,79.84) = 5.920; p < 0.0001; two-way ANOVA). Again, eIPSC PPR was unchanged after high-K+ washout in control (t(5) = 0.9374; p = 0.0.3932; paired t test) and RB (t(4) = 0.1342; p = 0.8997; paired t test) treated slices (Fig. 7J), indicating that neither high-K+ nor RB had a lasting effect on the initial probability of GABA release. Collectively, data in Figure 7 indicate that increasing PVN synaptic glutamate release increases the amplitude of PNZ-evoked GABA IPSCs and does so with a magnitude and time course similar to elevated bath glutamate.
PVN GGS can be induced by local synaptic glutamate release. A, Representative sEPSCs recorded in the presence of the GABA-A receptor antagonist picrotoxin (PTX, 100 µM) at baseline (left) and during bath perfusion with high-K+ (+10 mM) aCSF (right). B, Summary data showing that bath high-K+ significantly increased synaptic glutamate release (**p = 0.0050; n = 6 cells). C, Representative sEPSCs recorded in the presence of the GABA-A receptor antagonist picrotoxin (PTX, 100 µM) before (baseline) and during bath perfusion with the noncompetitive VGLUT inhibitor Rose Bengal (RB, 30 µM, 35 min). D, Summary data showing that RB nearly abolished synaptic glutamate release (*p = 0.0209; n = 6 cells). Summary data in B and D are plotted as cumulative sEPSC area (total charge transfer) of all sEPSCs recorded over a 1 min period at baseline and during treatment because high-K+ and RB shift the distribution of sEPSC frequency as well as amplitude. E, Representative traces of Im responses to puff applied (black bars) glutamate (Glu, 100 µM, 200 ms) at baseline (gray) and 35 min after bath RB (red). F, Summary graph of the Im amplitude response to puffed Glu in the presence of RB (p = 0.4336; n = 4 cells). G, H, Experimental protocols (top) with representative traces (bottom) at baseline (a) and 10 min after bath perfusion (10 min) with 10 mM KCl containing aCSF (high-K+) (b) alone (blue, left) and in the presence of 30 µM Rose Bengal (RB, red, right). Traces are the average of 8 consecutive eIPSC responses to paired-pulse electrical stimulation of the PNZ. I, Time course of high-K+ effect on eIPSC amplitude in control and RB treated neurons. Data points plotted every 2 min are the average of eight consecutive IPSCs evoked every 20 s and reveal that high-K+ increased eIPSC amplitude relative to baseline (***p = 0.0047–0.0002; n = 8 cells), an effect prevented by RB (n = 8 cells). J, PPR of eIPSCs was unaffected by high-K+ alone (p = 0.3932; n = 6 cells) or high-K+ with RB (p = 0.8997; n = 5 cells).
PVN GGS suppresses PVN neuronal discharge
Data so far indicate that increased synaptic glutamate spillover strengthens PVN GABAergic IPSCs by increasing presynaptic GABA synthesis and quantal release. The functional significance of this was investigated by recording discharge of PVN neurons at baseline, during bath glutamate (100 µM, 10 min) and 10 min after glutamate washout. For a group of six cells, action potential firing at baseline averaged 2.58 ± 0.60 Hz and increased during bath glutamate to 3.76 ± 1.63 Hz. Notably, firing frequency after glutamate washout fell significantly below baseline to 1.32 ± 0.88 Hz (t(5) = 4.274; p = 0.0079; paired t test; Fig. 8A,F), consistent with GGS increasing GABAergic inhibition. In support of this, time control experiments indicated that delayed suppression of cell firing was not attributable to cell rundown (Fig. 8B; n = 5 cells) and positive current injections to mimic glutamate excitation ruled out post-excitatory depression (Fig. 8C; n = 5 cells). In further keeping with GGS causing delayed strengthening of GABA inhibition, the reduction of discharge below baseline after glutamate washout was not observed during GABA-A receptor blockade with picrotoxin (Fig. 8D; n = 5 cells) or during inhibition of glutamate uptake with TBOA (Fig. 8E; n = 7 cells). Summary data comparing firing rate changes across treatment groups indicate that bath glutamate alone caused significant delayed suppression of cell firing through a mechanism consisted with PVN GGS (F(4,27) = 12.29; p < 0.0001; two-way ANOVA; Fig. 8F).
PVN GGS suppresses cell firing. A–E, Representative traces of PVN neuronal discharge before (left), during (middle), and 10 min after (right; A) bath glutamate (Glu, 100 µM, 10 min; n = 7 cells), (B) passage of time (10 min, n = 5 cells), (C) injection of depolarizing current to mimic excitatory effects of Glu (n = 5 cells), or bath perfusion with Glu (100 µM) during (D) GABA-A receptor blockade (PTX, 100 µM, n = 5 cells) or (E) glutamate uptake inhibition (TBOA, 100 µM; n = 7 cells). F, Summary changes in PVN cell firing relative to baseline 10 min after treatments. Firing rate was significantly decreased compared with baseline after washout of bath Glu alone (****p < 0.0001), but not in other treatment groups.
Discussion
Glutamatergic and GABAergic interactions dynamically regulate neuronal activity in the adult brain (Chapman et al., 2022). Here we found that increasing local glutamate availability rapidly strengthens GABAergic transmission in the hypothalamic PVN. This GGS phenomenon relies on neuronal glutamate uptake ostensibly into local GABA neurons/terminals with subsequent enzymatic conversion of glutamate to GABA and packaging of new GABA molecules into synaptic vesicles at an apparent density exceeding normal. PVN GGS is functionally robust, suppressing neuronal firing. A model is proposed in which PVN GGS buffers excitability, protecting against excitotoxicity while restraining responses to stress.
PVN GGS reflects presynaptic plasticity
Our findings indicate that PVN GGS does not involve iGluR activation and therefore does not represent a common form of glutamate-driven heterosynaptic plasticity (Sunstrum and Inoue, 2019; Herstel and Wierenga, 2021; Chapman et al., 2022). GGS also does not involve mGluR activation, consistent with PVN mGluR signaling acutely decreasing, not increasing, GABA IPSCs (Colmers and Bains, 2018). Although mGluR signaling can strengthen PVN GABA transmission, this involves membrane insertion of GABA-A receptors, a response developing over several hours following stress activation of β-adrenoceptors (Inoue et al., 2013). Notably, mGluR signaling is also not required for a GGS-like response in the hippocampus (Mathews and Diamond, 2003).
According to the binomial model of synaptic transmission, the number of vesicle release sites (n) is fixed on the time scale of PVN GGS (Silver et al., 1998; Regehr, 2012). This is consistent with our finding that the 1/CV2 value of peak eIPSC amplitudes, which varies in proportion to n (Brock et al., 2020), was unchanged by PVN GGS. Whereas sustained synaptic potentiation normally involves increased density of postsynaptic transmitter receptors (Regehr, 2012; Brock et al., 2020), short-term potentiation most often reflects a presynaptic increase in vesicle release probability and/or RRP size (Regehr, 2012; Brock et al., 2020).
Our synaptic vesicle depletion experiments indicate that PVN GGS did not change RRP size or vesicle recycling dynamics. Estimating RRP size by high-frequency stimulation-induced vesicle depletion is susceptible to error, for example, due to partial vesicle replenishment between evoked synaptic responses. Extrapolation of the linear phase of the cumulative eIPSC amplitude plot compensates for pool replenishment, but this assumes a constant rate of replenishment that might not always be true (Kaeser and Regehr, 2017). Although this opens the possibility that we may have underestimated RRP size at GABAergic PNZ-PVN synapses, the key point of interpretation is that pool size appeared to be unchanged after bath glutamate. Hence whatever the RRP size might be, a size increase does not appear to explain PVN GGS.
Notably, PVN GGS also did not affect eIPSC paired-pulse ratio under low, normal, or high bath Ca2+. This, together with no effect on peak amplitude 1/CV2 (Brock et al., 2020), argues against GGS being dominantly mediated by increased GABA vesicle release probability. A postsynaptic mechanism is also unlikely since GGS did not change the Im response to GABA-A receptor agonist (Fig. 2A,B) or estimated GABA-A receptor single-channel conductance (γ; Fig. 2E,F). Notably, peak-scaled decay-phase nonstationary fluctuation analysis can overestimate γ (De Koninck and Mody, 1994; Nusser et al., 1998, 2001), and although some of our estimates exceed the range typically reported (De Koninck and Mody, 1994; Nusser et al., 2001), literature values are available only for a limited number of GABA-A receptor subunit configurations. Regardless of their accuracy, our γ estimates were unchanged by bath glutamate, arguing against PVN GGS being mediated by this postsynaptic mechanism.
Fluctuation analysis indicated that PVN GGS activated a larger population of postsynaptic GABA-A receptors at the peak of mIPSCs (Fig. 2G). But, as noted previously, bath glutamate had no effect on the Im response to GABA-A receptor agonist activation. This argues against GGS reflecting increased membrane receptor density. This conclusion is supported by our use of the negative allosteric GABA-A receptor modulator quercetin to ensure GABA release by baseline eIPSCs fully saturated available receptors. Failure of IPSC strengthening under these conditions argues against postsynaptic membrane insertion of GABA-A receptors as the primary driver of GGS. Instead, GGS appears to involve increased synaptic GABA release and activation of spare GABA-A receptors. Recruitment of spare receptors is consistent with membrane insertion of new GABA-A receptors being slow relative to the onset latency of PVN GGS (Jacob et al., 2008).
Our interpretation that PVN GGS reflects increased unitary quantal size is largely based on observing no change in other synaptic parameters. We cannot entirely rule out that approaches used to quantify synaptic parameters could lack sufficient sensitivity to detect other changes. In the case of 1/CV2 analysis, this approach is entirely insensitive to changes in quantal size and so cannot be used to rule in or out a quantal mechanism. Likewise, our studies with quercetin indicating that GGS fails when postsynaptic GABA-A receptors are fully saturated cannot discriminate among mechanism that increase GABA release. These issues notwithstanding, literature evidence indicates that GABA uptake into synaptic vesicles by VGAT is typically slow (τ > 20 s) relative to the rate of GABA release (Egashira et al., 2016). This indicates that GABA vesicles are likely to be incompletely filled, especially under high release conditions, providing an avenue for quantal strengthening of GABA transmission.
PVN GGS relies on glutamate uptake by local GABA neurons
A key finding of this study was that the pan-specific glutamate uptake inhibitor TBOA effectively prevented PVN GGS, unlike inhibitors specific for glial glutamate transporters (i.e., UCPH101-GLAST/EAAT1 and DHK-GLT1/EAAT2). Therefore, PVN GGS appears to require neuronal glutamate uptake by EAAT3, consistent with literature evidence (Mathews and Diamond, 2003; Stafford et al., 2010). We further found that EAAT3 in the PVN was immunolocalized with the GABA terminal marker GAD65 and the vesicle-associated membrane protein synaptophysin, consistent with GGS reflecting direct glutamate uptake into PVN GABA terminals. EAAT3 localization to GABA neurons and their synaptic terminals has been previously reported (Rothstein et al., 1994; Berger and Hediger, 1998; Kugler and Schmitt, 1999; He et al., 2000). Interpretation of immunolocalization data relies on antibody specificity (Holmseth et al., 2012). Here we affirmed specificity by observing near complete loss of immunoreactivity in EAAT3 (SLC1a1) gene knock-out mice. Moreover, literature evidence shows that antisense knockdown of EAAT3 decreases brain GABA, impairs new GABA synthesis, and reduces GABAergic IPSC amplitude (Sepkuty et al., 2002). Further, D-aspartate is taken up into hippocampal GABAergic varicosities even when glial uptake is fully inhibited (Stafford et al., 2010). Collectively, evidence supports a model in which PVN GGS relies on direct EAAT3-mediated glutamate uptake into GABA neurons and possibly their PVN terminals.
Consistent with previous reports (Mathews and Diamond, 2003; Rowley et al., 2012; Bianchi et al., 2014), inhibition of GABA synthesis prevented PVN GGS, indicating that GABA vesicles are preferentially filled with newly synthesized, not recycled GABA (Patel et al., 2006). Collectively, our results indicate that PVN GGS relies on clearance of extracellular glutamate by local GABA neurons and terminals where new GABA molecules are synthesized and packaged into synaptic vesicles at an apparent density greater than normal, thereby strengthening GABA transmission largely through increased quantal size and recruitment of spare postsynaptic GABA-A receptors. Whether PVN GGS is exclusively mediated by a quantal mechanism is uncertain since, as noted, some of our analytical methods could have lacked sufficient sensitivity to detect small changes in other synaptic parameters, especially vesicle release probability (Foster et al., 2002) and RRP size (Thanawala and Regehr, 2013).
Is PVN GGS onset slow?
Given the above model of PVN GGS, its 12–15 min onset latency might appear unexpectedly slow. In hippocampal slices, a GGS-like phenomenon was reported to tonically support the resting strength of GABA IPSCs (Mathews and Diamond, 2003; Stafford et al., 2010). Interfering with this process was reported to reduce GABA IPSC amplitude ∼2.5× faster than the onset latency of PVN GGS. This disparity might reflect a species or age difference since hippocampal studies used 10–14-day-old rats while we used adult mice. Alternatively, the off response to interrupting GGS may be faster, especially at high vesicle release rates, than the time needed for GGS to develop. Slower GGS onset in PVN might also involve preferential strengthening of GABA vesicles comprising a reserve pool. Accordingly, vesicles of the RRP that formed prior to high glutamate exposure might have to be exhausted before strengthened vesicles become available for release. This, however, does not appear to be the case in hippocampus where GGS strengthens vesicles of the RRP (Mathews and Diamond, 2003).
Another factor that could slow GGS onset is the probability of GABA vesicle release relative to the size of the RRP. A high release probability together with a small RRP would reduce the time needed to exhaust preformed vesicles, enabling strengthened vesicles to be released sooner. Of special interest is that GABA vesicles are transiently hyper-acidified as they recycle (Egashira et al., 2016), enabling VGAT to fill vesicles to greater concentration. Thereafter, vesicle pH rises, reducing VGAT activity and lowering GABA concentration due to GABA leakage (Egashira et al., 2016). These findings indicate that during rapid turnover, GABA vesicles may be overfilled if a sufficient supply of GABA molecules is available. Collectively, available evidence is consistent with GGS being largely due to a quantal mechanism occurring specifically among recycling GABA vesicles.
Consistent with a preferential impact of GGS on rapidly recycling vesicles, we noted that whereas GGS increased the amplitude of evoked IPSCs by ∼30%, it strengthened spontaneous miniature IPSCs by only half as much. Whether this reflects some or all the regulatory influences noted above is unknown. Alternatively, a smaller impact of GGS on mIPSCs could reflect challenges associated with setting the amplitude threshold for mIPSC detection to perfectly exclude noise at baseline while simultaneously detecting all mIPSCs that will eventually undergo strengthening. Below threshold mIPSCs at baseline that undergo strengthening to an amplitude detectable after GGS will likely be small, causing low-end skewing of the post-GGS mIPSC amplitude distribution and falsely indicating that GGS has less impact on mIPSCs than eIPSCs. The latter issue appears to have had little impact here since it would have increased mIPSC frequency relative to baseline, and this was not observed.
Functional implications
For PVN GGS to function in vivo requires sufficient synaptic glutamate release to drive robust uptake into GABA neurons. Extracellular glutamate in the brain is normally low (1–5 µM; Moussawi et al., 2011) but increases sharply (0.1–5 mM) as glutamatergic activity rises (Attwell et al., 1993; Montana et al., 2006). Here, high-K+ aCSF increased synaptic glutamate release and the amplitude of GABA IPSCs (Fig. 7) to a similar extent and with a similar onset time as bath glutamate. This suggests our bath glutamate concentration (100 µM) may have saturated the GGS mechanism and that PVN neurons receive sufficient glutamatergic input for spillover to induce a potentially maximal GGS response. Because the VGLUT inhibitor Rose Bengal (Fig. 7H–J) prevented high-K+ PVN GGS, it follows that the latter requires synaptic glutamate release.
Many homeostatic challenges activate glutamatergic inputs to PVN (Oliet, 2002; Tasker et al., 2002; Stocker et al., 2006; Brunton et al., 2008; Bardgett et al., 2014), which creates conditions conducive to GGS development. During pregnancy and lactation, stress axis activation is strongly blunted (Brunton et al., 2008). The extent to which enhancement of the GGS mechanism contributes to this is unknown. During chronic dehydration, synaptic glutamate release increases while GABA release decreases (Tasker et al., 2002), which suggests uncoupling of the GGS mechanism, ostensibly preventing glutamate from strengthening GABA inhibition, an action that could ensure robust PVN-driven homeostatic responses such as pituitary vasopressin/adrenocorticotropic hormone (ACTH) release and sympathetic activation (Toney and Stocker, 2010).
PVN GGS appears functionally significant in that it causes delayed suppression of cell firing (Fig. 8), an action prevented by GABA-A receptor blockade. A contribution of AMPA receptor desensitization to discharge suppression seems unlikely because reduced spiking was not observed when slices were exposed to bath glutamate during continuous uptake inhibition (Fig. 8E), which would have increased ambient glutamate and therefore increased, not decreased, AMPA receptor desensitization.
Here we observed GGS in all major PVN neuronal subtypes, suggesting it can regulate all major PVN output systems, including stress-coping behaviors, release of pituitary hormones, and increased sympathetic activity (Herman et al., 2004; Toney and Stocker, 2010; Wamsteeker Cusulin and Bains, 2015). Studies not only show robust glutamate uptake in PVN (Bardgett et al., 2014; Pandit et al., 2015) but that uptake inhibition strongly enhances sympathetic responses to ionotropic glutamate receptor activation (Park et al., 2009; Bardgett et al., 2014). The presence of avid PVN glutamate uptake suggests GGS might be tightly regulated. For example, protein kinase C (PKC) activation rapidly (≤30 min) reduces membrane EAAT2 and increases EAAT3 (Kalandadze et al., 2002; Fournier et al., 2004). Given that PVN is targeted by numerous transmitter/modulator systems coupled to PKC (Hazell et al., 2012), the EAAT3-to-EAAT2/1 ratio itself may be dynamically regulated. Conditions that increase this ratio could elicit a more robust and rapid expression of GGS. Moreover, specific activation of PKC epsilon activates GAD65 (Chou et al., 2017), which could increase the rate at which cleared glutamate is converted to GABA, leading to faster onset and greater amplitude GGS.
In sum, GABA inhibition of the hypothalamic PVN rapidly strengthens when glutamate availability increases. This process depends on EAAT3-mediated glutamate uptake, apparently into local GABA neurons/terminals, and synthesis of new GABA molecules. PVN GGS ostensibly operates to mitigate neuronal hyperactivation during strong glutamatergic excitation. The extent to which PVN GGS is dynamically regulated and sustained during prolonged homeostatic challenges or disease conditions warrants further study.
Footnotes
This work was supported by National Institutes of Health grants R01MH093320 and R01NS115072 (GMT). Images were generated using the UTHSA Optical Imaging Core Facility support by NIH-NCI P30 CA54174. Brain tissue from EAAT3 knock-out mice was generously provided by Dr. Zhiyi Zuo, Department of Anesthesiology, University of Virginia.
The authors declare no competing financial interests.
- Correspondence should be addressed to Glenn M. Toney at toney{at}uthscsa.edu.
