Homeostatic scaling ensures behavioural stability during corticosterone negative feedback

Abstract

Maintaining appropriate behavioral and physiological responses in the face of challenge is essential for survival. The persistent increase in corticosteroids (CORT) during chronic stress blunts the endocrine response to any subsequent stressors. But the impact of prolonged CORT on behaviors that promote survival in the face of an acute stress is not well understood. Here we used an aerial predator threat model combined with in vivo calcium imaging, whole-cell electrophysiology, chemogenetics and computational modeling to evaluate the effects of short and long-term CORT. We show that in the short term, the activity of the corticotropin releasing hormone neurons of the paraventricular nucleus of the hypothalamus (CRH^PVN) and innate defensive behaviors that rely on these cells, are sensitive to the negative feedback effects of CORT. In response to long-term increases in CORT, however, behaviors recover, even though intrinsic CRH^PVN activity remains low. This escape from negative feedback requires local, homeostatic scaling of glutamate synapses that overcomes the inhibitory effects of CORT. This scaling is sufficient to maintain the output of this system in vivo and preserves innate defensive responses to threat. We propose that homeostatic synaptic scaling functions as a local adaptive mechanism to preserve the reliability of essential survival circuits during times of chronic stress.

Introduction

Acute stress triggers rapid behavioral and physiological changes essential for an organism survival. Those changes include but not limited to activation of the sympathetic nervous system, the selection of the appropriate defensive response and the activation of the endocrine stress axis that culminates in the increase of the circulating corticosterone (CORT). Persistent elevation of CORT poses negative feedback on the central controllers of CORT release, the corticotropin releasing hormone neurons of the paraventricular nucleus of the hypothalamus (CRH^PVN) [1, 2]. This negative feedback is associated with blunted CORT response to an acute stressor [3], but its impact on survival behaviors in response to an acute stress is not well understood.

CRH^PVN neurons are not only responsible for the endocrine response to stress; their activity also regulates stress-related behavioural response. Recent work indicates these cells also play an essential role in the choice of defensive behavior in response to predatory threat [4]. In response to the advance of a presumptive aerial predator, CRH^PVN neurons in mice show an increase in activity prior to and during active escape behavior, but not during passive freezing responses [4]. Furthermore, optogenetic silencing CRH^PVN decreases escape probability and increases passive defensive responses, suggesting the activity of these cells is strongly linked to the choice between active and defensive behavior.

Chronic stress, in which circulating CORT is persistently elevated, affects physiology, cognition and decision making [1, 3, 5–7]. Whether there is CORT-induced consequences that affect behaviors that rely on the activity of CRH^PVN neurons has not been investigated. Increased circulating CORT in a diurnal fashion has been shown to elicit depressive-like phenotypes [3, 7–9]. These observations allow us to predict that an increase in CORT, by decreasing the activity of CRH^PVN neurons, should promote a switch from active to passive defensive strategy in response to threat.

Here, we used in vivo single cell calcium imaging to evaluate CRH^PVN activity in freely behaving mice, assessed defensive behavior in response to perceived aerial attack, and employed chemogenetics and slice electrophysiology to provide mechanistic explanations of functional changes at the behaviour level. To model the homeostatic pressure caused by prolonged CORT increase, mice were given free access to CORT in the drinking water for one or seven days. As expected, one day of CORT decreases the activity of CRH^PVN neurons in vivo and in vitro. This negative activity pressure is accompanied by a decrease in escape response to perceived aerial attack. Following seven consecutive days of CORT, this negative feedback pressure is maintained at the level of individual CRH^PVN neurons, but the in vivo activity and escape behavior are restored. This restoration is due to an upscaling of glutamate synapses on CRH^PVN neurons which is sufficient to overcome the CORT negative feedback. Furthermore, local chemogenetic inhibition of these cells provides support for a local mechanism and does not require involvement of upstream circuits. Our findings show that CRH^PVN neurons use homeostatic synaptic plasticity to escape negative activity pressure and maintain essential survival behaviors.

Materials and methods

Animals

All animal protocols were approved by the University of Calgary Animal Care and Use Committee (AC21-0067). Male and female mice were housed on a 12-h:12-h light:dark cycle (lights on at 7:00 a.m.) with ad libitum access to food and water in whole litters. They were then individually housed during the experimental phase Crh-IRES-Cre (B6(Cg)-Crhtm1(cre)Zjh/J; stock number 012704) mice or these mice crossed with Ai14 (Ai14(tm14(CAG-tdTomato)Hze); stock number 007914) were used for electrophysiological experiments and behavior experiments. Mice aged P22-30 were used in electrophysiological experiments, while those undergoing surgery were P42-56 at the time. For the behavioral and fiber photometry experiments, animals were P60-P80. For the experiments involving miniature microscopy, Crh-IRES-Cre mice, crossed with Ai148 (Ai148(TIT2L-GC6f-ICL-tTA2)- D; stock number 030328) were P80-P120. Mice were obtained from Jackson Laboratories. Animals from different litters were randomly assigned to experimental groups.

CORT and EtOH solutions

Corticosterone (Sigma) was dissolved in 95% EtOH and sonicated for 2 min to ensure homogenous dissolution before being diluted to a final solution of 25 μg/ml CORT in 0.95% EtOH with tap water. For the EtOH solution, 95% EtOH was diluted to 1% with tap water. During experiments, mice had their regular drinking water replaced with either solution. Solutions were replaced every 2–3 days and the amount of solution consumed was recorded. All water bottles were wrapped with foil to prevent light exposure.

Corticosterone immunoassay

Plasma was extracted at the distal portion of the tail by making a small incision of the tail vein. A microvette tube (CB 300 Z, Sarstedt AG & Co) was used to collect plasma, prior to centrifugation (8,000 RPM for 20 mins, at 4 °C). Plasma was separated from the hematocrit and was stored in an Eppendorf tube at −20 °C until further analysis with the DetectX Corticosterone Immunoassay Kit (Arbor Assay). Samples were run in triplicate and averaged to obtain the animal’s plasma CORT concentration. Samples from each experimental group were run on the same plate to control for deviations in the standard curve generated for each plate.

Looming shadow

The looming shadow paradigm was performed similar as described previously by our lab [4, 10]. Briefly, mice were placed in a 50 × 30 × 40 cm plastic arena with a 21-inch LCD monitor positioned above the arena facing downward to display a visual stimulus. At one end of the arena, a shelter was provided (13 × 12 × 10 cm) under which the mice could hide. A frontal-angled camera was placed to record the behavioral responses. Mice were habituated to the test arena for three days before the test (15 min a day). On the test day, following 3 min of habituation to the arena, 5 overhead looming visual stimuli were presented with a 1 min inter-trial interval. The looming stimulus was a 2 cm black disk that expanded to 20 cm in 3 distinct phases: (i) 2 cm disc presented for 3 s, (ii) expanded to full size (20 cm) in 2 s and, (iii) stable for 3 additional seconds.

Looming shadow analysis

The following criteria were used to discriminate the different behaviors: 1) A flight response during the 8 s visual stimulus that resulted in the subject reaching the shelter, was classified as an escape; 2) A response identified as (i) one episode with absence of any movement except that related to breathing was classified as a freeze; (ii) multiple freezing episodes, interspersed with brief periods of movement; 3) If the subject did not freeze, failed to alter ongoing behavior, or made an escape attempt, but did not reach the shelter during the 8 s stimulus presentation, this was classified as a non-responder. Additional measures such as escape reaction time and distance to shelter were analyzed using a custom-made analysis tool. If the animal did not leave the shelter for 10 consecutive minutes, the test was terminated, and the remaining trials were classified as no trial. Behavioral analysis was conducted by an experimenter who was blind to the treatment group.

Novel environment

Following the baseline recording in the home cage, mice were placed in a 41 cm × 19 cm × 20.5 cm clean and transparent plastic arena for 5 min. The same environment was used for the other 2 exposures.

Electrophysiology

Mice were anesthetized with isoflurane and decapitated. The brain was rapidly extracted and submerged in freezing slicing solution (0 °C, 95% O₂/5% CO₂ saturated) containing 87 mM NaCl, 2.5 mM KCl, 0.5 mM CaCl₂, 7 mM MgCl₂, 25 mM NaHCO₃, 25 mM D-glucose, 1.25 mM NaH₂PO₄ and 75 mM sucrose. A vibratome (Leica) was used to take coronal slices (thickness = 250 µm). The PVN was hemisected along the 3^rd ventricle, and sections were incubated for 1 h in artificial cerebral spinal fluid (aCSF) (30 °C, 95% O₂/5% CO₂ saturated) containing 126 mM NaCl, 2.5 mM KCl, 26 mM NaHCO₃, 2.5 mM CaCl₂, 1.5 mM MgCl₂, 1.25 mM NaH₂PO₄ and 10 mM glucose. Immediately prior to recording, slices were transferred to a temperature-controlled recording chamber superfused with aCSF (1 ml/min, 30–32°C, 95% O₂/5% CO₂). Slices were visualized using an Olympus BX51WI upright microscope fitted with infrared differential interference contrast optics. Pulled borosilicate glass pipettes (3–6 MΩ tip resistance) were filled with internal solution containing 108 mM potassium gluconate, 2 mM MgCl₂, 8 mM sodium gluconate, 8 mM KCl, 1 mM K2-EGTA, 4 mM K2-ATP, 0.3 mM Na3-GTP and 10 mM HEPES. CRH^PVN neurons were identified by CRE-dependent expression of tdTomato fluorescent marker [11].

Traces were amplified (MultiClamp 700 A), digitized at 100 kHz (DigiData 1322 A), low-pass (Bessel) filtered at 1 kHz, and recorded (Clampex 9.2, Molecular Devices) for analysis. mEPSCs were recorded for 3 mins at −63 mV in aCSF solution containing tetrodotoxin (1 μM, Alamone) and picrotoxin (100 μM, Sigma). For sIPSCs, the recording chamber contained DNQX (10 µM, Tocris) and the membrane potential was clamped at 0 mV to produce robust IPSCs. Evoked currents and current-clamp recordings were analyzed with Clampfit, while mEPSCs and sIPSCs were analyzed with Minianalysis. For mEPSCs, a detection threshold of 8.5 pA was used based on 5x RMS noise analysis (RMS = 1.7 pA). For experiments requiring evoked EPSCs, AMPA currents were electrically evoked at −70 mV in picrotoxin (100 μM, Sigma) with a stimulating electrode located in the paraventricular aspect as previously described [12]. To determine Paired-pulse Ratio (PPR), a protocol with paired stimulation 50 ms apart at 0.1 Hz intervals was used. PPR was calculated as the peak amplitude of P₂/P₁. For all recordings, access resistance during voltage-clamp recordings was monitored. Cells were discarded if access resistance was >20 MΩ, or if access resistance changed >15% during recordings. Each neuron’s capacitance and membrane resistance were recorded upon entering the cell. For current clamp experiments, cells were held at −70 mV, and incremental current steps of 10 pA (from −40 pA to 70 pA, at 0.5 s) were delivered at 1 Hz. The F-I curves were generated using the number of action potentials evoked during each current step. The membrane I-V curve was created using the average membrane voltage at the plateau of the negative current steps. For experiments involving γ-DGG, mEPSCs were first recorded for 3 min. Then, γ-DGG was infused into the bath (bath concentration was 500 µM) for 5 min, and mEPSCs were recorded for another 3 min. The first 150 events were normalized to the largest event before and after the antagonist was added to generate the cumulative distributions. A ratio of mean amplitude before and after addition of the antagonist was calculated for comparison.

For local CNO experiments, Crh-IRES-Cre mice that expressed the hM4DG_i-mCherry construct were used. CRH^PVN neurons were identified by the presence of mCherry. A recording pipette was used to gently approach the cell and land on the soma. Gap free recording was performed in bridge-mode to record spontaneous action-potentials. A second pipette connected to a Picospritzer II microcellular injector (General Valve) that contained 10 µM CNO dissolved in CSF was placed adjacent to the cell. 20 psi was used to deliver a 1 s pulse of CNO from the pipette adjacent to the cell while simultaneously recording spontaneous action potentials with the recording pipette. The recordings were continued well after the delivery of CNO to ensure the force from local delivery did not disrupt the recording electrode.

Synaptic scaling analysis

To assess the synaptic scaling, we compiled the amplitudes of the first 250 mEPSCs from each cell. We then employed a method characterized by Kim et al., [13] to assess for synaptic scaling [13]. To assess for the scaling factor, the amplitude distribution from CORT-treated animals was divided by several factors to achieve the highest P-value, when compared against the distribution from naive animals, using the Kolmogorov-Smirnov (KS) test. As the distributions from CORT-treated animals were scaled back, mEPSCs that fell below the 8.5 pA detection threshold were discarded. The factor that achieved the highest P-value was considered to be the scaling factor.

Stereotaxic injection

6–8 weeks old Crh-IRES-Cre mice were maintained under isoflurane anesthesia in the stereotaxic apparatus. A glass capillary containing a Cre-dependent AAV construct with hM4D(Gi)-mCherry or mCherry (AAV-CAG-DIO-hM4D(Gi)-mCherry ;AAV-CAG-DIO-mCherry; Penn Vector Core) was lowered into the brain (anteroposterior (AP), −0.7 mm; lateral (L), −0.3 mm from the bregma; dorsoventral (DV), −4.5 mm from the dura. The virus was pressure injected with a Nanoject II apparatus (Drummond Scientific Company) in a total volume of 210 nl. Two weeks were allowed for recovery before any manipulation. Animals in which inhibition of activity was observed in CRH^PVN cells in response to a CNO puff during electrophysiological recordings were included in the final analyzed sample for each group (1-day CNO and 7-days CNO groups).

GRIN Lens implantation

The lens implantation procedure was performed as previously described [14, 15]. Mice were 6–8 weeks old at the time of ferrule or lens implantation. Crh-IRES-Cre;Ai148 mice were maintained under isoflurane anesthesia in the stereotaxic apparatus. A midline incision was made along the scalp and retracted to expose the suture lines. A high-speed drill was used to make a craniectomy centered at 0.7 mm posterior, and 0.3 mm lateral to bregma. A small incision was made to penetrate the dura and leptomeninges. A motorized stereotactic apparatus was used to implant a GRIN lens (Length = 7.3 mm, Inscopix) at a depth of 4.5 mm deep to the dura. The lens was targeted to 200 µm dorsal to the PVN to accommodate for the focal distance of the GRIN lens. Once in position, the lens was fastened to the skull with METABOND® and dental cement. The camera baseplate was installed at minimum of 4 weeks after lens implantation. Experiments began after 2 weeks of recovery, followed by 5 episodes of habituative handling. This was then followed by 5, 20 min periods where animals were habituated to the weight of the miniscope camera.

Miniature microscopy recordings

Activity of CRH^PVN neurons was recorded continuously for at least 15 min at 20 FPS, 40% LED power and 2.5 gain with an nVista or nVoke miniature microscope (Inscopix) using the nVista Acquisition Software (Inscopix). On the day of recordings, the camera was fixed to the baseplate while animals were in their homecage. The recordings began with animals in their homecage. After 5 mins, animals were transferred to the novel environment for 5 mins and allowed to freely explore the space. They were then transferred back to the homecage where recordings continued for another 5 mins before the camera was removed from the baseplate affixed to the head.

Miniature microscopy analysis

The video files were cropped with ImageJ and the data analysis was performed by the MIN1PIPE script in MATLAB [16]. The raw data was downscaled by a factor of 0.25 spatial and 0.5 temporal. Regions of interest corresponding to CRH^PVN neurons were identified by the software, and signal intensity from each ROI was calculated by averaging the dF/F results sampled at 0.1 s over a 3-min long period, corresponding to when the animal was in the novel environment, and multiplied by 100. Longitudinal alignment of recordings from different days was made as described previously [14].

Computational model

The modified adaptive exponential integrate-and-fire model

CRH^PVN were fit to a modified Adaptive Exponential Integrate-and-Fire Model [17, 18]. The modified model is given by the equations:

$$C\frac{dV}{dt}=-g_L\left(V-E_L\right)+g_L\mathrm{\Delta }\exp \left(\frac{V-V_T}{\mathrm{\Delta }}\right)-w+I$$ 1

$${\tau }_w\frac{dw}{dt}=a\left(V-E_L\right)-w$$ 2

$${\tau }_{\phi }\frac{d\phi }{dt}={\phi }_0-\phi ,$$ 3

where $V\left(t\right)$ serves as the membrane potential of the neuron, $w\left(t\right)$ serves as the adaptation variable, which helps set the steady state firing rate of a neuron (among other features), and $\phi =\left[V_T,V_r,V_p\right]$ represents the voltage threshold, the voltage reset and the peak voltage, respectively, for the neuron. The voltage threshold partially influences the shape of a spike and the rheobase of a neuron, while the voltage reset and peak voltages define the amplitude of a spike and its immediate after-spike behaviour. The set of new equations were added based on empirical observations that the threshold and spike amplitudes vary according to neuronal spike frequency in previous work [19]. The other parameters for the AdEX model include; the membrane capacitance $C$, the external or applied current $I$, the leak conductance $g_L$, the resting membrane potential $E_L$, a slope factor $\mathrm{\Delta }$ for the threshold, an, adaptation time constant ${\tau }_w,$ and finally, the adaptation strength parameter $a$. Since $V_T,V_r$ and $V_p$ are now variables, ${\phi }_0$ and ${\tau }_{\mathrm{\phi }}$ stand for their respective resting values ${\phi }_0=\left[V_{T0},V_{r0},V_{p0}\right]$ and time constants ${\tau }_{\phi }=\left[{\tau }_T,{\tau }_r,{\tau }_p\right]$.

When the membrane potential $V\left(t\right)$ reaches the peak $V\left(t^{-}\right)\ge V_p$ the state variables are all updated according to the following rules:

$$\begin{array}{c}\hfill V\left(t^{+}\right)\to V_r,\hfill \\ \hfill w\left(t^{+}\right)\to w\left(t^{-}\right)+b,\hfill \\ \hfill V_T\left(t^{+}\right)\to V_T\left(t^{-}\right)+q,\hfill \\ \hfill V_r\left(t^{+}\right)\to V_r\left(t^{-}\right)+r,\hfill \\ \hfill V_p\left(t^{+}\right)\to V_p\left(t^{-}\right)+p\hfill \end{array}$$ 4

The parameters associated with Eq. (4) are the adaptation increment $b$, threshold increment $q$, the reset increment $r$ and, the voltage peak increment $p$. From our previous work in modelling CRH^PVN neurons [19], we also empirically observed a linear dependence of the resting $V_{p0}$ and $V_{r0}$ on the external current $I$ given by:

$$\begin{array}{c}\hfill V_{p0}={\alpha }_1+{\beta }_{1}I\hfill \\ \hfill V_{r0}={\alpha }_2+{\beta }_{2}I,\hfill \end{array}$$ 5

where $\alpha$ and $\beta$ are the intercept and slope coefficients of the linear dependence

As the current $I$ was determined by the patching current-step protocol the final set of parameters to be determined were $x=\left[C,g_L,E_L,\mathrm{\Delta },{\tau }_w,a,h,{\tau }_T,{\tau }_r,{\tau }_p,b,q,r,p,{\alpha }_1,{\alpha }_2,{\beta }_1,{\beta }_2\right]$. The fits were performed in Matlab2020a using Euler Method with 0.05 ms integration step. The parameters were fit using particle swarm optimization, with the details reported in [19].

Synaptic currents

To correctly simulate the synaptic currents all the miniature Excitatory Post-Synaptic Currents (mEPSC) were extracted from data (black lines in Fig. 4c). The mean pulse was calculated (blue line in Fig. 4b) and modeled by a double exponential function (dashed red line in Fig. 4b) [20, 21] as follows:

$$EPSC=E_{Amp}\frac{\left(e^{-t/{\tau }_d}-e^{-t/{\tau }_r}\right)}{E_{norm}}$$ 6

Fig. 4. Computational modelling reveals precise tuning of excitatory currents and intrinsic excitability.

a AdEX modelling of empirically recorded voltage traces elicited by current clamps. b Quantification of rheobase current distribution from neurons modelled after control and 7-Day CORT animals. The rheobase current is estimated from the model parameters (U = 331, P = 0.0006, Mann Whitney test). c Double-exponential modelling after empirically recorded mEPSCs. The double exponential model features a rise time and decay time for EPSCs. d Schematic showing construction of network models CRH^PVN neurons. The neuron and synaptic models are those fit to either naïve (black) or CORT exposure (orange). e PSTH activity of the 4 models generated with using either naïve/CORT synapses and neuron models. f Steady state activity ratio of control model and experimental models with varying synaptic strength and EPSC input frequency. g Steady state activity ratio of control model and experimental models in response to varying intensity of rheobase current and EPSC input frequency.

Here $E_{amp}$ represents the pulse amplitude. The distribution of amplitudes extracted from both Control and CORT mEPSCs are shown in Fig. 4c. The decay/rise time constants ${\tau }_d/{\tau }_r$ were calculated by fitting the mean EPSC pulse to Eq. (6) where $E_{norm}$ normalizes the term inside the parenthesis and is given by:

$$E_{norm}=\frac{{\tau }_d{\tau }_r}{{\tau }_d-{\tau }_r}\ln \left(\frac{{\tau }_d}{{\tau }_r}\right)$$ 7

To fit the mean pulse waveform we used the function “lsqcurvefit” in Matlab2020a with initial conditions $\left[E_{amp},{\tau }_d,{\tau }_r\right]=\left[25pA,1.5ms,0.5ms\right]$, and the respective lower $L_b=\left[8,0.01,0.01\right]$ and upper $U_b=\left[200,20,20\right]$ boundary conditions. The final fit parameters for the Control condition were ${\tau }_d=1.234\mathrm{ms}$, ${\tau }_r=0.276\mathrm{ms}$ and for the CORT condition ${\tau }_d=1.460\mathrm{ms}$, and ${\tau }_r=0.293\mathrm{ms}$. To generate the synaptic currents in the network simulation we fixed the decay and rise times to the empirically fit values and the pulse amplitudes were randomly generated according to the empirical distribution (as depicted on Fig. 4c).

Network simulations

Here we describe the details of the network simulation. We considered a network composed of $N=5000$ uncoupled neurons, where each neuron $i$ receives a constant current $I$ (same for all) and an independently generated time series ${EPSC}_i\left(t\right)$. This time series is randomly generated from a Poisson distribution with frequency ${\gamma }_{EPSC}$. The EPSC current represents all the excitatory synaptic inputs, therefore controlling its frequency can be associated with an increase/decrease of inputs. The neuronal model is given by:

$$C_i\frac{d{V}_i}{dt}=-{g_L}_i\left(V_i-{E_L}_i\right)+{g_L}_i{\mathrm{\Delta }}_i\exp \left(\frac{V_i-{V_T}_i}{{\mathrm{\Delta }}_i}\right)-w_i+I+{EPSC}_i$$ 8

$${{\tau }_w}_i\frac{d{w}_i}{dt}=a_i\left(V_i-{E_L}_i\right)-w_i$$ 9

$${{\tau }_{\phi }}_i\frac{d{\phi }_i}{dt}={{\phi }_0}_i-{\phi }_i$$ 10

The choice of parameters for each neuron $i$ was made by a random selection from the empirically estimated parameter sets of CRH^PVN neurons, in other words the network is composed of randomly selected, but parameter fit neurons from one of the conditions, Control or CORT.

We performed network simulations covering 4 distinct scenarios, regarding the type of neuron (Control or CORT) and EPSC waveform (Control or CORT). To better understand the conditions for a homeostatic response figure (F) left gives the ratio between the steady state PSTH of Control-network/Control-EPSC over Cort-network/Cort-EPSC as a function ${\gamma }_{EPSC}$ for several $I$. The homeostatic responses are reached when the ratio is close to 1. The homeostatic response can also be reached in the network simulation by a scaling factor $s_f=1.33$ (Fig. 4f) where the CORT probability distribution function (PDF) of EPSC amplitudes was given by a simple linear PDF transformation over the Control distribution. Let $X$ be a random variable representing the Control amplitudes with a PDF $f\left(X\right)$, and let $Y=s_{f}X$ be a variable representing the CORT amplitudes where $s_f>0$, $Y$ has a probability density function $g\left(Y\right)$ given by:

$$g\left(Y\right)=\frac{1}{s_f}f\left(\frac{Y}{s_f}\right)$$ 11

The network simulations were performed in Matlab2020a using Euler integration method with $0.005\mathrm{ms}$ integration step.

Model fitting

The parameters for the AdEx Model fitting was performed as previously described [19].

Statistics

GraphPad Prism 10.0 software was used for statistical analysis. When comparing means from two independent groups, different time points unpaired t-test (two-tailed) were used. When comparing the means of multiple groups, one-way or two-way ANOVA were used, followed by Tukey’s corrections for multiple comparisons. When analyzing mEPSC amplitudes from each condition plotted on a cumulative distribution, Kolmogorov-Smirnov (K-S) statistics was used. The behavioral data for the looming shadow task were analyzed using Fisher’s exact test to compare escape proportions between groups.

Replication: The observation that PVN CRH neuron activity is linked to innate escape behavior has been reproduced in multiples cohorts of mice. Experiments have been conducted by multiple lab members.

Results

Acute CORT feedback decreases CRH^PVN neuron excitability and promotes passive defensive strategies

We tested the effects of acute (1 day) and long (7 days) CORT treatment on the intrinsic excitability of CRH^PVN neurons and defensive behavior selection in the looming shadow task. Mice were placed in an arena with an overhead screen that projects a rapidly enlarging shadow to the underlying surface, which mimics the shadow of an advancing aerial predator [4, 10, 22]. In response to the stimulus, mice either freeze in place or flee to a shelter in the corner of the arena (Fig. 1a, b, c). We previously showed that escape is the preferred response in this paradigm (80%), but this is decreased if CRH^PVN neurons are optogenetically silenced [4]. Feedback from CORT blunts CRH^PVN activity by augmenting inhibitory potassium currents [6]. Therefore, we aimed to test whether the inhibitory CORT feedback of CRH^PVN neurons would disrupt behavioral responses to the looming shadow task.

Fig. 1. CORT decreases CRH^PVN neuron excitability and promotes passive defensive strategies during acute feedback.

a Animals were exposed to 5 trials of looming shadow while in the arena. Each trial started with a small, static shadow projected from above to the surface of the arena for 3 s. The shadow then rapidly increases to its final size in 2 s and remains static for another 3 s. b The animal’s decision to escape or freeze was recorded after each trial. c (Above) Whole-cell current clamp recordings from naïve (N = 24 cells) and 1-Day CORT (N = 23 cells) animals (20 pA and 40 pA steps shown). (Below) F-I plots showing CRH^PVN neuron response to varying levels of current injection in naïve and 1-Day CORT animals. A two-way ANOVA with the following factors was used for comparison: Current step (F(7315) = 547.3, P < 0.0001), treatment (F(1,45) = 18.75), P < 0.0001, interaction (F(7315) = 13.39, P < 0.0001). d Behavioral results from the two cohorts. The grid represents the five trials (across) for each animal (down) color matched to the behavioural response on each trial (Naïve N = 10 mice, 1-Day CORT N = 8 mice). e (Above) Current-clamp recordings from naïve (N = 20 cells) and 7- days CORT mice (N = 12 cells; 20 pA and 40 pA steps shown). (Below) F–I plot showing response to successive current steps in naïve and 7-Day CORT animals. A two-way ANOVA with the following factors was used for comparison: Current step (F(7189) = 266.2, P < 0.0001), treatment (F(1,27) = 7.618, P = 0.0103), interaction (F(7189) = 2.773, P = 0.0091). f Grid showing color matched behavioural response from mice given CORT for 7 days (N = 8 mice). g Circular plot showing percentage of escape, freeze, and no response. Compared to naïve, 1-Day CORT forced a preference towards freezing in lieu of escape behavior (P < 0.0001, Fisher exact test). The probability of escape was rescued after 7 days of CORT when compared to control (P = 0.4570, Fisher’s exact test). Data shown as mean ± sem.

We replaced the regular drinking water with a solution containing 25 µg/ml CORT [3, 7] for 1 day or 7 days. This increases plasma CORT during the nocturnal phase (when animals are consuming the solution) and is comparable with CORT responses in experimental models of stress [3, 7] (SFig. 1A, B). In the first set of experiments, we prepared acute brain slices containing the PVN from 2 groups of Crh-IRES-Cre;Ai148 mice that express the tdTomato reporter in CRH-expressing neurons [11]: mice with access to water + 1% EtOH (vehicle-control) and mice with access to the CORT solution (1-Day CORT). Whole-cell current-clamp recordings of CRH^PVN neurons were performed to evaluate intrinsic excitability in each group. Following 1 day of CORT, CRH^PVN neurons showed a decrease in intrinsic excitability compared to control mice (Fig. 1c). The EtOH vehicle had no effect on the F-I relationship (SFig. 2). CORT had no effect on passive membrane properties (SFig. 2). The reduction in intrinsic excitability of CRH^PVN neurons following 1-Day CORT is consistent with previous work showing a decrease in CRH^PVN neuron excitability following direct CORT application or behaviorally-induced increases in CORT that were less than one day [6]. We then evaluated the impact of 1-day CORT on defensive behaviors in the looming shadow. Consistent with reduced CRH^PVN neuron excitability, CORT-treated mice showed a significant increase in freezing behavior and decrease in escape in comparison to control animals (Fig. 1d). Further analysis confirmed that the variation in behavioral outcomes was not influenced by the distance to the shelter, and escape time remained unaffected when the animals exhibited escape responses (SFig. 3A,B). Next, we performed similar experiments on a new cohort of animals treated with CORT for 7 days. Whole-cell current-clamp recordings revealed a similar decrease in intrinsic excitability, indicating that negative feedback was maintained during the 7-Day CORT exposure (Fig. 1e). In addition, on-cell recordings were performed to measure spontaneous activity of CRH^PVN neurons. Cells from CORT-treated mice had a significantly lower firing rate when compared to naïve, and 1% EtOH controls (SFig. 4). However, the 7-Day CORT mice showed robust escape behavior at rates consistent with those observed in control mice (Fig. 1f, g, SFig. 3a). Since the intrinsic activity of CRH^PVN neurons was still reduced, the restoration of canonical behavioral response to aerial threat strongly hinted at the existence of a compensatory mechanism that allowed CRH^PVN neurons to circumvent the chronic CORT-induced inhibitory activity pressure.

Prolonged CORT feedback induces synaptic scaling in CRH^PVN neurons

Persistent decreases in neuronal activity have been shown to provoke compensatory changes in the strength of synapses onto CRH^PVN cells. This process, described as homeostatic synaptic scaling, has been implicated as one of several mechanisms neuronal networks may employ to achieve stability and maintain functional output [23–30]. We hypothesized that a homeostatic process, such as synaptic scaling, may overcome negative activity pressure from CORT that ultimately will ensure behavioral stability.

To interrogate synaptic changes during CORT feedback, we performed whole-cell voltage-clamp experiments to assess miniature excitatory post-synaptic currents (mEPSCs) onto CRH^PVN neurons from mice given CORT for 1 or 7 days. Excitatory quantal currents were isolated by addition of picrotoxin (100 µM) and tetrodotoxin (1 µM) to the recording chamber. 1 day of CORT had no effect on the amplitude or frequency of mEPSCs (SFig. 3). Following 7 days of CORT, we observed an increase in mEPSC amplitude (Fig. 2a). No change in frequency of mEPSCs was observed (SFig.4A). There was no significant change in membrane resistance, capacitance, or access resistance (SFig. 2C); 7-Day EtOH exposure (1% EtOH) had no effect on the amplitude of mEPSCs (SFig. 4B). Finally, 7 days of CORT had no effect on the amplitude or frequency of spontaneous inhibitory post-synaptic currents (sIPSCs) (SFig. 5). These findings demonstrate that prolonged exposure to CORT is associated with an increase in mEPSC amplitude, and is suggestive of a homeostatic process involving excitatory synapses.

Fig. 2. Prolonged CORT feedback induces synaptic scaling in CRH^PVN neurons.

a (Left) Representative voltage-clamp recording mEPSCs from CRH^PVN neurons in naïve (N = 11 cells) and 7-Day CORT (N = 12 cells) animals. (Right) quantification of mEPSC amplitude (t = 3.844, df = 21, P = 0.0009, t-test). b Rank-order plot of mEPSC amplitudes from naïve (slope = 1) vs 7-Day CORT treated animals (y = 1.31X – 2.01, R² = 0.9990, P < 0.0001, linear regression). c Cumulative distribution showing mEPSC amplitudes from naïve, 7-Day CORT animals and scaled values (naïve vs 7-Day CORT, KS D = 0.2864, P < 0.0001, KS test). Inset: The amplitude distribution of mEPSCs from CORT animals was scaled by several factors to define the scaling factor as determined by the KS. The scaling factor, 1.33 yielded the highest P-value (0.7368) (naïve vs scaled, KS D = 0.0569, P = 0.9885, KS test). d Average mEPSC from naïve (N = 12 cells) overlaid with scaled mEPSC after 7 days CORT (N = 12 cells). e Quantification of 10–90 rise time with average mEPSC from naïve and 7-Day CORT (t = 0.6818, df = 22, P = 0.5025, t-test). Whole-cell patch-clamp configuration with stimulating electrode along the paraventricular aspect. f (Left) Paired evoked post-synaptic glutamate currents from naïve (N = 9 cells) and 7-Day CORT (N = 8 cells) animals. (Right) Quantification of PPR (t = 0.7137, df = 15, P = 0.4863, t-test). g (Left) Cumulative distributions of mEPSC/mEPSCmax amplitude before and after addition of γDGG in naïve (N = 5 cells) and 7-Day CORT (N = 4 cells) animals. (Right) quantification of amplitude fraction of mean mEPSC amplitude before and after addition of γDGG (t = 0.0184, df = 7, P = 0.9858, t-test). Data shown as mean ± sem.

We employed a validated approach [13] to determine whether the increase in mEPSC amplitude after 7-Day CORT was consistent with synaptic scaling. First, we took the first 250 mEPSCs recorded from 12 cells in each group. The events were rank-ordered and plotted against each other as a function of X,Y. A linear regression yielded a slope of 1.31 (R² = 0.99) (Fig. 2b). This is consistent with a multiplicative increase across the entire amplitude distribution. An iterative approach was then used to determine the precise scaling factor. Here, the amplitude distribution from the CORT-treated group was divided by several factors (between 1.15 – 1.48, in increments of 0.01) and compared to the control distribution using Kolmogorov-Smirnov (KS) statistics. A factor of 1.33 yielded the highest P-value (Fig. 2c, d). This analysis indicates that prolonged CORT feedback results in a multiplicative synaptic scaling of excitatory synapses by a factor of 1.33. Interestingly, this reduced excitability after 7 days of CORT was not mantained if the treatment was removed. Excitatory synapses mEPSC apmplitude were back to control levels after 6 days following the termination of the CORT treatment (SFig. 8).

Multiplicative scaling typically results from a coordinated increase in post-synaptic receptor density [31], but alterations in receptor activation kinetics or pre-synaptic mechanisms may also be involved [29, 32]. We conducted a number of electrophysiological experiments to evaluate whether changes in receptor kinetics and pre-synaptic involvement could account for increased mEPSC amplitude. We observed no difference in the mean 10–90 rise-times (Fig. 2e) or decay time (SFig. 6c) of the events from each cell recorded in control and 7-Day CORT animals indicating channel kinetics were not altered by CORT.

Next, to evaluate presynaptic release probability, we performed voltage-clamp recordings from CRH^PVN neurons while electrically stimulating the adjacent neuropil. In the presence of picrotoxin (to isolate synaptic glutamate currents), we delivered two pulses (P₁ and P₂), 50 ms apart, to elicit glutamate release, and recorded the resulting post-synaptic currents. We observed no differences in paired pulse ratio (PPR) between cells from control and 7-Day CORT mice (Fig. 2f), indicating CORT had no effect on vesicular release probability. This does not, however, rule out any changes in quantal content that may be a result of increased glutamate packaging or multi-vesicular release.

Increased vesicular glutamate packaging has been demonstrated in CRH^PVN neurons after exposure to noradrenaline [33]. This increased quantal content could potentially result in synaptic scaling if coordinated across all presynaptic terminals. To evaluate changes in quantal packaging after CORT, we used sub-saturating concentrations of the low-affinity α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR) antagonist, γ-D-glutamaglycine (γ-DGG) and assessed mEPSC amplitude. The rationale was that since glutamate has a higher affinity for the AMPAR binding site, γ-DGG will be less effective at blocking AMPAR when cleft concentrations of glutamate are higher [33, 34]. If, however, the glutamate concentration in the cleft is unchanged by CORT, then the relative block of mEPSCs by γ-DGG should be similar in control and CORT-treated animals. We first recorded baseline mEPSCs from control and CORT treated animals. Then, γ-DGG (500 µM) was infused into the recording chamber for 5 mins, and a second set of mEPSCs, from the same neuron was recorded. Events from before and after infusion of γ-DGG were normalized to the largest event in each condition. A cumulative distribution of mEPSC amplitude (expressed as the fraction of mEPSC/mEPSC_max) from before and after addition of the antagonist was made to assess the effect of γ-DGG across the entire distribution (Fig. 2g). As previously demonstrated, the cumulative distributions show γ-DGG had a preferential effect on lower amplitude events [33]. This confirmed the presence of γ-DGG in the recording chamber in this experiment. The amplitude fraction before and during addition of the antagonist was not different between groups, indicating that glutamate concentration within the synaptic cleft is unaffected by CORT (Fig. 2g). Altogether, these experiments demonstrate that pre-synaptic properties and post-synaptic receptor conduction kinetics are not influenced by CORT. This strongly argues against multivesicular release accounting for increased mEPSC amplitude after 7 days of CORT. Therefore, we conclude that post-synaptic receptor accumulation is the most plausible mechanism for synaptic scaling.

Synaptic scaling rescues CRH^PVN neuron activity

To interrogate the effects of CORT feedback on CRH^PVN neuron activity in vivo, gradient index of refraction (GRIN) lenses were implanted immediately dorsal to the PVN in Crh-IRES-Cre;Ai148 mice that express GCAMP6f specifically in CRH neurons (Fig. 3a). We hypothesized that CRH^PVN neuron activity would be blunted after 1 day CORT, but synaptic scaling would rescue activity at 7 days. After implantation, animals were allowed 2–4 weeks for recovery and were then habituated to the weight of the camera prior to beginning the experiment (Fig. 3b). On the recording days, a miniature camera was affixed to the animals while in their homecage. After 5 min, mice were placed in a novel environment (NE) for 5 min, which we have recently shown evokes a reliable increase in CRH^PVN neuron activity [14]. This allowed us to record activity levels from multiple, individual CRH^PVN neurons simultaneously. We first recorded from mice prior to CORT exposure (Day 0). Immediately after the NE, animals were returned to their homecage, and their drinking water was replaced with the CORT solution. We then repeated the recordings on the following day (Day 1), and again on Day 7 (Fig. 3c). On Day 1, there was a significant reduction in CRH^PVN neuron activity as measured by the mean GCaMP6f signal intensity of each cell over the course of the recording (Fig. 3d). This is consistent with CORT negative activity pressure. On Day 7 however, the mean fluorescence of the neurons had returned to a level that was statistically unchanged from Day 0, indicating a return to baseline activity after 7 days of CORT feedback (Fig. 3d). To test whether these changes in neuronal activity are apparent in individual neurons we tracked the cells over the 7 days. Either quantified mean GCaMP6f signal (Fig. 3e) or with event detection (SFig. 9), the CORT induced reduction of neuronal activity on Day 1 and the subsequent recovery on Day 7 was recorded in the same cells over the days. These observations show a recovery of CRH^PVN neuron activity in vivo and confirm our hypothesis that synaptic scaling re-establishes the activity of CRH^PVN neurons.

Fig. 3. Synaptic scaling rescues CRH^PVN neuron activity.

a Crh-IRES-Cre:Ai148 mice expressing GCaMP6f in CRH^PVN were used for in vivo imaging. GRIN lens were unilaterally implanted dorsal to PVN (N = 5 mice) b Timeline from date of implantation to recovery, habituation, and experimental recordings. c Representative miniature microscope traces showing in vivo CRH^PVN neuron activity on Day 0, Day 1, and Day 7 of CORT treatment. d Quantification of mean neuronal fluorescence (F(2377) = 9.358, P = 0.0001, Day 0 vs Day 1: P < 0.0001; Day 0 vs Day 7: P = 0.4078; Day 1 vs Day 7: P = 0.0096; one-way ANOVA, Tukey’s multiple comparisons test). e Quantification of mean neuronal fluorescent in tracked cell over days (F(2114) = 13.77, P < 0.0001, Day 0 vs Day 1: P < 0.0001; Day 0 vs Day 7: P = 0.0506; Day 1 vs Day 7: P = 0.0135; one-way ANOVA, Tukey’s multiple comparisons test, N = 4 mice).

Computational modelling reveals tuning of excitatory currents and intrinsic excitability

To test this hypothesis further, we developed a computational model that allowed us to manipulate activity parameters and assess the precision required for synaptic scaling to offset changes in intrinsic excitability. A modified adaptive exponential integrate-and-fire (AdEX) model was fitted to a population of multiple single-neuron outputs over a range of input currents based on CRH^PVN recordings from control and 7-Day CORT animals. Particle swarm optimization was used to fit the AdEX biophysical parameters [19] (Fig. 4a). This allowed us to test whether synaptic scaling can precisely offset decreased intrinsic excitability in model networks constructed with biophysical properties empirically derived from CRH^PVN neurons. The simulations modelled on CORT-treated neurons showed increased rheobase current in comparison to those modelled on control neurons, which complies with a rheobase shift in intrinsic excitability (Fig. 4b). A double exponential fit was employed to model mEPSCs recorded from control and CORT CRH^PVN (Fig. 4c). Notably, we did not include inhibitory synapses in the simulations as there was no difference in the amplitude or frequency of inhibitory synaptic currents between control and CORT animals (SFig. 5). We first tested the model with simulations of hypothetical CRH^PVN constructed with varying permutations of intrinsic properties and synaptic strengths (Fig. 4d, e). Here, simulated CRH^PVN activity was assessed with model neurons constructed with synaptic and intrinsic properties from both CORT-treated and control neurons. The steady state activity was nearly identical from networks modelled on control and 7-Day CORT treated animals. There was a decrease in activity from networks modelled using intrinsic properties from 1-day CORT animals that were paired with synaptic input from control animals. Conversely, there was an increased activity when intrinsic properties from control CRH^PVN were paired with increased synaptic strength from 7-Day CORT animals. These results confirmed that the model behaved as expected.

Next, we compared the activity of a CRH^PVN population modelled after control neurons and compared the activity with CRH^PVN populations with reduced intrinsic excitability, but with synapses scaled by various factors. To compare activity, a ratio of the peristimulus time histograms (PSTH) from a network modelled on the control condition (PSTH_control) and experimental networks modelled with various scaling factors (PSTH_exp). The experimental networks were modelled with intrinsic properties from CORT CRH^PVN neurons. Here, a ratio of 1 indicates homeostasis. We found that the models achieved homeostasis with a scaling factor of 1.30 (Fig. 4f-g). A PSTH ratio of 1 was achieved when the experimental population’s synapses were scaled by a factor of 1.30. The PSTH ratio deviated further above or below 1 as the scaling factor was incrementally decreased or increased respectively. The ratio also approached 1 as synapses were scaled to 1.33 (the empirically derived scaling factor). These results indicate that CRH^PVN neurons must precisely scale excitatory synapses to offset negative feedback pressure induced by CORT.

Homeostatic plasticity following prolonged chemogenetic inhibition rescues functional output of CRH^PVN neurons

Although our data is consistent with a homeostatic rescue of CRH^PVN neuron activity and accompanying behavior, we are cognizant that CORT has multiple signaling targets within the central nervous system. Therefore, it is possible that the behavioural effects observed are due to CORT effects on extra-hypothalamic pathways. We designed an experiment to inhibit CRH^PVN neurons specifically to probe changes in synaptic strength and behavior. A viral vector containing AAV8-hSyn-DIO-hM4DG_i-mCherry was injected immediately dorsal to the PVN of Crh-IRES-Cre mice (Fig. 5a). Animals were given 2 weeks for recovery and DREADD expression. We first confirmed the function of hM4DG_i using on-cell electrophysiology in acutely prepared brain slices containing CRH^PVN neurons. A brief 1 s, localized pulse of CNO (10 µM dissolved in the extracellular recording solution) was administered adjacent to hM4DG_i-expressing CRH^PVN neurons (Fig. 5b). The CNO pulse temporarily dampened activity, consistent with transient membrane hyperpolarization induced by the G_i signaling cascade (Fig. 5c, SFig. 6).

Fig. 5. Homeostatic plasticity rescues behavioral responses to threat.

a hM4DG_i expression in CRH^PVN from Crh-IRES-Cre mice 2 weeks after viral injection. b Electrophysiological recording arrangement. A pipette containing CNO was placed adjacent to CRH^PVN neurons expressing hM4DG_i while a patch-pipette recorded on-cell activity. c Representative trace, and histogram showing average activity before, during, and after the CNO pulse. Scale bars 2 mV, 20 s. d Results from looming shadow experiments showing 1 day of inhibition with hM4DG_i resulted in a preference towards escape behavior (P < 0.0001, Fisher’s exact test). Following 7 days of CNO there was no significant difference in rate of escape between control and hM4DG_i animals (P = 0.2507, Fisher’s exact test) (mCherry control + 7 days CNO N = 5 mice, hM4DG_i + 1-Day CNO N = 6 mice, hM4DG_i + 7-Day CNO N = 7 mice) e (Left) Representative mEPSC traces from hM4DG_i-expressing CRH^PVN from animals given water (control) or those given CNO for 7 days. f There was an increase in mEPSC amplitude in CNO treated animals (control N = 9 cells, CNO N = 10 cells, t = 2.670, df  = 18, P = 0.0156, t-test). g Average mEPSC from both cohorts. h Rank-order plot and linear regression analysis of mEPSC amplitudes. The linear regression yielded a line with slope of 1.38 (Y = 1.38x – 2.61, R² = 0.9946, P < 0.0001). i Cumulative distribution showing mEPSCs from control, 7-Day CNO animals (control vs 7-Day CNO, KS D = 0.2984, P < 0.0001, KS test) and scaled values (control vs scaled, KS D = 0.1613, P = 0.0794, KS test).

We next tested whether inhibition of CRH^PVN neurons with DREADDs would shift the defensive strategy towards freezing in the looming shadow task. Here, mice expressing hM4DG_i or mCherry (control) were allowed ad libitum access to 0.1 mg/ml CNO in their drinking water for 1 day [35]. After 1 day of CNO, the mice were subject to 5 trials of looming shadow. Control mice showed preferential escape behavior at a rate consistent to what has been previously documented [4]. Meanwhile, DREADD-expressing mice showed a preference for freezing, and opted for escape at a rate significantly lower than control mice (Fig. 5d). Subsequent analysis of the behavioral responses revealed no significant differences in escape latency between experimental groups (SFig. 10A). Additionally, no significant differences were observed in the distance to the shelter during the 1-Day CNO trials (SFig. 10B). This indicates that inhibition of CRH^PVN neurons with DREADDs is sufficient to occlude the signal required to initiate escape behavior.

We then asked whether prolonged inhibition with hM4DG_i would result in synaptic scaling. A new cohort of animals were given either regular drinking water (control) or CNO dissolved in the drinking water for 7 days. On the 7^th day, acute brain slices were prepared for electrophysiological recordings. There was an increase in mean mEPSC amplitude (Fig. 5e, f) and no change in frequency (SFig. 7A). The increase in mEPSC amplitude was consistent with homeostatic synaptic scaling as demonstrated by linear regression and cumulative distribution (Fig. 5g-i, SFig. 7B). Control experiments were performed to ensure hM4DG_i expression did not alter passive membrane properties that would affect recording conditions (SFig. 7C-E). We also performed additional control experiments in mice expressing mCherry to ensure that CNO, a derivative of clozapine, did not alter mEPSC amplitude (SFig. 8). These experiments indicate that the scaling observed in CRH^PVN neurons after hM4DG_i activation is indeed homeostatic as it is induced by a prolonged decrease in activity. We then asked whether this local homeostatic scaling would rescue escape behavior. A separate group of animals expressing hM4D_i were given CNO dissolved in drinking water solution for 7 days. On Day 7, mice were placed in the looming shadow arena. In line with re-established CRH^PVN neuron activity, there was no difference in the fraction of animals that opted to escape to shelter between the two groups (Fig. 5d, SFig. 10). Our findings indicate that CRH^PVN neurons respond to inhibitory activity pressure by scaling up afferent excitatory synapses. This is sufficient to rescue active escape behavior in response to threat.

Discussion

When faced with prolonged elevations in circulating CORT, CRH^PVN neurons leverage homeostatic synaptic scaling to restore innate escape behavior in response to predatory threat. Consistent with this, we see a rescue of in vivo CRH^PVN neuron activity despite ongoing negative activity pressure. This represents an endogenous adaptive mechanism that ensures behavioral stability and maximizes survival probability when an animal’s internal state is perturbed for a prolonged period.

Animals may display a gamut of defensive behaviors in the face of threat. Choosing which behavioral response to employ requires the animal to rapidly integrate sensory information regarding the nature of the threat and the surrounding environment and to relay it among various nodes involved in behavioural selection. Visual information arrives to CRH^PVN neurons by direct projections from the retina, and is also relayed by the lateral hypothalamus (LH) and medial amygdala (MeA) via the superior colliculus (SC) [36, 37]. Upon detection of the looming shadow, these inputs evoke a ramping of activity in CRH^PVN neurons that provides a key signal to initiate the escape behaviour [4, 10, 36]. This signal is transmitted by efferent projections to the LH, and external segment of the globus pallidus (GPe), which provides an entry point for CRH^PVN neurons to influence higher-order circuits that participate in defensive behaviour selection. Short-term CORT feedback, by decreasing the excitability of CRH^PVN neurons, disrupts this signal, effectively eliminating escape behaviour from the array of defensive behaviors available to the animal. During prolonged CORT exposure, however, the participation of CRH^PVN neurons in influencing these behaviours is rescued, providing the animal with wider array of potential survival behaviours. Presumably, this grants the animal with greater survival probability upon confrontation with environmental threats.

This rescue mechanism appears to involve a global and multiplicative increase in excitatory synaptic strength, reminiscent of the original descriptions of homeostatic synaptic scaling in cultured neurons [24, 28, 38]. Neurons have a multitude of homeostatic mechanisms at their disposal. Many studies have shown that synaptic scaling demonstrate alterations in post-synaptic AMPAR density [31, 39, 40]. Other studies have shown variations in presynaptic quantal content [29], and various presynaptic mechanisms [32]. In our experiments, we were unable to find presynaptic changes such as increased release probability or increased quantal content after 7 days of CORT. The most parsimonious explanation for our observations is an increase in the number of post-synaptic AMPARs. Since CORT signaling has also been shown to alter glutamatergic synaptic transmission in the pre-frontal cortex [41, 42], which could indirectly influence CRH^PVN neuron activity and behavioural responses, we used a chemogenetic approach to specifically silence CRH^PVN neurons over 7 days and assess synaptic strength. Using the inhibitory DREADD, hM4DG_i, we show that inhibition of CRH^PVN neurons induced synaptic upscaling of glutamate synapses. This suggests the system is indeed responding in a true homeostatic fashion to a persistent decrease in neuronal firing. We then show that inhibition with hM4DG_i initially shifts the animal’s preferred behavior to freezing. However, the escape behavior is rescued after prolonged hM4DG_i activation, which coincided with an engagement of synaptic scaling. Although synaptic scaling had previously been shown to rescue basal activity of cortical networks [26, 43], our work is the first to suggest that this form of plasticity can rescue specific functional and behavioural outputs.

This study adds to a growing body of literature placing CRH^PVN neurons as a key node in stress processing [2] and behavioural/physiological outputs including social interaction [44], encoding of valence [5, 14], and wakefulness [45]. The relevance of synaptic scaling to these aspects of CRH^PVN neurons will require further investigation. For example, prolonged CORT feedback suppresses CRH expression in CRH^PVN neurons [3]. Therefore, although synaptic scaling may rescue activity, CRH^PVN neurons may be incapable of releasing CRH. This would result in a functional “rewiring” of the circuit as glutamatergic output from CRH^PVN neurons to the LH would remain intact, but CRH projections (pituitary and GPe) would be hindered. Furthermore, paracrine CRH release activates a periventricular CRH-receptor 1 (CRHR1)-expressing neuron population [46, 47] that sends glutamatergic projections to various brainstem nuclei [47]. An inability of CRH^PVN neurons to release CRH may alter activity of these neurons as well. It is possible that these changes may reflect an adaptive state of the neural stress circuitry that preserves key survival behaviors, while allowing integration of CORT feedback. Another aspect requiring investigation is whether synaptic scaling interferes with typical plasticity of CRH^PVN neurons. Stress results in various types of meta plasticity at GABA and glutamate synapses [48–52], that are presumed to alter CRH^PVN neuron outputs. Post-synaptic receptor accumulation in the post-synaptic density during synaptic scaling may interfere with glutamatergic plasticity after stress. Therefore, exposure to repeated stress after synaptic scaling has occurred may disrupt other forms of synaptic plasticity. Additionally, GABA synapses are in close proximity to glutamate synapses in CRH^PVN, and GABA synapses demonstrate time-dependent plasticity in response to CORT feedback [51]. Whether scaling of glutamate synapses alters plasticity at GABA synapses requires further evaluation. Another important observation is that CRH^PVN neurons do not show concomitant opposing scaling of inhibitory and excitatory synapses. Other examples of isolated excitatory synaptic scaling have been described in vivo [26], which likely reflect an optimal adaptation given constraints inherent to the network.

Altered HPA axis function is a common feature of many neuropsychiatric disorders. Anxiety, depression and post traumatic stress disorder (PTSD) are associated with dysregulated HPA function and linked to altered neuroendocrine and behavioral stress response [53]. Dysfunction in CRH^PVN neurons [1], persisting elevations in circulating CORT [54, 55], and altered threat processing and behavioural responses [56], are implicated with major depressive disorder (MDD). Exogenous CORT administration, including low doses of CORT in the drinking water, has been extensively used as a preclinical model to study the effect of HPA axis disruption on neuropsychiatric disorders [56]. This model allows for CORT fluctuation that follows the natural circadian rhythm but increases basal CORT levels during the active phases. As a result, increased circulating CORT is obtained without drastically altering circadian rhythm. Previous studies have consistently demonstrated that chronic CORT administration induces behavioral, endocrine and synaptic changes in glutamatergic signaling [56, 57]. Our study indicates that CRH^PVN neurons undergo homeostatic changes to respond to the constant negative pressure. These homeostatic changes occur in a brain node that is crucial for proper stress response and animal survival [36]. To our knowledge, this is the first demonstration that homeostatic plasticity is recruited to restore a specific behavioral output. By showing that synaptic scaling rescues CRH^PVN activity, and the defensive behaviours they mediate, our study provides evidence that the nervous system relies on homeostatic plasticity to not only ensure stable basal activity, but that key aspects of network function are preserved amidst an unyielding barrage of destabilizing stimuli.

Author contributions

N.R, T.F., N.D designed and conducted the experiments, analyzed the data, and wrote the paper. D.R., T.S., S.L, N.L. and I.A conducted experiments and analyzed data. T.K. provided the viral vector use for the chemogenetic experiments and contributed to the paper preparation. E.L. and W.N performed the computational modeling and contributed to paper preparation. N.D. and J.S.B. designed experiments, prepared the paper, and supervised the project.

Data availability

The raw data that support the findings of this study are available from the corresponding author upon request.

Competing interests

The authors declare no competing interests.

Ethics approval

All experimental procedures were approved by the University of Calgary Animal Care and Use Committee in accordance with Canadian Council on Animal Care guideline (protocol number AC210067).

Supplementary information

The online version contains supplementary material available at 10.1038/s41380-025-03391-5.