Glucocorticoid feedback uncovers retrograde opioid signaling at hypothalamic synapses.

Stressful experience initiates a neuroendocrine response culminating in the release of glucocorticoid hormones into the blood. Glucocorticoids feed back to the brain, causing adaptations that prevent excessive hormone responses to subsequent challenges. How these changes occur remains unknown. We found that glucocorticoid receptor activation in rodent hypothalamic neuroendocrine neurons following in vivo stress is a metaplastic signal that allows GABA synapses to undergo activity-dependent long-term depression (LTDGABA). LTDGABA was unmasked through glucocorticoid receptor-dependent inhibition of Regulator of G protein Signaling 4 (RGS4), which amplified signaling through postsynaptic metabotropic glutamate receptors. This drove somatodendritic opioid release, resulting in a persistent retrograde suppression of synaptic transmission through presynaptic μ receptors. Together, our data provide new evidence for retrograde opioid signaling at synapses in neuroendocrine circuits and represent a potential mechanism underlying glucocorticoid contributions to stress adaptation.

Exposure to stress results in two prominent hormonal responses: central and peripheral catecholamine release and a surge of glucocorticoids into the blood stream. Through temporally and mechanistically distinct pathways, both mediators are essential for appropriate behavior and mood regulation[1,2]. One unique and critical function of glucocorticoids on stress circuits is that they feedback to curtail hormone release in response to subsequent challenges. This serves a self-limiting homeostatic function in the face of diverse and repeated stress challenges[3]. Despite this fundamental role for glucocorticoids in shaping endocrine function with experience, relatively little is known about how it might be accomplished.

Adaptive control of the neuroendocrine response to stress resides with a small cluster of neurons in the paraventricular nucleus of the hypothalamus (PVN). These parvocellular neuroendocrine cells (PNCs) at the head of the hypothalamic-pituitary-adrenal (HPA)/stress-axis are positioned as the definitive point of neural stress integration; their activity is a function of both synaptic drive and negative feedback by glucocorticoids[3]. The dominant share of synapses onto PNCs are GABAergic[4]. GABA transmission onto PNCs restrains basal stress axis output[5] and is, itself, sensitive to stress[6,7]. Importantly, stress exposure causes diminished chloride extrusion capacity in PNCs, resulting in a situation in which GABA is excitatory during stress[5,8]. Thus, although it is counterintuitive, dampening GABA transmission alleviates the activation of the endocrine response[8].

In addition to corticotropin-releasing hormone (CRH) and vasopressin, PNCs synthesize proenkephalin-derived opioid peptides[9]. Enkephalins have been implicated as putative mediators of adaptive change to stress-axis function[9]. Consistent with this idea, mice lacking proenkephalin exhibit prolonged GC elevation to stress[10], suggesting opioids may participate in GC negative feedback. The cellular actions of endogenous opioid signaling have not been explored in PNCs; in other systems, they function as retrograde signals to inhibit neurotransmitter release[11,12]. We hypothesized that opioids are intermediaries of glucocorticoid actions in the PVN. Using whole-cell patch-clamp recordings of PNCs from naïve and stress-exposed rats, we examined GABA synapse strength and responses to patterned afferent activity. We report that a single stressful experience, followed by a 90-min temporal delay unmasks activity-dependent, heterosynaptic long-term depression of GABA (LTDGABA) synapses that is mediated by retrograde opioid signaling.

Results

Glucocorticoid receptor activation during stress unmasks LTDGABA

In response to an acute stress, plasma corticosterone (CORT, the major rodent glucocorticoid) rapidly rises; peak concentrations are reached 15–30 min from stress onset, persist during the stress, and subside slowly thereafter[1]. Subsequent access of CORT to the brain is regulated and time of peak elevation lags that of plasma CORT[13]. To investigate potential effects of CORT exposure resulting from stress, we examined PNCs in in vitro hypothalamic slices prepared from rats exposed to 30 min of immobilization stress followed by incrementally increasing periods of recovery before sacrifice (Fig. 1a). Naïve (unstressed) rats served as our age-matched controls. In whole-cell voltage-clamp recordings at −80 mV, we electrically evoked inhibitory postsynaptic currents (eIPSCs; in 10 μM DNQX)). eIPSC amplitude was used as an indicator of synaptic strength. We did not observe any significant alterations in cellular or synaptic properties between cells obtained from naïve (n = 142) versus stressed (n = 40) animals (Supplementary Fig. 1). Following 10 min baseline recording, we paired afferent, 10-Hz synaptic stimulation with subthreshold depolarization to −40 mV for 5 min: a protocol reminiscent of those used at various synapses to induce activity-dependent plasticity[14,15]. In naïve slices, pairing transiently suppressed of eIPSC amplitude (84.9 ± 6.6% baseline amplitude; Fig. 1b), which recovered quickly (104.4 ± 4.5% baseline at 30 min, Fig. 1b). Pairing in slices prepared immediately following the stress potentiated of eIPSCs (LTP: to 126.6 ± 10.2% baseline at 30 min one-sample t-test P = 0.039; Fig. 1c), similar to that observed by Inoue et al. (see accompanying paper). In slices from animals allowed to recover 30 or 60 min following the end of stress, we failed to observe any persistent changes to eIPSC amplitude (Fig. 1d–e). Further extending the post-stress recovery period to 90 min produced both an initial depression (42.4 ± 5.2%; P < 0.0001; Fig. 1f) and unmasked a long-term depression of eIPSC amplitude (LTDGABA) that persisted at least 30 min after pairing (69.4 ± 8.3% baseline; P = 0.0042). LTDGABA was not evident in naïve slices when pairing protocol duration was increased (92.7 ± 5.6%; P = 0.24; Supplementary Fig. 2f), suggesting that a threshold change does not underlie differences in responses between naïve and stressed rats. These results demonstrate that acute stress, with varied temporal delay, uncovers both conditional activity-dependent LTP and LTDGABA in PNCs.

Figure 1

Stress unmasks long–term plasticity of GABA synapses

a) Overview of experimental paradigm. b–f) Above: sample traces of eIPSCs recorded from individual PNCs either before (1) or 30 min after (2) a pairing stimulation protocol consisting of 5 minutes of afferent stimulation (10 Hz) and postsynaptic depolarization (to −40 mV). Below: summary graphs for each treatment group show normalized eIPSC amplitudes before and after pairing (Naïve: n = 14 cells, 10 rats; Stress + 0 min: n = 7 cells, 4 rats; Stress + 30 min: n = 4 cells, 3 rats; Stress + 60 min: n = 9 cells, 4 rats; Stress + 90 min: n = 10 cells, 7 rats). Scale bars are 50 pA/10 ms. Data expressed as mean ± s.e.m.

The temporal delay in unmasking of LTDGABA following acute immobilization is consistent with exposure to an in vivo associative signal, like CORT, which canonically has a slow onset of action compared with noradrenaline[2]. Consequently we tested whether activation of glucocorticoid receptors is an obligatory permissive factor for LTDGABA. Animals were given an intraperitoneal injection of either the glucocorticoid receptor antagonist RU-486 (25 mg·kg−1) or vehicle (DMSO) 15 min prior to immobilization and allowed to recover for 90 min afterwards (Fig. 2a). In vivo RU-486 pre-treatment completely prevented LTDGABA (102.9 ± 5.0% baseline; P = 0.58; Fig. 2b). Vehicle injection had no effect (67.8 ± 8.1% baseline; P < 0.0001). These data demonstrate that glucocorticoid receptor activation is necessary for stress-associated LTDGABA. They do not, however, provide information about anatomical specificity, nor do they indicate whether glucocorticoid receptor activation is sufficient in the absence of stress. Thus our next experiments probed the actions of local CORT administration to in vitro hypothalamic slices. Individual slices from naïve rats were incubated in CORT (100 nM) either with or without RU-486 (500 nM) for one hour followed by an additional 30-min recovery period prior to recording (Fig. 2c). As with stress, we did not observe any changes in basal cellular and synaptic properties in CORT-treated PNCs (Supplementary Fig. 1). We did observe LTDGABA in response to pairing in CORT-exposed cells (69.4 ± 4.9 % baseline; P < 0.0001; Fig. 2d). These changes were prevented by co-incubation with RU-486 (104.3 ± 5.5% baseline; P = 0.46). Next, we asked whether other stressors, which activate the HPA axis and elevate CORT, could also unmask LTDGABA. We observed LTDGABA in response to pairing in slices obtained 90 min following either forced swim or predator odor exposure (swim: 66.6 ± 9.2 % and predator: 73.8 ± 7.0 % baseline; P = 0.015 and P = 0.020; Fig. 2e–f). Together, these findings indicate that local glucocorticoid receptor activation in PNCs following stressful experience is necessary and sufficient to permit the induction of activity-dependent LTDGABA.

Figure 2

Glucocorticoid receptor activation is necessary and sufficient to unmask LTDGABA

a) Overview of experimental paradigm in which either the glucocorticoid receptor antagonist RU-486 (25 mg · kg−1) or vehicle were administered i.p. 15 min prior to stress. b) left: sample eIPSC traces from individual cells of stressed and RU-486− or stressed and vehicle-treated rats before and 30 min after pairing. Right: graph summarizes normalized eIPSC amplitudes in these groups (RU-486: n = 8 cells, 5 rats; vehicle: n = 7 cells, 3 rats). c) Overview of experimental paradigm in which hypothalamic slices from naïve rats were incubated in vitro with corticosterone (CORT; 100 nM) in the presence or absence of RU-486 (500 nM). d) left: sample eIPSC traces from individual cells of CORT− and CORT+RU-486- incubated slices before and 30 min after pairing. Right: graph summarizes normalized eIPSC amplitudes in these groups (CORT: n = 17 cells, 16 rats; CORT+RU-486: n = 7 cells, 4 rats). e,f) Left: sample eIPSC traces before and 30 min after pairing Right: Graphs summarizing normalized eIPSC amplitude in cells from rats exposed to forced swim (n = 6 cells, 3 rats) or predator odor (n = 5 cells, 5 rats) followed by a 90 min recovery period. Scale bars are 50 pA/10 ms. Data expressed as mean ± s.e.m.

We next probed for a locus (presynaptic vs. postsynaptic) of expression for LTDGABA. To assess GABA release probability (p) during these experiments, we examined variability in eIPSC amplitude (inverse squared coefficient of variation: CV−2) and the ratio between a pair of eIPSCs delivered in brief succession (paired-pulse ratio: PPR). 30 min after pairing, CV−2 was reduced in stressed cells (to 59.7 ± 10.8% baseline; P = 0.004; Fig. 3c), but remained unchanged in naïve cells (118.3 ± 15.2% baseline; P = 0.26). PPR was unchanged by pairing in naïve cells (99.8 ± 4.7% baseline; P = 0.97; Fig. 3d), but it was significantly increased in stressed cells (118.2 ± 6.8 % baseline, P = 0.021; from 0.62 ± 0.14 to 0.75 ± 0.21 un-normalized PPR; P = 0.009 paired t-test). Next we analyzed the inter-event interval/frequency and amplitude of spontaneous IPSCs (sIPSCs) from these recordings (Fig. 3e–g). sIPSC frequency decreased in cells from stressed (74.7 ± 4.6% baseline; P < 0.0001; Fig. 3e–f) but not naïve animals (106.9 ± 9.0% baseline; P = 0.46). sIPSC amplitude remained unchanged in both conditions (Fig. 3g). These data are consistent with decreased presynaptic release during LTDGABA. Similarly, we noted that LTD in CORT treated slices was also accompanied by an increase in PPR (to 119.2 ± 6.5% baseline; P = 0.010; Supplementary Fig. 2a), a decrease in CV−2 (62.7 ± 7.4% baseline; P = 0.0002; Fig. 3h), and a reduction in sIPSC frequency, but not amplitude (frequency to 77.5 ± 6.9% baseline; P = 0.016; Supplementary Fig. 2b–e). Indeed, across in vivo and in vitro experimental conditions, changes to CV−2 were consistently related to changes in eIPSC amplitude (Fig. 3h). Taken together, our data strongly indicate that glucocorticoid-associated LTDGABA is a consequence of a decrease in presynaptic GABA p.

Figure 3

LTDGABA is expressed presynaptically

a–b) Sample paired-pulse traces and amplitude time courses of eIPSCs before and after pairing from individual cells in slices prepared from naïve or stressed rats. Scale bars are 50 pA/10 ms. c–d) Summary graphs of normalized eIPSC variability (CV−2) and paired pulse-ratio (PPR) responses to the pairing protocol in cells from naïve (n = 14 cells, 10 rats) and stressed rats (n = 10 cells, 7 rats). e) Sample traces of sIPSCs and corresponding cumulative probability distribution plots of inter-event interval and amplitude before and 30 min following pairing in a cell from a stressed rat. Scale bars 20 pA/0.25 s. f–g) Summary graphs of normalized sIPSC frequency and amplitude response in naïve and stressed cells. h) Relationship between eIPSC amplitude and CV−2 changes at 30 min. Data expressed as mean ± s.e.m.

LTDGABA is induced heterosynaptically

Since electrical stimulation of synaptic inputs could recruit axons non-specifically, we used an optogenetic tool to test whether exclusive activation of GABA synapses was sufficient for LTD induction. Using CORT-treated slices from vGAT–mhChR2–YFP mice expressing channelrhodopsin under the vesicular GABA transporter promoter, we found that pairing delivered with light-evoked stimulation did not elicit LTDGABA, while electrical stimulation in wild-type mice did(ChR2: 119.6 ± 12.3% baseline, electrical: 74.5 ± 5.8% baseline; P = 0.17 and P = 0.003 respectively; Fig. 4a).

Figure 4

LTDGABA induction is heterosynaptic and requires a retrograde signal

a) Response to pairing in CORT-treated slices with exclusive activation of GABA synapses in vGAT-ChR2 mice (open circles; n = 6 cells, 3 mice). Comparison with electrical stimulation in slices from WT mice (filled circles; n = 8 cells, 5 mice). Sample eIPSCs from individual PNCs are shown to the left and a summary graph to right. b) Effects of mGluR5 antagonist MTEP (10μM; n = 5 cells, 3 rats) or c) mGluR1 antagonist JNJ16259685 (750 nM; n = 6 cells, 5 rats) on LTD. Prevention of LTDGABA by inclusion, in the patch pipette of: d) GTP-ase inhibitor GDPβs (2 mM; n = 6 cells, 3 rats), or e) calcium chelator, BAPTA (10 mM; n = 5 cells, 4 rats). f) Effect of bath application of the L-type calcium channel blocker nimodipine on LTDGABA (20 μM; n = 6 cells, 4 rats). g–h) Effects of CB1R-antagonist AM251, in rats, or genetic deletion of the CB1R (CB1R), in mice, on LTDGABA with eIPSC traces (left) and summary time course (right) (g: 3 μM; n = 8 cells, 5 rats or h: n = 5 cells, 3 mice). i) Prevention of LTDGABA by intrapipette inclusion of SNARE-dependent exocytosis inhibitor BoNT/C (5μg · mL−1; n = 6 cells, 3 rats). Control LTDGABA replotted in filled grey squares (rat) or circles (mouse). Scale bars are 50 pA/10 ms. Data expressed as mean ± s.e.m.

Metabotropic glutamate receptors (mGluRs) are important for GABA synapse plasticity requiring heterosynaptic induction[14-16]. We conducted experiments to test for the mGluR contributions in LTDGABA. In CORT treated slices we failed to induce LTDGABA in the presence of the non-selective group I/II mGluR antagonist MCPG (200 μM, 97.9 ± 7.5% baseline; P = 0.93; Supplementary Fig. 3a). We next tested group I mGluR subtypes 1 and 5. Treatment with mGluR5 antagonist MTEP (10 μM) completely abolished LTDGABA in CORT treated slices (eIPSC: 101.0 ± 5.0% baseline; n = 5; P = 0.84; Fig. 4b). By contrast, inclusion of selective mGluR1 antagonist JNJ-16259685 (750 nM) did not prevent LTDGABA (73.2 ± 7.6% baseline; P = 0.017; Fig. 4c). Preventing activation of NMDA receptors with intracellular MK801 (1 mM) also failed to impact the expression of LTDGABA (73.6 ± 5.0% baseline; P = 0.006; Supplementary Fig. 3b). These data indicate that group I mGluRs, in particular mGluR5, are part of a heterosynaptic mechanism involved in LTDGABA following GC exposure.

A postsynaptic, vesicle-based retrograde signal mediates LTDGABA

We next tested whether the mGluR responsible for induction of LTDGABA is postsynaptic. We interfered with G-protein signaling only in the postsynaptic PNC by including the non-hydrolysable GDP analogue GDPβs (2 mM) in the intrapipette solution. Under these conditions we failed to observe LTDGABA in CORT treated cells (104.8 ± 7.7% baseline; P = 0.56; Fig. 4d). Since mGluR5 is coupled to Gαq–type intracellular pathways and exerts many effects through elevations in intracellular calcium, we next assessed the effect of fast calcium buffer BAPTA (10 mM), also intrapipette. This, too, prevented expression of LTDGABA (102.0 ± 8.1% baseline; P = 0.81; Fig. 4e). Since postsynaptic depolarization was necessary LTDGABA induction, we tested the involvement of voltage-dependent calcium channels. Consistent with this idea, the L-type calcium channel antagonist nimodipine prevented LTDGABA (99.5 ± 12.3 % baseline; P = 0.97; Fig. 4g). These results provide evidence that a post-synaptic mGluR- and calcium-dependent signaling pathway is required for LTDGABA following CORT exposure.

LTDGABA requires heterosynaptic activation of postsynaptic mGluR5, but manifests as a presynaptic decrease in release probability, suggesting the presence of a retrograde signal. One widely described form of mGluR-dependent LTDGABA requires retrograde signaling by endocannabinoids (eCBs)[16-18]. We have previously characterized short-term retrograde eCB signaling at GABA synapses onto PNCs[7] indicating that eCBs are functional at these synapses; since we found that short-term eCB signaling is enhanced by acute exposure to CORT, we hypothesized that recruitment of eCBs and activation of CB1Rs may contribute to GC-LTDGABA. Following exposure to CORT, slices were incubated in aCSF containing CB1R antagonist AM251 (3 μM), for a minimum of 30 min. CB1R blockade, however, failed to prevent LTDGABA (72.6 ± 3.2 % baseline; P < 0.0001; Fig. 4g). To further test this idea, we using genetic deletion assessing LTDGABA in mice lacking CB1Rs (CB1R). We found that that LTDGABA persisted in CORT treated slices from CB1R mice (72.7 ± 5.1 % baseline; P = 0.0032; Fig. 4h). A TRPV antagonist, capsazepine, also failed to prevent LTDGABA (66.4 ± 8.5 % baseline; P = 0.016; Supplementary Fig. 3c). Based on these data, we conclude that eCBs are not the retrograde signal responsible for expression of LTDGABA at these synapses.

In addition to lipid-derived retrograde messengers, neurons, including PNCs[19], also release conventional and peptide transmitters that are packaged in vesicles in the somatodendritic compartment[17]. To test for the contribution of a vesicularly packaged retrograde transmitter, we conducted experiments in which the soluble NSF attachment protein receptor (SNARE)-dependent exocytosis inhibitor botulinum toxin C (BoNT/C: 5 μg·ml−1) was included in the patch pipette. Inclusion of BoNT/C prevented LTDGABA following pairing (105.3 ± 7.0% baseline; P = 0.49; Fig. 4i). Collectively, these observations indicate that LTDGABA requires activation of postsynaptic mGluRs, an increase in intracellular calcium and the fusion of neurotransmitter-filled vesicles postsynaptically. Given that these events underlie presynaptic reduction of GABA release, a retrograde signal is likely recruited by this mechanism.

Glucocorticoids alter mGluR signaling via RGS4

We next tested whether pharmacological activation of mGluRs was sufficient to recapitulate suppression of GABA transmission, and whether this mechanism was altered by glucocorticoid exposure. As LTDGABA requires high voltage activated L-type calcium channels and is evident only when afferent stimulation and depolarization to −40 mV are paired together (Depol. Alone: 118.3 ± 10.8% baseline, P = 0.15; Stim. Alone: 92.6 ± 6.8% baseline, P = 0.31; Supplementary Fig. 2g–h), we tested the hypothesis that LTDGABA results from membrane state-dependent activation of mGluRs. We performed recordings of eIPSCs at either −40 mV, or −80 mV and bath-applied the group I mGluR agonist DHPG (100 μM) for 5 min. At −40 mV eIPSCs are outward currents; we lowered intracellular chloride (4 mM) to increase the inward driving force through the GABAA receptor. We first confirmed that LTDGABA was still readily observed with reversed chloride driving force (70.4 ± 9.2% baseline; P = 0.02; Supplementary Fig. 2i). Surprisingly, DHPG potentiated eIPSC amplitude in naïve slices under these conditions (132.3 ± 11.6% baseline at 10 min; P = 0.049; Fig. 5a). By contrast, in CORT-treated slices, DHPG elicited long lasting depression of eIPSCs (63.0 ± 4.7 % baseline; P = 0.0006; Fig. 5a), which was accompanied by increased PPR (119.6 ± 2.7% baseline; P = 0.0020; Fig. 5b) and a decrease in eIPSC CV−2 (49.5 ± 7.6% baseline; P = 0.0012; Fig. 5b). Similar results were obtained at −80 mV; following CORT treatment DHPG no longer enhanced eIPSC amplitude, as it did in naïve cells, although no significant depression was observed (76.6 ± 10.6% baseline; P = 0.069; Supplementary Fig. 2k). From these data we conclude that mGluR activation, at a depolarized membrane potential is sufficient to recapitulate LTDGABA. Furthermore, CORT exposure unmasks LTDGABA by functionally altering the outcome of mGluR signaling.

Figure 5

Glucocorticoid receptor activation modifies mGluR signaling via RGS4

a) Suppression of GABA transmission by mGluR agonist after glucocorticoid exposure. Left: sample eIPSC traces at a holding potential of −40 mV from individual naïve and CORT-treated cells before and 30 min after bath-applied mGluR1/5 agonist DHPG (100 μM, 5 min). right: summary of the effects of DHPG on eIPSC amplitude at −40 mV in naïve (n = 5 cells, 3 rats) and CORT-treated (n = 6 cells, 4 rats) slices. b) Above: summary time course graph of normalized PPR in response to DHPG in naïve or CORT-treated slices. Below: relationship between eIPSC amplitude and CV−2 changes 30 min post-DHPG for each treatment group. c) mGluR supression of GABA transmission in naïve slices with inhibition of post-synaptic RGS4. eIPSC traces from an individual cell (left) and summary data (right) from cells in a naïve slice treated with DHPG when the RGS4 inhibitor CCG63802 (100 μM) is included in the patch pipette (n = 6 cells, 4 rats). d) mGluR modulation of GABA transmission in CORT-treated slices with post-synaptic RGS4. eIPSC traces from an individual cell (left) and summary data (right) from cells of CORT-treated slices in which the pipette solution contained recombinant RGS4 (50 pM; n = 6 cells, 3 rats). Scale bars are 50 pA/10 ms. Data expressed as mean ± s.e.m.

We next sought to examine how glucorticoids alter mGluR signaling. Regulator of G-protein signaling (RGS) proteins, in particular RGS4, associate with group I mGluRs and stifle Gq mediated signaling through GTP-ase acceleration[20]. RGS4 is abundantly expressed in the PVN and potently down-regulated by stress/glucocorticoid receptor activation[21,22]; this provides a compelling and testable potential mechanism. To test the hypothesis that RGS4 restrains mGluR signaling in naïve PNCs, we included the RGS4 inhibitor CCG63802 (100 μM) in the pipette solution and bath applied DHPG. Postsynaptic inhibition of RGS4 was sufficient to unmask a DHPG-mediated LTDGABA that was similar to that seen with CORT treatment (to 71.2 ± 7.0 % baseline; P = 0.0093; Fig. 5c). We next conducted the corollary experiment and included recombinant RGS4 in the patch pipette when recording from cells in CORT-treated slices. This completely prevented eIPSC depression following DHPG (132.0 ± 11.3 %; P = 0.036; Fig. 5d). These data indicate that RGS4 downregulation by glucocortioids is sufficient to enhance mGluR5 signaling and allow for the expression of LTD.

Persistent μ-opioid receptor signaling underlies LTDGABA

PVN neuroendocrine cells release neurotransmitters from vesicles in their somatodendritic compartment[19,23]. Opioid peptides released from magnocellular neurosecretory cells (MNCs) cause presynaptic LTD at glutamate synapses[11,24]. PNCs produce many peptides in a stress-dependent manner; this includes pro-enkephalin opioid gene products such as met-/leu-enkephalin[25-27]. We hypothesized that vesicular somatodendritic release of an opioid peptide is responsible for LTDGABA following CORT exposure. In CORT-treated slices, continuous bath application of the broad-spectrum opioid receptor (OR) antagonist naloxone (5 μM) prevented pairing induced depression of eIPSC amplitude (100.7 ± 7.9% baseline; P = 0.93; Fig. 6a). Naloxone also prevented LTDGABA associated changes to PPR (95.5 ± 4.4% baseline; P = 0.33), CV−2 (131.1 ± 16.4 %; P = 0.10), and sIPSC frequency (112.4 ± 13.3%; P = 0.38). Similarly, the μOR subtype antagonist CTAP (1 μM) prevented LTDGABA (99.0 ± 9.3% baseline; P = 0.92; Fig. 6b). Neither the δOR antagonist (Naltrindole; 1 μM) nor the κORs antagonist (nor-Binaltorphimine; 1 μM) prevented LTDGABA following pairing (71.1 ± 8.2 % and 71.5 ± 5.6 % baseline respectively; P = 0.017 and P = 0.0037; Supplementary Fig. 4c). We did, however, note suppressive effects of a κOR agonist U69593 (1 μM), but not δOR agonist DPDPE (1 μM) on eIPSC amplitude (U69593: 21.7 ± 7.6% baseline, P = 0.0005; DPDPE: 100.8 ± 5.8% baseline, P = 0.09; Supplementary Fig. 4a–b). These pharmacological data suggest that μ-ORs are necessary for induction of LTDGABA following CORT exposure. Finally, we assessed LTD in μOR mice,[28]. We failed to observe any lasting depression of eIPSCs (106.4 ± 9.3 % baseline; P = 0.52; Fig. 6c). These data confirm that μORs are necessary for LTDGABA.

Figure 6

Presynaptic μ-type opioid receptors mediate LTDGABA

a,b) Effect of μOR antagonism on LTDGABA. eIPSC traces (left) and summary time course (right) showing the effects of pairing in CORT-treated slices in the presence of non-specific OR antagonist naloxone (5 μM; n = 8 cells, 5 rats) or μOR antagonist CTAP (1 μM, n = 6 cells, 4 rats). c) Effect of genetic deletion of μORs on LTD. eIPSC traces (left) and summary time course (right) shows the effects of pairing in CORT-treated slices from mice lacking μORs (n = 6 cells, 3 mice). d) Sample recording from an individual PNC (left) and summary graphs (right) of the frequency and amplitude of mIPSCs recorded in TTX (1 μM) showing the reduction of mIPSC frequency elicited by μOR agonist DAMGO (1 μM; n = 9 cells, 7 rats). e) Occlusion of LTD by μOR agonist DAMGO. eIPSC traces (above left) and eIPSC amplitude time course (below left) from a single neuron (CORT-treated) during baseline recording, following bath perfusion of μOR agonist DAMGO (500 nM), and 25 min after pairing. Summarized time course graph (right) showing the effects of pairing on normalized eIPSC amplitude (above) and PPR (below; n = 7 cells, 4 rats) following DAMGO treatment. Control LTDGABA re-plotted in filled grey squares (rat) or circles (mouse). Scale bars are 50 pA/10ms in a–d, g and 25 pA/0.5 s in e. data expressed as mean ± s.e.m.

μOR subtypes are commonly located on GABA neurons and their synaptic terminals[29-31]. If an endogenous opioid were released from PNCs, its actions would likely be spatially restricted to local terminals, particularly since cell bodies of afferent GABA neurons reside outside the PVN[4]. In line with this hypothesis, we found that DAMGO (1 μM) significantly reduced the frequency of miniature IPSCs (1 μM TTX; from 2.1 ± 0.6 to 0.7 ± 0.2 events · sec−1; P = 0.0002 paired t-test; Fig. 6d) but not their amplitude (21.6 ± 1.1 pA before, 21.6 ± 0.7 pA after; P = 0.99), suggesting that μORs on terminals contacting PNCs regulate GABA p. Next we asked whether activation of presynaptically-located μORs would occlude the induction/expression of LTD. In CORT-treated cells, DAMGO (500 nM) depressed eIPSC amplitude (48.4 ± 8.4% baseline; PPR to 136.1 ± 11.5%; Fig. 6e). Once eIPSC amplitude had stabilized, we delivered the pairing protocol. This had no additional effect on either eIPSC amplitude or PPR. At 25 min after pairing, eIPSC amplitude was 52.8 ± 4.5% of pre-DAMGO baseline (paired t-test P = 0.48 vs. pre-DAMGO) and PPR: 133.8 ± 9.8% (paired t-test P = 0.89). We failed to observe significant changes to PNC holding current during DAMGO treatment (pooled 500 nM − 1 μM DAMGO; Ihold before: −17.3 ± 2.1 pA, after: −13.7 ± 2.7 pA; n = 27; paired t-test P = 0.13; not shown). Taken together these results demonstrate that μORs located at synaptic terminals suppress GABA release, and that their activation by an exogenous ligand occludes subsequent induction of LTDGABA.

Although necessary for LTDGABA expression, it is not clear whether μOR activation is necessary for its maintenance. We applied OR antagonist/inverse agonist naloxone (5 μM) 20 min following either DAMGO or induction of LTDGABA by the pairing protocol. Transient μOR activation by DAMGO (1 μM, 7 min) caused a long-lasting depression of eIPSCs (65.6 ± 9.0 % baseline at 35 min; P = 0.0088; Fig. 7a). This depression was completely reversed by naloxone (111.7 ± 14.1 % baseline at 35 min; P = 0.44; Fig. 7a). These results, suggest that transient μOR activation is capable of eliciting a long-lasting synaptic change, that requires persistent OR signaling. Next, following pairing, we established that eIPSC amplitude was suppressed (60.9 ± 10.4 % baseline; P = 0.0093; Fig. 7b). Subsequent application of naloxone caused a recovery of eIPSCs to near-baseline level (108.5 ± 12.9 % baseline at 35 min; P = 0.53; Fig. 7a–b). PPR also returned to baseline (146.1 ± 16.8 % at 20 min; P = 0.033; to 106.4 ± 7.3 % at 35 min; P = 0.42; Fig. 7b). This was not due to pre-existing OR tone as naloxone application to CORT-treated PNCs (in the absence of pairing) had no effect on eIPSC amplitude (108.0 ± 4.0 % baseline; n = 5; P = 0.12; data not shown). In summary, LTDGABA requires the μOR for both expression and maintenance of suppressed GABA release. This could be due to either persistent effects of μOR activation and/or sustained vesicular release of the opioid peptide.

Figure 7

LTDGABA is reversible by OR antagonism

a) Reversal of μOR agonist suppressed transmission by OR antagonist chase. Left: Sample eIPSC traces and right: plot of eIPSC amplitudes from a neuron treated with 1 μM DAMGO for 7 min (above) and another neuron with DAMGO treatment followed by naloxone (5 μM) at 10 min (below). Summary graph (right) showing effects of DAMGO on eIPSC amplitude alone (n = 7 cells, 6 rats) or followed by naloxone (n = 6 cells, 5 rats). b) Reversal of LTD by an OR antagonist. Sample eIPSC traces (above) and plot of eIPSC amplitude (below) taken from an individual cell in CORT-treated slices subjected to pairing followed by naloxone 20 min later. Right: Summary of the effects of naloxone (5 μM) applied following induction of CORT-LTD on eIPSC amplitude (above) and PPR (below) from n = 7 cells, 6 rats. Scale bars are 50 pA/10 ms. Data expressed as mean ± s.e.m.

LTDGABA does not display synapse specificity

Opioid release and signaling may occur across the entire somatodendritic axis, or alternatively at locally recruited segments of the dendrite. Furthermore, presynaptic activity or μOR expression could be restricted to certain inputs. Thus we probed whether LTDGABA exhibited synapse specificity. Given that mIPSC frequency is sensitive to the μOR agonist DAMGO and that sIPSCs are also suppressed during LTDGABA, we hypothesized that release, spread, and/or efficacy of endogenously release opioids would not be limited to synapses active during pairing. To test this, we electrically activated two distinct GABAergic inputs onto PNCs, s1 and s2, verifying their independence by confirming that the synaptic strength and release probability of one pathway was unaffected by recruiting the other pathway. Delivering the 10Hz stimulation during pairing through s1, depressed eIPSC amplitude at both s1 and s2 inputs (s1: 64.7 ± 4.4% baseline, s2: 71.8 ± 10.2% baseline; P = 0.005 and P = 0.039 respectively; Fig. 8a).

Figure 8

LTDGABA is not synapse specific

a) Left: a schematic of experiment in which one PNC is recorded and two independent synaptic inputs (S1 and S2) onto the cell are stimulated. Only S1 is activated during the pairing protocol. eIPSC traces and eIPSC amplitude graph (center) from two inputs onto a single neuron before and after pairing. Right: summary data from this experiment shows pairing-induced depression of eIPSC amplitude of S1 and S2 (n = 6 cells, 4 rats). b–c) OR-mediated CORT-LTD at excitatory synapses. b) Left: eEPSC traces isolated by picrotoxin (100 μM) before and 30 min after pairing in a cell from a CORT-incubated slice (above) and a cell recorded in the presence of naloxone (below). Right: Summary time course of the effects of pairing on eEPSC amplitude with (n = 6 cells, 5 rats) or without naloxone (n = 7 cells, 5 rats). c) plot of normalized PPR response to pairing (above) and relationship between eEPSC and CV−2 (below) in CORT-treated cells with or without naloxone. Scale bars are 50 pA/10 ms. Data expressed as mean ± s.e.m.

Finally, given this finding and with demonstrations that GABA and glutamate synapses on PNCs are intermingled[32], we hypothesized that somatodendritically released opioids may also depress glutamate synapses. First, we tested for the presence of functional μORs at glutamate synapses. In slices incubated in vitro with CORT, evoked excitatory post-synaptic currents (eEPSCs) were suppressed by DAMGO (40.8 ± 6.2% baseline; P < 0.0001; Supplementary Fig. 4d). Next, we applied the pairing protocol used above and observed a long-lasting depression of glutamate transmission. eEPSC amplitude at 30 min was suppressed to 62.3 ± 9.4% baseline (P = 0.0073; Fig. 8b), which was accompanied by an increased PPR (125.6 ± 6.6% baseline; P = 0.0081) and a decrease in CV−2 (57.0 ± 15.0% baseline; P = 0.028), suggesting a presynaptic locus of expression. Naloxone completely prevented expression of LTD (102.4 ± 9.0% baseline; P = 0.802; Fig. 8b), changes in PPR (100.2 ± 9.9% baseline; P = 0.99) and changes in CV−2 (92.8 ± 12.6% baseline; P = 0.59). These results demonstrate that LTD mediated by ORs in PNCs following GC exposure occurs in a synapse-independent fashion.

Discussion

Here we show that glucocorticoids, elevated in response to a stress experience, are instructive signals in the hypothalamus that allow for subsequent correlated synaptic and cellular activity to suppress GABA p. By suppressing RGS4 in PNCs, glucocorticoids functionally alter the outcome of post-synaptic mGluR signaling during synaptic stimulation culminating in calcium-dependent vesicle exocytosis and the liberation of a retrograde opioid signal from the somatodendritic compartment. Activation of presynaptic μORs is necessary for the expression and maintenance of decreased neurotransmitter release, implicating an endogenous opioid as the most likely candidate for this retrograde signal.

Glucocorticoid-associated LTDGABA requires heterosynaptic recruitment of mGLuR5 located on PNCs themselves. This finding is consistent with reports of enhanced mGluR1/5 signaling following stress/CORT exposure[33]. Importantly, pairing of afferent stimulation with a postsynaptic depolarization was necessary for LTDGABA suggesting that Gq-linked mGluRs in our system may behave as voltage-dependent “coincidence detectors”[34,35]. Membrane depolarization has been shown to amplify mGluR signaling by enhancing contributions of voltage-gated calcium channels[36] which can synergize with and sustain calcium sourced by mGluRs from IP3-receptor gated stores[34]. Although the mechanisms regulating somatodendritic exocytosis are not well defined[23], neuronal activity and Gq-coupled receptors cooperatively drive calcium-dependent dendritic peptide release[37]. For example, synaptic mGluR activation during burst firing in MNCs[27] and L-type channels in dentate granule cells play important roles in dendritic release of the opioid dynorphin[24,38]. In accordance with these previous studies, our findings indicate that calcium entry through L-type voltage-gated calcium channels is obligatory for LTDGABA. While somatodendritic vesicular release from PNCs can also occur following calcium influx through NMDARs[19], we found that LTDGABA persisted after NMDAR blockade.

Our data suggest that μOR activation is necessary for expression of LTDGABA. Intriguingly, we also found that ongoing OR activation is required for LTD maintenance, which is unconventional as an expression mechanism for long-term plasticity. μORs are functionally expressed within the PVN, and influence PNC activity and HPA function in a stress-state dependent manner[10,39,40]. In other brain regions μORs are widely expressed on GABAergic neurons and terminals[31]. μOR agonists hyperpolarize inhibitory neurons[29-31] and interfere with inhibitory synapse plasticity[41]. Agonist activation of μORs locally expressed at synaptic terminals also suppresses GABA p[42] and can induce LTD at both GABA and glutamate synapses[43,44]. In spite of this, there are only a few demonstrations of functional synaptic actions of endogenously produced and retrograde acting opioids[11,12]. One might conjecture that a likely candidate for the endogenous μOR ligand produced by PNCs and mediating LTDGABA is an enkephalin-like peptide. PNC enkephalins are a compelling candidate for experience-dependent control of neuroendocrine function and adaptation. Proenkephalin transcripts are incrementally upregulated by acute and repeated stress[45] in a glucocorticoid-dependent manner[46,47]. Notably, proenkephalin is also increasingly colocalized with c-fos and/or CRH following stressful conditions[9,48], suggesting that enkephalin-containing neurons may be relevant to stress-related PNC plasticity, and that enkephalin-derived peptides may exist in PNCs as adaptogenic signaling molecules.

Although LTDGABA reported here is not mediated by eCBs, it shares many similarities with eCB-LTD, which also occurs at synapses throughout the brain[16,18]. Gq-coupled metabotropic receptor activation is a strong stimulus for eCB production, and required for eCB-LTD[14-16,18]. Additionally, glucocorticoids enhance both eCB-mediated short- and long-term plasticity at GABA synapses[7,49]. We found that the switch in mGluR signaling necessary for LTDGABA following CORT exposure is likely RGS4, a molecule which has recently been shown to regulate eCB-LTD through gating mGluR signaling in the striatum[50]. Despite these common features, our experiments indicate that LTDGABA occurs independently of CB1Rs and, to our knowledge, is the first demonstration of an eCB-independent presynaptic LTD at mature GABAergic synapses.

PNC activity is known to be a function of both synaptic drive and circulating glucocorticoid levels. The CORT actions we observe here emerge within the time period classically defined as the “delayed” domain of glucocorticoid feedback[3]. During this time, endocrine responses to any subsequent stressors are blunted in proportion to the levels of CORT produced by the first exposure[3]. This period conforms with the time estimated for both the entry of CORT into the brain[13] and slow emergence of genomic glucocorticoid receptor-dependent actions[3]. Since GABA transmission onto PNCs during stress is excitatory[5,8], we propose that a retrograde opioid suppression of both GABA and glutamate release during a sustained period of PNC activity represents a synaptic correlate of the glucocorticoid-induced “refractory period” imposed onto PNCs[3]. This mechanism may act to mask or compete with the “priming” mechanisms imparted to PNCs during a stress[3,19]. One such mechanism, set in place by the metaplastic actions of the other major stress mediator noradrenaline, is detailed in the accompanying study by Inoue et al. Our findings, together, provide mechanistic underpinnings for bidirectional synaptic adaptations that can occur during different temporal windows after a single stress experience. We observed that these two forms of plasticity also exhibit different thresholds for induction. For example, unlike LTPGABA reported here at 0 min after stress, and extensively detailed by Inoue et al, LTDGABA was only evident following a relatively longer period of sustained synaptic and postsynaptic activity. While speculative, given the paucity of data regarding firing patterns of PNCs or their afferents during in vivo stress, this induction requirement suggests LTDGABA may preferentially serve a homeostatic function, imposing a ceiling on HPA activation and limiting systemic exposure to pathological levels of glucocorticoids during prolonged periods of stress. Together, our two studies suggest that polarity of synaptic metaplasticity on PNCs is a function of the time domain over which the body’s two principal stress mediators elicit their actions, and hint at the complex dynamics that allow stress circuits to respond and evolve with experience.

Methods

Animal handling and stress procedure

All protocols were approved by the University of Calgary Animal Care & Use Committee, in accordance with the Canadian Council for Animal Care. Group-housed juvenile male Sprague Dawley rats (postnatal day 22–31, Charles Rivers), wild-type C57BL6/J (Jackson Laboratories), μOR (Jackson Stock #007559), vGAT–mhChR2–YFP BAC transgenic (Jackson Stock #014548), or CB1R (From Dr. K. Sharkey) mice (bred to C57BL6/J background, postnatal day 28–50) were kept on a 12:12 light dark cycle with ad libitum access to food and water. Stress was carried out 2–3 hours after the onset of light during the trough of circadian fluctuation in plasma CORT. Immobilization stress consisted of cervical and caudal immobilization and confinement within a plastic cylinder for 30 min. Forced swim stress was carried out for 20 min in a plastic bucket (40 cm internal diameter) and 30–32 °C water at a depth where the bottom could not be touched by the rat. To expose rats to predator odor they were placed in an empty cage for 30 min with a tissue soaked with 2,5-dihydro-2,4,5-trimethylthiazoline (TMT, Contech), a compound isolated from fox feces[19]. In some experiments, an intraperitoneal injection of RU-486 (25 mg·kg−1) or DMSO vehicle preceded stress by 15 min. Following stress, the rat was placed alone, in a fresh cage, until slice preparation.

Slice preparation and electrophysiology

Animals were anesthetized with isoflurane and decapitated. The brain was quickly removed; it was submerged and coronally sectioned on a vibratome (Leica) to 300 μM in slicing solution (0°C, 95% O2/5% CO2 saturated) containing (in mM): 87 NaCl, 2.5 KCl, 0.5 CaCl2, 7 MgCl2, 25 NaHCO3, 25 D-glucose, 1.25 NaH2PO4, 75 sucrose. After placement into aCSF (30 °C, 95% O2/5% CO2 saturated) containing (in mM):126 NaCl, 2.5 KCl, 26 NaHCO3, 2.5 CaCl2, 1.5 MgCl2, 1.25 NaH2PO4, 10 glucose, hypothalamic slices recovered for at least 1 hour. Subsequently, some slices were placed for 1 hour into aCSF containing 100 nM corticosterone and/or 500 nM RU-486 (Sigma; final DMSO vehicle: < 0.0001%). Once transferred to a recording chamber superfused with aCSF (1 mL·min−1; 30–32 °C; 95% O2/5% CO2), slices were visualized using an AxioskopII FS Plus (Zeiss) upright microscope fitted with infrared differential interference contrast optics. Pulled borosilicate glass pipettes (3–6 MΩ) were filled with a solution containing (in mM) 108 K-gluconate, 2 MgCl2, 8 Na-gluconate, 8 KCl, 1 K2-EGTA, 4 K2-ATP, 0.3 Na3-GTP, and 10 mM HEPES. In indicated experiments KCl was reduced to 4 mM or the following were added: 10 mM 1,2-Bis(2-aminophenoxy)ethane-N,N,N′,N′-tetraacetic acid (BAPTA; Sigma), 5μg·mL−1 Botulinum Neurotoxin Type C (Light Chain Recombinant BoNT/C; List Biological), 1mM (+)-5-methyl-10,11-dihydro-5H-dibenzo[a,d]cyclohepten-5,10-imine maleate (MK801), 2 mM GDPβs (Na3GTP-free solution), CCG63802 (Tocris), or recombinant RGS4 (Genway). All other drugs were bath applied by perfusion pump. MCPG, MTEP, JNJ 16259685, capsazepine, and DHPG were obtained from Tocris, [D-Pen2,5] Enkephalin, [D-Pen2,D-Pen5]Enkephalin (DPDPE) was from Bachem, and nimodipine. Picrotoxin, U69593, [D-Ala2, NMe-Phe4, Gly-ol5]-enkephalin (DAMGO), CTAP, naltrindole, and naloxone were from Sigma.

Whole-cell patch-clamp recordings were performed from PNCs identified by location, morphology, and current clamp fingerprint, as previously described[5,7,19]. Of the 2–4 PVN slices obtained from each animal, one cell was recorded per slice. Slices were randomly assigned to treatment/no-treatment groups; a minimum of 2 cells per litter were used as no-treatment control. Each group consists of data obtained from at a minimum 3 animals from 2 different litters. Sample sizes were determined post-hoc based on those used in previous studies[5,7,19]. Experimenters were not blinded to treatment. PNCs were voltage-clamped at −80 mV with constant perfusion of 6,7-dinitroquinoxaline-2,3-dione (DNQX; 10 μM; Tocris) or picrotoxin (100 μM; Sigma). Pairs of post-synaptic currents (IPSCs) were evoked 50 milliseconds apart at 0.2 Hz intervals using a monopolar aCSF-filled glass electrode placed about 25 to 50 μm ventromedially from the recorded cell. To activate ChR2, a fiber optic cable (105 μm core diameter) was placed 1–2 mm from the PVN and a blue-light laser (473 nm, OptoGeni 473, IkeCool corporation) delivered 3–5 millisecond light pulses at 0.2Hz. The protocol used to elicit LTD consisted of 10 Hz synaptic stimulation paired with a voltage-clamp step to −40 mV for 5 min. Access resistance was continuously monitored; recordings in which values exceeded 20 MΩ or 15% change were excluded from analysis.

Data analysis and statistics

Signals were amplified (Multiclamp 700B, Molecular Devices), low pass filtered at 1 kHz, digitized at 10 kHz (Digidata 1322, Molecular Devices), and recorded (pClamp 9.2, Molecular Devices) for offline analysis. PSC amplitudes were calculated by subtraction of peak synaptic current from pre-stimulation baseline current. sIPSC events, with eIPSCs and stimulus artifacts removed, were detected using variable thresholds and confirmed by eye (MiniAnalysis, Synaptosoft). For each cell, mean eIPSC/eEPSC amplitude, paired-pulse ratio (2nd evoke/1st evoke), or sIPSC event frequency/amplitude obtained over a 2-min recording interval were normalized and expressed as a percent of baseline recording values. Coefficient of variation (CV−2) was analyzed with a 5-min interval, and expressed as percent baseline. Gaussian distribution of the data was confirmed by a D’Agostino & Pearson omnibus normality test (GraphPad Prism 4). A one-sample t-test (vs. 100%) was used to assess deviation in normalized values from baseline, and a paired two-tailed student’s t-test (where stated) to assess deviation in non-normalized values. P < 0.05 was considered the level of statistical significance.