Behavioral appraisal by implementing a short sequence of stress resolves adaptively changed stress gains

Chronic stress produces adaptive changes in the brain via the cumulative action of glucocorticoids, which causes psychiatric illnesses such as depression. Here we show that a behavioral method implementing weak stress does not strengthen but resolves existing stress gains. Chronic stress produces persistent depressive behaviors in mice, and repeated daily treatment with 5-min restraint produces antidepressive effects. Repeated treatment with low-dose glucocorticoids mimics the anti-depressive effects of weak stress. Repeated weak stress or low-dose glucocorticoid treatment distinctively activates the prelimbic cortex (PL), and reverses the stress-induced altered gene expression proles. Chemogenetic inhibition of PL outputs projecting to the nucleus accumbens, basolateral amygdala, or bed nucleus of the stria terminalis (BNST) dissipates antidepressive effects of weak stress, but only the PL-to-BNST circuit produces changes in dysregulated glucocorticoid release. Our results suggest that behavioral appraisal by implementing weak stress can resolve adaptively altered stress gains and rectify stress-induced depressive behaviors.


Introduction
Behavioral appraisals are used to treat the emotional dysfunction of psychiatric disorders 1 . Cognitive reappraisal is a cognitive method that regulates emotion by reinterpreting an emotion-provoking situation or reframing emotional expression 2,3 . A related but more complex form of behavioral appraisals includes cognitive behavioral therapy (CBT) 4,5 and exposure therapy 6 , which involve stress responses. Although the behavioral appraisals are bene cial for post-traumatic stress disorder 6,7 and depression 4,8,9 , the underlying neural mechanisms remain largely unknown, and it remains unclear if stress responses are required for therapeutic effects of the behavioral methods or they are an impediment to be properly controlled.
Chronic stress is a potent risk factor for various psychiatric illnesses, including depression 10,11,12 . Stress has been described as "the non-speci c response of the body to any demand for change" 13,14 . The stress response proceeds by activating the hypothalamus-pituitary-adrenal (HPA) axis, which causes release of glucocorticoids (GCs, cortisol in humans and corticosterone in rodents) from the adrenal glands 15,16 . GCs normally stimulate energy metabolism and general physiological activity and increase vigilance, memory, and other cognitive functions 17,18 . Therefore, normal physiological stress has necessary and bene cial effects in daily life. Basal blood GC levels are low, but they exhibit circadian oscillation, with the highest levels in the early morning and the lowest levels at midnight. The daily uctuation of GCs might be important for resetting physiological activities of the stress coping system, but the biological signi cance of low GC levels has not been carefully investigated.
Chronic stress produces changes in various brain regions beyond their homeostatic capability 16,18,19 . One of the brain regions that undergo functional and structural changes following chronic stress is the medial prefrontal cortex (mPFC) in rodents 19,20,21 . Chronic stress reduces glutamate transmission and related 5-min restraint (Fig. 2a,g). Repeated injection of CRST mice with 0.1 mg/kg of CORT, or even 0.5 or 1.0 mg/kg of CORT, produced anti-depressive effects in the SIT, SPT, TST, and FST, and those behavioral changes were comparable to those induced by RS5 (Fig. 2h-l). Repeated injection of low-dose CORT in CRST mice also restored their increased basal CORT and reduced the increased weight of their adrenal glands to those of the naïve controls (Fig. 2m,n). PCA and K-Means clustering of behavioral and physiological factors in the [SIT x SPT] x [TST x FST] x [adrenal gland weight] matrix indicated that behavioral recovery with CORT treatment proceeds with restoration of adrenal gland weight at the individual level (Fig. 2o,p).
RS5 effects required activation of the HPA axis NBI27914 (an inhibitor of CRH receptor 1) or RU486 (an inhibitor of GR) treatment in CRST mice unexpectedly heightened the increased basal CORT level, but it produced a partial suppression of depressive-like behaviors in the SIT, SPT, TST, and FST. RS5 treatment while being treated with NBI27914 or RU486 in CRST mice did not fully suppress the increased basal CORT level and did not change the effects of NBI27914 or RU486 in the behavioral tests (Extended Data Fig. 3a-h).
Mice that received CRST treatment followed by adrenalectomy (ADX) surgery had a low basal CORT level and showed no CORT release after a 5-min restraint (Fig. 3a,b). RS5 treatment in mice with CRST plus ADX did not improve their stress induced behavioral de cits in the SIT, SPT, TST, and FST ( Fig. 3c-h). ADX alone in CRST mice did not change their depressive behavior ( Fig. 3c-h). These results suggest that HPA axis activation is necessary for RS5 to produce therapeutic effects, whereas simple CORT depletion is ineffective.
RS5 treatment activated core parts of the limbic system We investigated the brain regions recruited by RS5 using a stimulus-induced c-Fos mapping strategy. Repeated treatment with 5-min restraint in CRST mice increased c-Fos expression in the PL, ventral subiculum (vSub), dorsal bed nucleus of the stria terminalis (dBNST), PVN, and other parts of the limbic system (Fig.4a-c and Supplementary Table 1). Interestingly, the c-Fos levels induced by repeated treatment with 5-min restraint in those regions were higher than those induced by repeated treatment with 15-min restraint or a single 5-min or 15-min restraint treatment. Among the c-Fos-positive neurons in the PL, 74.4% were glutamatergic, and 17.6 % were GABAergic. Among the glutamate neurons, 46.8% were c-Fos-positive, whereas 44.2% of GABA neurons were c-Fos positive ( Fig. 4d-g).
RS5 treatment restored altered expression of genes in the PL Next, we investigated whether RS5 affected genome-wide responses in the PL (Fig. 5a). Microarray analysis revealed that CRST treatment or RS5 in CRST-treated mice up-and downregulated 264 genes and 722 genes, respectively, by ≥1.2-fold (Supplementary Tables 2 and 3). A heatmap presentation of the expression pro les of those genes that the CRST-induced altered gene expression pro les underwent a cross reversal after RS5 treatment (Fig. 5b,c).
Gene Ontology (GO) enrichment analysis combined with K-Means clustering using the STRING database 22 showed that the genes identi ed above could be grouped into multiple clusters that presented various protein-protein interaction (PPI) networks. A serial K-Means clustering indicated that grouping those genes into 5-10 clusters (k = 5-10) built up PPI networks with functional modules relevant to the assumption of stress or glucocorticoid-related responses (Extended Data Fig. 5a). In the classi cation with 8 clusters, clusters 5 and 7, which contained 28 and 25 members, respectively, carried functional modules labeled with the GO terms "response to stress" or "response to glucocorticoid" (Fig. 5d,e). The remaining clusters are summarized in Extended Data Fig. 5b,c. Using the STRING database, the functional PPI networks formed by the clusters 5 and 7 were expanded by an additional 16 proteins that could participate as nodes (Fig. 5f).

RS5 treatment upregulated GR and GluN subunits in PL neurons
Chronic stress decreases GR expression and increases FK506-binding protein 5 (Fkbp5), a cellular factor that suppresses GR nuclear translocation 32,33 in the mPFC 34 . Of the genes for stress or glucocorticoidrelated responses (Fig. 5f), GR expression in the PL was downregulated by CRST, whereas its expression was reversed to the control after RS5 or CORT (0.1 mg/kg) treatment. Fkbp5 and Fkbp4 were upregulated and downregulated, respectively, by CRST, whereas their altered expression was restored after RS5 or CORT (0.1 mg/kg) treatment (Fig. 5g). Transcript levels of ERK1 and ERK2 declined after CRST, and their expression was reversed by CORT (0.1 mg/kg) treatment, but not by RS5. The transcript levels of mineralocorticoid receptor (MR), heat shock protein 90α, member A1 (Hsp90aa1), and dual speci city protein phosphatase 1 (Dusp1) tended to be changed by CRST, RS5, or CORT (0.1 mg/kg), although their changes overall were subtle (Fig. 5g,h). Immunohistological analysis followed by K-Means clustering indicated that CRST increased Fkbp5 expression and decreased GR nuclear distribution at the level of single cells, whereas RS5 and CORT (0.1 mg/kg) treatment reverted their altered expression to the control. Interestingly, the positive correlation between Fkbp5 and GR expression was apparent in the CRST group, but not in the CON, RS5, or CORT groups ( Fig. 5i-n). Small interfering RNA (siRNA)-mediated inhibition of GR increased Fkbp5 expression in the PL (Fig. 5o-q), increased CRH and AVP expression in the PVN (Fig. 5r), and induced depressive behaviors in the TST and FST (Fig. 5s,t). These results suggest that GR expression declined in PL neurons after CRST, and Fkbp5 levels increased, whereas RS5 and low-dose CORT treatment restored the altered GR and Fkbp5 expression.
Repeated stress suppresses the expression of the GluA and GluN subunits in the prefrontal cortex 23,35 .
Consistently, CRST treatment reduced the expression of the GluN subunits NR1, NR2A, and NR2B in the PL, whereas repeated weak stress or repeated treatment with CORT (0.1 mg/kg) restored their reduced expression to the control level (Extended Data Fig. 6a-h).
RS5 treatment recti ed the increased expression of CaMKIIα in the PL Western blot analyses indicated that p-CaMKIIα and p-ERK1/2 levels in the PL were upregulated and downregulated, respectively, after CRST, and that the altered expression of p-CaMKIIα, but not p-ERK1/2, was restored to the control after RS5 treatment (Fig. 6a-c). p-CaMKIIα was expressed mostly in glutamatergic neurons (Fig. 6d,e). Immunohistological analysis followed by K-Means clustering indicated that CRST increased p-CaMKIIα expression in PL neurons while decreasing GR expression, and that their altered expression was restored to the control after RS5 or CORT (0.1 mg/kg) treatment. p-CaMKIIα and GR expression levels were positively correlated at the single-cell level throughout CRST, RS5 or CORT (0.1 mg/kg) treatment ( Fig. 6f-k). siRNA-mediated inhibition of CaMKIIα increased GR expression in the PL, whereas inhibition of ERK1 or ERK2 did not. Fkbp5 expression was partially affected by the inhibition of ERK2 (Fig. 6l-p). In fact, siRNA-mediated inhibition of CaMKIIα but not ERK1 and ERK2, in the PL of CRST-treated mice restored the reduced sociability and increased immobility time in the TST and FST (Fig. 6q-s).
CRST decreased the expression of the GABAA receptor subunits GABRα1 and GABRβ2 in the PL, and RS5 or CORT (0.1 mg/kg) treatment increased their reduced expression (Fig. 7a). In mice exposed to CRST followed by RS5, inhibition of GABAA receptors in the PL by infusion of picrotoxin through a preimplanted cannula upregulated p-CaMKIIα levels and promoted depressive-like behavior in the TST and FST ( Fig. 7b-g). These results suggest that p-CaMKIIα upregulation in the PL can be induced by a decrease in GABAA receptor function.
Interestingly, local inhibition of CaMKIIα in the PL by stereotaxic infusion of siRNA-CaMKIIα or KN62 (an inhibitor of CaMKIIα) in normal mice produced depressive-like behaviors in the SIT, TST, and FST, suggesting that reduced levels of CaMKIIα in the PL are also pro-depressive ( Fig. 7h-m,p-s). In contrast, siRNA-mediated or U0126-mediated inhibition of ERK1 in the PL in normal mice reduced immobility time in the TST and FST, although it produced no effect in the SIT (Fig. 7h-m). p-ERK1/2 was expressed mostly in glutamatergic neurons (Fig. 7n,o).These results suggest that CRST upregulates p-CaMKIIα in PL neurons, which promotes depressive behavior, whereas RS5 produces anti-depressive effects by downregulating p-CaMKIIα in PL neurons.

Activation of PL neurons was required for RS5 effects
Given that RS5 upregulated GR, GluN, and GABAA receptor subunits in PL neurons (Figs. 5g and 7a and Extended Data Fig. 6a) and that the PL had distinct c-Fos expression following repeated 5-min restraint (Fig. 4a-c), we investigated whether the neural activity of PL neurons was required for RS5 to have antidepressive effects. Mice injected in the PL with the AAV8-CaMKIIα-hM3D(Gq) or AAV8-hSyn-hM3D(Gq) vector were subjected to CRST. Clozapine N-oxide (CNO) injection in CRST mice carrying an excitatory vector increased c-Fos expression in PL neurons (Extended Data Fig. 7a-f). CNO treatment in mice with hM3D(Gq) expression driven by the CaMKIIα promoter, but not the human synapsin (hSyn) promoter, produced increased sociability in the SIT and decreased immobility time in the TST and FST (Extended Data Fig. 7g-k). After 7 days of CNO washout, mice with CaMKIIα-hM3D(Gq) showed a relapse of depressive behaviors and had a substantial increase in their basal CORT level relative to the control (Extended Data Fig. 7l-q). These results suggest that activation of PL glutamatergic neurons can suppress depressive phenotypes in CRST-treated mice, but those rescue effects are transient.
Mice injected in the PL with the AAV8-CaMKIIα-hM4D(Gi) inhibitory vector were subjected to CRST and then treated with RS5 or RS5+CNO. CNO injection in those mice partially blocked the RS5-induced suppression of the basal CORT level and dissipated the antidepressive effects of RS5 in the SIT, SPT, TST, and FST ( Fig. 8a-k). These results suggest that RS5 produces anti-depressive effects by activating PL glutamatergic neurons.

Multiple PL outputs mediated the effects of RS5
The PL sends collaterals to the dBNST, BLA, and NAcc, the brain regions recruited by repeated treatment with 5-min restraint (Fig. 4a,b). We investigated whether the effects of RS5 require multiple PL output pathways or a speci c efferent circuit.
The dBNST, which receives glutamatergic inputs from the PL (Extended Data Fig. 8a-d and 8h-k) and vSub (Extended Data Fig. 8l), contains GABA neurons that project to the PVN 36,37 . To test whether the PLàdBNST circuit mediated the effects of RS5, the PLàdBNST circuit was labeled with the AAV-DIO-hM4D(Gi)-mCherry inhibitory vector using a retrograde Cre vector (Fig. 9a-c). CNO-mediated inhibition of the PLàdBNST circuit suppressed RS5-induced c-Fos expression in dBNST neurons, which were GAD67positive. However, that inhibition did not change RS5-induced c-Fos expression in the PVN (Fig. 9d-f). CNO-mediated inhibition of the PLàdBNST circuit in CRST mice dissipated the antidepressive effects of RS5 in the SIT, SPT, TST, and FST, and the normalization of the basal CORT level by RS5 (Fig. 9g-m). These results suggest that activation of the PLàdBNST is required for RS5 to have antidepressive effects and restore defective basal CORT release.
Next, we examined whether PLàBLA and PLàNAcc neurons played a role in regulating the effects of RS5. The BLA and NAcc receive glutamatergic inputs from the PL (Extended Data Fig. 8e,f). The PLàBLA and PLàNAcc circuits were speci cally labeled with the AAV-DIO-hM4D(Gi)-mCherry inhibitory vector and the retrograde Cre vector (Fig. 9n,o). CNO-mediated suppression of the PLàBLA or PLàNAcc circuit decreased RS5-induced c-Fos expression in the BLA and NAcc, respectively ( Fig. 9p-s). Inhibiting the PLàBLA or PLàNAcc circuit in CRST mice with CNO during RS5 treatment blocked the anti-depressive effects of RS5 in the SIT, SPT, TST, and FST, whereas inhibition of those circuits did not change the RS5-induced normalization of basal CORT levels ( Fig. 9t-x). These results indicate that activation of the PLàBLA and PLàNAcc circuits facilitates the anti-depressive effects of RS5, but the activation of those pathways is not essential for the recovery of the altered basal CORT level.

Discussion
The nding that behavioral method implementing a short sequence of stress arousals and its resolutions can rectify persistent behavioral changes in multiple models of depression highlights the feasibility of ghting adaptively altered stress gains with behavioral stress. Chronic restraint produces depressive behaviors that last for more than 3 months 38,39 . Despite those adaptive changes, RS5, but not RS10 or RS15, recti ed depressive behaviors as did the antidepressant imipramine ( Fig. 1). RS5 treatment did not produce anti-depressive effects when the HPA axis was blocked ( Fig. 3 and Extended Data Fig. 3). Repeated injection with low-dose GC (0.1 mg/kg) recapitulated the effects of RS5 ( Fig. 2h-l). Paradoxically, however, repeated injection with GC at a dose higher than that induced by 10-min or 15-min restraint, and repeated injection with even 1.0 mg/kg of GC, which was comparable to the GC level induced by 2-h restraint (Fig. 2a,g), also produced anti-depressive effects ( Fig. 2h-l). These results suggest that behavioral appraisal by implementing a short sequence of stress produces anti-depressive effects via a GC-dependent mechanism. However behavioral method has a limited window to produce anti-depressive effects.
Chronic stress changes various brain regions beyond the homeostatic range by activating the HPA axis 16,18,19 . Patients with depression have increased basal serum GC levels 40,41 . In contrast, RU486 (mifepristone), a GR antagonist, is bene cial for patients with psychotic depression 42 and bipolar disorder 14 . Mice exposed to chronic stress have increased basal serum GC levels (Fig. 2). Administration of high-dose GC in drinking water (35 μg/ml/day) for 4 weeks 43 or subcutaneous injection of GC at a dose of 10, 20 or 40 mg/kg/day for 21 days 44 in rats mirrors the stress-induced dysfunction of the HPA axis and produces depressive-like behaviors. Therefore, GC is regarded as a mediator of chronic stress 15,16,17 .
In the present study, we demonstrated that repeated treatment with a short sequence of behavioral stress or repeated injection with GC (0.1 -1.0 mg/kg/day) produced anti-depressive effects (Figs. 1 and 2h-l) and reversed stress-induced molecular changes (Fig. 5) as did imipramine. The key ndings are summarized in Fig. 10. These results raise the following important and related points. First, the fact that GC induction by behavioral stress and exogenous GC resolved existing stress gains suggests that GC functions as a stress modi er, which contradicts the classical conception that GC is a stress mediator. Although when and how GC functions as a stress modi er need to be studied in more detail, we speculate that repeated weak stress can restore stress coping ability, presumably by repeatedly boosting the feedback and feedforward regulatory mechanisms of the HPA axis (Fig. 10b,c,e). This possibility does not con ict with the classical conception that chronically imposed GC produces cumulative effects on stress gains due to the points described below. It will be worth studying whether the circadian oscillation of basal GC levels 45,46 , which is disrupted in patients with depression 40,41 , functions as a stress modi er by performing a daily reset of the stress coping system. In healthy individuals, the basal serum GC levels vary throughout the day, with the highest in the early morning and then falling throughout the day to the late evening. Second, repeated treatment with the behavioral stress or repeated treatment with GC could be used as an antidepressant strategy. Although the behavioral method has a narrow window to afford therapeutic effects, it could have an advantage over the pharmacological method. On the other hand, challenging with GC provides a more wider and more versatile therapeutic window to resolve existing stress gains. The profound therapeutic effects of behavioral stress and exogenous GC demonstrated in this study warrant further investigation. Third, the nding that repeated weak stress or treatment with even high-dose GC did not strengthen existing stress gains but resolved them (Figs. 1 and 2), raises the possibility that prior stress gains might become transiently deconsolidated and labile upon new stress inputs or GC ux. As stress-induced adaptive changes are deconsolidated by GC, brain cells appear to restore their normal homeostatic stability and physiological function to produce normal behavioral outputs. It will be worth studying the key factors and underlying signaling networks that regulate GCdependent changes and the mechanisms of homeostatic restoration.
Chronic stress increased the p-CaMKII level, which downregulated GR expression in PL neurons ( Our results indicate that PL neurons and their associated neural systems, including the NAc, BLA, and BNST, compose the critical neuronal nodes and edges that support stress-induced depression and the modi cation of stress-induced changes by strategic weak stress, as summarized ( Fig. 10b-e). Glutamate/glutamine levels and the neural activity of glutamatergic neurons in the mPFC are reduced in CUMS, CRST, and CSDS-induced depression models in mice 48,49,50 . Chemogenetic activation of PL neurons facilitated anti-depressive effects and suppressed the stress-induced increase in basal GC levels, whereas chemogenetic inhibition of PL neurons during RS5 dissipated its anti-depressive effects (Figs. 8 and Extended Data Fig. 7). Furthermore, chemogenetic inhibition of the PLà NAc, PLà BLA or PLà dBNST circuits during RS5 blocked its anti-depressive effects ( Fig. 9), suggesting that activation of those circuits is required for RS5 to produce anti-depressive effects. Previous studies reported that activation of the PL 51 , the PLà NAc circuit 52 or the PLà BLA circuit 53 produced anti-depressive effects, which is partly consistent with our results. Interestingly, however, the PLà dBNST circuit played a role in the recovery of basal GC levels by RS5, whereas the PLà NAc and PLà BLA circuits did not ( Fig. 9), raising the question if the latter cases, which did not recover normal HPA function, produce stable recovery from stress-induced changes. Overall, our results suggest that although the PL is the key area recruited by RS5, the PLà NAc, PLà BLA, and PLàdBNST pathways comprise the critical neural nodes and circuits that support the behavioral effects of RS5, and the PLàdBNST circuit supports the recovery of the HPA axis by RS5 ( Fig. 10b-e).
Stress inoculation is a pretreatment strategy that improves subsequent stress coping and emotional regulation 45 . Stress inoculation in mice, by placing them behind a mesh-screen barrier in a cage containing an aggressor mouse for 15 min, enhances subsequent stress coping behavior and cognitive function 55 . Another form of stress inoculation, called predictable chronic mild stress (PCMS), uses 5-min restraint for 28 days to improve mood, hippocampal neurogenesis and memory in rats 56 . PCMS treatment of rats in their early adolescence increases resilience to chronic unpredictable mild stress in adulthood 57 . In those studies, stress inoculation or PCMS is regarded as an immunization to enhance coping ability for future stress 54 . PCMS treated for 28 days induced anti-depressive effects in rats 56 . In contrast, our RS5 treatment (daily 5-min restraint for 14 days) in normal mice did not induce depressive behaviors ( Fig. 1a-g), Therefore, it will be worth studying that a pretreatment paradigm of daily 5-min restraint for 7-14 days in normal mice would also produce resiliency to future chronic stress. It is possible that stress inoculation/PCMS and RS5 could commonly have a certain neural mechanism. Nonetheless, our experimental procedure provoking weak stress deals with a therapeutic strategy, whereas stress inoculation/PCMS is preventative.  h-I, Immobility time in the TST (h) and FST (i) for the indicated groups on post-stress days 43 and 44 (n = 8-10 per group). j, Experimental design. The susceptible mice were treated with RS5, and then given behavioral tests.
k-o, Mice susceptible or resilient to CSDS were separated by the sociability ratio (k). % time spent in the target chamber in the two-chamber SIT (l), sucrose preference in the SPT (m), and immobility time in the TST (n) and FST (o) for the indicated groups (n = 8-11 per group).
Extended Data Fig. 1 Extended Data Fig. 1 Dosing analysis for the repeatability of weak stress that produces anti-depressive effects.
b-f, Time spent in the target chamber in the SIT (b), and immobility time in the TST (c) and FST (d) for the indicated groups. K-Means clustering (k = 2) of individuals in the SIT x TST x FST matrix (e) and % composition of each group in the clusters (f) (n = 8-10 per group).
Extended Data Fig. 2 Extended Data Fig. 2 Repeated treatment with a short sequence of behavioral stress produces antidepressive effects in ICR mice.
Extended Data Fig. 4 Extended Data Fig. 4 Low-dose CORT treatment activates the brain regions regulating stress coping in CRST mice.
Extended Data Fig. 5 Extended Data Fig. 5 Analysis of gene expression pro les and protein-protein interaction networks in the PL of mice exposed to CRST and RS5 treatment.
a, Serial K-Means clustering was used to group the 986 identi ed genes into featured clusters. Each increase in k value added a new cluster, and its members were mostly supplied from the largest cluster. The cognate inter-clusters are marked with the same color code. C, cluster. Concerning the classi cation with k = 8, cluster 1 (gray) contained 622 genes; cluster 2 (scarlet) contained 71 genes; cluster 3 (yellow) contained 39 genes; cluster 4 (light green) contained 29 genes; cluster 5 (orange) contained 28 genes; cluster 6 (blue) contained 27 genes; cluster 7 (green) contained 24 genes; and cluster 8 (violet) contained 19 genes.
b, Functional protein-protein interaction (PPI) networks constructed with the 237 genes that belonged to clusters 2 to 8. The members in each cluster are coded with the same colors as indicated above (a).
c, In the classi cation with k = 8, clusters 1,2,3,4,6, and 8 are shown with the number of cluster members and selective modules representing speci c GO terms. Clusters 5 and 7 are shown in Fig. 4d,e.
Extended Data Fig. 8 Extended Short-term restraint stress Short-term restraint was delivered to give weak restraint stress. Mice were individually placed in a 50-ml polypropylene conical tube with many ventilation holes and restrained for 5 min, 10 min, 15 min, or the indicated time beginning at 10 AM, and this treatment was repeated for the indicated number of days.
After each restraint session, the mice were returned to their home cages.
Chronic restraint stress (CRST) CRST was carried out as described previously 58 . In brief, mice were individually restrained in a wellventilated 50-ml conical tube for 2 h daily from 10 AM to 12 PM, and this procedure was repeated for 14 days. After each daily restraint session, the mice were placed in their home cages with free access to food and water.
Chronic social defeat stress (CSDS) CSDS was carried out as described previously 59,60 . Brie y, mice were individually exposed to a novel ICR aggressor for 10 minutes to produce physical defeat stress and then housed for the remainder of the day in a compartment of the ICR aggressor's cage partitioned with a perforated acrylic divider. This procedure was repeated for 10 days with a different ICR aggressor each day. ICR aggressors were preselected: if ICR aggressors showed ≤30 s attack latencies on three consecutive screening tests, or if any aggressor exhibited overly aggressive attacks during social defeat exposure, they were excluded from the next experiment. Stress-susceptible and resilient mice were selected on the basis of sociability in the open-eld version of the SIT on day 11 by following the procedure described previously 59,60 .

Maternal stress (MS) and postnatal stress treatments
Male and female ICR mice at 8-9 weeks of age were crossed, and pregnant females were randomly assigned to the maternal stress (MS) or normal (N) control group. MS pregnant females were treated with restraint daily for 2 h (10 AM -12 PM) from 8.5 days post coitus (dpc) to delivery (at 19.5 -20.5 dpc).
Normal (N) pregnant females were maintained in parallel to the MS females. The offspring of both groups were weaned at postnatal day 20 (PN20) and reared in pairs with the same sex in cages under standard conditions. For mouse groups assigned for CRST treatment, on PN49, offspring from normal (N) control mothers or MS mothers were housed in same-sex pairs from different litters to prevent possible litter effects. They were then treated with daily 2-h restraint for 14 days (PN50 -PN63).
Afterward, the half of the CRST-treated mice were subjected to the RS5 regimen, and the remaining half was used as the CRST-treated group.
Immunohistochemistry Immunohistochemistry was performed as described previously 38,39 . Brie y, mice were perfused with 4% paraformaldehyde via the trans-cardiac method and isolated brains were post xed at 4°C overnight. Each brain was coronally sectioned into 40-μm thicknesses using a vibratome (VT1000S, Leica Instruments, Nussloch, Germany). The collected sections were blocked for 1 h with 4% bovine serum albumin in phosphate buffered saline (PBS) containing 0.1% Triton X-100 (PBST) and then incubated with primary antibody at 4°C overnight. After washing, sections were reacted for 90 min with biotinylated secondary antibody: anti-rabbit IgG (BA-1000, Vector Laboratories, Burlingame, CA, USA) or anti-mouse IgG (BA-9200, Vector Laboratories) diluted at 1:200 in PBST. Signals were visualized using an ABC Elite kit (PK-6200, Vector Laboratories).
For the analysis of c-Fos expression induced by weak restraint stress, CRST mice were treated with a single 5-min restraint or 15-min restraint (S5 and S15, respectively) on post-stress day 8, and they were sacri ced, 20 min and 10 min later, respectively. Other groups of CRST mice were treated with 5-min restraint or 15-min restraint for 8 days (S5x8d and S15x8d, respectively), and sacri ced 20 min and 10 min, respectively, after the last restraint. Sigma-Aldrich) in tert-amyl alcohol (24,048-6, Sigma-Aldrich) diluted to 2.5% avertin with saline. Blood was collected from the abdominal aortas of sacri ced mice in the morning (8 AM -12 PM), and serum was obtained by centrifugation at 3,000 g and 4°C for 15 min and then stored at -80°C until use. Each diluted serum sample (10 μl) was mixed on a 96-well plate with an equal volume of a steroid displacement reagent solution and sera diluted at 1: 40 or 1:80 in the ELISA assay buffer provided in the corticosterone ELISA kit (ADI-901-097, Enzo Life Sciences, NY, USA). The reaction was incubated for 2 h at room temperature on an orbital shaker rotating at 120 rpm. The reaction mixture in each well was discarded, and the plate was rinsed with washing buffer. The pNpp substrate solution provided in the kit was added at 200 μl/well, and the plate was incubated for 1 h at room temperature without shaking.

Chemogenetic manipulation of the activity of speci c neurons
Chemogenetic manipulation of neurons was carried out using a DREADDs system as described

Drug infusion through a cannula
Cannula implantation and drug infusion were performed as described previously 63 . In brief, mice were anesthetized with ketamine and xylazine (

Western blot analyses
Western blot analyses were performed as described previously 38,58 . Brie y, PL tissue (n = 2 -4 mice) was . The right side operation followed the same procedure as the left. Adrenalectomized mice were subsequently given 0.9% saline solution instead of drinking water, but they were given no corticosterone replacement. Sham-operated mice underwent the same surgical procedure, but although the adrenal glands were grabbed with the forceps, but they were not removed. After 7 recovery days, half of the ADX and sham-operated mice were subjected to RS5 treatment, and the remaining mice served as controls.

Microarray analysis
Microarray analysis was carried out as described previously 38 . In brief, two independent sets of control (CON), CRST, and CRST+RS5 groups were prepared. Mice were subjected to CRST treatment followed by RS5, and then sacri ced on post-stress day 22. Total RNA was extracted from the pooled PL of the 6 -7 animals in each group using Trizol (Invitrogen Life Technologies), followed by puri cation using RNeasy columns (Qiagen, Valencia, USA). The puri ed RNA had an A260/280 ratio of 1.9 -2.1 as determined by an ND-1000 Spectrophotometer (NanoDrop, Wilmington, DE, USA) and an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA).
A functional protein-protein interaction (PPI) network (PPI enrichment p-value, <1.0e-16) consisting of 8 clusters (k = 8) were chosen as a representative grouping: Cluster 1 contained the most members (622 genes). However, the members of cluster 1 were poorly interactive and scattered outside the functional networks formed by the remaining 7 clusters, which included 238 genes in total. Therefore, the cluster 1 was excluded from further analyses. Clusters 2 to 8 contained 71, 39, 29, 28, 27, 25 and 19 genes, respectively. Clusters 5 and 7, which contained 28 and 25 genes, respectively, covered the GO terms "response to stress" and "response to glucocorticoids", along with the cell communication, the estabilishment of localization, and response to endogenous stimuli.
Potential interactors which were missed from the microarray screening but that could exist in the functional PPI networks were sought in the STRING network database. Using an edge con dence of 0.7 and higher to classify nodes in the PPI networks, that effort produced 16 new interactors. Afterwards, functional networks were constructed to include the members of clusters 5 and 7 and the 16 new members. The nodes and edges for interactions with members were color-coded using Cytoscape StringApp v3.7.2 57 .
Real-time PCR was used to examine the expression pro les of the Nr3c1(GR), Nr3c2(MR), Fkbp5, Fkbp4, Hsp90aa1, and Hsp90ab1 genes, whch were arbitrarily chosen from cluster 7, and the expression of Dusp1, CaMKIIα, Mapk3, and Mapk1, which were selected from cluster 5 for the indicated groups.

Behavioral assessments
Behavioral tests were carried out as described previously 38,39 . In brief, all behavioral tests were recorded with a computerized video tracking system (SMART; Panlab S.I., Barcelona, Spain) or a webcam recording system (HD Webcam #C210; Logitech, USA). All behavioral tests among groups and within groups were conducted in a randomized fashion or in an alternating manner with respect to test order and position within the testing equipment or test eld (e.g., left vs. right side, between positions in the test room). CRST-induced persistent behavioral changes, and RS5 effects were veri ed in a blind manner by two experimenters. All animals were housed in pairs unless otherwise indicated.

Two-chamber social interaction test
The two-chamber social interaction test (SIT) was developed as a modi ed version of the three-chamber test 58 and the U-shaped two-choice eld test 68 . In brief, the two-chamber apparatus consisted of two symmetrical chambers (26 x 26 cm rectangular oor with 40 cm walls) separated by an intermediate path (8 cm wide x 10 cm long x 26 cm in height), with an empty circular grid cage (12 cm in diameter x 33 cm in height) in each chamber. A subject mouse was placed in the middle path and allowed to freely explore the two chambers for 5 min as habituation. While the subject mouse was returned to its home cage for 2 min, a social target (C57BL/6) was placed in one of the grid cages. As soon as the social target was stabilized, the subject mouse was placed in the middle of the two choice chambers, with a social target on one side and an empty grid cage on the other side, and allowed to freely explore the chambers for 5 min. The time spent and the trajectory taken in each chamber were recorded. The side containing the social target was called the target eld, and the opposite side was called non-tarhget eld. Detailed procedures for preparation of subject animals, habituation, social targets, sociability and social interaction tests, and handling of the behavior test room were carried out as described in a previous report 68 .
U-shaped two-choice eld sociability test The U-shaped two-choice eld sociability test was conducted as described previously 38,39,68 . Brie y, the U- habituation, a subject mouse was allowed to freely explore the U-shaped eld containing an empty grid cage on each side for 5 min. While the subject mouse was returned to its home cage for 2 min, a social target was loaded into one of the grid cages. Afterward, the subject mouse was again allowed to explore the U-shaped eld for 5 min, and the time spent and trajectory taken were recorded. The side containing the social target was called the target eld, and the opposite side was called non-target eld.

Open-eld social interaction test
The open eld version of the social interaction test was carried out as described previously 59,60 . Brie y, a mouse was placed in an open eld (45 cm x 45 cm x 40 cm) containing an empty perforated interaction box at one side and allowed to explore it for 2.5 min for habituation. While the subject mouse was returned to its home cage for 1 min, a social target was placed in the interaction box, and then the subject mouse was allowed to freely explore the open eld for 2.5 min. The time spent interacting with the social target and locomotive activity in the eld were recorded.

Sucrose preference test
The sucrose preference test (SPT) was carried out as described previously 69 . Brie y, mice were habituated to two water bottles for seven days, followed by the presentation of two bottles containing a 1% sucrose solution for 24 h. The mice were then deprived of water and sucrose solution for 9 h (10 AM -7 PM). Beginning at 5 PM, the mice were singly housed in a cage provided with food but no water. At 7 PM, each mouse was given two bottles, one containing water and the other containing a 1% sucrose solution. The positions of the water and sucrose bottles were changed at 8 PM, 9 PM, and 10 PM. The amounts of water and sucrose consumed were measured by weighing the bottles. The sucrose preference (%) was calculated as the amount of sucrose solution consumed over the amount of water for 2 h (8 PM -10 PM).

Tail suspension test
The tail suspension test was carried out as described previously 38,39 . Brie y, mice were individually suspended for 6 min by xing their tails with adhesive tape to the ceiling of a rectangular box, leaving them 50 cm above the table surface. During the 6 min period, the cumulative immobility time was counted. Immobility was de ned as the total time in which all limbs and the body did not move.

Forced swim test
The forced swim test was carried out as described previously 38,39 . Brie y, mice were placed for 6 min in a transparent Plexiglass cylinder (15 cm in diameter x 27 cm in height) containing water (a depth of 15 cm) at 24 °C. The cumulative immobility time was counted for the last 5 min. Immobility was de ned as oating with all limbs motionless.

Principal component analysis and K-Means clustering
The behavioral features of individual animals in multiple behavior tests were analyzed using K-Means clustering, an unsupervised machine learning algorithm that can portion individual data points into distinct subgroups without reference to known or labeled outcomes 70  individual data points were randomly selected to nd two centroids by computing the sum of the squared distance between data points. The two centroids used the following objective function (J) to subsequently assign each data point to the closest cluster center.
The objective function is:

Statistical analysis
Statistical analyses were conducted using GraphPad Prism 6 software (GraphPad Software, Inc., CA, USA). Two-sample comparisons were performed using Student's t-test, and multiple comparisons were performed using one-way ANOVA and the Newman-Keuls post-hoc test. All data are presented as mean ± SEM, and statistical differences were accepted at p < 0.05 unless otherwise indicated. The statistical details of the results of all gures are provided in Supplementary Table 4.

Data availability
Further information and requests for resources and reagents should be directed to and will be ful lled by the corresponding author upon request.

Competing interests
The authors have no competing nancial interests to declare.
Author Contributions