Cue- versus reward-encoding basolateral amygdala projections to nucleus accumbens

In substance use disorders, drug use as unconditioned stimulus (US) reinforces drug taking. Meanwhile, drug-associated cues (conditioned stimulus [CS]) also gain incentive salience to promote drug seeking. The basolateral amygdala (BLA) is implicated in both US- and CS-mediated responses. Here, we show that two genetically distinct BLA neuronal types, expressing Rspo2 versus Ppp1r1b, respectively, project to the nucleus accumbens (NAc) and form monosynaptic connections with both dopamine D1 and D2 receptor-expressing neurons. While intra-NAc stimulation of Rspo2 or Ppp1r1b presynaptic terminals establishes intracranial self-stimulation (ICSS), only Ppp1r1b-stimulated mice exhibit cue-induced ICSS seeking. Furthermore, increasing versus decreasing the Ppp1r1b-to-NAc, but not Rspo2-to-NAc, subprojection increases versus decreases cue-induced cocaine seeking after cocaine withdrawal. Thus, while both BLA-to-NAc subprojections contribute to US-mediated responses, the Ppp1r1b subprojection selectively encodes CS-mediated reward and drug reinforcement. Such differential circuit representations may provide insights into precise understanding and manipulation of drug- versus cue-induced drug seeking and relapse.


Introduction
During the development of substance use disorder (SUD), the drug as an unconditioned stimulus (US) with its primary rewarding effects drives drug taking and seeking (1, 2). Meanwhile, drug use is often associated with conditioned stimuli (CS), such as environmental cues, which also acquire incentive salience to promote drug seeking and relapse (1-3). For anti-relapse treatment, an effective strategy would be to coordinatively intervene the US-and CS-reinforced drug responses.
However, the cellular and circuit underpinnings that differentiate the US versus CS processing are not well understood, preventing the precise and targeted manipulations.
The nucleus accumbens (NAc) is a key brain region where medium spiny neurons (MSNs) process both US and CS inputs for behavioral outputs (4). Using region-specific lesion, pharmacological inhibition, and optogenetic manipulation, extensive previous studies demonstrate that the basolateral amygdala (BLA) provide both US-and CS-associated input to the NAc during drug taking and seeking (5)(6)(7). BLA pyramidal neurons heterogeneously respond to US, CS, as well as stimuli with appetitive or aversive valence (8)(9)(10). These BLA neurons are genetically separable into two populations, R-spondin2-positive (Rspo2) magnocellular neurons and protein phosphatase 1 regulatory subunit 1B-positive (Ppp1r1b, also known as DARPP-32) parvocellular neurons (11,12). Here, we explored whether Rspo2 and Ppp1r1b neurons project to the NAc to form two distinct BLA-to-NAc subprojections that differentially contribute to US-versus CSreinforced drug responses.
Our results show that optogenetic stimulation of either Rspo2 or Ppp1r1b presynaptic terminals in the NAc established intracranial self-stimulation (ICSS), indicating that both subprojections contribute to US-mediated responses. However, after establishment of the subprojection-specific ICSS, only Ppp1r1b mice exhibited cue-induced ICSS seeking, suggesting that the Ppp1r1b-to-NAc subprojection selectively encodes CS-mediated reinforcement.
Furthermore, chemogenetically increasing versus decreasing the transmission efficacy of Ppp1r1b-to-NAc, but not Rspo2-to-NAc, subprojection increased versus decreased cue-induced cocaine seeking after withdrawal from cocaine self-administration. These results characterize the differential behavioral roles of two major BLA-to-NAc sub-projections and provide a circuit mechanism through which US-versus CS-reinforced reward and drug seeking can be selectively manipulated.

BLA Rspo2 and Ppp1r1b sub-projections to the NAc
To selectively examine the Rspo2-to-NAc subprojection, we crossed the Rspo2-Cre and D1-tdTomato mice. In Rspo2-Cre x D1-tdTomato mice, we selectively expressed channelrhodopsin-2 (ChR2)-YFP in BLA Rspo2 neurons via an AAV-DIO vector and differentiated dopamine receptor D1-versus D2 NAc MSNs based on the presence versus absence of tdTomato signals ( Fig. 1A-C). Eight weeks after intra-BLA viral injection, we observed YFP signals in neuronal somas in the BLA as well as in axon fibers in the NAc core and shell ( Fig. 1D-I). We prepared brain slices containing the NAc and performed whole-cell voltage-clamp recordings of D1 and D2 MSNs in response to optogenetic stimulation of ChR2-expressing Rspo2 axon fibers (laser power, 2 mW; laser duration, 1 ms; interpulse interval, 7 sec; Fig. 1J). In both D1 and D2 MSNs, such stimulations evoked fast-activating and fast decaying synaptic responses that were inhibited by the antagonists of glutamatergic receptors (NBQX 10 µM, D-AP5 50 µM), indicating that these synaptic responses were excitatory postsynaptic currents (EPSCs) mediated by Rspo2-to-NAc monosynaptic transmission (Fig. 1KL).
Ppp1r1b BLA neurons can be genetically targeted through cocaine-and amphetamineregulated transcript protein (Cartpt)-Cre mice (11). We crossed Cartpt-Cre and D1-tdTomato mice to generate the mice in which we selectively expressed Ch2R in Ppp1r1b BLA neurons ( Fig. 1M-O). After 4-8 weeks, the BLA was enriched in ChR2-YFP-expressing neuronal somas, and YFPexpressing axon fibers were observed extensively in the NAc (Fig. 1P-U). Optogenetic stimulation of these Ppp1r1b fibers evoked EPSCs in both D1 and D2 NAc MSNs (Fig. 1V-X). Thus, both Rspo2 and Ppp1r1b BLA neurons project to the NAc and form monosynaptic connections with both D1 and D2 MSNs.
To examine the synaptic properties of Rspo2-and Ppp1r1b-to-NAc synapses, we optogenetically stimulated presynaptic terminals using a paired-pulse protocol, with interpulse intervals ranging from 25 to 200 ms (Fig. 1Y). The paired-pulse ratio (PPR), which is inversely correlated with the presynaptic release probability (Pr), was generally lower at Ppp1r1b synapses than Rspo2 synapses, suggesting a higher basal Pr at Ppp1r1b synapses (Fig. 1Z). Furthermore, for either Rspo2-or Ppp1r1b-to-NAc transmissions, EPSCs recorded in D1 and D2 MSNs exhibited similar PPRs, suggesting that each subprojection synapses on D1 and D2 MSNs with a similar Pr (Fig. 1Z).
For each of the subprojections, we recorded MSNs from both the NAc shell and core and did not detect preferential innervation of core versus shell MSNs. Taken together, the Rspo2 and   Ppp1r1b subprojections formed monosynaptic synapses with NAc MSNs, with presynaptic properties exhibiting between-projection difference.

Subprojection-specific optogenetic ICSS
ICSS has been extensively used in animal models to examine the reinforcing properties of specific brain regions, neuronal types, and neural circuits. For subprojection-specific ICSS, we prepared the mice in which BLA Rspo2 or Ppp1r1b neurons expressed ChR2 such that their ChR2-expressing axon terminals in the NAc could be self-stimulated bilaterally by laser through preinstalled optical fibers ( Fig. 2A). In the operant chamber, pressing one (active), but not the other (inactive), lever resulted in a 10-pulse train of laser stimulation (pulse duration, 1 ms), and the light cue was selectively conditioned to the active lever pressing ( Fig. 2A). Thus, if the stimulation is reinforcing, ICSS will be established.
We tested the mice over several daily sessions (1 h/session/day). In control mice in which Rspo2 or Ppp1r1b neurons expressed YFP alone without ChR2, intra-NAc laser stimulation at either 20 or 40 Hz did not induce ICSS (Fig. 2BC). In Rspo2-ChR2 or Ppp1r1b-ChR2 mice, stimulations at 40 Hz, but not 20 Hz or lower (data not shown), established ICSS (Fig. 2DE).
Thus, strong, but not moderate, activation of the axonal fibers of Rspo2-or Ppp1r1b-to-NAc subprojection reinforces operant responding.
After the ICSS was established, we tested the mice in an extinction procedure, in which pressing the active lever resulted in presentation of light cues but no stimulation. Over 3d of such testing, Rspo2-ChR2 mice decreased their operant responses to the level as in Rspo2-YFP control mice (Fig. 2F). In contrast, Ppp1r1b-ChR2 mice maintained high levels of responses to the active lever throughout the 3d testing (Fig. 2G). Thus, in addition to inducing ICSS, activation of Ppp1r1b-to-NAc subprojection may also assign incentive salience to conditioned cues, providing a circuit-based encoding for CS-mediated operant responses in the absence of US.
To determine whether this CS-mediated reinforcement is long-lasting, we retrained both Rspo2-and Ppp1r1b-ChR2 mice with ICSS for 3 d and then kept them in home cages for 21 d.
After this abstinence, we place the mice back into operant chambers for cue-induced seeking tests. Ppp1r1b-ChR2, but not Rspo2-ChR2, mice exhibited persistent cue-induced lever pressing, indicating the durable feature of the CS-mediated reinforcement established through the Ppp1r1bto-NAc subprojection (Fig. 2HI).
Despite ICSS-related caveats (see Discussion), the above results indicate that stimulation of either the Rspo2-or Ppp1r1b-to-NAc subprojection may serve as a US for establishing ICSS, while the Ppp1r1b subprojection functions as key component within the BLA-to-NAc input that encodes CS-mediated reinforcement. These differences between the two subprojections predict that they play differential roles in cue-conditioned drug seeking, which was tested below.

Acquisition and performance of cocaine self-administration
The BLA-to-NAc projection is one of the primary circuit underpinnings for cue-conditioned cocaine seeking (6,7). We first examined whether the two BLA subprojections contribute to the acquisition of cocaine self-administration, a procedure through which contextual and discrete cues are conditioned to cocaine experience. Our experimental strategy was to examine whether experimentally enhancing or decreasing the transmission of a subprojection influences cocaine taking during cocaine self-administration (SA).
To enhance the transmission efficacy, we selectively expressed a Stabilized Step Function Opsin (SSFO) in Rspo2 or Ppp1r1b BLA neurons and installed optical fibers in the NAc (Fig.   3AB). Activation of SSFO upon 473-nm laser stimulation generates a low but persistent (deactivation tau at ~29 min) conductance to cations upon 473-nm laser stimulation (13). With these features, activation of SSFO does not evoke action potentials but instead moderately depolarizes the membrane potential to facilitate presynaptic release (13). After viral-mediated expression of SSFO in the prefrontal cortex or BLA, our previous studies verify that laser stimulation (473 nm, 50 ms x 10 pulses at 10 Hz) turns on SSFO, which, persistently (>40 min) selectively facilitates presynaptic release from these projections to NAc MSNs without generating back-propagating action potentials (14,15). To further verify this SSFO approach, we prepared NAc slices containing SSFO-expressing BLA presynaptic terminals and recorded spontaneous EPSCs (sEPSCs) in MSNs. A 473-nm laser train (10 pulses at 10 Hz) of stimulation increased the frequency (F2,10 = 6.0, p = 0.02) without affecting the amplitude (F2,10 = 1.3, p = 0.33) of sEPSCs, and this effect of SSFO was terminated upon application of a 590-nm laser train (10 pulses at 10 Hz), indicating an effective and reversable presynaptic upregulation by SSFO (Fig. 3CD).
After 8 weeks of viral expression, we trained the mice with a 7-d cocaine SA procedure (0.75 mg/kg/infusion, 2 h/d, fixed ratio 1 schedule). Mice established cocaine SA over the first two days and their daily responses to active lever stabilized afterwards. On d 6, we applied intra-NAc laser stimulation bilaterally to induce enhanced presynaptic release of the Rspo2-or Ppp1r1b-to-NAc transmission and then placed the mice to the operant chamber for cocaine SA training (Fig. 3E).
With this manipulation, neither the response to active lever nor cocaine intake was altered in Rspo2 or Ppp1r1b mice on d 6, compared to the day before or after ( Fig. 3FG). Thus, upregulation of Rspo2-or Ppp1r1b-to-NAc transmission does not affect the performance of cocaine SA.
To decrease the transmission efficacy, we selectively expressed inhibitory DREADDs (hM4D) in Rspo2 or Ppp1r1b BLA neurons and infused 500 nL of CNO (3 µM) into the NAc ~15 min before the SA training on each training day ( Fig. 3H-J). As such, the activity of Rspo2 or Ppp1r1b presynaptic terminals within the NAc was locally decreased. Over 10 d of cocaine SA, both Rspo2-CNO and Ppp1r1b-CNO mice exhibited similar levels of active lever responding and cocaine intake compared to control mice, which expressed DREADDs but with daily intra-NAc infusion of 500 nL of saline ( Fig. 3K-N). Thus, downregulation of Rspo2-or Ppp1r1b-to-NAc transmission does not affect the acquisition or performance of cocaine SA.
Taken together, neither the Rspo2-nor Ppp1r1b-to-NAc subprojection seems to be essential for acquiring and maintaining the operant response for cocaine, leading our subsequent experiments to explore the role of these subprojections in cue-conditioned responses after cocaine SA was established.

Subprojection-specific regulation of cue-induced cocaine seeking
Through cocaine SA, conditioned cues acquire reinforcing properties that drive cocaine seeking.
We next examined whether the two BLA subprojections regulated cue-induced cocaine seeking after cocaine SA was established. We prepared the mice such that excitatory (hM3D) or inhibitory (hM4D) DREADDs was selectively expressed in Rspo2 versus Ppp1r1b neurons in the BLA bilaterally for subsequent chemogenetic manipulations of these two subprojections in vivo ( Fig.   4A-L).
Without chemogenetic manipulations, mice with intra-BLA expression of hM3D were trained with 10d cocaine SA to establish stable operant responding to the active lever for cocaine intake ( Fig. 4MNPQ). On withdrawal d 7, 21 and 45, mice received intra-NAc injections of CNO or saline control and were placed into operant chambers immediately. Rspo2 mice with hM4D-CNO exhibited similar levels of cue-induced cocaine seeking compared to the hM4D-saline control mice ( Fig. 4O). However, cue-induced cocaine seeking was decreased in Ppp1r1b mice with hM4D-CNO, compared to hM4D-saline controls (Fig. 4R).
These results suggest the Ppp1r1b subprojection as a key component of the BLA-to-NAc circuit that regulates the behavioral expression of cue-conditioned cocaine seeking.

Discussion
Our current study characterizes the basic circuit and behavioral properties of Rspo2-and Ppp1r1b-to-NAc subprojections in the context of cocaine seeking and relapse. The two BLA-to-NAc subprojections may be part of the circuit mechanism through which US-and CS-mediated reinforcements are differentially encoded and, thus, can be targeted for selective manipulation of drug-versus cue-induced drug relapse.

The two BLA-to-NAc subprojections
The BLA projection to NAc is critically implicated in both US-and CS-mediated reinforcement, including drug-and cue-induced drug seeking (13,14). In behaving animals, BLA neurons tend to be activated in a population-specific manner in response to US versus CS as well as stimuli with different valences, suggesting the functional and populational heterogeneity of BLA neurons (9,15,16). Based on morphological and anatomical properties, BLA pyramidal neurons can be divided into two populations, magnocellular versus parvocellular neurons that are preferentially located in the anterior versus posterior BLA, respectively (17,18). Genetic profiling reveals that these two neuronal types correspond to Rspo2-and Ppp1r1b-expressing neurons, which are differentially implicated in appetitive versus aversive affective responding (11,12). Our current results show that both Rspo2 and Ppp1r1b BLA neurons project to the NAc, forming monosynaptic connections with both D1 and D2 MSNs (Fig. 1). Although our approach cannot determine whether the same MSNs receive both Rspo2 and Ppp1r1b inputs, stimulation of Rspo2-or Ppp1r1b-to-NAc presynaptic fibers evokes EPSCs in most randomly sampled MSNs, suggesting co-innervation of individual MSNs by both BLA subprojections. If this is the case, each MSN may function as a basic NAc unit to integrate Rspo2 and Ppp1r1b inputs. Potentially through different downstream targets, D1 and D2 NAc MSNs play differential and, on many occasions, opposing roles in cue-conditioned responses (7,19). In conditioned responses, the driving force and inhibitory control are the two essential components that cooperate to ensure a smooth behavioral output. We speculate that the co-innervation of D1 and D2 MSNs by Rspo2 or Ppp1r1b provides a circuit basis on which the dynamic balance between the drive and inhibition can be achieved.
A prominent functional difference between the Rspo2-and Ppp1r1b-to-NAc synapses is their different PPRs (Fig. 1). Because both subprojections are assessed by stimulation of ChR2expressing presynaptic terminals with the same set of parameters, systemic errors are normalized. These results suggest that the presynaptic Pr of Ppp1r1b synapses is higher than Rspo2 synapses. In response to conditioned stimuli, BLA neurons fire action potentials with heterogenous patterns (20). With these synaptic features, the Rspo2-versus Ppp1r1b-to-NAc transmissions are expected to be more effective in activating NAc MSNs upon bursting versus tonic activations, respectively.

ICSS
Optogenetic stimulation of either Rspo2 or Ppp1r1b presynaptic terminals within the NAc establishes ICSS, revealing the reinforcing capacity of these two subprojections (Fig. 2).
However, recent results show that Rspo2 and Ppp1r1b BLA neurons may encode opposite valence such that they selectively express activity biomarkers in response to aversive versus appetitive stimuli, respectively (11,12). Despite the different experimental approaches, these seemingly discrepant results may reveal the projection-specific functions of Rspo2 neurons.
Specifically, results from the retrograde tracing show that, compared with the NAc, a larger portion of Rspo2 neurons send much denser projections to the central nucleus of amygdala (CeA), a brain region associated with negative behaviors (11). BLA neurons that project to the CeA versus NAc are preferentially activated by aversion-versus reward-predictive cues, and activation of BLA-to-CeA versus BLA-to-NAc promotes aversive versus appetitive behaviors (12,21). We speculate that Rspo2 BLA neurons projecting to the CeA versus NAc are largely nonoverlapping, mediating distinct behavioral responses.
Without differentiating Rspo2-or Ppp1r1b-to-NAc subprojections, previous studies show that mice establish instrumental responding for self-stimulation of the BLA-to-NAc projection as a whole (22). Taken together with our current results, activation of each of the BLA-to-NAc subprojections may function as an unconditioned reinforcer to drive the instrumental responding.
However, an interpretational caveat should be considered. Specifically, low-frequency (i.e., 5 or 10 Hz over ~5 sec) stimulations of BLA neurons do not establish ICSS (23). Similarly in our experiments, relatively low frequency (i.e., 20 Hz) optogenetic stimulation of either of the subprojections does not establish ICSS; ICSS is established only when the stimulation frequency is increased to 40 Hz (Fig. 2). Through potential backpropagation, such high-frequency stimulations of presynaptic terminals can strongly activate the soma of BLA neurons, which, in turn, activate other BLA-projected brain regions that are involved in US-mediated reinforcement.
Indeed, direct or indirect BLA projections are detected to innervate neurons in the ventral tegmental area, prefrontal cortex, hippocampus, and lateral hypothalamus, all implicated in behavioral reinforcement (8). Thus, in addition to BLA-to-NAc subprojections, other divergent projections from the BLA may also be activated to substantiate the US-mediated reinforcement during ICSS training.
It has long been known that the BLA-to-NAc projection is essential for cue-conditioned incentive processing (24)(25)(26). Our results show that stimulation of Ppp1r1b-, but not Rspo2-to-NAc, subprojection, induced a prolonged cue-conditioned responding, highlighting the Ppp1r1b subprojection in cue conditioning (Fig. 2). These results are consistent with the topographic features of the BLA-to-NAc projection. Specifically, compared to the anterior BLA, the posterior BLA is more enriched in Ppp1r1b neurons and send denser projections to the shell than the core of NAc (11,(27)(28)(29). Disconnecting the BLA and NAc shell impairs outcome-specific Pavlovian-toinstrumental transfer and the maintenance of cue-conditioned responding in the absence of reward (30)(31)(32). Thus, among the NAc sell and core projections, the Ppp1r1b-to-NAc shell branch may be particularly important for cue-conditioned reinforcement.

Acquisition and expression of cue-induced cocaine seeking
Our results show that manipulating either of the BLA subprojections does not affect the acquisition of cocaine SA (Fig. 3). This is in line with the findings that lesion of the BLA does not prevent the acquisition of instrumental responding to rewards (26,(33)(34)(35). Indeed, the BLA is also not essential for first order cue conditioning under many experimental conditions (36)(37)(38). Albeit nonessential, our ICSS results suggest that activation of the Ppp1r1b subprojection is sufficient to establish cue conditioning (Fig. 2). Thus, BLA inputs to the NAc are likely among several parallel and potentially redundant circuits that collectively contribute to cue conditioning during motivated behaviors.
After withdrawal from cocaine SA, chemogenetically enhancing versus inhibiting the Ppp1r1bto-NAc subprojection increases versus decreases cue-induced cocaine seeking, respectively (Fig.   4). Thus, instead of the acquisition (Fig. 3), the Ppp1r1b subprojection preferentially regulates the execution of cue-reinforced cocaine seeking. This is consistent with previous results that disconnection and chemo-or optogenetic inhibition of BLA-to-NAc projection compromise the performance of cue-controlled reward seeking (20,22,39), including cue-induced cocaine seeking under a second-order schedule of reinforcement (14). Our current study extends these findings by pinpointing the Ppp1r1b subprojection as the key circuit component within the BLA-to-NAc projection for cue-induced cocaine responses.
Emerging evidence suggests that the BLA-to-NAc projection is one of the primary circuit targets for drug experience, with the resulting adaptive changes promoting drug seeking. For example, after withdrawal from cue-conditioned cocaine SA, the overall strength of randomly sampled BLA-to-NAc synapses is increased, and reversing this cocaine-induced strengthening decreases cue-induced cocaine seeking (7,40,41). In these studies, the behavioral reversal is observed upon manipulation of BLA inputs to the NAc shell, where the Ppp1r1b presynaptic terminals are enriched (see discussion above). Thus, among the heterogenous BLA-to-NAc projections, the Ppp1r1b subprojection is likely the primary bearer of cocaine-induced adaptations that promote cue-induced cocaine seeking.
The BLA-to-NAc projection plays multi-faceted roles in motivated behaviors. Our current study pinpoints the differential roles of two major BLA-to-NAc subprojections in invigorating US-versus CS-mediated reinforcement. These results may provide a circuit basis through which drug-and drug-conditioned cues invigorate drug craving and seeking.

Subjects and reagents
Both Rspo2-Cre and Carpt-Cre mice with the C57B/6J background were obtained from the Tonegawa lab (11). The D1-tdTomato mouse line (strain number 016204) was purchased from the Jackson Lab (Bar Harbor, ME) and bred in house. The two Cre lines were crossed with the D1-tdTomato line to generate Rspo2-Cre-D1-tdTomato and Ppp1r1b-Cre-D1-tdTomato mice.
Male mice were used in all experiments. Mice were grouped and allocated for planed experiments at the age of 6-8 weeks, and were singly housed on a regular 12-h light/dark cycle (light on/off at 7:00/19:00), with food and water available ad libitum. After acclimation, surgery, and post-surgery recovery, mice at ~3-month-old were used to prepare brain slices for

In vivo viral injection
Mice were anesthetized with i.p. injection of ketamine (100 mg/kg)-xylazine (10 mg/kg) mixture and were held in a stereotaxic frame (Stoelting Co., Wood Dale, IL). Through a 10-μl NanoFil syringe with a 33-gauge needle controlled by UMP3 and Micro4 system (WPI, Sarasota, FL), the AAV solution (500 nl per side) was infused bilaterally at a rate of 100 nL/min into the BLA of

Optogenetic electrophysiology
We recorded optogenetically evoked EPSCs in a projection-and cell type-specific manner (42,43). Voltage-clamp whole-cell recordings were performed using an Axon MultiClamp 700B Lenses, Québec, Canada) mounted on the top of the exterior box was used to connect the splitter branching fiber-optic patch cords and the input patch cable attached to the laser system, allowing the cable-tethered mice to move freely during operant responding. To minimize the interference of laser flash with animals' behavior, ceramic sleeves connecting the ferrules and patch cord were covered with similarly sized black tubing. Scheduling of experimental events and data collection were accomplished using MED-PC IV software from Med-Associates Inc. The experiment was consisted of the following sequential phases: 1) Initial conditioning: During a 1-h session (Fixed ratio 1 schedule, FR1), mice were allowed to move freely within the operant chambers. A single press on the active lever triggered the bilateral delivery of a train of laser pulses into the NAc, concurrent with the cue light illuminating above the active lever over the duration of laser stimulation (0.25 or 0.75 sec). There was no timeout period during the training session. Pressing the inactive lever had no consequence. The criterion of ICSS was set to be a minimum of 100 active lever presses with the ratio of active/inactive lever presses ≥ 2:1 over the 1-h session. The mice were first trained for 3 sessions with a weak ICSS protocol (20 Hz, 30 times, pulse duration of 5 ms). During following sessions, the stimulation frequency was increased to 40 Hz with reduced pulse duration (1 ms; 10 mW; 10 times).
2) Extinction: After the cue-conditioned ICSS was established, mice were placed back into the operant chambers, where pressing the active lever only resulted in the presentation of cue without stimulation. Pressing the inactive lever did not result in stimulation nor cue.
3) Cue-induced ICSS seeking: After retrained with the 4 Hz ICSS protocol for 4 sessions, the mice were kept within their home cages for 21 days. The mice were then placed back into the same operant chambers, in which pressing the active lever resulted in the presentation of cue without stimulation. Pressing the inactive lever did not have any consequence.

Intravenous cocaine self-administration
The jugular vein catheterization was completed concurrently with the guide cannulae implantation 1 week before the cocaine self-administration training. Briefly, an incision was made in the neck to illuminating the house light (continuously on throughout each session). Intravenous injection was triggered by a single active lever press, enabling the mice to receive cocaine solution at a dose of 0.75 mg/kg/infusion and in a volume of 0.5 µL/g/infusion over ~2 sec, depending on the bodyweight of the subject. The infusion was accompanied by a cue light illumination over the course of infusion. To prevent cocaine overdose, there was a 10-sec timeout period between two consecutive infusions, during which pressing the active lever was recorded but did not result in cocaine infusion. Pressing the inactive lever had no programmed consequence.
For the test of cocaine seeking, mice underwent withdrawal after the 10d cocaine selfadministration procedure and were re-exposed to drug-associated cues in the same operant chambers on withdrawal d 7, 21, and 45. In the 1-h cue-induced cocaine-seeking session, both the active and inactive levers were extended. Pressing the active lever resulted in the presentation of light cues without cocaine infusion, while pressing the inactive levers resulted in nothing. The number of active lever pressing was used to assess cue-induced cocaine seeking.

Chemogenetic manipulations of BLA-to-NAc pathways
The CNO solution was freshly prepared by dissolving the compound in saline and filled in a 10-μL Hamilton syringe (22-g blunt tip) connected to the customized internal cannulae (C315DCMN/SPC, P1 Technologies) by cannula-tubing (C313CT, P1 Technologies, Roanoke, VA). To infuse CNO into the NAc, the mice were gently held in hand without anesthesia before the infusion. After the dummy cannula inserts were removed, needles were gently inserted into the cannulae. Saline or CNO (3 μM, 500 nL) was infused into NAc through the needle at 150 nL/min driven by a syringe pump (Pump 11 Elite, Harvard Apparatus). The injection needle was left in place for 5 min after injection to reduce backflow. The guide annulae were then recapped and mice were placed in operant chambers for testing.

Tissue histology and imaging
Mice were anesthetized with a ketamine/xylazine mixture and then transcardially perfused with 0.9% saline, followed by 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer (BP, pH 7.4).
After being removed from the skull and post-fixed in the same fixative overnight at 4°C, the brain tissues were embedded in 4% low melting point agarose (IBI Scientific, Dubuque, IA, USA) and sectioned coronally at room temperature using vibratome (VT1200S, Leica, Wetzlar, Germany) at a thickness of 50 µm and collected in PB. Free-floating sections were either mounted onto the slide for viewing the fluorescence signal. In some experiments, slices were processed with subsequent immunonistochemistry to enhance the signal of mCherry by antibodies: Rabbit anti-RFP (1:500, 600-401-379, Rockland) and Alexa Fluor 568, Donkey Anti-Rabbit IgG (1:500, ab175470, Abcam). In this case, the floating tissue were incubated with the primary antibody in blocking buffer (4% bovine serum albumin +0.3%Triton X-100 in BP) at 4°C overnight and washed three times with PB, followed by 2h incubation with the secondary antibody at room temperature. Images were acquired using a Leica SP5 laser-scanning confocal microscope.

Data acquisition and statistics
All results were expressed as mean ± s.e.m. Most experiments were replicated in 4-13 mice. All data collection was randomized. No statistical methods were used to predetermine sample sizes, but our sample sizes were similar to those reported in previous publications with similar experimental designs (45)(46)(47)(48). All data were analyzed offline, and investigators were not blinded to experimental conditions during the analyses. Statistical analyses were performed in GraphPad Prism (v9) and SigmaPlot (13.0). Statistical significance was assessed using two-tailed two-way mixed or three-way ANOVA followed by posttests, as specified in the related text. Differences were considered significant when the p value <0.05.