Male or female C57BL6/J mice (8 weeks old; weighing 20 ± 2 g at the beginning of the experiments) were obtained from SPF (Beijing) Biotechnology Co. Unless otherwise specified, mice were housed in groups of four per plastic cage in a 12-h light–dark cycle (from 7 a.m. to 7 p.m.). Food was provided with a standard chow diet, and water was available ad libitum. All experimental procedures were approved by the Beijing Institute of Pharmacology and Toxicology. Laboratory animal care was conducted in accordance with the Chinese Veterinary Medicine Association guidelines.

The experimental apparatus consisted of one star arm and four choice arms (two baited arms and two empty arms), which were modified by a standard radial arm maze (each arm was 40 cm long, 10 cm wide, and 10 cm high; Shanghai Xinruan Information Technology Co., Ltd., China). Unused arms were blocked by a black plastic door. The central platform and arms were light gray. Each end of the arm contained a clear plastic water bottle fixed to the arm ∼3 cm high from the bottom (Figure 1A). The overhead CCD camera allowed detection of the position of animals, and the location of the apparatus was fixed throughout the entire experiment. The apparatus was put in a sound-attenuated testing room.

Before the start of the experiment, each mouse was habituated to and pretrained in the apparatus for 10-min free explorations for 1 week, during which water bottles were placed in all four choice arms that the mice were free to drink. All mice were subjected to 12-h mild water deprivation each day before trials.

In the risk decision-making task, the reward was sucrose water, and quinine water was used for the adverse outcomes. Mice were placed at the end of the start arm and allowed to choose one of the four choice arms when the experiment started. Mice could get different flavors of water in the baited arms and nothing in the empty arms. When a mouse drank water from the bait arm during a trial, it was immediately taken back to the start arm for the subsequent trial to prevent the mice from getting full. When a mouse entered the empty arm, it was taken back to the start arm to continue this trial until it entered the baited arm and drank water. Each mouse was subjected to 16 trials per day, with intertrial intervals of 5 s between each trial, and repeated for 14 consecutive days.

The baited arms contain one low-risk/low-reward (L-L) and one high-risk/high-reward (H-H) arm. In the beginning, we used a 12.5% risk/reward probability as the standard trial condition. In the L-L arm, a mouse has a 12.5% (2/16 trials) probability of drinking quinine water and an 87.5% (14/16 trials) probability of drinking regular water. In the H-H arm, a mouse has a 12.5% (2/16 trials) probability of drinking sucrose water and an 87.5% (14/16 trials) probability to drink quinine water. Quinine water in the L-L arm and sucrose water in the H-H arm were provided randomly at a 12.5% probability during 16 trials per day. L-L arm or H-H arm choice (%) was calculated as follows:

Mice were randomly assigned to a control or stress group. Mice in the stress group were restrained in a 50-ml conical tube with an air vent for 2 h, followed immediately by 20 min of forced swimming in water (20–24°C). Then, after 15 min of recuperation, mice were exposed to ether in a glass desiccator until fully anesthetized without toe or tail pinch response (total time less than 5 min). Mice were immediately returned to their home cages for 7 days without further disturbance until decision-making experimental procedures commenced. Mice in the control group remained in their home cages and undisturbed for 7 days.

Fluorescence immunohistochemistry was performed as in the previous study (Wei et al., 2021). Briefly, animals were perfused using heparinized saline under deep anesthesia and followed by cold 4% paraformaldehyde in 0.1 M phosphate-buffered saline (PBS). Mice brains were dissected and continuously fixed for 12 h at 4°C. Then, the brains were dehydrated with 30% sucrose in PBS. Then, 40-μm-thick coronal sections were made using a freezing microtome (CM 1950, Leica, Germany), and sections were stored at 4°C in PBS until use. For c-Fos staining, brain slices were made from the mice 1.5 h after the SPS procedure and then incubated with rabbit anti-c-Fos antibody (1:1,000; sys-226-003; Synaptic Systems) for 24 h at 4°C. Then, sections were incubated with Alexa Fluor 488-conjugated donkey anti-rabbit IgG (1:1,000) in PBS for 1 h at room temperature. Finally, the slices were mounted on slides with the fluorescent mounting medium containing DAPI, covered with coverslips, and stored at –20°C for further examination. To visualize and quantify the number of c-Fos positive cells, we use a Zeiss 880 confocal microscope to scan the slices (Zeiss, Oberkochen, Germany). Then, 6–8 different sections from each animal were selected for counting. To examine the effect of CNO treatment on neural activity in the AIC, mice were killed 1.5 h after CNO treatment.

During surgery, mice were anesthetized with sodium pentobarbital (50 mg/kg). For chemogenetic manipulation, 0.2 μl of AAV-CaMK2α-hM4D(Gi)-mCherry (AAV2/8, 2.57 × 10¹² genomic copies per ml, OBiO Technology, China) was bilaterally infused through a 33-gauge needle attached to a 10-μl Hamilton syringe at a rate of 0.02 ml/min. The injection site is AIC (AP: 0.74 mm; ML: ± 2.75 mm; DV: 3.75 mm). After injection, the needle was left in place for an additional 10 min to allow total dispersion of the virus. For selective AIC-BLA circuit inhibition, 0.1 μl of anterograde trans-neuronal virus AOV-hSyn-EGFP-Cre (AAV2/1, 5.24 × 10¹², OBiO Technology) was bilaterally injected into AIC, and 2 weeks later AAV-EF1α-DIO-hM4Di-mCherry (AAV2/8, 9.39 × 10¹² genomic copies per ml, OBiO Technology, China) was injected into BLA (AP: –1.22 mm; ML: ± 2.80 mm; DV: 4.80 mm). For all experiments, the accuracy of injection coordinates was confirmed by visualization of fluorescent in the injection needle tracks in 40-μm tissue sections.

Mice were deeply anesthetized with isoflurane and decapitated. Coronal brain slices (300 μm) containing the anterior insular cortex were cut with a vibratome in oxygenated artificial cerebrospinal fluid (ACSF). For electrophysiological experiments, the brain slices were then transferred to a submerged recovery chamber with oxygenated ACSF at room temperature. This used standard methods for whole-cell voltage-clamp electrophysiology from our previous study (Shi et al., 2018). Briefly, for AIC neuron recording, the excitatory postsynaptic currents (EPSCs) were recorded from layer II/III neurons with an Axon 700B amplifier (Molecular Devices, USA), and the stimulations were delivered by a bipolar tungsten electrode placed in layer V/VI. The recording pipettes (3-5MΩ) were filled with a solution containing (in mM) 124 K-gluconate, 5 NaCl, 1 MgCl₂, 0.2 EGTA, 10 HEPES, 2 MgATP, 0.1 Na₃GTP, and 10 phosphocreatine disodium (adjusted to pH 7.2 with KOH). Picrotoxin (100 mM) was invariably used to block GABAA receptor-mediated inhibitory synaptic currents in all experiments. The amplitudes of EPSCs were adjusted between 50 and 100 pA to obtain a baseline. For long-term potentiation (LTP) induction, a TBS protocol was used. Data were discarded if the access resistance changed > 15% during the experiment. Data were filtered at 1 kHz and digitized at 10 kHz.

All data are represented as mean ± SEM throughout the text and figures. Primary statistical analyses were conducted using Prism 5 (GraphPad, La Jolla, CA, USA). Differences between groups were analyzed using t-tests and one-way or two-way analysis of variance (ANOVA) for multigroup comparisons or repeated-measures ANOVA. In all cases, P < 0.05 was considered statistically significant.

A modified eight-arm maze was used for the risk decision-making task of mice (Figure 1A; Method Details). After 12 h of water deprivation per day, habituated mice were urged to enter the baited arm to find water. It presented the mice with a conflicting choice between sucrose water (high reward) paired with a high probability of quinine water (high risk) and normal water (low reward) paired with a low probability of quinine water (low risk). To find an attractive enough reward and aversive enough condition, preference experiments were performed on different concentrations of sucrose or quinine. It showed that the high concentration of sucrose water did not cause preference but also caused aversive in mice, whereas mice repelled all different concentrations of quinine (Supplementary Figure 1). In this study, 2% sucrose water and 0.5-mM quinine water were chosen as reward or risk factors. The risk/reward probability was set at 12.5% (2/16 trials) at the beginning of the experiment (Figure 1B). Choosing the low-risk/low-reward (L-L) arm may result in randomly encountering bitter water two times in 16 trials but can drink normal water with a high probability (14/16 trials, 87.5%). Choosing the high-risk/high-reward (H-H) arm means that mice can get sweet water with a low probability (2/16 trials, 12.5%), but they are very likely (14/16 trials, 87.5%) to drink quinine water. During 14 days of the gambling test, mice gradually showed a preference for the L-L arm. The L-L arm choice rate reached 82.7 ± 3.7%, while the H-H arm choice was 11.6 ± 3.0% on day 14 (Figure 1C). The number of entries into empty arms, which could be considered an index of the learning ability of mice, dropped rapidly after a few days of task training and is stably maintained (Figure 1D). Arm choice preferences changed with the probability of risk/reward. When the risk probability increased from 12.5% to 50%, the number of L-L arm entries decreased. Conversely, when the reward probability increased from 12.5% to 50%, the number of H-H arm entries increased (Supplementary Figure 2). These results suggest that mice preferred the L-L arm over the H-H arm, similar to healthy humans in the Iowa gambling task (IGT).

To determine the effect of chronic stress on the risk decision-making task, mice were treated with single prolonged stress (SPS), a well-established mice model to mimic behavioral characteristics of post-traumatic stress disorder PTSD (Figure 2A). Stressed mice showed enhanced anxiety-like behavior, such as decreased exploratory time in the center of the open field [t ₍₁₄₎ = 2.097, P = 0.0546, n = 8, unpaired t-test] and open arm of EPM [t ₍₁₄₎ = 2.718, P = 0.047, n = 8, unpaired t-test] (Figure 2B). Previous studies from our laboratory revealed that the activation of the anterior insular cortex mediated fear and anxiety-like behavior (Shi et al., 2020, 2022). Significantly increased c-Fos expression was also investigated in AIC after SPS [t ₍₁₄₎ = 2.194, P = 0.0456, n = 8, unpaired t-test] (Figures 2C, D). The stressed mice gradually performed more likely to choose the H-H arm and less frequently to choose the L-L arm than matched controls in the decision-making task. Repeated two-way ANOVA for the L-L arm revealed significant effects on treatment [F_(1,14) = 19.81, P < 0.001], time [F_(13,182) = 25.07, P < 0.001], and their interaction [F_(13,182) = 1.561, P > 0.05]. For the H-H arm, it also revealed significant effects on treatment [F_(1,14) = 10.93, P < 0.001], time [F_(13,182) = 8.088, P < 0.001], and their interaction [F_(13,182) = 2.235, P < 0.05] (Figure 2E). The number of entries into empty arms did not differ between groups (Figure 2F). These findings suggest that stress-induced changes in AIC activity may play a crucial role in decision-making.

To determine whether inhibition hyperactivation of the AIC after stress was necessary to alter the risky choice strategy of mice, we used the chemogenetic method and bilaterally injected an AAV vector containing an hM4Di DREADD in the AIC. The gambling test is performed after a 30-min intraperitoneal (i.p.) injection of clozapine-N-oxide (CNO) or vehicle daily. CNO treatment decreased c-Fos expression in AIC (Figures 3A, B). During the 14-day test, CNO-treated mice showed an increased L-L arm choice and a decreased H-H arm choice probability compared with the vehicle group. Meanwhile, there was no significant difference in the number of empty arm entries (Figures 3C, D). This improvement in risk-taking behavior supported a causal relationship between alterations in the activity of the AIC and arm choice after stress.

Furthermore, previous studies have found that the BLA contributes to value computation and the encoding of the valence of positive or negative outcomes (Zhang et al., 2013). The BLA receives prominent projections from the AIC by injecting trans-neuronal virus AOV-hSyn-EGFP-Cre (Figure 3E), and this has also been confirmed in other studies (Lavi et al., 2018; Kayyal et al., 2019). We next searched for the neural circuit pathway of AIC-BLA that supports the risk decision-making behavioral mechanism. After injection of the trans-neuronal virus in the AIC, AAV-EF1α-DIO-hM4Di-mCherry was bilaterally stereotactically injected into the BLA 2 weeks later. The neurons labeled with red fluorescence are innervated by projection from the AIC (Figure 3F). Then, the mice were given i.p. injections of CNO (0.5 mg/kg) or vehicle (PBS) 30 min before the gambling test each day. Compared with the vehicle group, CNO treatment resulted in a gradual increment for L-L arm choice (F_(1,10) = 57.86, P < 0.001; Figure 3G), which further demonstrated that the activation of the AIC-BLA circuit is necessary for stress-induced changes in decision-making behavior. There was no difference in the number of entries into empty arms (Figure 3H).

Gender influences decision-making differently. Men tend to make riskier decisions than women (Charness and Gneezy, 2012). In this study, we first observed the decision-making behavior under standard conditions (12.5% risk/reward probability). Male and female mice did not exhibit a significant difference in decision-making behaviors, both showed more preference for the L-L arm and aversion for the H-H arm [for L-L arm, F_(1,20) = 0.2194, P = 0.6446; for H-H arm, F_(1,20) = 0.0126, P = 0.9118, Figure 4A]. There were no differences in empty arm choice ratio between male and female mice [F_(1,20) = 2.454, P = 0.1329, Figure 4B]. When the probability is increased to 25% (4/16 trials), male mice showed higher risk preference, whereas female mice did not exhibit a similar trend compared with male mice [for L-L arm, F_(1,20) = 12.90, P < 0.001; for H-H arm, F_(1,20) = 18.90, P < 0.001, Figure 4C]. There were significant effects on empty arm entries between male and female mice under 25% condition [F_(1,20) = 23.16, P < 0.001, Figure 4D]. The number of entries into empty arms is considered to represent an index of learning and memory of the task rules. These results may suggest that male mice need more time to learn the rules of the task or prefer to spend more time taking risks than female mice. By averaging the L-L choice rate acquired over 14 days, it showed that higher reward did not make female mice more risk-taking. Instead, the female mice chose more the L-L arm than the male mice under the 25% condition [two-way ANOVA, F_(1,40) = 5.560, P < 0.05; Figure 4E]. Furthermore, we evaluated the arm preferences of different sexes after stress. Female mice show less risk-taking than male mice [for the L-L arm, F_(1,14) = 6.470, P < 0.05; for the H-H arm, F_(1,14) = 20.71, P < 0.001, Figure 4F]. Moreover, there were no differences in the empty arm choice ratio [F_(1,14) = 3.816, P = 0.0711, Figure 4G]. Although there was a certain degree of L-L arm aversion, female mice did not show significant risk preference after stress [F_(1,34) = 1.608, control/female vs. stress/female, P = 0.0546, Figure 4H].

Stabilized estrogen levels are important for regulating and improving effect (Wharton et al., 2012). To test whether the endogenous estrogen in the AIC is also crucial for stress-induced alteration of decision-making behavior, we examined the effects of an intra-AIC injection ICI (ER inhibitor) through cannula during a gambling task (Figures 5A, B). ER antagonist increased the risk-taking behavior of female mice after stress [for L-L arm, F_(1,10) = 9.234, P < 0.05; for H-H arm, F_(1,10) = 10.97, P < 0.001, Figure 5C]. Whereas for mice that did not suffer stress, ICI had no significant effect on the decision-making behavior of normal female mice (Supplementary Figure 3). There was no significant effect on empty arm entries between the vehicle and ICI group [F_(1,10) = 0.4639, P = 0.5113, Figure 5D]. Previous studies have shown that estrogen, as a neuroprotective agent, is essential for synaptic plasticity and normal expression of LTP (Lu et al., 2019). To explore the role of estrogens on synaptic plasticity in decision-making behavior, whole-cell recordings were performed to determine the functional synaptic transmission and LTP in visually identified pyramidal neurons in layers II and III of the AIC. Pyramidal neurons fired repetitive action potentials with frequency adaptation (Figure 5E). Evoked EPSCs were obtained by delivering focal electrical stimulation to layer V/VI of AIC, and neurons were voltage clamped at -40 mV. IGT rapidly potentiated EPSC amplitude within minutes in most of the recorded neurons than the control group [F_(3,20) = 15.58, ^***P < 0.001, vs. control group, Figure 5F]. This increment can be partially occluded in the presence of ICI (^#P < 0.05, vs. IGT group), whereas perfusion of ICI alone had no effect on EPSCs. Next, to test the effect of endogenous estrogen on LTP induction, we used TBS to induce LTP. This protocol induced a significant long-lasting potentiation of synaptic responses in the AIC. The LTP amplitude was significantly occluded by ERs antagonist ICI (1 μM) (Figure 5G). Then, we analyzed the amplitude of EPSCs 5 min before TBS (pre-TBS) and 5 min after TBS (post-TBS) to examine the effect of ICI on LTP. ICI perfusion had no effect on the baseline of EPSCs, but significantly suppressed LTP induction immediately after TBS [one-way ANOVA, F_(1,40) = 18.09, ^**P < 0.01, vs. ACSF group of post-TBS, Figure 5H]. Together, these findings suggest that estrogen mediated-excitatory synaptic plasticity in AIC plays an essential role in decision-making behavior.