Male offspring of C57BL/6J mice (the Academy of Chinese Military Medical Science) were used as the intruder subjects and obtained at weaning (PND 21) from our in-house breeding program (Center of Experimental Animal, Institute of Psychology, Chinese Academy of Sciences). At PND 21, male siblings were housed in groups of 2–4 mice/cage until the start of the experiment. Adult (2–2.5 months old) CD1 mice (Vital River Laboratories) were used as the resident subjects and housed individually. All animals were kept under controlled environmental conditions (ambient temperature 20 ± 2°C, 12 h light–dark cycle with lights on at 07:00 a.m.) with free access to water and food except during the attentional set-shifting task (AST). All experimental procedures were approved by the Institutional Review Board of the Institute of Psychology, Chinese Academy of Sciences and were in compliance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

The Experiment was Conducted in Two Parts:

Experiment 1: at PND 28, male siblings were separately assigned to two groups: stress (STR) or control (CON). Mice in the stress and CON groups were exposed to social defeat stress or manipulated for 10 days (PND 28–37). After that, all mice were individually housed. Some mice in each group underwent behavioral testing 1 week after the last stress session. Twenty-four hours after the behavioral assessment, the mice were sacrificed and brain samples were collected. Other mice were individually housed until 6 weeks later during adulthood (PND 80), for the second test; the experimental procedure was replicated. Single housing after the stress is necessary for the adult deficit of cognitive function induced by social defeat during early adolescence, and this effect can be ameliorated by social housing (with siblings) after the stress (for details, see Zhang et al., 2016).

Experiment 2: the stressed and CON groups were similar to those described for experiment 1. Four weeks after the last stress, mice in the STR and CON groups were further divided into two subgroups: saline (SAL) or duloxetine (DUL). They were intraperitoneally injected with saline or duloxetine at a volume of 0.1 ml/25 g body weight daily over a period of 14 days (PND 65–79). At PND 80, the same behavioral tests and brain sampling procedures outlined for experiment 1 were performed.

Duloxetine [(S)-Duloxetine Hydrochloride] (Sigma-Aldrich, St. Louis, MO, USA) used in experiment 2 was dissolved in saline solution (0.9%) at a concentration of 2.5 mg/ml. Beginning at PND 65, stress and CON mice were treated with duloxetine or saline at a dose of 10 mg/kg for 14 days. This dose was chosen based on the stable antidepressant action of duloxetine reported in previous studies (Calabrese et al., 2007; Engel et al., 2013).

The “resident-intruder” paradigm similar to that in previous studies by us and others was performed to mimic social defeat (Golden et al., 2011; Zhang et al., 2016). In brief, a daily stress protocol consisted of two steps: first, the intruder C57BL/6J mouse in the defeat group was placed in the cage of the resident CD-1 mouse and was attacked for 5 min or for 3 min if the attack by the CD-1 mouse was intense (attack latency shorter than 60 s, multiple or continuous attack). Thereafter, the intruder was separated from the resident by a clear perforated Plexiglas divider; sensory contact with the resident was maintained for the remainder of the 24 h period. The C57BL/6J intruders were rotated every day across nine defeat days to avoid habituation to a single aggressor. CON mice were pair housed with other C57BL/6J mice but kept on the opposite side of the cage by the divider. All CON mice were rotated on a daily basis in a manner similar to that of mice undergoing defeat, but they were never allowed physical contact with their cage mate.

On the last day of the social defeat or control manipulation, the mice were placed on food restriction for 7 days to maintain 80–85% original body weight with free access to water. AST testing was similar to the method described for a previous study (Bissonette and Powell, 2012; Yuan et al., 2014). In brief, a 4-day training and testing protocol was initiated on the fourth day of food restriction. The mice were initially trained to locate a buried reward by digging into two sawdust-filled pots placed in the home cage. On the fifth day, mice were transferred to the testing arena and trained to retrieve the reward from both sawdust-filled pots. On the sixth day of food restriction, mice were required to perform two separate simple discrimination (SDs) using two sets of exemplar pots scented with different odors (lemon vs. rosewood, both pots filled with sawdust) or filled with different digging media (foam sheets vs. shredded paper, no odor). On the seventh day of testing, a five-stage discrimination task, including SD, compound discrimination (CD), intra-dimensional shifting (IDS), RL and extra-dimensional set shifting EDS were performed. Animals proceeded from one stage to the next after the criterion of six consecutive correct trials was achieved. A typical task sequence and a description of the test stimuli are included in Supplementary Table S1. The number of trials to achieve the criterion and the number of errors to achieve the criterion were recorded at each stage for each mouse.

For qPCR, the mice were decapitated 24 h after the end of behavioral testing, and the brain was quickly removed. The mPFC was rapidly dissected on an ice plate pretreated with RNA stabilization solution. The tissue was immediately frozen in liquid nitrogen and then stored at –80°C for subsequent analysis. In particular, according to the atlas of Paxinos and Franklin (2003), the mPFC was dissected from a 1-mm-thick coronal section using a mouse brain mold (approximately 2.34–1.34 mm from the bregma, containing the cingulate, prelimbic and infralimbic (IL) subregions) under microscopic observation (Leica, Germany).

Total RNA was isolated by homogenization in TRIzol Reagent (Invitrogen, Carlsbad, CA, USA), according to the manufacturer’s instructions. The concentration and purity of the isolated RNA were determined by measuring the absorbance (NanoDrop 2000, Thermo, Wilmington, DE, USA). Next, cDNA was prepared from RNA with the SuperScript^TM III reverse transcription kit (Invitrogen, USA) according to the manual. The following forward (F) and reverse (R) primers (Invitrogen) were used in this study. For GAPDH, F: TGCACCACCAACTGCTTA; R: AAGTGTACAAGTCCGCGTCC; for BDNF, F: TGGCTGACACTTTTGAGCAC; R: AAGTGTACAAGTCCGCGTCC. The housekeeping gene GAPDH was used as an internal CON and was amplified from each sample. Real-time quantitative PCR using cDNA from individual animals was performed. The amplification protocol was as follows: 95°C for 10 min, 95°C for 30 s and 60°C for 1 min (40 cycles), followed by signal detection for 10 s. Each sample was assessed in duplicate, and the average value was calculated for each sample. Relative BDNF mRNA expression levels were calculated according to the 2(−ΔΔC (T)) method.

The target tissue, mPFC, was collected as described for qPCR analysis and homogenized in lysis buffer supplemented with protease inhibitor cocktail (Roche Diagnostics, Indianapolis, IN, USA). The protein concentration was determined using a BCA Protein Assay Kit (Cwbiotech, Beijing, China). Equal amounts of protein were added with loading buffer and denatured at 95°C for 8 min before separation by SDS-PAGE (12% polyacrylamide gel). After transfer to a PVDF membrane (Millipore, Bedford, MA, USA), nonspecific binding sites were blocked with 5% fat-free milk for 1 h at RT. The membranes were subsequently incubated overnight at 4°C with a primary rabbit polyclonal anti-BDNF antibody (1:1000, Abcam, Cambridge, UK) and a mouse β-actin antibody (1:1000, TA-09, Zhongshan Golden Bridge, Beijing, China). After being washed in TBST (3 × 10 min), the membranes were incubated with an HRP-conjugated goat anti-rabbit IgG secondary antibody (1:10,000, Jackson ImmunoResearch, West Grove, PA, USA) for 40 min at RT and then washed again. The images were visualized with enhanced chemiluminescence (ECL, Millipore, Bedford, MA, USA) using a FluorChem E System (Protein Simple, Santa Clara, CA, USA). Quantity One software (Bio-Rad, Hercules, CA, USA) was used to quantify the data to determine protein levels, which were normalized to the levels of β-actin.

The protocols were adapted from our previous studies (Wang et al., 2015). Twenty-four hours after the final behavioral test, the mice were deeply anesthetized with 10% chloral hydrate (0.3 ml, intra-peritoneal injection) and transcardially perfused with pre-cooled (4°C) 0.1 M PBS followed by 4% paraformaldehyde in 0.1 M PBS (pH 7.4). Brains were removed and post-fixed in 0.01 M PBS containing 4% PFA for 24 h. A 4-mm segment of interest containing mPFC (3.56–0.72 mm from the bregma) was dissected using a mouse brain mold. The samples were dehydrated, cleared and embedded in paraffin at 60°C for later analysis. Prior to analysis, the samples were sectioned on a microtome (Leica Microsystems, Wetzlar, Germany) at RT, and coronal paraffin sections (5 μm thick) containing the target area of mPFC (1.98–1.34 mm from the bregma, Figure 1) were prepared (Paxinos and Franklin, 2003). The sections were transferred onto coated slides (2 sections/slide) and dried on a heating plate at 60°C for 30 min.

For immunohistochemical analysis, the sections were dewaxed 3 × 10 min in xylene and then dehydrated by serial ethanol rinsing (2 × 2 min in 100% ethanol, 2 × 2 min in 95% ethanol and 1 × 2 min in 85% ethanol). After washing in 0.01 M PBS (3 × 2 min), the sections were heated in a 0.1 M citrate buffer solution in a microwave oven. The sections were then washed again in 0.01 M PBS (3 × 2 min), incubated in 3% hydrogen peroxide (15 min at RT) to eliminate endogenous peroxidase and then washed again and incubated in a blocking solution (0.01 M PBS with 5% goat serum) for 25 min at RT. The sections were then incubated with primary rabbit anti-BDNF polyclonal antibody (1:200, Santa Cruz Biotechnology, CA, USA) overnight at 4°C, washed in 0.05 M PBS (3 × 5 min), incubated in secondary goat anti-rabbit antibody (1:1000, Santa Cruz Biotechnology, CA, USA) for 1.5 h at RT, and then washed again. Finally, the sections were stained with 0.05% 3,3′-diaminobenzidine (DAB; ZLI-9018, Zhongshan Golden Bridge, Beijing, China), dehydrated and transparentized by gradient ethanol (1 × 2 min 85%, 2 × 2 min 95%, 2 × 2 min 100%) and xylene treatment (3 × 10 min), and then cover-slipped with neutral balsam.

Quantification of the staining density (optical density (OD)) was performed in a blinded manner using a light microscope (Olympus BX-51 with a Camedia Master C-3040 digital camera) and ImageJ software (version 1.37 V, NIH, Bethesda, MD, USA). For each subject, the staining intensity in the mPFC was analyzed. Each region of interest (ROI) within the mPFC was analyzed in three consecutive sections. For each section, the staining density of nonspecific background labeling was measured in a cell-free area and digitally subtracted so that any background staining was eliminated. The ROIs for mPFC remained uniform for each section. For analysis, OD values of the three bilateral sections were averaged into a single score for each animal.

The commercially available program SPSS 16.0 was used for the statistical analysis. All data are presented as the mean ± standard error of the mean (M ± SEM). The AST data analysis was performed using two-way ANOVA (stress × stage) with repeated measures in experiment 1 and three-way ANOVA (stress × drug × stage) with repeated measures in experiment 2. The BDNF mRNA and protein data from experiment 1 were analyzed using the t-test, and the BDNF immunochemistry data in experiment 2 were analyzed using two-way ANOVA (stress × drug). When significant main effects or interactions were detected, t-test or LSD post hoc comparisons were conducted. Significance for all analyses was set at p < 0.05.

As shown in Figure 4A, when tested 6 weeks after the last stress, adult mice that experienced social stress during adolescence exhibited lower levels of BDNF mRNA in the mPFC than the adult controls (t_(1,14) = 4.103, p < 0.01). In addition, western blot analysis showed that previously stressed mice had less BDNF protein in the mPFC than the corresponding controls during adulthood (t_(1,11) = 2.433, p < 0.05; Figure 4B).

In adult animal, lesions, pharmaceutical and chronic stress studies have demonstrated that the mPFC plays a critical role in set-shifting performance (Birrell and Brown, 2000; McAlonan and Brown, 2003; Liston et al., 2006). This suggested that impaired EDS performance in the AST, as seen in this and the previous study after adolescent social stress, might be attributable to abnormal mPFC functioning. In rodents, significant structural and functional remodeling of the mPFC occurs throughout the adolescence, with progressive and regressive changes in synapses and receptor (Giedd, 2008). It is thought that at this stage, neural structures undergoing rapid development are more susceptible to stress, which may alter the trajectories of critical developmental events, resulting in the expression of behavioral symptoms after their maturation in adulthood (Andersen and Teicher, 2008). Consistent with this, several studies have indicated that adolescent stress, particularly early adolescent stress, can cause long-term changes in synaptic density and morphology of the mPFC in adult animals (Leussis and Andersen, 2008; Eiland et al., 2012). This study provided additional evidence that alterations in mPFC function mediated the impaired cognitive performance induced by adolescent social defeat stress. Moreover, the data suggest that the mPFC alterations develop over time as they were not present 1 week after the social stress.

In parallel with the development of cognitive dysfunction after adolescent social defeat exposure, there had corresponding changes in mPFC BDNF expression in this study. Brain BDNF expression can be affected by multiple factors, including age and stress (Fanous et al., 2010; Taylor et al., 2011; Bath et al., 2013). In the rodent, PFC BDNF expression initially changes in a regionally specific manner across postnatal developmental stages, characterized by very low levels during prenatal and neonatal periods, then a rapid increase to peak levels during pre- and early adolescence, followed by a gradual decline over the lifespan (Kolbeck et al., 1999; Bath et al., 2013; Luoni et al., 2014). This study showed there was no difference in the mPFC BDNF mRNA levels between CON mice at PND 45 and at PND 87 (t_(1,16) = 0.802, p = 0.434, data not shown). This result is consistent with findings by Guo et al. (2010), who reported that the protein expression of BDNF in the rat PFC was comparable at PND 42 and PND 77. These results suggest that mPFC BDNF expression decreased to a similar level from late adolescence to early adulthood in mice.

Given that BDNF is an activity-dependent molecule, previous studies by our group and others have shown that early stress (maternal separation) may disturb the normal developmental pattern of PFC BDNF and its later function (Roceri et al., 2004; Lee et al., 2012; Wang et al., 2015). We found that social stress during early adolescence induced a transient increase in mPFC BDNF mRNA levels 1 week after the last exposure at PND 45, while a long-term decrease in mPFC BDNF mRNA levels was observed 6 weeks later, at PND 87, suggesting a more obvious decline trajectory in previously stressed mice compared to the normal developmental decline tendency from late adolescence to early adulthood in CON mice, as described above. Coppens et al. (2011) reported that social defeat in adolescent rat led to up-regulation of BDNF mRNA expression in hippocampus, but not in the PFC. This discrepancy may be attributable to different experimental conditions in the two studies, including stress duration and age at exposure to stress (Bath et al., 2013). Furthermore, western blot and immunohistochemical analyses also showed a protracted effect of adolescent social stress on BDNF protein expression in the mPFC. This effect was region-specific, given that the decrease in BDNF expression was evident in the mPFC but not in the OFC, as shown by immunohistochemical data. Consistently, human adults who have experienced abuse during childhood and adolescence demonstrate an increased susceptibility to the development of depression and also present with lower salivary BDNF levels (Grassi-Oliveira et al., 2008). It has been shown that BDNF and its receptor TrkB are critically involved in the modulation of neural development and structural and synaptic plasticity in the mPFC (Ward and Hagg, 2000; Finsterwald et al., 2010). Long-term inhibition of BDNF signaling in the rodent PFC by genetic and stress manipulations can alter the pattern of neural connectivity, the growth and complexity of dendrites and synaptic plasticity, and these effects can be reversed by direct BDNF rescue (Lu et al., 2008; Sakata et al., 2013). Thus, the long-term decrease in BDNF expression in adult mice subjected to adolescent social defeat might exert detrimental effects on mPFC plasticity and the cognitive performance mediated by this area (Liston et al., 2006; Ragozzino, 2007).

We demonstrated that chronic treatment with the antidepressant duloxetine improved the cognitive dysfunction in EDS and the decreased expression of BDNF in the mPFC of previously stressed adult mice. Clinical evidence has demonstrated that the improvement of depressive symptoms and the recovery of neuroplasticity decrease usually occur concomitantly after chronic antidepressant treatment (usually 2~4 weeks; Pompili et al., 2013). Consistent with this, BDNF has emerged as a critical mediator for the efficacy of various antidepressants by reversing the reduced neuroplasticity in the PFC and hippocampus in depression; the same improvement in neuroplasticity also occurs after chronic antidepressant treatment (Castrén and Rantamäki, 2010; Duman and Monteggia, 2006). For example, chronic duloxetine administration increases PFC BDNF mRNA and protein levels in animal models for depression, correlating with an improvement in emotional behaviors (Calabrese et al., 2007; Mannari et al., 2008; Engel et al., 2013). Chronic but not acute duloxetine treatment can also increase BDNF levels in the blood of patients with depression (Fornaro et al., 2015; Hashimoto, 2015). Furthermore, such effects of antidepressant treatments can be prevented by the abolishment or inhibition of BDNF signaling (Adachi et al., 2008). Thus, our study provides extensive evidence that chronic duloxetine treatment during adulthood can rescue the impairment in cognitive flexibility and the decrease in mPFC BDNF expression induced by adolescent social stress.

In summary, our study demonstrates that adolescent social stress impaired cognitive flexibility in adulthood. Moreover, we show that this cognitive dysfunction was accompanied by reductions in the mRNA and protein levels of BDNF in the mPFC. The long-term cognitive and molecular alterations following adolescent stress could be reversed by chronic antidepressant treatment during early adulthood. These findings suggest that adolescent social stress could be used to model depression-related psychopathologies during adulthood, including cognitive impairment and inhibition of BDNF signaling in the mPFC. The identification of molecular targets that can enhance BDNF signaling in the mPFC may reveal the mechanisms underlying cognitive dysfunction and contribute to effective strategies to improve the treatment of depression.