NMDA-receptor-dependent plasticity in the bed nucleus of the stria terminalis triggers long-term anxiolysis

Anxiety is controlled by multiple neuronal circuits that share robust and reciprocal connections with the bed nucleus of the stria terminalis (BNST), a key structure controlling negative emotional states. However, it remains unknown how the BNST integrates diverse inputs to modulate anxiety. In this study, we evaluated the contribution of infralimbic cortex (ILCx) and ventral subiculum/CA1 (vSUB/CA1) inputs in regulating BNST activity at the single-cell level. Using trans-synaptic tracing from single-electroporated neurons and in vivo recordings, we show that vSUB/CA1 stimulation promotes opposite forms of in vivo plasticity at the single-cell level in the anteromedial part of the BNST (amBNST). We find that an NMDA-receptor-dependent homosynaptic long-term potentiation is instrumental for anxiolysis. These findings suggest that the vSUB/CA1-driven LTP in the amBNST is involved in eliciting an appropriate response to anxiogenic context and dysfunction of this compensatory mechanism may underlie pathologic anxiety states.

A nxiety is a physiological negative emotion that triggers a state of alert to possible threat and promotes survival. The bed nucleus of the stria terminalis (BNST) belongs to a neuronal network of interconnected limbic regions and exerts a pivotal role in the expression of anxiety both in humans 1,2 and in rodent models 3,4 . Previous studies using electrical 5 , lesioning 6,7 , pharmacological 8 or optogenetic 3,4 manipulations targeting directly specific BNST nuclei or cell types have been informative but have not been able to determine precisely how different BNST inputs influence anxiety. It has been shown that the ventral subiculum/ CA1 (vSUB/CA1) and infralimbic cortex (ILCx) massively projects to the BNST 9 . However, the integrative properties of BNST neurons at the single-cell level have never been explored. One hypothesis is that cortical and hippocampal information is processed at the single-cell level in BNST to trigger anxiolysis.
One of the most likely neural mechanisms underlying persistent anxiety is long-lasting plasticity in the neuronal network 10 . Probably owing to their anatomically convergent and segregated excitatory inputs from the infralimbic Cortex (ILCx) and the vSUB/ CA1 and their high content in stress-related neuromodulators, BNST circuits display morphological or synaptic plastic adaptations in response to stress and anxiety [11][12][13] . Ex vivo studies in slice have correlated plasticity at glutamate synapses within the BNST with alterations in anxiety levels 11 , but it remains unclear how the BNST integrates diverse inputs to modulate anxiety-related behaviors. Synaptic homeostasis seems to be a crucial process to compensate a long-lasting enhancement in signal transmission and maintain the stability of neuronal activity 14,15 . Recent computational modeling of synaptic plasticity have shown that the homeostatic processes that control the network stability are supported by the interaction of homosynaptic plasticity with heterosynaptic plasticity 16 . To unravel the integrative properties of BNST neurons at the single-cell level and the mechanism of their plastic changes in response to specific input stimulation, we combined conventional tracing, transsynaptic tracing from single-electroporated neurons, in vivo single-cell recordings, pharmacological and behavioral techniques. Here, we demonstrate: (1) that both vSUB/CA1 and ILCx converge on the same amBNST neurons; (2) that in vivo stimulation of the vSUB/CA1 promotes an N-methyl-D-aspartic acid or N-methyl-Daspartate (NMDA)-dependent long term potentiation (LTP) at the vSUB/CA1-amBNST synapses (LTP vSUB/CA1 ), associated with an NMDA-independent long-term depression (LTD) at the ILCx-amBNST synapses (LTD ILCx ); and (3) that the induction of in vivo NMDA-R-dependent plasticity in the amBNST triggers long-term changes of anxiety state.

Results
BNST neurons are connected to both vSUB/CA1 and ILCx inputs. It has been shown that the BNST could integrate information from the vSUB/CA1 or the ILCx (refs 9,12,17-19). However, it was unknown whether these inputs were integrated separately by different BNST neurons or if the same BNST neuron integrates both inputs. To address this question, we first injected a retrograde tracer, the cholera-toxin-B subunit (CTb), into the BNST, and confirmed that the BNST receives strong innervations from the vSUB/CA1 and the ILCx ( Supplementary Fig. 1a-c). We then injected two different anterograde tracers, Phaseolusvulgaris-leucoagglutinin (PHAL) and biotinylated-dextran-amine (BDA) in the vSUB/CA1 and in the ILCx, respectively (Fig. 1a,b). We used the Fox3 protein as a neuronal marker in the BNST. Using confocal imaging, we showed strong convergences of ILCx and vSUB/CA1 terminal fibers on the same Fox3-positive amBNST neuron (Fig. 1c). By using in vivo electrophysiology in anesthetized rats, we investigated whether amBNST neurons were controlled by both vSUB/CA1 and ILCx projections at the single-cell level. When we tested the evoked responses of all recorded amBNST neurons to the stimulation of the ILCx and the vSUB/ CA1, we observed that 85.2% of the recorded neurons responded  to the stimulation of one of these inputs (Fig. 1d). We quantified that 70% of these responding neurons responded to both ILCx and vSUB/CA1 stimulation, (Fig. 1e,f), whereas only 17% and 13% responded to either ILCx or vSUB/CA1 stimulation, respectively (Fig. 1e). The basal firing rate of the responding amBNST neurons was 0.7 ± 0.2 Hz. The onset of the evoked responses for the ILCx-amBNST pathway was 6.94 ± 0.59 ms and 9.75 ± 0.77 ms for the vSUB/CA1-amBNST pathway (Po0.01). Durations of evoked excitatory responses did not differ between these two pathways (ILCx-amBNST: 12.06±1.06 ms; vSUB/CA1-amBNST: 12.78 ± 1.16 ms). The intensity of stimulation used to trigger evoked responses in the amBNST was in the same range between the two inputs (0.2-1 mA). To further dissect connectivity of amBNST inputs onto single-amBNST neurons, we performed in vivo single-cell electroporation of amBNST neuron followed by retrograde monosynaptic tracing with pseudotyped rabies virus strategy (Fig. 2a,b) 20 . Five days after electroporation, retrogradely infected neurons were detected in the ILCx and in the vSUB/CA1 (Fig. 2b). Unfortunately, starter cells were not detected in the amBNST, most probably because rabies virus infection will eventually induce cytotoxicity in infected neurons 21 . Together, our findings confirm that a large population of amBNST neurons was synaptically controlled, at the single-cell level, by both vSUB/CA1 and ILCx inputs.
HFS vSUB/CA1 promotes in vivo input-specific LTD and LTP. To unravel the integrative properties of BNST neurons at the singlecell level, we combined in vivo single-cell recordings and pharmacological approaches. Given that vSUB/CA1 neurons fire at hundreds of Hertz in basal condition ( Supplementary Fig. 2a-c), we first assessed the impact of a high-frequency stimulation (HFS; Supplementary Fig. 2d,e) of the vSUB/CA1 (HFS vSUB/CA1 ), on the ability of the vSUB/CA1-amBNST and ILCx-amBNST synapses to undergo plasticity ( Fig. 3a-d). This HFS vSUB/CA1 , originally described by Abraham et al. 22 , induced an extremely long-lasting (48 days), robust and stable LTP in the projection from vSUB/ CA1 to medial prefrontal cortex (mPFC) 23 . Here, we established that HFS vSUB/CA1 triggers input-specific neuroplastic changes in the BNST (Fig. 3).  Fig. 3g). In addition, HFS protocol applied in the ILCx (HSF ILCx ) was ineffective to trigger plasticity in the ILCx nor in the vSUB/CA1 ( Supplementary Fig. 3).
In vivo LTP in the amBNST triggers anxiolysis. Finally, we conducted behavioral experiments to test our hypothesis that vSUB/CA1-driven NMDA-receptor-dependent LTP in the amBNST triggers anxiolysis (timeline Fig. 4a), as both regions are implicated in anxiety-related behaviors 24,25 . To test whether HFS vSUB/CA1 promotes anxiolytic effect in basal but also in anxiogenic situation, we used two complementary anxiety-tests based on the innate aversion of rodents to brightly illuminated areas, with the light-dark test ( Fig. 4b-f) or to open spaces with the elevated plus maze (EPM; Fig. 4e,g,h) 26 . In the light-dark test, rats exposed to HFS vSUB/CA1 spent more time in the light compartment in basal situation compared with SHAM group (percentage of time in light; HFS vSUB/CA1 group, 36.8 ± 3.8%; SHAM group, 27.5± 2.7%; Po0.05, Fig. 4c). Intra-BNST AP5 infusion prevented the anxiolytic-like effect induced by HFS vSUB/ CA1 . AP5 injection into the BNST decreased the % of time spent in light only in the HFS vSUB/CA1 group (Po0.05 for the effect of AP5 injection, Fig. 4c,d). Neither novelty-induced locomotor activity ( Supplementary Fig. 4a) nor circadian rhythms of general activity ( Supplementary Fig. 4b) were altered after HFS vSUB/CA1 or after manipulating the activity on the NMDA receptors in the BNST, thereby reinforcing the specific role of homeostatic plasticity on behavioral anxiety. There was no significant difference in the number of transitions between groups (light-dark test: SHAM/aCSF: 15.20 ± 0.60 transitions, SHAM/AP5: 16 Fig. 4c). In the light-dark test or in the EPM, the light serves as an anxiogenic stimulus 27 . To create an anxiogenic situation, the lighting was increased from 560 Lux to 1,230 Lux, and rats were restrained for 5 min before the light-dark test and the lighting was set-up at 260 Lux in open arms for the EPM (Fig. 4e) 28 . In the light-dark test, as expected, when rats where submitted to anxiogenic situation, they spent less time in the light compared with the SHAM group in basal situation (SHAM group in basal situation, 27.5 ± 2.7%; SHAM group in anxiogenic situation, 10.16 ± 3.1%, Fig 4c,f). In addition, when rats are exposed to anxiogenic situation in the light-dark test or EPM (Fig. 4e- Supplementary Fig. 4e). Taken together, these data indicate that (1) HFS vSUB/CA1 decreases the steady-state anxiety level through a NMDA-receptor-dependent mechanism in the BNST (2) acute HFS vSUB/CA1 diminishes the anxiety induced by an anxiogenic situation.

Discussion
Using tract-tracing, trans-synaptic tracing from single-electroporated neurons and in vivo recordings, we report that the majority of neurons of the anteromedial BNST integrates information from the vSUB/CA1 and the ILCx at the single-cell level. Considering the differences in the straight-line distances between the ILCx-BNST (E4 mm) and vSUB/CA1-BNST (E6 mm), the 3 ms difference in the latency to the onset of the stimulation response between the two inputs is probably not supported by a difference in conduction velocity, but by the length of the axonal projections. We next found that HFS vSUB/CA1 triggers in vivo an evoked spike potentiation (LTP vSUB/CA1 ), which requires the activation of NMDA-Rs in the BNST. This HFS protocol was efficient to trigger plasticity in the amBNST when applied in the vSUB/CA1, but not in the ILCx (Supplementary Fig. 3). This is probably due to the fact that vSUB/CA1 is one of the few major output structures of the hippocampal formation and transmits learning and memoryrelated signals in a high-frequency bursting mode (Supplementary Fig. 2) repeated at a low frequency (0.5-2 Hz) 29 . A pioneering study in the hippocampus demonstrated that changes in synaptic strengths affect network activity and shape neuronal integration in an input-specific manner 30 . Here, we report that in response to an excessive activity of the vSUB/ CA1-amBNST inputs (LTP vSUB/CA1 ), amBNST neurons at the single-cell level down-regulate the efficacy of their ILCx-amBNST inputs (LTD ILCx ) and maintain their basal activity stable. This interaction of homosynaptic plasticity (LTP vSUB/CA1 ) with heterosynaptic plasticity (LTD ILCx ) occurring at the single-cell level at two amBNST excitatory synapses could be considered as homeostatic plasticity 16 . In fact, these neuroplastic changes correspond to a form of homeostasis necessary to maintain  stability in the amBNST, in response to the strengthening of the vSUB/CA1-amBNST synapses (LTP vSUB/CA1 ) associated with a negative feedback at the ILCx-amBNST synapses (LTD ILCx ). Regardless of the cellular explanation, the functional effect of LTD ILCx is complementary to that of LTP vSUB/CA1 , that is it optimizes the signal-to-noise ratio by reinforcing the functional weight of the recently potentiated synapses. Interestingly, in presence of AP5 in the amBNST, the direction of plasticity at the vSUB/CA1-amBNST inputs elicited by HFS vSUB/CA1 , was switch from an LTP vSUB/CA1 to an LTD vSUB/CA1 (Fig. 3f). One possibility is that AP5, locally infused in the amBNST, only partially blocks the NMDA receptors, and this partial blockade reverses the direction of plasticity elicited HFS vSUB/CA1 (refs 31,32). Another unexpected result was that the intra-amBNST blockade of NMDA-Rs potentiates LTD ILCx elicited by HFS vSUB/CA1 (Fig. 3g). One possibility is that is that HFS vSUB/CA1 triggers concomitant activation of NMDA and metabotropic glutamate receptors in amBNST neurons, leading to a more profound LTD in the presence of AP5 (ref. 33). Further experiments are necessary to determine the molecular mechanisms by which in the absence of NMDA-Rs stimulation, HFS vSUB/CA1 triggers LTD vSUB/CA1 and potentiates LTD ILCx... Finally, we can not exclude that HFS vSUB/CA1 also triggers plasticity in the mPFC 23 or the basolateral amygdala 34 , but we provide behavioral evidence that vSUB/CA1-driven NMDA-R-dependent LTP in the amBNST triggers anxiolytic-like effects. This is in line with pioneer studies showing that changing the activity in the amBNST has a direct impact on the perception of aversive contextual stimuli 35 or production of stress hormones 5 . In fact, here we have demonstrated, using two different anxiety assays that HFS vSUB/CA1 induced an anxiolytic effect in basal situation but also in anxiogenic situation (Fig. 4). Together, these data support the conclusion that the amBNST plays a crucial role in integrating and sending information related to anxiety 9,36 . Previous studies have shown that anxiety is controlled by multiple circuits in the brain, many of which share robust and reciprocal connections with the BNST 4,37 . These circuits include projections from the basolateral nucleus of the amygdala (BLA) to the ventral hippocampus 38 , from BLA to the central nucleus of the amygdala (CeA) 39 , from the ventral hippocampus to the medial prefrontal cortex (mPFC) 40 , from mPFC to BLA 41 and from BLA to BNST 4 . Our anatomical and functional characterization of the vSUB/CA1-amBNST projection on a circuit and synaptic level furthers the understanding of the role played by amBNST in the modulation of anxiety 4,37 . In conclusion, we show that in response to HFS vSUB/CA1 , homeostasis in amBNST neurons is guaranteed at the single-cell level by an NMDA-R-dependent up-scaling of the vSUB/CA1-amBNST synapses associated with an NMDA-R independent down-regulation of the efficacy of its ILCx-amBNST inputs (LTD ILCx ; Supplementary  Fig. 5). Together these findings elucidate the molecular targets that contribute to changes in synaptic functions in the amBNST, and highlight important future directions where manipulation of inputs to the amBNST using opto-or chemogenetic tools may be critical for changing network output, physiological manifestations of anxiety and anxiety-associated disorders 42 .

Methods
Animals. Male sprague Dawley rats (275-300 g; 10 weeks old; Elevage Janvier, France) were used. Rats were housed three or four per cage under controlled conditions (22-23°C, 40% relative humidity, 12 h light/dark illumination cycle; lights on from 07:00 hours to 19:00 hours), were acclimatized to laboratory conditions 1 week before the experiment, with food and water ad libidum. All procedures were conducted in accordance with the European directive 2010-63-EU and with approval from Bordeaux University Animal Care and Use Committee (N°50120205-A).
amBNST recordings. A glass micropipette (tip diameter, 1-2 mm; 10-15 MO) filled with a 2% pontamine sky blue solution in 0.5 M sodium acetate was lowered into the amBNST. The extracellular potential was recorded with an Axoclamp-2B amplifier and filter (300 Hz/0.5 Hz). Spikes were collected online (CED 1401, SPIKE 2; Cambridge Electronic Design; UK). During electrical stimulation of the ILCx or vSUB/CA1, cumulative peristimulus time histograms (PSTH, 5 ms bin width) of amBNST activity were generated for each neuron recorded.
Pharmacological treatment. For local delivery of 100 mM AP5, double barrel pipettes were used 43 . For behaviour, a mixture of 180 nl of AP5 (100 mM) and 0.2% Fluorogold (to mark the injection site) was injected bilaterally in the BNST.
Histology. At the end of each recording experiment, the recording pipette placement was marked with an iontophoretic deposit of pontamine sky blue dye ( À 20 mA, 30 min). To mark electrical stimulation sites, þ 50 mA was passed through the stimulation electrode for 90 s. After, brains were removed and snap-frozen in a solution of isopentane stored at À 80°C.
Plasmid solution. A plasmid encoding for the rabies glycoprotein, the avian virus receptor TVA and a fluorescent marker (tdTomato) was used in this study. For recording followed by electroporation experiments, the electrode was filled with the plasmid (pAAV-EF1a-G-TVA-tdTomato, 17.5 ng ml À 1 ) diluted in standard intracellular solution.
In vivo single-cell electroporation. Single-cell electroporation was performed as described previously 44 . After recording amBNST neurons responding to both ILCx and vSUB/CA1 stimulations, they were electroporated with a solution containing a plasmid DNA (pAAV-EF1a-G-TVA-tdTomato). We applied À 10 V square-pulses delivered at 50 Hz for 1 s. Only one cell per rat was electroporated. After 2 days, an EnvA pseudotyped G-deleted rabies virus (EnvA-SADJG-GFP) was injected into the amBNST. After 5 days, the electroporation protocol, rats were killed for immunohistological experiments.
Tract-tracing method. Tracer injections were performed as described previously 43 with the following modifications. For retrograde tracing, 30 nl of 0.5% CTb (Sigma Aldrich, France) were infused by pressure into the amBNST (n ¼ 5). Animals received a single iontophoretic injection of a 2.5% solution of an anterograde tracer PHAL (Vector Laboratories; UK) in the vSUB/CA1 (n ¼ 5). Animals received a single iontophoretic injection of a 2.5% solution of an anterograde tracer BDA in the ILCx (n ¼ 3). Animals were allowed to survive 7-14 days.
Elevated plus maze. One week later SHAM and HFS stimulations, it has been performed an EPM test. The apparatus is in plexiglass, with four elevated arms arranged in a cross-like disposition, it consisted in two opposite enclosed arms and two open arms, having at their intersection a central square platform which gave access to any of the four arms. Each arm is 50 cm; the wall for the closed arm is 40 cm height. All floor surfaces were grey and the open arms were under an anxiogenic illumination of 260 Lux. In brief, the rats were placed individually in the central square of the EPM facing an open arm and then allowed to start exploring the maze freely during 5 min test. Video recordings were analysed offline using video tracking software (Videotrack from Viewpoint, Lyon, France). The following behaviours were measured: time spent in each compartments and number of transitions between compartment. The percentage of time spent in open arms was evaluated to assess anxiety.
Light-dark test. Eight days after the application of HFS in the vSUB/CA1 in a new group of anesthetized rats, we performed the light-dark test. The test lasted 5 min and was performed in a two-compartment box (40 Â 40 Â 35 cm) with two equal compartments that limit exploratory behavior. An aperture enabled the rats to pass from one compartment to the other. One was completely enclosed by black opaque plastic sides, with a lid of the same material, while the other was white, had no lid, and was brightly illuminated (560 lux). At the start of the experiment a rat was placed in the center of the lighted box with its head facing the wall opposite to the door. The latency for the first emergence from the dark to the light compartment, time spent in each compartment and frequency of explorations of the light compartment were recorded. Time in a zone was considered when the animal placed its four paws in that zone. Rats performed only one time the light-dark test.
A separate group of rats was exposed to anxiogenic situation in order to increase the anxious state of the rats. The intensity of the lighting was increases from 560 Lux to 1,230 lux, and animals were restrained for 5 min, and after 2 min of recovery were exposed to the light-dark test 45 . The same behavioral parameters were measured in anxiogenic situation as in normal situation.
Data analysis. For in vivo electrophysiological experiments, cumulative PSTHs of aBNST activity were generated during stimulation of ILCx or vSUB/CA1. Excitatory magnitudes (R mag values) were normalized for different levels of baseline impulse activity. Baseline activity was calculated on each PSTH, during the 500 ms preceding the stimulation. For each PSTH, R mag values for excitation were calculated according to: excitation R mag ¼ (number of spikes in excitatory epoch)-(mean number of spikes per baseline bin x number of bins in excitatory epoch).
The cortical or hippocampal excitation strength onto amBNST neurons was determined as the amount of current needed to obtain a 50% spike probability for ILCx-evoked responses or vSUB/CA1-evoked responses (R mag ranging from 30 to 60). Results are expressed as mean ± s.e.m. For statistic, two-group comparisons were achieved using Student's t-tests or Mann-Whitney when necessary. For multiple comparisons, values were subjected to a two-way Anova followed if significant by Bonferroni post hoc tests or to Kruskall-Wallis Anova for the behavioral part.
Data availability. The data that support the findings of this study are available from the corresponding author upon reasonable request.