A new GABAergic projection from the BNST onto accumbal parvalbumin neurons controls anxiety

The prevailing view is that parvalbumin (PV) interneurons play modulatory roles in emotional response through local medium spiny projection neurons (MSNs). Here, we show that PV activity within the nucleus accumbens shell (sNAc) is required for producing anxiety-related avoidance when mice are under anxiogenic situations; sNAcPV neurons exhibited high excitability in chronically stressed mice model, which generated excessive maladaptive avoidance behavior in an anxiogenic context. We also discovered a novel GABAergic projections from the anterior dorsal bed nuclei of stria terminalis (adBNST) to sNAcPV neurons; optogenetic activation of these afferent terminals in sNAc produced an anxiolytic effect via GABA transmission. Next, we further demonstrated that chronic stressors attenuated the inhibitory synaptic transmission at adBNSTGABA sNAcPV synapses, which in turn explains the hyperexcitability of sNAcPV neurons on stressed models; therefore, activation of these GABAergic afferents in sNAc rescued the excessive avoidance behavior related to anxious state. Our findings reveal the coordination between BNST and NAc through an inhibitory architecture in controlling of anxiety-related response and provide a neurobiological basis for therapeutic interventions in pathological anxiety.


Introduction 1
Stressors and stress responses are critical for guiding both approach and 2 avoidance behaviors in animals and humans. Exposure to chronic, 3 unpredictable stressors leads to increased anxiety responses, including 4 excessive avoidance behavior, and this exposure has been adopted to study 5 anxious state-related behaviors 1, 2 . The bed nucleus of the stria terminalis 6 (BNST), a subregion of the extended amygdala, is a critical node in the stress 7 response 3, 4 . Recent work on human drug addiction has also demonstrated a 8 role of BNST in withdrawal-related anxiety and relapse, indicating an intrinsic 9 link between this stress response region and the reward system. The nucleus 10 accumbens (NAc) is a vital component in the reward circuitry 5-7 , which 11 responds to stress signals 8,9 and has a dominant effect on anxiety regulation 10 . 12 However, with the exception of one 15-year-old anatomical observation 11 , the 13 connectivity between NAc and BNST, and whether it is necessary in producing 14 anxiety-related behavior, remains unexplored. Furthermore, GABAergic 15 efferents originating from the BNST are predominantly sent downstream 12, 13 ; 16 however, the nature and function of any GABAergic input to NAc is unknown. because other work points to a role in reward seeking, but not anxiety-related 2 behavior 17, 18 . Based on these very different findings, we predicted that there is 3 another neuronal type within the NAc that contributes to anxiety-related 4 behavior. One possibility is PV GABAergic interneurons, which comprise only 5 3-4% of all neurons in the NAc 19,20 . In other brain regions, these neuron 6 regulate fear response 21 , anxiolysis 22 , alcohol addiction 23 , and reward seeking 3, 7 24 . However, we know less about the function of accumbal PV neurons and the 8 inputs they receive, and nothing about any possible role in anxiety related 9 behavior. 10 We addressed these important questions regarding the neural mechanisms GABAergic terminals in sNAc produced an anxiolytic effect, which was 1 mediated by sNAc PV cells; activation of these inhibitory inputs from adBNST to 2 sNAc rescued the excessively anxious state of the stressed mice.
3 Therefore, our results reveal a previously undescribed circuit mechanism, 4 defined by neuronal type, that shapes the coordination between BNST GABA and 5 NAc PV cells in response to anxiogenic stimuli under both physiological and 6 pathological conditions. was performed under full anesthesia, and every effort was made to minimize 17 animal suffering. Male and female mice (6-14 weeks) were used in this study. 18 We used the following mouse lines : PV-Cre (B6;129P2-Pvalbtm1(cre)Arbr/J,

23
Behavioral tests 24 All behavioral tests were performed blind to mice genotypes. Groups of mice 1 were age-matched (8-14 weeks). All mice were handled for 15-30 min per day 2 for three days before behavioral assays to reduce stress introduced by contact 3 with experimenter.  2) Open-field test 12 A plastic open field chamber (50×50 cm) was used and conceptually divided 13 into a central field (25×25 cm) and a peripheral field for analysis. Each mouse 14 was placed in the peripheral field at the start of each test. The open field test 15 consisted of a 10 min session and mice locations were monitored/tracked 16 using Anymaze software. 18 The UCMS protocol was performed as previously described 51, 52 with 19 modification. Mice were exposed to environmental stressors for three weeks.

20
One of the following stressors were presented during each daily session in a 21 random order over three weeks: (i) restraint, where each mouse was placed in 22 a tube (50 mL) for two hours without access to food or water, (ii) a wet 1 environment where water was added (such that bedding was damp but not 2 overly wet) to a housing cage containing mice for twelve hour sessions, and (iii) 3 squeezing, where four mice were housed into a box (3 × 5 ×7 cm) for two 4 hours, without access to food or water. Inc. Filter 525/39), and detected by the sensor of an CMOS camera (Thorlabs,19 Inc. DCC3240M).

20
A Labview program was developed to control the CMOS camera which 21 recorded calcium signals at 50 Hz. Behavioral event signals were recorded by 22 a DAQ card (NI, at 1000 Hz using the same program. Single-unit and local field potential (LFP) recordings in freely-moving 2 mice 3 Both naive and chronically stressed mice (aged 8-12 wk) were anesthetized 4 with isoflurane (4.0% for induction and set-up on the animal bed, 0.8%-1% 5 during experiments) in a 20% O 2 /80% air mixture). Body temperature was 6 maintained at 36-37 °C with a heating pad. For single-unit recording, mice 7 were secured in a stereotaxic apparatus and a custom-made screw-driven 8 microdrive containing eight tetrodes (four wires wound together) was 9 unilaterally implanted in the left NAc shell; and for LFP recordings, 10 custom-made stereotrodes (two wires wound together) were unilaterally 11 implanted in both left NAc shell and aBNST. Each stereotrode was housed in a 12 silica tube and consisted of two individually insulated platinum-iridium wires 13 (17 µm inner diameter). The electrodes were modified by electrochemical 14 deposition of platinum to reduce their impedance to ~500 KΩ. The skull was 15 leveled using bregma and lambda landmarks, and screws were implanted on 16 the anterior and posterior portions of the skull to serve as reference.

17
Coordinates were measured from bregma and depth was calculated from the 18 brain surface. The electrodes were implanted through burr holes in the skull 19 aimed at the following coordinates: AP 1.35 mm, ML 1.35 mm, and DV -4.85 20 mm for NAc shell and AP 0.20 mm, ML 0.80 mm, and DV -4.05 mm for aBNST.

21
The microdrive electrode was attached to a micromanipulator and moved 22 gradually to a position about 400 μm above the desired depth. The electrodes 1 were anchored to the microdrive that made it possible to advance along the 2 dorsal-ventral coordinates. Following surgery, mice were allowed to recover for 3 at least one week and were then habituated to experimenter handling. During 4 recording, electrodes were connected to a unitary gain headstage (Plexon,5 Dallas, TX) connected to a 32-channel preamplifier (Plexon, Dallas, TX). Once 6 mice were familiar with the recording setup they were connected to the 7 headstage preamplifier in their home cages for two daily sessions of 20 min 8 each. Neurophysiological signals were recorded with a 64-channel 9 Multichannel Acquisition Processor (Plexon, Dallas, TX) and mouse positions 10 were tracked using an overhead camera (30 Hz). Wideband signals were 11 recorded at 40 kHz and LFP signals were acquired at 1 kHz.

12
At the end of each experiment, each mouse was deeply anesthetized with 10% 13 chloral hydrate (0.4 mg/kg) and transcardially perfused with PBS, then 4%

21
All mice were initially anesthetized with isoflurane (4.0% for induction and 22 set-up on the animal bed, 0.8%-1% during experiments) in a 20% O 2 /80% air 1 mixture. Body temperature was maintained at 37 °C using warm water circuitry.
2 MRI experiments were conducted using a Bruker Biospec70/20USR small 3 animal MR system (Bruker BioSpin MRI, Ettlingen, Germany) operating at 300 4 MHz (7T). Breathing rate, heart rate and blood oxygen saturation were 5 monitored using a pulse oximeter positioned at the hind limb and a  Resting state data sets were then acquired using single shot gradient echo EPI 16 (Echo Planar Imaging) with TR 1000 ms and TE 16 ms. Twenty coronal slices 17 (using the same procedure as the T2 anatomical images above) were done using SPM12 (http://www.fil.ion.ucl.ac.uk/spm/software/spm12/) to 1 eliminate head movement and image shift by co-registering with the 2 anatomical image, and then Gaussian smoothing was performed for every 3 slice to improve the signal to noise ratio. To estimate functional connectivity, 4 4-5 voxels in each bilateral NAc image were selected as ROI (seed points), 5 using REST (http://restfmri.net/forum/index.php) and home-written algorithms 6 using Matlab2014a (www.mathworks.com). Cre-mice (6-12 weeks) were used for stereotactic viral injections in the NAc 10 shell. During isoflurane anesthesia (as above), the skull was exposed via small  Glycoproteindeleted rabies virus for retrograde tracing (RV-ENVA-ΔG-dsRed 7 2.0× 10^8 IFU/mL) was also purchased from BrainVTA. In vivo anesthetized electrophysiology 10 Adult mice (8-12 weeks) were anesthetized with isoflurane (4.0% for 11 induction and set-up on the animal bed, 0.8%-1% during experiments) in a 20%

12
O 2 /80% air mixture). Once anesthetized, mice were placed into a stereotactic 13 frame and body temperature was maintained at ~ 37 °C using a heating pad. A  the NAc shell using the same coordinates. Approximately seven days after the 5 second injection, mice were anaesthetized with 10% chloral hydrate (0.4 6 mg/kg) and transcardially perfused with PBS, then 4% PFA (wt/vol) and then 7 brain slices were prepared for tracing with dsRed. In situ hybridization 10 We used single-probe in situ hybridization on fixed frozen sections. Coding 11 region fragments of Gad1, Gad2, Vglut1, Vglut2 and CRH were isolated from 12 mouse brain cDNA using PCR and cloned into the pCR4 Topo vector (Thermo   and open arms, Bonferroni post hoc test, ***P < 0.001) 53 . EPM scores were 21 used to quantify the degree to which a neuron can distinguish the structure of 22 the EPM 49, 53 ; EPM scores were calculated as previously described 49,54 .
Horizontal and vertical arms represent closed and open arms, respectively. F R , 5 F L , F D and F U are the % differences from mean firing rate in right, left, down 6 and up arms, respectively; A is the mean difference in normalized firing rate 7 between different-type arms and B is the mean difference for same-type arms.

8
Neurons with firing patterns related to the EPM task have a high EPM score, 9 as neurons will have similar firing rates in the same arm types (resulting in a 10 small B value) and large differences in rates between different arm types 11 (resulting in a large A value). The maximum EPM score of 1.0 shows no 12 difference in firing rate across arms of the same type (B = 0). Negative EPM 13 scores show that firing rates were more similar across arms of different types 14 than across arms of the same type. 15 We calculated whether there was a statistically significant difference between 16 the population of experimentally observed EPM scores from that expected by 17 chance using a bootstrapping method. For each unit that had n spikes, 500 18 simulated EPM scores were generated by calculating the EPM score of n 19 randomly chosen time stamps 500 times. 500 × 98 EPM scores were 20 generated from 98 units recorded. Statistical differences between 21 experimentally observed EPM scores of all neurons and chance were calculated by comparison to the simulated distribution of EPM scores using the 1 Wilcoxon rank-sum test 49, 53 .

2
The firing pattern of NAc neurons at transitions between different types of 3 EPM arms and Z-scores of firing rate were calculated for each unit for 10 s and 4 averaged over total transitions for each unit. We identified a point where there 5 was a change in the slope of the averaged z-scores. The averaged z-scores 6 were divided into two parts by using this identified change point and a 7 nonparametric Kolmogorov-Smirnov test was used to evaluate whether there 8 were statistically significance differences between the means from these two 9 data segments. This was calculated using 0.25 s bins.  Single-unit spike sorting was performed using Offline sorter (Plexon). respectively 56 . Units with L ratio higher than 0.2 and Isolation Distance lower 9 than 15 were excluded from the following analysis. Classification of NAc 10 neurons were as described in previous studies and two features used for this, 11 peak-to-peak width and average firing rate, were calculated for each unit 29, 57 .

12
An unsupervised cluster algorithm based on Ward's method was used to

6
Chronic stressors increase functional brain connectivity between the 7 BNST and NAc 8 To gain a circuit-level understanding of anxiety-related behavior, we adopted 9 a chronic stress model to investigate specific brain regions in the regulation of 10 avoidance behavior under anxious states.

11
After exposure to chronic stressors (Fig. 1a), mice showed higher anxious  (Fig. 1d). These data indicate that chronic stressors disrupted normal 17 anxiety-related behavior and resulted in a maladaptive and excessive 18 avoidance coping response. 19 We further tested global functional brain connectivity by quantifying the 20 synchronization of blood oxygen level-dependent (BOLD) fMRI signals across 21 brain regions in anesthetized naive and chronically stressed mice. The 22 synchronization of BOLD signals in the BNST and NAc was significantly higher 1 in chronically stressed mice compared to their naive littermates (Fig. 1e-f).
2 Moreover, functional connectivity was higher in stressed mice compared to 3 their naive littermates between the basolateral amygdala (BLA) and BNST, and 4 lower between the periaqueductal gray (PAG) and locus coeruleus (LC) (Fig.   5 S1a, left and middle), whereas connectivity between NAc and PFC was not 6 significantly different between groups (Fig. S1a, right). Consistent with the 7 fMRI synchronization data, fMRI heat maps in coronal sections generated from 8 stressed mice show a higher correlation of resting-state fMRI BOLD signal 9 than those generated from their naive littermates, with a seed in NAc across 10 brain regions (Fig. S1b, top) between BNST and NAc (Fig. S1b, bottom).

11
To quantify long-range functional connectivity, we measured local field 12 potential (LFP) coherence between the BNST and NAc in awake behaving 13 stressed and naive mice (Fig. 1g). Compared to naive littermates, the stressed 14 mice showed higher BNST-NAc coherence in low gamma, alpha and beta 15 bands, but did not show any significant coherence in theta band in these two 16 structures (Fig. 1h-i). We then analyzed the local alpha, beta, gamma and 17 theta rhythms, respectively, in these two brain regions and found that local 18 theta power was significantly lower in stressed mice compared to their naive 19 littermates in both structures, whilst there was a trend towards lower gamma 20 power in stressed mice in the BNST (Fig. 1j-k). in stressed mice, we tested sNAc PV neuronal firing properties in stressed mice.

11
We selectively expressed ChR2-mCherry in sNAc PV cells of PV-Cre mice to  (Fig. 2j-k). These results imply that hyper-excited sNAc PV neurons in 2 stressed mice contributed to their excessive avoidance behavior. could not be classified as FS or non-FS) or 4) unclassified, the units could not 20 be identified as neurons (Fig. 3A). A total of five putative FS neurons (5.1%) 21 and 28 putative Non-FS neurons (28.6%) were clearly identified (Fig. 3a).
These two classes of neurons had significantly different mean firing rates and 1 burst numbers (Fig. 3b, ***P < 0.001, respectively). Individual FS units showed 2 a firing preference for the open arms over the closed arms (Fig. 3c). The 3 z-scores of FS unit firing rates increased when mice entered an open arm (Fig.   4 3d) and decreased when they moved to closed arms (Fig. S2a). Note that FS 5 unit firing rates were not influenced by locomotion speed (Fig. 3e). By contrast,  we calculated the mean spike-field coherence for both FS and Non-FS units. 21 We found that FS, but not Non-FS, had strong coherence between their spikes and theta oscillations at 4-8 Hz (Fig. S2f, **P = 0.007). These results imply that 1 NAc FS activity was inversely correlated with theta power, and that a reduction 2 of the accumbal theta activity reflects a higher stress load during exploration of 3 the threatening environments, which can promote adaptive avoidance 4 behavior. These findings are consistent with the above result showing that a 5 decrease in local theta power within either NAc or BNST is reflected by the 6 anxious state of the stressed mouse (see also Fig. 1j-k).  (see also Fig. 3c). Immunostaining results indicated that the majority of ChR2-mCherry labeled neurons expressed PV (Fig. 4a, right). We confirmed 1 that opto-tagged PV + cells, recorded from patch-clam experiments in the acute 2 brain slices, were steadily activated by 5-80 Hz light stimulation at 470 nm (Fig.   3 4b). Selective light stimulation of these PV neurons in the NAc during the EPM 4 task led to more avoidance behavior in PV-ChR2 mice compared to 5 PV-mCherry controls, illustrated by a significantly lower number of entries and 6 markedly less time spent in the anxiogenic open arms compared (Fig. 4c-d). compared to the PV-mCherry control mice (Fig. 4e) without any difference in 10 locomotion between the two groups during each five-min epoch (Fig. 4f).

11
In order to further determine the causal role of accumbal PV activity in 12 avoidance behavior related to anxiety states, we expressed hM 4 D i in PV 13 neurons by bilateral injection of Cre-dependent AAV-DIO-hM 4 D i -mCherry into 14 the NAc shell of PV-Cre mice. Mice were given a 10-min OFT followed by a 15 5-min EPM test (Fig. 4g). Co-localization of the majority PV cells with hM 4 D i 16 was verified by immunostaining (Fig. 4h). Representative OFT and EPM heat 17 maps for both PV/mCherry (control) and PV/hM 4 D i groups are shown in Fig.   18 4i-k: hM 4 D i mice showed significantly greater center exploration, relative to an inappropriate avoidance coping behavior (Fig. 4l-n).
Taken together, these data suggest that activity of PV neurons within the NAc 1 is required for execution of appropriate avoidance behavior to buffer the stress 2 response evoked by anxiogenic environments.

4
Inputs to NAc PV neurons originate predominantly from the anterior dorsal 5 BNST (adBNST) 6 We next investigated whether the BNST was upstream of the NAc PV neurons 7 and if it is the region where anxiety-related avoidance coping behavior is 8 integrated. Cre-dependent, rabies-virus-based whole brain monosynaptic 9 tracing was performed to analyze upstream regions that innervate NAc PV 10 neurons. We injected PV-Cre mice with Cre-dependent AAVs expressing the 11 avian EnvA receptor (TVA) and rabies virus envelope glycoprotein (RG) in 12 combination with theΔG-dsRed (EnvA) rabies virus (RV) (Fig. 5a). We found 13 that a Cre-dependent helper virus combined with RV expressing dsRed 14 labeled 84% of NAc PV neurons (GFP + cells, Fig. 5b). Our results showed that 15 the dominant inputs to NAc PV neurons were from the anterior dorsal part of the 16 nucleus of the stria terminalis (adBNST) (52%, Fig. 5c, d), a classic GABAergic 17 anxiety-associated region. Other brain regions that provided inputs to NAc PV 18 neurons included the basolateral amygdala (BLA, 4.34%), central amygdala 19 (CeA, 10.27%), media prefrontal cortex (mPFC, 5.7%) as well as the ventral 20 tegmental area (VTA, 6.07%). NAc PV neurons also received monosynaptic 21 inputs from reward-associated components, such as the paraventricular 22 thalamus (PVT, 19.6%) and lateral hypothalamus (LH, 16.07%) 36 ; no 1 projections from the Hippocampus (Hippo, 0%) were found (Fig. 5c, d). In situ 2 hybridization experiments demonstrated the co-expression of Gad 1/2 and 3 dsRed in aBNST (96.1%, Fig. 5e, f). These findings indicate that NAc PV 4 neurons were modulated under a GABAergic network. No RV expressing 5 dsRed labeled cells were found in the above brain regions (Fig. S4). We tested 6 further and found no other neuronal markers in the BNST except sparse CRH 7 co-expressed with RV signals in adBNST (Fig. S5). Next, we investigated the impact of the connection between the adBNST 12 GABAergic neurons and NAc PV cells on producing anxiety-related behavior. 13 We virally expressed GAD67-Cre and Cre-dependent channelrhodopsin-2 14 (ChR2) in adBNST GABAergic neurons and visualized NAc PV neurons by 15 injection of adeno-associated viruses (AAVs) encoding the fluorophore 16 mCherry into PV-Cre mice (Fig. 6a). Co-staining results revealed that the 17 majority of neurons expressing GAD + also co-expressed ChR2 (Fig. 6b-c). We 18 recorded evoked IPSPs from PV neurons within the NAc shell by illumination 19 of adBNST afferent axon fibers, which was completely blocked by 20 μM 20 bicuculline, implying a GABAergic monosynaptic input to the NAc PV neurons 21 (Fig. 6d-e). The mean latency was 4.07± 0.7 ms, in line with monosynaptic transmission (Fig. 6f, right). These data suggest a direct functional GABAergic 1 input to the sNAc PV neurons. We further targeted the function of this 2 GABAergic input to the NAc PV neurons by virally expressing Cre-dependent 3 ChR2 and GAD67-Cre in adBNST neurons followed by light stimulation of the 4 terminals within sNAc (Fig. 6g). Blue light stimulation resulted in decreased 5 avoidance of both EPM open arms and OFT center area (Fig. 6i). We then 6 tested whether, under similar conditions, we would obtain similar results 7 without adBNST neurons innervating NAc PV . AAV2/9-FLEX-taCasp3-TEVp 8 and PV-Cre viruses were injected into the sNAc to selectively kill PV + neurons; 9 Cre-dependent ChR2 and GAD67-Cre were both injected into the adBNST of 10 mice to specifically activate this new BNST-NAc GABAergic circuit (Fig. 6k); 11 immunostaining confirmed that most of the PV + neurons were killed by 12 taCasp3 compared with the control virus (Fig. 6l). Terminals in the NAc shell 13 were stimulated once again and we found that reduced avoidance of EPM 14 open arms and OFT center was now effectively blocked during the light ON 15 phase ( Fig. 6m-n). These findings suggest a crucial role of adBNST 16 GABAergic inputs to the NAc PV neurons in distinguishing safety and risk to 17 facilitate avoidance of anxiogenic locations. To examine the presynaptic effect 18 of GABA release on the NAc PV neurons, we recorded two consecutive eIPSPs, 19 which were separated by varying interspike intervals to calculate the 20 paired-pulse ratio (PPR), upon light stimulation of the adBNST GABAergic 21 afferents to sNAc. We found that the PPR was significantly increased in 22 stressed mice compared to naive ones at both 50-and 100-ms interstimulation 1 interval ( Fig. 6p-q). This increased PPR in stressed mice suggests an impaired 2 presynaptic GABA release at adBNST to sNAc synapses, further indicating 3 these sNAc PV neurons are disinhibited by GABAergic inputs from the 4 adBNST under a chronic stress state.

5
Based on these findings, we then tested whether activation of the adBNST 6 GABAergic afferents to sNAc PV could rescue the pathological anxiety-related 7 behavior in stressed mice. We injected AAV-GAD67-Cre and either DIO-ChR2 8 or DIO-mCherry into the adBNST of stressed PV-Cre mice and implanted an 9 optical fiber above the sNAc (Fig. 6r). Light stimulation of adBNST GABAergic 10 afferents in sNAc significantly increased the time spent in the OFT center and 11 open arms of the EPM (Fig. 6s-t), which implies that excessive avoidance 12 behavior related to anxious state was rescued. Taken together, these findings 13 indicate that in anxious states, the adBNST may send a disinhibition input to 14 the sNAc PV , leading to excessive avoidance of the threatening locations. Chronic stress leads to long-term changes in brain structure and function, 2 which increases the incidence of stress-related disorders, such as anxiety 4 .

17
Our resting fMRI findings indicate that besides the high correlation among the 18 stress response regions (Fig. S1a), increased connectivity was observed 19 between BNST and NAc under a chronic-stress-induced anxious state (Fig.   20 1e-f); the LFP coherences in alpha, beta and gamma bands between these 21 two regions were also significantly increased (Fig. 1i), confirming an intrinsic link between the stress response region and the reward circuit component 1 under anxious state. We can envision that the increased synchronization of 2 BOLD signals in BNST and NAc may potentially be used as an imaging marker 3 for the diagnosis of anxiety disorders in the future.

4
LFP recordings further showed a marked decrease in the local theta power 5 both in the NAc and BNST of stressed mice (Fig. 1j-k), indicating PV cell 6 involvement within these two regions as previous studies have suggested a 7 correlation between PV activity and local theta changes 25, 26 . However, only the 8 NAc shell contained PV cell bodies (Fig. 2a) and the patch-clamp data found 9 that these accumbal PV cells were hyper-excitable under an anxious state ( Fig.   10 2d-e), confirming the surprising role of accumbal PV interneurons in behavior 11 related to anxiety, given their low occurrence in the region. According to a 12 previous report, the anxious state is reflected by the reversed correlation anxiogenic EPM open arms (Fig. 3n).

17
Indeed, other observations in our present study also support this hypothesis.  Fig. 3c-d, Fig. S2a). Because 22 most fast-spiking neurons within the NAc have previously been identified as 1 PV neurons, it confirms that accumbal PV neurons were engaged in the 2 anxiogenic-related behavior. Second, our results (Fig. 3k) were consistent with 3 a recent study showing that excitation of D2R cells in the NAc was necessary 4 for normal, innate risk-avoidance 16 ; however, we further demonstrated that PV 5 neurons fired earlier than D2R cells (Fig. 3l-m), which implies that the 6 accumbal PV neurons are informed of the risk and then modulate the activity of 7 nearby cells that, in turn, initiate avoidance coping behavior. Further study is 8 needed to investigate the accumbal neuronal firing dynamics in orchestrating 9 risk avoidance in response to anxiogenic stimuli. Third, in vivo genetic 10 manipulation of PV-neuronal activity in both healthy and chronic stress models 11 further highlighted their importance in encoding anxiety-related behavior ( Fig.   12 2i-k and Fig. 4). Taken together, these findings significantly extend previous 13 observations of NAc in anxiety-related information processing; this is the first 14 study that has begun to reveal real-time PV activity in the accumbens during 15 the encoding of anxiety-related behavior in free exploration of aversive spaces 16 without prior training and suggests that the underlying neuronal mechanism 17 has been evolutionarily programmed.

18
Previous studies have showed that the NAc receives intermingled 19 glutamatergic and dopaminergic inputs from a variety of forebrain regions, 20 including the amygdala 23 , hippocampus 22 , thalamus 21, 33 , ventral tegmental 21 area 19 and the prefrontal cortex 34 . Using Cre-dependent, rabies-virus-based whole brain monosynaptic tracing strategy and electrophysiological recordings 1 from brain slices, we demonstrated that NAc PV cells were specifically 2 innervated by the GABAergic afferents stemming from the adBNST (Fig. 5 and 3 Fig. 6d). To our best knowledge, this is the first study to map novel neural 4 circuitry specifically innervating NAc PV neurons. Although a recent study 5 found that light-evoked activation of vHPC inputs to NAc resulted in NAc PV cell 6 regulation of cocaine-seeking behavior, a cell-specific monosynaptic tracing 7 strategy was not used in their study to show the anatomic connection 22 . 8 Therefore, we speculate that the PV response to vHPC afferent activation may 9 be indirect.

10
A previous study has shown that anterior BNST-associated activity exerts 11 anxiolytic influence on anxious states 49 . Several of our current findings are 12 consistent with this conclusion: specific activation of these afferents from the 13 adBNST resulted in a robust inhibition of accumbal PV activity (Fig. 6d) and 14 reduced avoidance coping behavior in response to anxiogenic stimuli ( Fig. 6i-j); 15 when PV function was ablated, the previously observed reduced avoidance 16 behavior was also abolished (Fig. 6k-n); PPR has been reported to reflect the 17 state of presynaptic input at synapses 50 . In our chronic stress model, we found 18 in the adBNST--sNAc circuit, there was an increase in the PPR, which 19 indicating that an impaired release of GABA onto sNAc PV cells unpon light 20 activation of these GABAergic terminals in the sNAc (Fig. 6o-q); meanwhile, 21 elevating the activity of GABAergic inputs to the NAc PV neurons rescued 22 anxiety-related excessive avoidance behavior, representing an anxiolytic effect 1 (Fig. 6r-t). Combine these findings, we summarised that adBNST sents 2 GABAergic inputs to sNAc to control avoidance behaviour, which is mediated 3 by sNAc PV neurons. 4 PV activity has been implicated in contributing to the theta rhythms in the 5 mPFC and hippocampus 25, 26 and our results also showed a marked decrease 6 in theta rhythm on stressed mice both in NAc and BNST (Fig. 1j-k); additionally, states and therefore, highly activated PV cells contribute to theta oscillations 13 changes either in BNST or NAc. Further study is needed to determine whether 14 PV neuronal activation in the accumbens is the main driver for the decrease in 15 theta oscillation within these two structures ( Fig. 1j-k), although in the present 16 study the increase in the coherence between FS spikes and LFP in theta range 17 does suggest that this may be the case (Fig. S2f).

18
In conclusion, our results provide strong evidence for accumbal PV neurons Anxiety represents a brain state: our study uncovered a new circuit mechanism, 3 precisely defined by the neuronal types involved, by which the stress response 4 brain region orchestrates the reward circuit component to exert direct effects 5 on anxious states. Our findings may help to explain why anxiety and addiction 6 are highly comorbid, although these two common psychiatric disorders engage 7 emotion and reward circuits, respectively.  The authors have declared that no conflict of interest exists. FosB isoforms throughout the brain by fluoxetine and chronic stress.   (n =9 mice per group, unpaired t test, *P < 0.05; **P < 0.01; ***P < 0.001). 2 3 4