Genetically identified amygdala-striatal circuits for valence-specific behaviors.

The basolateral amygdala (BLA) plays essential roles in behaviors motivated by stimuli with either positive or negative valence, but how it processes motivationally opposing information and participates in establishing valence-specific behaviors remains unclear. Here, by targeting Fezf2-expressing neurons in the BLA, we identify and characterize two functionally distinct classes in behaving mice, the negative-valence neurons and positive-valence neurons, which innately represent aversive and rewarding stimuli, respectively, and through learning acquire predictive responses that are essential for punishment avoidance or reward seeking. Notably, these two classes of neurons receive inputs from separate sets of sensory and limbic areas, and convey punishment and reward information through projections to the nucleus accumbens and olfactory tubercle, respectively, to drive negative and positive reinforcement. Thus, valence-specific BLA neurons are wired with distinctive input-output structures, forming a circuit framework that supports the roles of the BLA in encoding, learning and executing valence-specific motivated behaviors.

Introduction

A prevailing model posits that the basolateral amygdala (BLA) contains valence-specific neurons that are intrinsically responsive to stimuli of either positive or negative valence (unconditional stimuli, USs), and can drive valence-specific behaviors once activated[1-3]. During learning, initially neutral environmental cues (conditional stimuli, CSs) acquire the capacity to selectively activate these neurons and hence the ability to induce valence-specific behavioral responses[2]. Indeed, previous studies have shown that BLA neurons activated by either appetitive or aversive stimuli, which were targeted on the basis of expression of the immediate early gene c-fos, are critical for valence-specific behaviors[1, 4]. In addition, activation of a population of BLA pyramidal neurons labeled by Rspo2 (R-spondin 2) or Cck (Cholecystokinin), two genes that are co-expressed in this population, drives aversive behaviors[5, 6], whereas activation of another BLA pyramidal population expressing Ppp1r1b (Protein phosphatase 1 regulatory inhibitor subunit 1B) drives appetitive behaviors[5]. These findings are consistent with the model and strongly suggest the existence of valence-specific neurons in the BLA.

Recent study has also started to measure the in vivo activities of presumptive valence-specific BLA neurons in behaving animals[7]. However, the moment-to-moment relationships between animal behavior and the activity of valence-specific BLA neurons, identified on the basis of either c-fos or other genetic markers, remain uncharacterized. Such characterization, especially in naturalistic settings where multiple sensory stimuli are available and different behavioral responses can be observed[8], is essential for further understanding the encoding properties of BLA neurons and how these neurons participate in learning or expressing valence-specific behavioral responses. A major obstacle to elucidating the encoding properties and in vivo functions of valence-specific neurons has been a lack of methods to reliably identify these neurons. The c-fos-based targeting strategy may recruit neurons not relevant to valence, as BLA neurons are sensitive to multiple task features, including sensory properties of individual stimuli and measures of behavioral actions[9]. In addition, this strategy necessitates engaging animals with an aversive or rewarding experience, which may by itself change neuronal properties or responsiveness. Alternatively, previous studies have manipulated BLA neurons on the basis of their projection targets, and shown that activation of BLA neurons projecting to the nucleus accumbens (BLA→NAc), for example, drives appetitive behaviors[3, 10-12]. However, recent studies demonstrate that activating a subset of BLA→NAc neurons, which express Rspo2 or Cck, drives aversive behaviors[5, 6], suggesting that BLA→NAc neurons contain functionally opposing populations. Indeed, in vivo recording reveals that BLA→NAc neurons, or BLA neurons projecting to other targets, are heterogeneous, containing neurons responsive to stimuli of either positive or negative valence[13, 14]. Therefore, projection targets per se may not be sufficient to define valence-specific neurons in the BLA.

The Forebrain expressed zinc finger 2 (Fezf2) gene encodes a highly conserved master transcription factor that plays an important role in fate specification of layer 5b corticofugal neurons[15-17]. Notably, Fezf2 is selectively expressed in the BLA within the amygdala complex and is essential for the development of BLA pyramidal neurons[18]. As the BLA and the cortex are evolutionarily and developmentally related[19], it is possible that Fezf2 also specifies a unique class of pyramidal neurons in the BLA.

In this study, we took advantage of the observation that Fezf2 selectively labels the previously reported Rspo2-expressing pyramidal neurons in the anterior basolateral nucleus (BLa) of the BLA complex, and examined how these neurons might encode valences in real time during behavior. Surprisingly, we found that these neurons do not homogeneously represent aversive stimuli as expected from previous studies[5, 6]. Rather, these neurons contain two functionally distinct classes, which separately convey punishment or reward information through projections to different ventral striatal nuclei. Thus, genetic identity and projection specificity together allowed us to identify valence-specific BLA neurons, and to uncover how these neurons contribute to the establishment of valence-specific behaviors.

Results

Fezf2 labels a unique population in the BLA

To explore the role of Fezf2-expressing (Fezf2+) neurons in the BLA, we used our newly developed Fezf2-CreER and Fezf2-Flp knock-in mouse driver lines[20], in which the inducible Cre (CreER) or the flippase (Flp) recombinase, respectively, is expressed under the control of endogenous Fezf2 promoter (see Methods). We have previously shown that Cre-recombination patterns faithfully recapitulate endogenous Fezf2 expression patterns in Fezf2-CreER mice[20]. Consistent with this finding, single molecule fluorescent in situ hybridization (smFISH) showed that the expression of Flp also recapitulated that of Fezf2 in the BLA of Fezf2-Flp mice (Extended Data Fig. 1a, b). To characterize Fezf2+ cells in the BLA, we bred mice harboring the Fezf2-CreER allele and an Ai14[21] or H2B-GFP[22] reporter allele expressing the fluorescence protein tdTomato or H2B-GFP (nuclear GFP), respectively, in a Cre-dependent manner. After tamoxifen induction in these mice (see Methods), we found dense fluorescently labeled cells in the anterior basolateral nucleus (BLa), but only sparse labeled-cells in the posterior basolateral nucleus (BLp) of the BLA complex (BLa, 93.5±0.65%, BLp, 6.5±0.65% (4 mice); note that BLa and BLp can exist in the same anteroposterior sections; Fig. 1a; Extended Data Fig. 1c; Extended Data Fig. 2). The vast majority (92±2%) of Fezf2+ cells in the BLa expressed CaMKII, a marker for excitatory pyramidal neurons[23], but nearly none of them expressed GABA, a marker for inhibitory neurons (Extended Data Fig. 1d-f). In reverse, about 60% of CaMKII+ neurons in the BLa expressed Fezf2 (Extended Data Fig. 1g), indicating that Fezf2+ cells constitute a subset of pyramidal neurons in the BLa.

Extended Data Fig. 1 ∣

Fezf2BLa neurons constitute a subset of pyramidal neurons in the BLa.

a, Confocal images of a coronal brain section containing the BLA from a Fezf2-Flp mouse, showing the distribution of Flp and Fezf2 detected by smFISH, and overlay as indicated. In the right-most panel are high magnification images of the boxed area in the overlay image. b, Quantification of the fraction of Flp+ neurons that co-express Fezf2, and vice versa (n = 3 mice). c, Confocal images of coronal brain sections from Fezf2-CreER;Ai14 mice, in which Fezf2+ neurons in the BLA are labeled with tdTomato (red). The BLa has the highest density of Fezf2+ neurons within the BLA complex. This experiment was repeated in 4 mice. d, Representative confocal images showing Fezf2BLa neurons labeled with tdTomato (left), BLa pyramidal neurons labeled by antibodies recognizing CaMKII (middle), and overlay (right). Arrows indicate Fezf2BLa cells expressing CaMKII. e, Representative images showing Fezf2BLa cells labeled with tdTomato (left), BLa inhibitory neurons labeled by antibodies recognizing GABA (middle), and overlay (right). Arrows indicate Fezf2BLa cells that do not express GABA. f, Quantification of the fraction of Fezf2BLa neurons that are pyramidal neurons or GABAergic neurons (n = 4 mice). g, Quantification of the fraction of BLa pyramidal neurons that are Fezf2+ neurons or not (n = 4 mice). Data are presented as mean ± s.e.m.

Fig. 1 ∣

Fezf2 selectively labels a unique pyramidal neuron population in the BLA.

a, Confocal images of coronal brain sections containing the BLa from a representative Fezf2-CreER;Ai14 mouse, showing the distribution of Fezf2+ neurons labeled with tdTomato, and the distribution of PPP1R1B expression detected with antibodies. In the right-most panel are high magnification images of the boxed area in the corresponding overlay image. These images are the same as those at AP −1.34 mm in Extended Data Fig. 2. This experiment was repeated in 4 mice. b, Upper: confocal images of coronal brain sections containing the BLa from a representative Fezf2-CreER;H2B-GFP mouse, showing the distribution of Fezf2+ neurons labeled with H2B-GFP, and the distribution of Rspo2 and Ppp1r1b expression detected with smFISH. Lower: high magnification images of the boxed area in the upper overlay image. These images are the same as those at AP −1.34 mm in Extended Data Fig. 3a. This experiment was repeated in 2 mice. c, Same as (b), except that the images include BLp. These images are the same as those at AP −2.06 mm in Extended Data Fig. 3a. d, Quantification of the overlap between neurons that express Fezf2, Rspo2 and Ppp1r1b detected with smFISH in the BLa (left) and BLp (right). Each value in parentheses represents the number of neurons detected with the respective probe in 2 mice. Values in matrix represent the percent labeling with probes in columns among neurons labeled with probes in rows. For example, in the BLa (left), 98% of Fezf2-labeled neurons were also labeled with Rspo2.

Extended Data Fig. 2 ∣

Fezf2BLa neurons are not PPP1R1B+ neurons.

a, Confocal images of coronal brain sections containing the BLA along the antero-posterior axis from a representative Fezf2-CreER;Ai14 mouse, showing the distribution of Fezf2+ neurons labeled with tdTomato, and the distribution of PPP1R1B expression detected with antibodies. In the rightmost panel are high magnification images of the boxed area in the corresponding overlay images. b, Quantification of the distribution of Fezf2+ neurons and PPP1R1B+ neurons in the BLa and BLp (data from 4 mice). Note that the vast majority of Fezf2+ neurons are in the BLa, whereas essentially no PPP1R1B+ neurons are in the BLa. c, Quantification of the overlap between neurons that express Fezf2 and PPP1R1B in the BLa (left) and BLp (right) in the same mice as in (b). Values in matrix represent the percent labeling with markers in columns among neurons labeled with markers in rows. For example, in the BLa (left), 0% of Fezf2 (tdTomato)-labeled neurons were labeled with PPP1R1B.

Recent studies show that activating Rspo2+ or Cck+ pyramidal neurons in the BLa, which largely belong to the same population, drives aversive behaviors[5, 6], whereas activating neurons expressing Ppp1r1b, which reside in the BLp, drives appetitive behaviors[5]. We therefore examined the relationship between these populations and Fezf2+ neurons in the BLa and BLp using immunohistochemistry and smFISH. Notably, we found that in the BLa virtually all Fezf2+ neurons were Rspo2+, and none of them were Ppp1r1b+ (Fig. 1a, b, d; Extended Data Fig. 2 & 3). In the BLp, Fezf2+ neurons were very sparse and were still mainly Rspo2+, with only a small fraction being Ppp1r1b+ (21%, Fig. 1c, d; Extended Data Fig. 2 & 3). In particular, immunohistochemistry revealed that, in the BLp, only 8% of Fezf2+ neurons expressed PPP1R1B, and only 1% of PPP1R1B+ neurons were Fezf2+ (Extended Data Fig. 2c). Considering that only 6.5% of Fezf2+ neurons are in the BLp (Extended Data Fig. 2b) and the rest are in the BLa where PPP1R1B is essentially absent, the contribution of PPP1R1B+ neurons to the Fezf2+ population in BLa and BLp combined is minimal.

Extended Data Fig. 3 ∣

Fezf2BLa neurons are predominately Rspo2+ neurons.

a, Confocal images of coronal brain sections containing the BLA along the antero-posterior axis from a representative Fezf2-CreER;H2B-GFP mouse, showing the distribution of Fezf2+ neurons labeled with H2B-GFP, and the distribution of Rspo2 and Ppp1r1b expression detected with smFISH. This experiment was repeated in 2 mice. b, High magnification images of the boxed areas in the corresponding overlay images in (a). Arrows in the left three panels indicate that, in the BLa, Fezf2+ neurons expressing Rspo2. Arrows in the rightmost panel indicate that, in the BLp, many Ppp1r1b+ neurons express Rspo2, but not Fezf2. This experiment was repeated in 2 mice.

Surprisingly, we found that the majority (~70%) of Rspo2+ neurons in the BLp were Ppp1r1b+ (Fig. 1c, d; Extended Data Fig. 3), contradicting the previous report that Rspo2+ neurons and Ppp1r1b+ neurons are distinct and spatially segregated in the BLA[5]. Our results thus indicate that, in the BLa, Fezf2 and Rspo2 label the same population of pyramidal neurons, which constitute a major population that are distinct from Ppp1r1b+ neurons. By contrast, in the BLp, Fezf2 and Rspo2 label only sparse populations that are partially overlapping, with the vast majority of Fezf2+ neurons having no PPP1R1B or Ppp1r1b expression, whereas a substantial fraction of Rspo2+ neurons expressing Ppp1r1b. Thus, Fezf2 is superior to Rspo2 in avoiding labeling (and hence the contamination by) PPP1R1B+ neurons in the BLA.

Fezf2BLa neurons respond to appetitive or aversive stimuli

As recent studies show that optogenetic activation of Rspo2+ neurons in the BLa drives innate aversive reactions[5, 6], we hypothesized that these neurons homogeneously represent negative valence. To test this hypothesis, we set out to monitor the responses of these neurons to aversive and appetitive stimuli in vivo in behaving animals. To this end, we targeted Fezf2+ neurons in the BLa (hereafter referred to as Fezf2BLa neurons) in Fezf2-CreER mice, because these neurons are essentially the same as Rspo2+ neurons in this location (see above).

We expressed the genetically encoded calcium indicator GCaMP6[24] in Fezf2BLa neurons by injecting the BLa of Fezf2-CreER mice with an adeno-associated virus (AAV) expressing GCaMP6f in a Cre-dependent manner, followed by implanting a gradient-index (GRIN) lens into the same location. Four to six weeks after the surgery, the GCaMP6 signals in infected Fezf2BLa neurons were imaged through the GRIN lens with a wide-field microscope (Fig. 2a; Supplementary Fig. 1a, b, d, e, f). These mice were kept thirsty and were presented with different stimuli, including water and air-puff blowing to the face – as the appetitive and aversive USs, respectively – as well as tones of different frequencies, under naïve, awake and head-restraint conditions (Fig. 2b, c; Supplementary Fig. 1a). Hierarchical clustering based on Fezf2BLa neuron response profiles to these stimuli (Methods) revealed two major and distinct populations, with one excited by water but not air-puff, and the other excited by air-puff but not water (Fig. 2b, c; Supplementary Video 1). We termed the former positive-valence neurons (PVNs) and the latter negative-valence neurons (NVNs). Only few Fezf2BLa neurons responded to the tones (Fig. 2h). Interestingly, PVNs and NVNs, as well as all other Fezf2BLa neurons including a small population that was excited by both water and air-puff (Fig. 2d), were intermingled with no obvious spatial clustering (Fig. 2e).

Fig. 2 ∣

Characterization of Fezf2+ and Fezf2− neuron response properties.

a, Left: the field of view of GCaMP6f signals from Fezf2+ BLa neurons through a GRIN lens. Right: the spatial locations of individual extracted neurons. The contours of neurons showing excitatory response to water, air-puff or both stimuli, and the contours of the “other” neurons showing either inhibitory response to the two stimuli or showing only spontaneous responses, were outlined and color-coded. The numbers indicate the corresponding example neurons in Supplementary video 1. The letters indicate the corresponding representative neurons in (b). b, Trial-by-trial responses of three representative Fezf2+ neurons (NVN (N), negative-valence neurons; PVN (P), positive-valence neurons; Other (O), other neurons) to different stimuli. c, Heatmaps of the activities (z-scores) in individual Fezf2+ neurons from all mice (n = 10), in trials where the following stimuli were presented (from left to right): 4k-Hz tone, 12k-Hz tone, water and air-puff. Stimulus onset was indicated by dashed line. Each row represents the temporal activities of the same neuron. Neurons were sorted by hierarchical clustering based on their response profiles. d, Pie graph showing the percentage distribution of Fezf2+ neurons excited by different USs. e, Cumulative probability distributions of centroid distances between pairs of cells from the entire population (“all”), between pairs of cells that were excited by the water reward (“water”), and between pairs of cells that were excited by the aversive air-puff (“air-puff”) in each of the 10 mice (all vs. water, P = 0.16; all vs. air-puff, P = 0.19; K-S test). Inset, the corresponding probability densities. f, Trial-by-trial responses of five representative Fezf2− neurons to different stimuli. g, Same as (c), except that activities of Fezf2− neurons (n = 5 mice) were used for analysis. h, Quantification of the proportions of Fezf2+ neurons (n = 10 mice) or Fezf2− neurons (n = 5 mice) showing excitatory responses to different stimuli (F(3, 39) = 44.16, P < 0.0001, ****P < 0.0001, two-way repeated measures ANOVA followed by Bonferroni’s multiple comparisons test). Data are presented as mean ± s.e.m.

Importantly, we targeted the viral injections to the BLa, which is devoid of PPP1R1B expression. In addition, as described above, the contribution of PPP1R1B+ neurons to the Fezf2+ population is minimal even if both the BLa and BLp are considered. Thus, the PVNs we imaged unlikely represent PPP1R1B+ neurons, which have been reported to process appetitive stimuli[5, 25]. We conclude that Fezf2BLa (or Rspo2BLa) neurons are functionally heterogeneous; they encode either positive or negative valence, rather than homogeneously represent negative valence.

Fezf2 BLa neurons are sensitive to auditory stimuli

As a comparison, we also imaged the activities of Fezf2-negative (Fezf2−) BLa pyramidal neurons, which were labeled by injecting the BLa of Fezf2-Flp mice with an AAV expressing GCaMP6 only if Flp is not present (Flp-off)[26] (Supplementary Fig. 1a, c, d, h, i). Interestingly, in contrast to Fezf2+ BLa neurons, a major fraction of Fezf2− BLa neurons showed prominent responses specifically to the auditory stimuli, whereas only small fractions of these neurons showed valence-specific responses or mixed encoding (Fig. 2f-h). Our results suggest that Fezf2+ and Fezf2− BLa neurons have distinct functional roles, with the former sensitive to valence and the latter tuned in to auditory stimuli.

CS responses evolve in Fezf2BLa neurons during learning

To test how Fezf2BLa neurons might participate in the development of behaviors motivated by positive and negative valences, we designed an ‘active approach and avoidance’ (AAA) task, in which mice learned to both actively seek reward and actively avoid punishment (Fig. 3a) (Methods). There were three types of trials – reward, punishment and neutral – in this task, which were randomly interleaved. In the reward trials, a sound (CSR) indicated that water would be delivered, but only if mice licked the water spout during a decision window following the CS. In the punishment trials, another sound (CSP) announced that an air-puff would be delivered, but mice could avoid the air-puff by running during the decision window. In the neutral trials, a third sound (CSN) indicated that nothing would happen (Fig. 3a). Early during training, mice rarely licked or ran during the decision window, and thus failed to obtain water or avoid air-puff in most of the cases; late in training, however, mice were able to acquire appropriate actions, licking during the decision window only in reward trials to obtain water, and running during the decision window specifically in punishment trials to avoid air-puff (Fig. 3b; Extended Data Fig. 4a-d; Supplementary Fig. 2).

Fig. 3 ∣

Learning induces valence-specific predictive responses in Fezf2BLa neurons.

a, A schematic of the task. b, Quantification of licking (left) and running (right) during the decision window (n = 10; licking: F-stat = 28.04, P < 0.0001; CSR, *P = 0.014; CSN, P = 0.90; CSP, P = 0.68; running: F-stat = 24, P < 0.0001; CSR, P = 0.99; CSN, P = 0.99; CSP, *P = 0.012; Friedman test with Dunn’s post-hoc test). c, Responses of 6 representative neurons during early (top) and late (bottom) training stages. d, Quantification of percent neurons selectively excited (top) or inhibited (bottom) by different CSs (n = 10; excited: F-stat = 34.04, P < 0.0001; CSR, *P = 0.018; CSN, P = 0.99; CSP, ****P < 0.0001; inhibited: F-stat = 5.02, P = 0.41; CSR, P = 0.99; CSN, P = 0.99; CSP, P = 0.99; Friedman test with Dunn’s post-hoc test). e, Venn diagrams show the percent distributions of neurons at early (upper) and late (lower) training stages. Numbers on the side indicate the percentage of CS-excited, CS-inhibited or nonresponsive (NR) neurons. f & g, Data from the late stage. f, Top: activity traces of neurons excited by only CSR (left), both CSR and USR (middle) or only USR (right). The latter two belong to PVNs. Bottom: activity traces of neurons excited by only CSP (left), both CSP and USP (middle) or only USP (right). The latter two belong to NVNs. g, Pie graphs showing the percent distributions of different types of neurons. h, Raw image projections acquired in different sessions (S1, S12, S29 and S42) from a representative mouse (top), and positioning of extracted cells in the corresponding images after alignment (bottom). Neurons detected in all sessions are indicated in green. i, Responses of cell 1 (PVN) and cell 2 (NVN), which correspond to the two cells indicated by arrows in (h), in reward and punishment trials, respectively, across different training sessions. The corresponding behavioral responses (licking or running) of the mouse are also plotted. Data are presented as mean ± s.e.m. Shaded areas around traces represent s.e.m.

Extended Data Fig. 4 ∣

Characterization of behavior and Fezf2BLa neuron activity in the AAA task.

a, Top: licking (left) and running (right) events, sorted according to trial types, for a representative mouse in the early stage of training in the AAA task. Bottom: average licking rate (left) or running velocity (right) of this mouse in different types of trials as indicated. b, Average lick rate (left) and running velocity (right) of all the mice in the different types of trials in the AAA task during early stage of training. c, d, Same as (a, b), respectively, except that mice were in the late stage of training in the AAA task. e, Heatmaps of the activities (z-scores) in individual Fezf2BLa neurons from a representative mouse, obtained from different types of trials in the early stage of training in the AAA task. Dashed lines indicate the onset of CS, decision window and US, as indicated. Each row in each type of trials (reward, neutral and punishment) represents the temporal activities of one neuron. Neurons are sorted according to their average z-scores during the presentation of different CSs. f, Left: heatmaps of the activities (z-scores) in individual Fezf2BLa neurons from a representative mouse, obtained from different types of trials in the late stage of training in the AAA task. Dashed lines indicate the onset of CS, decision window and US as indicated. Each row represents the temporal activities of the same neuron in reward (left), neutral (middle) and punishment (right) trials. Neurons are sorted according to their average z-scores during the presentation of CSR in the reward trials. Right: same dataset as that in the left, except that neurons are sorted according to their average z-scores during the presentation of CSP in the punishment trials. g, Left: heat-maps of the responses of Fezf2BLa neurons to CSR (left) and CSP (right) in the late training stage in the AAA task. Responses from all Fezf2BLa neurons in all the mice (n = 10) are shown. Each row represents the responses of the same neuron to CSR and CSP. Right: PCA was applied to the CS responses in the left, followed by hierarchical clustering using the first three principal components (PCs) to sort Fezf2BLa neurons into three clusters. h, The average responses of all neurons in each of the three clusters to different CSs. Data are presented as mean ± s.e.m. Shaded areas represent s.e.m.

Early during training in the AAA task, only a small fraction of Fezf2BLa neurons responded to any of the CSs (Fig. 3c-e; Extended Data Fig. 4e), consistent with the observation that not many Fezf2BLa neurons responded to tones in naïve mice (Fig. 2h). At this stage, reward responses in these neurons were also rare, but this was because that animals failed to obtain water in most of the reward trials (Extended Data Fig. 4a, b;). Late in training, when the mice have achieved high performance in this task, the number of Fezf2BLa neurons excited by either CSR or CSP was substantially increased, whereas that excited by CRN remained unchanged (Fig. 3c-e; Extended Data Fig. 4f). At this stage, punishment responses in these neurons were rare, because animals successfully avoided air-puff in most of the punishment trials (Extended Data Fig. 4c, d). Very few Fezf2BLa neurons were inhibited by the CSs throughout training (Fig. 3d). Notably, the neurons excited by CSR or CSP after learning were largely nonoverlapping (Fig. 3e) and could be classified into two distinct groups corresponding to NVNs and PVNs by unsupervised hierarchical clustering (Extended Data Fig. 4g, h). These results indicate that learning increases the size of two functionally distinct Fezf2BLa populations, with one excited by the CS predicting reward, whereas the other excited by the CS predicting punishment.

To determine how the CS responses in Fezf2BLa neurons evolve during learning, we first examined if the neurons excited by CS after learning were also excited by the entraining US. We identified the neurons selectively excited by either CSR or CSP at the late training stage in the AAA task, and assessed how they responded to either signaled or un-signaled delivery of reward (USR) and punishment (USP). We found that the majority of the CS-excited neurons were specifically excited by the entraining US but not the other US, and thus belonged to either PVNs or NVNs (74% of the CSR-excited neurons were excited by USR but not USP; 72% of the CSP-excited neurons were excited by USP but not USR) (Fig. 3f, g). There was also a fraction of USR- or USP-excited neurons that remained nonresponsive to CSR or CSP (the USR- or USP-only neurons; Fig. 3g).

Next, we tracked the activities of the same neurons in multiple sessions spanning the early and late training stages (session 1 (S1), S12, S29 and S42; one session per day) (Fig. 3h, i; see Methods). There was a total of 45 Fezf2BLa neurons traceable in all the four sessions. Among these neurons, four developed excitatory responses selectively to CSR and six developed excitatory responses selectively to CSP during training, with the responses being detectable from S12 and increasing in the subsequent sessions. Notably, all the tracked CSR-excited neurons showed excitatory response to USR but not USP; in a similar fashion, all the tracked CSP-excited neurons showed excitatory response to USP but not USR (Fig. 3h, i). Moreover, the valence-specific US response in these neurons was maintained from before training to late in training. Together, these results suggest that, during learning, valence-specific CS response emerges in a subpopulation of Fezf2BLa neurons showing intrinsic, valence-specific response to the entraining US.

Fezf2BLa CS response correlates and predicts behavior

To determine if the CS responses of Fezf2BLa neurons can be used to guide behavior, we assessed their relationship to valence-specific actions. We found that, first, the appearance and development of CSR or CSP responses were accompanied by the emergence and development of reward-seeking (licking) or punishment-avoidance behavior (running), respectively, in the AAA task (Fig. 3i). Second, by separately analyzing the CS responses in success trials and failure trials in the AAA task, we found that PVNs had larger CSR responses in trials where mice successfully obtained reward than trials in which mice failed to obtain the reward (Fig. 4a). In parallel, NVNs had larger CSP responses in trials where mice successfully avoided punishment than trials in which mice failed to avoid the punishment (Fig. 4b). Third, change-point analysis[27, 28] revealed that the CSR responses in most PVNs preceded reward seeking, while the CSP responses in most NVNs preceded punishment avoidance (Fig. 4c, d). These results suggest that the CS responses of these neurons can be used to guide animal’s behavior.

Fig. 4 ∣

Fezf2BLa neuron activity predicts reward seeking and punishment avoidance.

a, Left: activity traces of a representative PVN in reward trials, with average activities in success and failure trials plotted separately. Right: quantification of CSR responses of all PVNs in success and failure trials (n = 26, t(25) = 5.8, ****P = 9.4e-6; paired t-test with Bonferroni correction). b, Left: activity traces of a representative NVN in punishment trials, with average activities in success and failure trials plotted separately. Right: quantification of CSP responses of all NVNs in success and failure trials (n = 34, t(33) = 8.7, ****P = 1e-9; paired t-test with Bonferroni correction). c, Left: the timing of increase in activity in a representative PVN, and that of increase in licking response of the corresponding mouse after CSR onset, as determined by change-point analysis. Right: quantification of the difference between neural and behavioral change points, for all PVNs and the corresponding mice. Values below zero indicate that neural responses precede behavioral responses. d, Same as in (c), except that neural responses in NVNs and running behavior after CSP onset were analyzed. e, Trial-by-trail population CS responses in a representative mouse in early (left) and late (right) training stages in the AAA task. The responses are plotted in a 2D space after dimensionality reduction with principal component analysis (PCA). Each line represents one trial. Dots indicate CS onset. f, Performance of an SVM classifier, trained with population CS responses, in predicting trial types (reward, green; punishment, red; neutral, blue). g, Learning increased decoding accuracy (n = 10 mice; t(9) = −6.4, ***P = 1.3e-4; paired t-test). Data are presented as mean ± s.e.m.

To further test this idea, we performed trial-by-trial analysis on Fezf2BLa neuronal population activities after dimensionality reduction[29] (Methods). We found that, after learning, the trajectories of the dynamic population activities during CS presentation in trials of different types (reward, punishment and neutral) were clearly separable (Fig. 4e, right), suggesting that these activities predict the trial type. Indeed, the population CS responses of Fezf2BLa neurons after learning could be used to accurately decode the three trial types (Fig. 4f, g). Such features of Fezf2BLa neurons were not present before learning (Fig. 4e, g). Together, these findings point to the possibility that the learning-induced CS responses in Fezf2BLa neurons can inform the type or the valence of an incoming US, and thus guide or promote reward seeking and punishment avoidance.

Fezf2BLa neurons project to distinct ventral striatal areas

We reasoned that the function of Fezf2BLa neurons in reward seeking or punishment avoidance is mediated by their projections to downstream targets. To identify the targets, we first used an anterograde tracing strategy in which we expressed GFP selectively in Fezf2BLa neurons of Fezf2-CreER mice, and searched for areas containing GFP-labeled axons in the whole brain (Fig. 5a-c; Extended Data Fig. 5a-c). We found four major areas that contained visible labeled axons, the NAc, olfactory tubercle (OT), medial prefrontal cortex (mPFC) and lateral septum (LS), with the NAc and OT having the densest labeling (Fig. 5c; Extended Data Fig. 5a-c).

Fig. 5 ∣

Predictive valence-specific signals evolve in Fezf2BLa→NAc and Fezf2BLa→OT neurons during learning.

a, A schematic of the approach. b, A representative confocal image showing the infection of Fezf2BLa neurons with an AAV expressing GFP. This experiment was repeated in 4 mice. c, Confocal images of GFP-labeled axons originating from Fezf2BLa neurons (left), expression of substance P (SP) in the ventral pallidum recognized an antibody (middle) and an overlay image (right). This experiment was repeated in 4 mice. d, Schematics of the approach. e, Upper panel: the responses of Fezf2BLa→NAc neurons in an example mouse to different stimuli. Lower panel: the responses of Fezf2BLa→OT neurons in an example mouse to different stimuli. f, Upper panel: quantification of the responses of Fezf2BLa→NAc neurons in all mice to different stimuli (n = 10 mice; F(4, 45) = 23.96, P < 0.0001; ****P < 0.0001, n.s. (nonsignificant), P > 0.05; one-way ANOVA followed by Tukey’s multiple comparisons test). Lower panel: quantification of the responses of Fezf2BLa→OT neurons in all mice to different stimuli (n = 9 mice; F(4, 40) = 14.1, P < 0.0001; ****P < 0.0001, n.s., P > 0.05; one-way ANOVA followed by Tukey’s multiple comparisons test). g, The responses of Fezf2BLa→NAc neurons in an example mouse in the AAA task during early (left) and late (right) training stages. h, Quantification of the responses of Fezf2BLa→NAc neurons in all mice to different CSs during learning in the AAA task (n = 10 mice; Friedman test (F-stat 22.7) with Dunn’s post-hoc test: P = 0.0004; CSR, P = 0.13; CSN, P = 0.99; CSP, **P = 0.004). i, The responses of Fezf2BLa→OT neurons in an example mouse in the AAA task during early (left) and late (right) training stages. j, Quantification of the responses of Fezf2BLa→OT neurons in all mice to different CSs during learning in the AAA task (n = 9 mice; Friedman test (F-stat 25.48) with Dunn’s post-hoc test: P = 0.0001; CSR, **P = 0.0018; CSN, P = 0.56; CSP, P = 0.35). Data are presented as mean ± s.e.m.

Extended Data Fig. 5 ∣

Anterograde and retrograde tracing of Fezf2BLa neurons.

a-i, Characterization of axonal projections from Fezf2BLa neurons in target areas. a, A schematic of the approach and the sections used to quantify GFP fluorescence in the major target areas (mPFC, medial prefrontal cortex; VS, ventral striatum; LS, lateral septum). b, Upper: diagrams of the four major target areas. Lower: representative confocal images of coronal brain sections containing the corresponding target areas. Areas selected for analysis are marked by red boxes. c, Quantification of axonal fluorescence signals as the fluorescence intensity in target areas normalized by the intensity in the NAc for each animal (n = 4 mice; Friedman test (F-stat 11.1) with Dunn’s post-hoc test: P = 0.0009; n.s., P > 0.05; *P < 0.05; **P < 0.01). d, Left: a schematic of the approach to selectively label Fezf2BLa→NAc neurons. Right: a confocal image showing injection location of AAVrg-fDIO-Cre in the NAc, as indicated by fluorescent beads (arrow). e, Representative confocal images showing axons of Fezf2BLa→NAc neurons in the mPFC (left), VS (middle) and LS (right). f, Quantification of Fezf2BLa→NAc neuron axonal fluorescence signals. Data are represented as the fluorescence intensity in target areas normalized by the intensity in the NAc for each animal (n = 4 mice; Friedman test (F-stat 11.15) with Dunn’s post-hoc test: P = 0.0006; n.s., P > 0.05; *P < 0.05). g, Left: a schematic of the approach to selectively label Fezf2BLa→OT neurons. Right: a confocal image showing injection location of AAVrg-fDIO-Cre in the OT, as indicated by fluorescent beads (arrow). h, Representative confocal images showing axons of Fezf2BLa→OT neurons in the mPFC (left), VS (middle) and LS (right). i, Quantification of Fezf2BLa→OT neuron axonal fluorescence signals. Data are represented as the fluorescence intensity in target areas normalized by the intensity in the OT for each animal (n = 4 mice; Friedman test (F-stat 11.1) with Dunn’s post-hoc test: P = 0.0009; n.s., P > 0.05; *P < 0.05; **P < 0.01). j-l, Retrograde tracing of Fezf2BLa→NAc and Fezf2BLa→OT neurons with CTB. j, A schematic of the approach (left) and a histology image (right) showing CTB-647 and CTB-555 injected into the NAc and OT of Fezf2-CreER;LSL-H2B-GFP mice, respectively. k, Representative confocal images of the BLa in the same mouse as that in (j), showing Fezf2BLa neurons labelled by H2B-GFP, CTB-555 and CTB-647. In the lower panel are high magnification images of the boxed area in the image in the upper panel. Arrowheads and arrows indicate the Fezf2BLa→OT and Fezf2BLa→NAc neurons, respectively; on the upper right corner (indicated by a vertical arrow) is a Fezf2BLa neuron projecting to both the NAc and the OT (triple positive for H2B-GFP, CTB-555 and CTB-647). l, Quantification of the overlap between Fezf2BLa→OT and Fezf2BLa→NAc neurons. Data are presented as mean ± s.e.m.

As the NAc and OT are two ventral striatum structures that have been extensively implicated in reward- or punishment-driven behaviors[30, 31], we further characterized the projections to these two structures. To selectively label Fezf2BLa neurons projecting to the NAc (Fezf2BLa→NAc) or OT (Fezf2BLa→OT), we used an intersectional strategy, in which we injected the NAc or OT of Fezf2-Flp mice with a retrograde AAV (AAVrg) expressing Cre in a Flp-dependent manner (fDIO), and injected the BLa of the same mice with an AAV expressing GFP in a Cre-dependent manner (Extended Data Fig. 5d, g). We found that Fezf2BLa→NAc neurons sent stronger projections to the NAc than OT, mPFC and LS (Extended Data Fig. 5e, f), while Fezf2BLa→OT neurons sent stronger projections to the OT than NAc, mPFC and LS (Extended Data Fig. 5h, i). These results suggest that Fezf2BLa→NAc and Fezf2BLa→OT neurons preferentially project to the NAc and OT, respectively.

To assess the degree of overlap between Fezf2BLa→NAc and Fezf2BLa→OT populations, we used Fezf2-CreER;H2B-GFP mice, in which Fezf2BLa neurons were labeled with the nuclear GFP, H2B-GFP, and injected the NAc and OT in the same animals with the retrograde tracer cholera toxin subunit B (CTB) conjugated with different fluorophores (Extended Data Fig. 5j). This approach, which allows identification of Fezf2BLa→NAc and Fezf2BLa→OT neurons in the same animals, revealed that there was little overlap between these two populations (Extended Data Fig. 5k, l). To verify this result with a different type of tracer, we injected the NAc and OT of the Fezf2-CreER mice with AAVrg expressing GFP or mCherry, respectively, in a Cre-dependent manner (Extended Data Fig. 6a, b). This approach, which again allowed identifying Fezf2BLa→NAc and Fezf2BLa→OT neurons in the same animals, confirmed that there were only small portions of overlapping neurons in these two populations (Extended Data Fig. 6c).

Extended Data Fig. 6 ∣

Characterization of Fezf2BLa→OT and Fezf2BLa→NAc neurons.

a, Confocal images of coronal brain sections containing the BLa or BLp along the antero-posterior axis from a representative Fezf2-CreER mouse, showing the distribution of NAc- or OT-projecting Fezf2+ neurons. In the rightmost panel are high magnification images of the boxed area in the corresponding overlay images. b, A schematic of the approach to target OT- or NAc-projecting Fezf2+ neurons in the BLA. c, Quantification of the proportion of overlapping neurons among Fezf2BLA→NAc neurons (13 ± 3.1%) or Fezf2BLA→OT neurons (11.8 ± 3.2%; 4 mice). d, Quantification of the OT- or NAc-projecting Fezf2+ neurons in the BLa and BLp. e, Quantification of the expression of PPP1R1B in NAc- or OT-projecting Fezf2+ neurons, in the BLa and BLp (n = 4; Friedman test (F-stat 10.1) with Dunn’s post-hoc test: P = 0.003; n.s., P > 0.05). Data are presented as mean ± s.e.m.

The retrograde tracing with AAVrg, combined with immunohistochemistry, also revealed that the vast majority of either Fezf2BLa→NAc (93±2.5%) or Fezf2BLa→OT (92±2.4%) neurons were localized to the BLa, and none of these neurons expressed PPP1R1B; moreover, even among the 7-8% of either Fezf2BLa→NAc or Fezf2BLa→OT neurons localized to the BLp, only 10% expressed PPP1R1B (Extended Data Fig. 6d, e). These results thus confirm that the vast majority of BLA Fezf2+ neurons are located in the BLa and do not express PPP1R1B, irrespective of their projection targets.

Next, we examined the functional connectivity between Fezf2BLa neurons and neurons in the NAc or OT (Extended Data Fig. 7). Using the intersectional strategy described above (Extended Data Fig. 5d, g), we expressed the light-gated cation channel channelrhodopsin (ChR2) selectively in Fezf2BLa→NAc or Fezf2BLa→OT neurons (Extended Data Fig. 7a, f). We prepared acute brain slices from these mice, in which we recorded monosynaptic responses (in the presence of TTX and 4-AP[32]) in randomly chosen NAc or OT neurons to photoactivation of the axons originating from Fezf2BLa→NAc or Fezf2BLa→OT neurons (Extended Data Fig. 7b, g). The majority of NAc or OT neurons showed detectable light-evoked synaptic responses, suggesting high levels of functional connectivity (Extended Data Fig. 7c, d & h, i). Notably, photoactivation of Fezf2BLa→NAc axons drove much larger synaptic responses in NAc neurons than OT neurons (Extended Data Fig. 7c, e). By contrast, photoactivation of Fezf2BLa→OT axons drove responses of equivalent amplitude in NAc or OT neurons (Extended Data Fig. 7h, j).

Extended Data Fig. 7 ∣

Functional connectivity between Fezf2BLa neurons and ventral striatum neurons.

a, A schematic of the intersectional approach to selectively target Fezf2BLa→NAc neurons for optogenetic activation. b, A schematic of the approach to record the synaptic responses of NAc and OT neurons to optogenetic activation of the axons originating from Fezf2BLa→NAc neurons. c, Traces of excitatory postsynaptic currents (EPSCs) recorded from two representative neurons, with one in the NAc (top) and the other in the OT (bottom), in the presence of TTX and 4-AP. The EPSCs were evoked by optogenetic activation of axon fibers originating from Fezf2BLa→NAc neurons. The upward square pulses (1 ms duration) in the blue traces (top) indicate the timing of photo-stimulation. d. Quantification of connection probability between Fezf2BLa→NAc neurons and neurons in the NAc or OT (36 out of 39 cells in the NAc, and 30 out of 36 cells in the OT had evoked EPSCs). e, Quantification of peak amplitudes of the EPSCs (NAc, 36 cells, OT, 30 cells; 6 mice; t(64) = 3.265, **P = 0.0018, unpaired t test). f-h, Same as a-c, respectively, except that Fezf2BLa→OT neurons were selectively targeted for optogenetic activation. i. Quantification of connection probability between Fezf2BLa→OT neurons and neurons in the NAc or OT (18 out of 20 cells in the NAc, and 17 out of 19 cells in the OT had evoked EPSCs). j, Quantification of peak amplitudes of the EPSCs (NAc, 18 cells, OT, 17 cells; 4 mice; n.s., t(33) = 0.2793, P = 0.78, unpaired t test). Data are presented as mean ± s.e.m.

Together, the anterograde and retrograde tracing results indicate that Fezf2BLa→NAc and Fezf2BLa→OT neurons represent two largely nonoverlapping populations that preferentially project to the NAc and OT, respectively. The functional circuit mapping results provide further support that Fezf2BLa→NAc neurons preferentially innervate NAc neurons (Extended Data Fig. 5d-f), but also reveal that Fezf2BLa→OT neurons drive synaptic responses of similar amplitude in OT and NAc neurons. The latter finding suggests that even sparse projections in the NAc can have functional impact, considering that only a small portion of Fezf2BLa→OT neurons send collaterals to the NAc (Extended Data Fig. 5j-l; Extended Data Fig. 6a-c) and the collaterals are relatively sparse (Extended Data Fig. 5g-i).

Fezf2BLa→NAc and Fezf2BLa→OT represent opposing valences

The capacity of retrograde tracing in labeling largely nonoverlapping Fezf2BLa→NAc and Fezf2BLa→OT neurons (Extended Data Fig. 5j-l; Extended Data Fig. 6a-c) allowed us to investigate whether these two populations are differentially involved in reward seeking and punishment avoidance. We first monitored the responses of each of these populations to appetitive and aversive stimuli. For this purpose, we expressed GCaMP6 selectively in Fezf2BLa→NAc or Fezf2BLa→OT neurons by injecting the NAc or OT, respectively, of Fezf2-CreER mice with an AAVrg expressing GCaMP6 in a Cre-dependent manner. This virus had relatively low expression levels of GCaMP6, resulting in dim fluorescence signals that precluded us from resolving single neurons with GRIN lens-based imaging. To address this issue, we used fiber photometry[33, 34] to record bulk GCaMP6 signals from the infected Fezf2BLa→NAc or Fezf2BLa→OT neurons through optical fibers implanted in the BLa in these mice (Fig. 5d; Supplementary Fig. 3). Notably, we found that Fezf2BLa→NAc neurons were activated by aversive stimuli, including air-puff and tail-shocks, but were nonresponsive to appetitive stimuli, such as water, fat or sucrose solution (Fig. 5e, f; top). By contrast, Fezf2BLa→OT neurons were excited by the appetitive stimuli but not the aversive stimuli (Fig. 5e, f; bottom). These results suggest that Fezf2BLa→NAc and Fezf2BLa→OT neurons represent negative or positive valence, respectively, and convey the respective valence information to the NAc and OT.

Next, we trained these mice in the AAA task as above (Fig. 3a), and recorded the activities of Fezf2BLa→NAc or Fezf2BLa→OT neurons at both early and late training stages with fiber photometry. We found that at the early stage, when the mice’s performance was poor, neither populations showed obvious responses to CSR and CSP (Fig. 5g-j). By contrast, at the late stage when mice had achieved high performance, Fezf2BLa→NAc neurons were excited by CSP but not CSR, whereas Fezf2BLa→OT neurons were excited by CSR but not CSP. Neither Fezf2BLa→NAc nor Fezf2BLa→OT neurons showed response to CSN (Fig. 5g-j). These results are consistent with the observation from GRIN lens imaging that valence-specific predictive responses evolve in the NVNs and PVNs, and moreover map the responses onto Fezf2BLa→NAc and Fezf2BLa→OT neurons.

Fezf2BLa→NAc and Fezf2BLa→OT have distinct synaptic inputs

Because Fezf2BLa→NAc and Fezf2BLa→OT neurons are strikingly different in their responses to stimuli of positive or negative valence, it is likely that these two populations are innervated by different synaptic inputs. To test this possibility, we mapped brain-wide monosynaptic inputs onto each of these populations (Fig. 6a-i; Supplementary Fig. 4 & 5). To achieve this goal, we first injected the NAc or OT of Fezf2-CreER mice with a retrograde virus CAV-Flex-Flp[35] (Fig. 6a, e), which allowed the Flp to be specifically expressed in Fezf2BLa→NAc or Fezf2BLa→OT neurons, respectively, following tamoxifen induction. We then infected these neurons with a Flp-dependent rabies virus system[34, 36-39] (Fig. 6a, e; Supplementary Fig. 4a-h; and see Methods).

Fig. 6 ∣

Fezf2BLa→NAc and Fezf2BLa→OT neurons receive distinct sets of monosynaptic inputs.

a, A schematic of the approach. b, A confocal image showing injection location in the NAc indicated by fluorescent beads (arrow). c, Confocal images of the BLa showing the Fezf2BLa→NAc neurons infected with the starter virus (left) and the rabies virus (middle), and the colocalization (right). d, Images showing the major areas labeled by the rabies virus. e, A schematic of the approach. f, A confocal image showing injection location indicated by fluorescent beads (arrow). g, Confocal images of the BLa showing the Fezf2BLa→OT neurons infected with the starter virus (left) and the rabies virus (middle), and the colocalization (right). h, Images showing the major areas labeled by the rabies virus. i, Whole-brain quantification of rabies labeled areas (n = 8 mice for each pathway; Friedman test with Dunn’s post-hoc test: F-stat = 304.3, P < 0.0001; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; P values between Fezf2BLa→NAc and Fezf2BLa→OT: PL, 0.2458; IL, 0.0932; Mop, 0.1805; SI, 0.5438; PIR, <0.0001; CoA, 0.0029; ACC, 0.6361; PVT: 0.0065; MD, 0.5320; CM, 0.3719; RE: 0.0549; BL, 0.9431; LA, 0.3173; CeA, 0.1291; ENT, 0.2421; AuD, 0.1080; vHC, 0.8723; MGv, <0.0001; MGd, 0.0727; MGm, 0.7210; PIL, 0.1268; PR, 0.0055; CLI, 0.0038; MRN, 0.0031). Data are presented as mean ± s.e.m.

PL, prelimbic area; IL, infralimbic area; MOp, primary motor area; SI, substantia innominata; PIR, piriform area; CoA, cortical amygdalar area; ACC, anterior cingulate cortex; PVT, paraventricular nucleus of the thalamus; MD, mediodorsal nucleus of the thalamus; CM, central medial nucleus of the thalamus; RE, nucleus of reuniens; BL, basolateral amygdalar nucleus; LA, lateral amygdalar nucleus; CeA, central amygdalar nucleus; ENT, entorhinal area; AuD, auditory areas; vHC, ventral hippocampus; MGv, medial geniculate complex, ventral part; MGd, medial geniculate complex, dorsal part; MGm, medial geniculate complex, medial part; PIL, posterior intralaminar thalamic nucleus; DR, dorsal nucleus raphe; CLI, central linear nucleus raphe; MRN, midbrain reticular nucleus.

This approach revealed marked differences between the inputs onto Fezf2BLa→NAc neurons and those onto Fezf2BLa→OT neurons (Fig. 6i; Supplementary Fig. 4i, j; Supplementary Fig. 5). In particular, Fezf2BLa→NAc neurons receive inputs from the paraventricular nucleus of the thalamus (PVT), the cortical amygdala (CoA) and dorsal raphe (DR) (Fig. 6d, i). These areas send only weak or no output to innervate Fezf2BLa→OT neurons (Fig. 6h, i). By contrast, Fezf2BLa→OT neurons receive substantial inputs from the piriform cortex (PIR) and the ventral part of the medial geniculate complex (MGv) (Fig. 6h, i), which send no or weak output to innervate Fezf2BLa→NAc neurons (Fig. 6d, i). There were several other areas (e.g., posterior intralaminar nucleus of the thalamus (PIL) and entorhinal area (ENT)) showing differential innervation of Fezf2BLa→NAc versus Fezf2BLa→OT neurons, although the differences did not reach significance when corrected for multiple comparisons (Fig. 6i). Thus, Fezf2BLa→NAc and Fezf2BLa→OT neurons not only preferentially project to different targets, but also receive inputs from distinct sets of sensory and limbic structures that may determine, at least in part, their distinct response properties and functions.

Fezf2BLa activity is required for approach and avoidance

To determine if the activity of Fezf2BLa neurons is causally linked to reward seeking or punishment avoidance, we began with inhibiting these neurons – irrespective of their projection targets – in mice that have learned the AAA task (Extended Data Fig. 8a-k). To achieve selective and reversible inhibition of Fezf2BLa neurons, we expressed in these neurons an inhibitory DREADD (Designer Receptor Exclusively Activated by Designer Drug) by bilaterally injecting the BLa of Fezf2-CreER mice with an AAV expressing KORD (a DREADD derived from the kappa-opioid receptor)[40] in a Cre-dependent manner (Extended Data Fig. 8a, b). These mice were first trained in the AAA task (Extended Data Fig. 8c), and subsequently tested in a session in which their Fezf2BLa neurons were inhibited by systemic application of salvinorin B (SALB), the agonist of KORD[40]. Remarkably, inhibition of Fezf2BLa neurons markedly and reversibly reduced the licking responses induced by CSR, but did not affect the subsequent licking triggered by USR in the same trial (Extended Data Fig. 8d, f, g). In a similar manner, the inhibition markedly and reversibly reduced the running responses induced by CSP, but did not affect the subsequent running triggered by USP in the same trial (Extended Data Fig. 8e, i, j). Control experiments confirmed that vehicle treatment, or treating animals with SALB alone did not affect animals’ behavior in the AAA task (Extended Data Fig. 8d-k).

Extended Data Fig. 8 ∣

Chemogenetic inhibition of Fezf2BLa neurons impairs both reward seeking and punishment avoidance.

a, A schematic of the approach. b, Representative confocal images showing the BLa neurons expressing KORD. c, A schematic of the AAA task. d, Top: licking events sorted according to trial types (which were randomly interleaved) for a representative mouse before (left), during (middle) and after (right) chemogenetic inhibition of Fezf2BLa neurons. Bottom: the average rate of the licking responses in the corresponding top panels. e, Top: running events sorted according to trial types (which were randomly interleaved) for a representative mouse before (left), during (middle) and after (right) chemogenetic inhibition of Fezf2BLa neurons. Bottom: the average velocity of the running responses in the corresponding top panels. f, Chemogenetic inhibition of Fezf2BLa neurons reduced the licking responses during the decision window (n = 7 mice; F(2, 18) = 5.13, P = 0.01; Pre-SALB and SALB, *P = 0.0211; SALB and DMSO, *P = 0.0299; Pre-SALB and DMSO, n.s. (nonsignificant), P = 0.97; one-way ANOVA followed by Tukey’s multiple comparisons test). g, Chemogenetic inhibition of Fezf2BLa neurons did not affect the licking responses evoked by water (US) delivery to the mouth (n = 7; t(6) = 1.4, P = 0.22; paired t-test). h, Treatment with SALB does not change the licking responses during the decision window in control mice in which the Fezf2BLa neurons expressed GFP (n = 6; t(5) = −0.33, P = 0.75; paired t-test). i, Chemogenetic inhibition of Fezf2BLa neurons reduced the running responses during the decision window (n = 7 mice; F(2, 18) = 3.95, P = 0.03; Pre-SALB and SALB, *P = 0.0386; SALB and DMSO, *P = 0.0048; Pre-SALB and DMSO, n.s. (nonsignificant), P = 0.6742; one-way ANOVA followed by Tukey’s multiple comparisons test). j, Chemogenetic inhibition of Fezf2BLa neurons did not affect the running responses evoked by air-puff (US) blowing to the face (n = 7; t(6) = 0.95, P = 0.38; paired t-test). k, Treatment with SALB does not change the running responses during the decision window in control mice in which the Fezf2BLa neurons expressed GFP (n = 6; t(5) = −0.02, P = 0.98; paired t-test). Data are presented as mean ± s.e.m.

These results indicate that Fezf2BLa neurons are essential for anticipating, or motivating actions to pursue the goal of obtaining reward or avoiding punishment, but are dispensable for generating the specific movements (licking and running) per se that let animals reach such goals.

Fezf2BLa→NAc and Fezf2BLa→OT have essential behavioral roles

To determine whether Fezf2BLa→NAc and Fezf2BLa→OT neurons are differentially required for learning or expressing valence-specific behaviors, we sought to selectively and transiently inhibit each of these populations during and after learning in the AAA task. To achieve this goal, we expressed the light sensitive Guillardia theta anion-conducting channelrhodopsin 2 (GtACR2)[41, 42] selectively in Fezf2BLa→NAc or Fezf2BLa→OT neurons using the intersectional viral strategy described above (Extended Data Fig. 5d-i; Extended Data Fig. 7). Specifically, we bilaterally injected the NAc or OT of Fezf2-Flp mice with the AAVrg expressing Cre in a Flp-dependent manner, and injected the BLa of the same mice with an AAV expressing GtACR2, or GFP (as a control), in a Cre-dependent manner (Extended Data Fig. 9a-d). Optical fibers were implanted in the BLa above the infected areas for photostimulation (Supplementary Fig. 6a, b). These mice were trained in the AAA task and received photostimulation in the BLa in a time window covering CS and US presentations in each trial throughout the training (Extended Data Fig. 9e, f). The mice were subsequently given additional training in the absence of photostimulation (to allow the GtACR2 and GFP groups to reach equal performance), followed by a test session (“after learning”) in which the photostimulation was delivered in a subset of randomly chosen trials (Extended Data Fig. 9f; Methods).

Extended Data Fig. 9 ∣

Optogenetic manipulation of Fezf2BLa→NAc and Fezf2BLa→OT neurons.

a-f, Optogenetic inhibition of Fezf2BLa→NAc and Fezf2BLa→OT neurons. a, A schematic of the approach to selectively inhibit Fezf2BLa→NAc neurons with optogenetics. b, Left: a confocal image showing Fezf2BLa→NAc neurons expressing stGtACR2. Locations of optical fibers for optogenetics are indicated. Right: a confocal image showing injection location of AAVrg-fDIO-Cre in the NAc, as indicated by fluorescent beads (arrow). c, A schematic of the approach to selectively inhibit Fezf2BLa→OT neurons with optogenetics. d, Left: a confocal image showing Fezf2BLa→OT neurons expressing stGtACR2. Locations of optical fibers for optogenetics are indicated. Right: a confocal image showing injection location of AAVrg-fDIO-Cre in the OT, as indicated by fluorescent beads (arrow). e, f, Schematics of the AAA task (e; blue bar on time axis indicates laser delivery) and experimental procedure (f). g-p, Optogenetic activation of Fezf2BLa→NAc and Fezf2BLa→OT neurons increases pupil size. g, A schematic of the approach to selectively activate Fezf2BLa→NAc neurons with optogenetics. h, Left: a confocal image showing Fezf2BLa→NAc neurons expressing ChR2. Locations of optical fibers for optogenetics are indicated. Right: a confocal image showing injection location of AAVrg-fDIO-Cre in the NAc, as indicated by fluorescent beads (arrow). i, A schematic of the approach to selectively activate Fezf2BLa→OT neurons with optogenetics. j, Left: a confocal image showing Fezf2BLa→OT neurons expressing ChR2. Locations of optical fibers for optogenetics are indicated. Right: confocal image showing injection location of AAVrg-fDIO-Cre in the OT, as indicated by fluorescent beads (arrow). k, Images of the pupil in a representative mouse, before (left) and after (right) photoactivation of Fezf2BLa→NAc neurons. l, Trial-by-trial pupil size changes in response to photoactivation (blue bar, 2 s) of Fezf2BLa→NAc neurons in an example mouse. m, Quantification of pupil size change in response to laser stimulation in the BLa in all the mice in which ChR2 (n = 9) or GFP (n = 7) was expressed in Fezf2BLa→NAc neurons (Kruskal-Wallis test (K-stat 18.88) with Dunn’s post-hoc test: P = 0.0003; ChR2: **P = 0.002; GFP: P = 0.99). n, o, same as (k, l), respectively, except that Fezf2BLa→OT neurons were photoactivated. p, Quantification of pupil size change in response to laser stimulation in the BLa in all the mice in which ChR2 (n = 10) or GFP (n = 7) was expressed in Fezf2BLa→OT neurons (Kruskal-Wallis test (K-stat 21.9) with Dunn’s post-hoc test: P < 0.0001; ChR2: ****P < 0.0001; GFP: P = 0.99).

We found that, in mice where Fezf2BLa→NAc neurons were targeted, the photostimulation during learning caused a decrease in performance in punishment trials, but not in reward trials in the GtACR2 group compared with the GFP group (Fig. 7a, b). By contrast, in mice where Fezf2BLa→OT neurons were targeted, the photostimulation during learning caused a decrease in performance in reward trials, but not in punishment trials in the GtACR2 group compared with the GFP group (Fig. 7c, d). Furthermore, in the “after learning” test session, photostimulation of Fezf2BLa→NAc neurons in the GtACR2 group caused a decrease in anticipatory running in punishment trials, but did not affect anticipatory licking in reward trials (Fig. 7e, g), whereas photostimulation of Fezf2BLa→OT neurons in the GtACR2 group caused a decrease in anticipatory licking in reward trials, but did not affect anticipatory running in punishment trials (Fig. 7f, h). The photostimulation did not affect the behavioral responses in the GFP groups (Fig. 7e-h).

Fig. 7 ∣

Fezf2BLa→NAc and Fezf2BLa→OT are differentially required for approach or avoidance.

a-d, During learning. a, Performance of mice having Fezf2BLa→NAc neurons infected (GtACR2, n = 11; GFP, n = 9), in reward trials with laser (left) (F(1, 378) = 3.34, P = 0.069), and without laser (right) (F(1, 54) = 0.5765, P = 0.451). b, Performance of the mice in (a) in punishment trials with laser (left) (F(1, 378) = 63.08, P < 0.0001; *P = 0.0126, **P = 0.0071), and without laser (right) (F(1, 54) = 1.024, P = 0.316). c, d, Same as (a, b), except that Fezf2BLa→OT neurons were infected (GtACR2, n = 9; GFP, n = 8). c, With laser (left), F(1, 315) = 65.11, P < 0.0001; *P = 0.013, **P = 0.009. Without laser (right), F(1, 45) = 0.5992, P = 0.443. d, With laser (left), F(1, 315) = 2.863, P = 0.092. Without laser (right), F(1, 48) = 0.5015, P = 0.4823. e-h, After learning. e, f, Mice having Fezf2BLa→NAc neurons infected (stGtACR2, n = 11; GFP, n = 9). e, Left: raster (upper) and average (lower) plots of licking behavior of a mouse in reward trials. Laser stimulation period is indicated by a blue box. Right: lick rate during decision window (F(1, 18) = 0.3164, P = 0.58; stGtACR2, P = 0.3530; GFP, P = 0.8796). f, Left: raster (upper) and average (lower) plots of running behavior of a mouse in punishment trials. Laser stimulation period is indicated by a blue box. Right: running velocity during decision window (F(1, 18) = 11.43, P = 0.0033; stGtACR2, **P = 0.0013; GFP, P = 0.661;). g, h, Same as (e, f), except that Fezf2BLa→OT neurons were infected (stGtACR2, n = 9; GFP, n = 8). g, F(1, 15) = 7.287, P = 0.0165; stGtACR2, **P = 0.0015; GFP, P = 0.9562. h, F(1, 15) = 3.63, P = 0.076; stGtACR2, P = 0.5779; GFP, P = 0.2042. Statistics were from two-way ANOVA with repeated measures followed by Bonferroni’s multiple comparisons test. Data are presented as mean ± s.e.m.

Together, these results indicate that Fezf2BLa→NAc neurons are selectively required for learning and expression of punishment-avoidance actions, while the Fezf2BLa→OT neurons are selectively required for learning and expression of reward-seeking responses.

Fezf2BLa→NAc and Fezf2BLa→OT differently instruct learning

To further understand how the signals from Fezf2BLa→NAc and Fezf2BLa→OT neurons might contribute to valence-specific behaviors, we tested the effects of activating these neurons with optogenetics. For this purpose, we again used the intersectional strategy to express ChR2, or GFP (as a control), selectively in Fezf2BLa→NAc or Fezf2BLa→OT neurons, followed by optical fiber implantation in the BLa for photostimulation (Extended Data Fig. 9g-j; Supplementary Fig. 6c-i). We found that brief photostimulation of either Fezf2BLa→NAc (Extended Data Fig. 9k-m) or Fezf2BLa→OT (Extended Data Fig. 9n-p) neurons in the ChR2 mice, but not GFP mice, reliably and drastically induced pupil enlargement. This result suggests that activities in either population promote affective or motivational arousal, a crucial factor determining learning and performance in motivated behaviors[43-45].

To probe if the arousal is related to perception of positive or negative valence, we tested these mice in a real-time place preference or aversion (RTPP/RTPA) test, in which the photo-stimulation was contingent on entering one side of a chamber containing two compartments. In stark contrast, stimulating Fezf2BLa→NAc neurons caused avoidance, whereas stimulating Fezf2BLa→OT neurons induced preference to the side paired with the stimulation (Fig. 8a-d). These results suggest that the activities in Fezf2BLa→NAc and Fezf2BLa→OT neurons promote arousal by signaling negative and positive valence, respectively.

Fig. 8 ∣

Fezf2BLa→NAc and Fezf2BLa→OT neurons differentially instruct learning.

a, Movement trajectory of a mouse at baseline (left), or in a situation whereby entering the left (middle) or right (right) side of the chamber triggered photoactivation of Fezf2BLa→NAc neurons. b, Quantification of behavior as shown in (a), for mice in which ChR2 (n = 9) or GFP (n = 7) was expressed in Fezf2BLa→NAc neurons (F(1, 28) = 202.5, P < 0.0001; ****P < 0.0001, n.s., P > 0.05; two-way ANOVA followed by Bonferroni’s multiple comparisons test). c, d, Same as (a, b), respectively, except that Fezf2BLa→OT neurons were targeted (ChR2, n = 10; GFP, n = 7; F(1, 30) = 73.05, P < 0.0001; ****P < 0.0001, n.s., P > 0.05; two-way ANOVA followed by Bonferroni’s multiple comparisons test). e, A schematic of the approach. f, Raster plot of nose-poking events at active (with laser) or inactive (without laser) port, for a mouse in which ChR2 was expressed in Fezf2BLa→NAc neurons. g, Quantification of nose-poking events in a 60-min session for mice in which ChR2 (n = 9) or GFP (n = 7) was expressed in Fezf2BLa→NAc neurons (F(1, 28) = 1.09, P = 0.3; n.s., P > 0.05; two-way ANOVA). h, Quantification of nose-poke events of the ChR2 mice in (g), in the first 2 mins of the reversal test, in which the active and inactive ports were switched (P > 0.99; Mann-Whitney test, U = 38). i, Same as (f), except that ChR2 was expressed in Fezf2BLa→OT neurons. j, Same as (g), except that Fezf2BLa→OT neurons were targeted (ChR2, n = 10; GFP, n = 7; F(1, 30) = 70.53, P < 0.0001; ****P < 0.0001, n.s., P > 0.05; two-way ANOVA followed by Bonferroni’s multiple comparisons test). k, Same as h, except that ChR2 was expressed in Fezf2BLa→OT neurons (***P = 0.0001; Mann-Whitney test, U = 4). l, A schematic of the approach. m, Top: trial-by-trial heat-maps of running velocity for an example mouse in the active avoidance task; bottom: average running velocity of the mouse in early (first 10 trials) and late (last 10 trials) trials. n, Quantification of running velocity during the decision window in early (first 10) and late (last 10) trials (ChR2 mice, n = 9, GFP mice, n = 7; F(1, 28) = 42.3, P < 0.0001; ****P < 0.0001, n.s., P > 0.05; two-way ANOVA followed by Bonferroni’s multiple comparisons test). o, p, Same as (m, n), respectively, except that Fezf2BLa→OT neurons were targeted (ChR2 mice, n = 10, GFP mice, n = 7; F(1, 30) = 0.24, P = 0.63; n.s., P > 0.05; two-way ANOVA). Data are presented as mean ± s.e.m.

If Fezf2BLa→NAc and Fezf2BLa→OT neurons are crucial for representing and conveying negative and positive valence information, respectively, activation of these neurons should differentially instruct valence-specific learning. To test this hypothesis, we trained the mice in which Fezf2BLa→NAc or Fezf2BLa→OT neurons were targeted in a two-port choice task, where poking one port (the active port) resulted in photo-stimulation of Fezf2BLa→NAc or Fezf2BLa→OT neurons, whereas poking the other (the inactive port) resulted in nothing (Fig. 8e). The mice in which Fezf2BLa→NAc neurons were targeted (including the ChR2 and GFP groups) did not make much effort to poke into either the active or the inactive port (Fig. 8f-h). However, the mice in the ChR2 group in which Fezf2BLa→OT neurons were targeted quickly learned to poke the active port while ignoring the inactive one, vigorously performing self-stimulation (Fig. 8i, j). Moreover, in a reversal test conducted 24 hours later, the mice persisted in choosing the initial active port even if the port had become inactive (Fig. 8k).

Next, we trained the mice in an optically-driven active avoidance task. We took advantage of the design for active avoidance in the AAA task (Fig. 3a), but substituted photostimulation of Fezf2BLa→NAc or Fezf2BLa→OT neurons for air-puff as a potential negative reinforcer (Fig. 8l). Remarkably, photostimulating Fezf2BLa→NAc neurons in the ChR2 group was sufficient to instruct learning, such that the mice acquired running response to CS and thus successfully avoided the photostimulation, in a manner similar to avoiding the air-puff in the AAA task (Fig. 8m, n). Photostimulation of Fezf2BLa→OT neurons failed to support such learning (Fig. 8o, p). Control experiments showed that photostimulation in the GFP groups did not affect animals’ behavior in all the tests performed (Fig. 8; Extended Data Fig. 9k-p). Importantly, essentially no PPP1R1B+ neurons could be found in the BLa locations where we targeted the Fezf2+ neurons (Supplementary Fig. 6c-h).

In a separate set of experiments, we repeated all the above-described photostimulation experiments based on ChR2, with the exception that the photostimulation was delivered to the NAc or OT, to stimulate the axon terminals originating from Fezf2BLa neurons (Extended Data Fig. 10; Supplementary Fig. 7). The results from these experiments, which targeted Fezf2BLa→NAc and Fezf2BLa→OT axon terminals, fully recapitulated those from the experiments described above that targeted the cell bodies of the respective neuronal populations (Extended Data Fig. 9g-p; Fig. 8; Supplementary Fig. 6c-i).

Extended Data Fig. 10 ∣

Fezf2BLa→NAc and Fezf2BLa→OT projections differentially instruct valence-specific learning.

a, A schematic of the approach to stimulate the Fezf2BLa→NAc pathway. b, Confocal images of the BLa (left) and ventral striatum (right) from a representative mouse, showing Fezf2BLa neurons expressing ChR2 and ChR2+ axon fibers originating from Fezf2BLa neurons, respectively. The placement of optical fiber in the NAc is also shown. Antibodies recognizing SP were used to label the ventral pallidum, which lies in between the NAc and OT. c, d, Same as a, b, respectively, except that the Fezf2BLa→OT pathway was the target for optogenetic stimulation (c), and the optical fiber was placed over the OT (d). e, Images of the pupil in a representative mouse, before (left) and after (right) photoactivation of the Fezf2BLa→NAc pathway. f, Trial-by-trial pupil size changes in response to photoactivation (blue bar, 2 s) of the Fezf2BLa→NAc pathway in an example mouse. g, Quantification of pupil size change in response to laser stimulation in the NAc in all the mice in which ChR2 (n = 8) or GFP (n = 7) was expressed in Fezf2BLa→NAc neurons (F(1, 26) = 32.52, P < 0.0001; ****P < 0.0001, n.s., P = 0.9999; two-way ANOVA followed by Bonferroni’s multiple comparisons test). h, i, same as e, f, respectively, except that the Fezf2BLa→OT pathway was photoactivated. j, Quantification of pupil size change in response to laser stimulation in the OT in all the mice in which ChR2 (n = 8) or GFP (n = 7) was expressed in Fezf2BLa→OT neurons (F(1, 26) = 30.34, P < 0.0001; ***P < 0.0001, n.s., P = 0.9999; two-way ANOVA followed by Bonferroni’s multiple comparisons test). k, Movement trajectory of a representative mouse at baseline (left), or in a situation whereby entering the left (middle) or right (right) side of the chamber triggered photoactivation of the Fezf2BLa→NAc pathway. l, Quantification of mouse activity as shown in (k), for mice in which ChR2 (n = 8) or GFP (n = 7) was expressed in Fezf2BLa neurons. The ChR2 mice, but not the GFP mice, avoided the side associated with the photo-stimulation (F(1, 26) = 89.75, P < 0.0001; ****P < 0.0001, n.s., P > 0.05; two-way ANOVA followed by Bonferroni’s multiple comparisons test). m, n, Same as k, l, respectively, except that the Fezf2BLa→OT pathway was targeted. The ChR2 mice (n = 8), but not the GFP mice (n = 7), preferred the side associated with the photo-stimulation (F(1, 26) = 63.28, P < 0.0001; ****P < 0.0001, n.s., P > 0.05; two-way ANOVA followed by Bonferroni’s multiple comparisons test). o, A schematic of the approach. p, Raster plot of nose-poking events at active or inactive port, for a mouse in which ChR2 was expressed in Fezf2BLa neurons and photo-stimulation was delivered to the NAc at the active port. q, Quantification of nose-poking events in a 60-min session for mice in which ChR2 (n = 8) or GFP (n = 7) was expressed in Fezf2BLa neurons and photo-stimulation was delivered to the NAc at the active port (F(1, 26) = 1.4, P = 0.25; n.s., P > 0.05; two-way ANOVA). r, Quantification of nose-poke events of the ChR2 mice in q, in the first 2 mins of the reversal test, in which the active and inactive ports were switched (P = 0.57; Mann-Whitney test, U = 24). s, Same as p, except that photo-stimulation was delivered to the OT at the active port. t, Same as q, except that the Fezf2BLa→OT pathway was targeted. The ChR2 mice (n = 8), but not the GFP mice (n = 7), vigorously poked the active port (F(1, 26) = 35.43, P < 0.0001; ***P < 0.0001, n.s., P > 0.05; two-way ANOVA followed by Bonferroni’s multiple comparisons test). u, Quantification of nose-poke events of the ChR2 mice in t, in the first 2 mins of the reversal test, in which the active and inactive ports were switched (***P = 0.0003; Mann-Whitney test, U = 0.5). v, A schematic of the approach. w, Top: trial-by-trial heat-maps of running velocity for an example mouse in the active avoidance task; bottom: average running velocity of the mouse in early (first 10 trials) and late (last 10 trials) trials. x, Quantification of running velocity during the decision window in early (first 10) and late (last 10) trials (ChR2 mice, n = 8, GFP mice, n = 7; (F(1, 26) = 29.66, P < 0.0001; ****P < 0.0001, n.s., P > 0.05; two-way ANOVA followed by Bonferroni’s multiple comparisons test). y, z, Same as w, x, respectively, except that the Fezf2BLa→OT pathway was targeted (ChR2 mice, n = 8, GFP mice, n = 7; (F(1, 26) = 2.976, P = 0.096; n.s., P > 0.05; two-way ANOVA followed by Bonferroni’s multiple comparisons test). Data are presented as mean ± s.e.m.

Together, these results indicate that the activities in Fezf2BLa→NAc neurons serve as a negative reinforcer to instruct avoidance learning, while activities in Fezf2BLa→OT neurons serve as a positive reinforcer that supports self-stimulation and drives formation of appetitive memories.

Fezf2BLa→mPFC and Fezf2BLa→LS have different functions

Finally, as Fezf2BLa neurons also send projections to the mPFC and LS (Extended Data Fig. 5a-i), we examined the effect of optogenetic activation of these projections in the RTPP/RTPA test (Supplementary Fig. 8), using the same strategy as described above (see Extended Data Fig. 10a-d & k-n). Activation of either Fezf2BLa→mPFC (Supplementary Fig. 8a-e) or Fezf2BLa→LS (Supplementary Fig. 8f-j) pathway did not induce any obvious behavioral effect in this test. Future research will need to elucidate the potential functions of these pathways.

Discussion

Valence-specific neurons in the BLA

Through imaging Fezf2BLa neurons during behavior, we discovered two functionally distinct populations, with one (PVNs) being excited by an appetitive but not aversive US, and the other (NVNs) excited by an aversive but not appetitive US. These PVNs and NVNs underwent plasticity during learning, such that they acquired excitatory responses to the CS entrained by appetitive or aversive US, respectively, after learning. As a result, the number of Fezf2BLa neurons excited by either of the CSs increased with learning. These results show that the BLA contains valence-specific neurons and provide, for the first time, direct in vivo evidence that CSs acquire the ability to activate valence-specific BLA neurons through learning.

The CS responses in PVNs and NVNs preceded and were coupled with reward-seeking and punishment-avoidance actions, respectively. At the population level, the CS responses of Fezf2BLa neurons predicted valence-specific behaviors on a trial-by-trial basis. This tight relationship after, but not before learning suggests an important role of these neurons in guiding or promoting goal-directed behaviors acquired through experience. Indeed, chemogenetic inhibition of these neurons prevented the CS-induced reward seeking and punishment avoidance in well-trained animals. Notably, this manipulation did not affect the very same action (licking or running) evoked by a US immediately following the CS. This remarkable specificity indicates that Fezf2BLa neurons were not required for generating the specific movements (licking and running) that ultimately let animals reach their goals; rather, they were required for anticipating or motivating such actions. We propose that Fezf2BLa neurons, in particular the PVNs and NVNs, play a critical role in generating positive and negative motivations or saliences, respectively, that underlie valence-specific behaviors.

Distinct BLa-ventral striatal circuits for opposing valences

Through circuit mapping, pathway-specific fiber photometry and optogenetics, we found that Fezf2BLa neurons had two major targets – the NAc and the OT – both of which belong to the ventral striatum[30]. Furthermore, the response properties and functions of Fezf2BLa→NAc and Fezf2BLa→OT neurons suggest that they are NVNs and PVNs, respectively.

The in vivo properties and behavioral functions of Fezf2BLa→NAc neurons that we describe here provide a mechanistic explanation for previous findings that the BLA→NAc pathway is involved in aversive behaviors[5, 6, 46, 47]. Of note, it has also been shown that the BLA→NAc pathway plays an important role in reward-related behaviors[10-12, 48]. These observations indicate that NAc-projecting BLA neurons are functionally heterogeneous. While Fezf2BLa→NAc neurons represent negative valence and are essential for negative reinforcement, other yet to be identified NAc-projecting BLA neurons likely have opposing or different functions.

The properties and functions of the BLa→OT pathway have not been described previously. It has been shown that the OT plays an important role in reward related behaviors[31, 49-52]. Our results from Fezf2BLa→OT neurons provide the first evidence that positive valence or motivation information is conveyed from the BLa to the OT, and establish a critical role for the OT in the BLA valence-specific circuits.

It is worth noting that our anatomical and functional circuit mapping experiments suggest that, although Fezf2BLa→NAc and Fezf2BLa→OT neurons preferentially project to the NAc and OT, respectively, they do send collaterals to the other targets. These collaterals may play an important role in coordinating the activities across brain areas during valence-specific behaviors.

A circuit framework for valence-specific behaviors

We show that Fezf2BLa→NAc and Fezf2BLa→OT neurons receive distinct sets of monosynaptic inputs. This result suggests that the two classes, and hence the NVNs and PVNs in the BLA, are “hard wired” in distinct input/output circuits. These circuits may represent the key actors that play the previously established roles of the BLA in encoding, learning and expressing valence-specific behaviors[2, 3, 48, 53].

The ventral striatum has long been considered an interface between limbic and motor areas[30, 31, 54]. Future studies need to elucidate how the negative and positive motivational signals from BLA NVNs and PVNs, respectively, are further transformed in the NAc and OT for guiding valence-specific behaviors. It will also be important to determine how neurons in the distinct input structures interact with the NVNs and PVNs, to give rise to the valence-specific motivational signals therein.

Methods

Animals

Male and female mice with age of 2-4 months were used in all the experiments. Mice were housed under a 12-h light/dark cycle (7 a.m. to 7 p.m. light) in groups of 2-4 animals, with the room temperature being 22 °C and humidity being 50%. Food and water were available ad libitum before behavioral training. All behavioral experiments were performed during the light cycle. Littermates were randomly assigned to different groups prior to experiments. All mice were bred onto a C57BL/6J background. All experimental procedures were approved by the Institutional Animal Care and Use Committee of Cold Spring Harbor Laboratory (CSHL) and performed in accordance to the US National Institutes of Health guidelines.

The Fezf2-CreER and Fezf2-Flp knock-in mouse driver lines, in which the expression of the inducible Cre recombinase (CreER) or Flp, respectively, is driven by the endogenous Fezf2 promoter, was generated as previously described[55, 56]. A gene-targeting vector for Fezf2-CreER or Flp was generated using a PCR-based cloning approach[55] to insert an in-frame 2A-CreER or 2A-Flp coding cassette, respectively, immediately before the STOP codon of the Fezf2 gene[20]. The targeting vector was linearized and transfected into C57/B6 ES cells. G418-resistant ES clones were first screened by PCR and then confirmed by Southern blotting using probes against the 5’ and 3’ homology arms of the targeted site. The Ai14 reporter mice[21] were purchased from Jackson Laboratory (Stock No: 007908). The H2B-GFP (Rosa26-stop-H2B-GFP) reporter mouse line[22] was generated by Z. Josh Huang’s lab at CSHL.

Viral vectors

The rAAV5-Ef1a-DIO-hChR2-eYFP, AAV9-CAG-Flex-GFP, and AAV8-hSyn-DIO-KORD-IRES-mCitrine were produced by the University of North Carolina vector core facility (Chapel Hill, North Carolina, USA). The pAAV-Syn-Flex-GCaMP6f-WPRE-SV40 (Catalog number 100833), AAV-Ef1a-DIO-GCaMP6f-P2A-nls-tdTomato (retrograde AAV; Catalog number 51083), pAAV-EF1a-fDIO-Cre (retrograde AAV; Catalog number 121675-AAVrg), pAAV_hSyn1-SIO-stGtACR2-FusionRed (Catalog number 105677-AAV1), pAAV-hSyn-DIO-EGFP (retrograde AAV; Catalog number 50457-AAVrg) and pAAV-hSyn-DIO-mCherry (retrograde AAV; Catalog number 50459-AAVrg) were produced by Addgene (Watertown, MA, USA). The CAV-Flex-FlpO was produced by Montpellier vector platform (Plateforme de Vectorologie de Montpellier (PVM), Biocampus Montpellier, Montpellier, France). The AAVDJ-CAG-fDIO-oG-WPRE-SV40pA was produced by the Viral Vector Core Facility at Salk Institute. The AAV2/8-Ef1a-fDIO-TVA-mCherry was produced by Z. Josh Huang’s lab at CSHL. The Rbv-CVS-N2c-dG-GFP (the optimized rabies virus)[36] was produced by HHMI Janelia Research Campus. The AAV8-CaMKIIa-FDO-GCaMP6f (Flp-off) virus was produced by K. Deisseroth’s lab at Stanford University. All viral vectors were aliquoted and stored at –80 °C until use.

Stereotaxic surgery

All surgery was performed under aseptic conditions and body temperature was maintained with a heating pad. Standard surgical procedures were used for stereotaxic injection and GRIN lens or optical fiber implantation, as previously described[57, 58]. Briefly, mice were anaesthetized with isoflurane (3% at the beginning and 1% for the rest of the surgical procedure), and were positioned in a stereotaxic injection frame and on top of a heating pad maintained at 35 °C. A digital mouse brain atlas was linked to the injection frame to guide the identification and targeting of different brain areas (Angle Two Stereotaxic System, myNeuroLab.com). We used the following coordinates for injections or implantations in the BLA: −1.6 mm from Bregma, 3.3 mm lateral from midline, and 4.5 mm vertical from skull surface; the NAc (shell and core): 1.5 mm from Bregma, 1.0 mm lateral from midline, and 4.6 mm vertical from skull surface; the OT: 1.5 mm from Bregma, 0.8 mm lateral from midline, and 5.0 mm vertical from skull surface; mPFC: 1.8 mm from Bregma, 0.3 mm lateral from midline, and 3 mm vertical from skull surface; and the LS: 0.3 mm from Bregma, 0.5 mm lateral from midline, and 3 mm vertical from skull surface.

For virus injection, we made a small cranial window (1–2 mm2) for each mouse, through which a glass micropipette (tip diameter, ~5 μm) containing viral solution was lowered down to the target. About 0.3 μl of viral solution was delivered with pressure applications (5–20 psi, 5–20 ms at 1 Hz) controlled by a Picrospritzer III (General Valve) and a pulse generator (Agilent). The speed of injection was ~0.1 μl/10 min. We waited for at least 10 min following the injection before slowly removing the injection pipette. For imaging experiments, we implanted the GRIN lens (diameter, 0.6 mm; length, 6.7 mm; Inscopix) two weeks after virus injection. The GRIN lens was carefully implanted 200 μm above the center of the injection using a GRIN lens holder (Inscopix). The speed for lowering the GRIN lens was constant and slow (~100 μm/min) until reaching the target. We secured the GRIN lens to the skull with C&B-Metabond Quick adhesive luting cement (Parkell Prod), and subsequently mounted a small piece of metal bar on the skull for head-fixation. We waited four to six weeks following GRIN lens implantation before starting the imaging experiments.

For optogenetic experiments, we bilaterally implanted optical fibers (diameter, 200 μm, NA, 0.22) into the BLa, NAc or OT. For fiber photometry experiments, we implanted the optical fiber (diameter, 200 μm, NA, 0.37) into the BLa. For the retrograde tracing of Fezf2BLa neurons with the retrograde tracer cholera toxin subunit B (CTB), we used the Fezf2;H2B-GFP mice and injected the NAc and OT in the same mice with CTB conjugated with either Alexa Fluor™ 647 or 555 (CTB-647 or CTB-555, respectively; Invitrogen, Thermo Fisher Scientific, Waltham, Massachusetts, USA; 0.2 μl, 1mg/ml in phosphate-buffered saline). The animals were perfused 5 days after CTB injection. For certain viral injection, where indicated, a small amount of blue fluorescent microspheres (“blue beads”) (ThermoFisher, Catalog number F8805, 2% solids, diluted to 0.1% with PBS) was added to the viral solution at 1:1 (volume-to-volume ratio) to mark the injection site for histologic verification.

Tamoxifen induction

All Fezf2-CreER mice underwent tamoxifen induction to activate the Cre-dependent expression of genes of interest. Tamoxifen (Sigma-Aldrich, Catalog number: WXBC6404V) was dissolved in corn oil at a concentration of 20 mg/ml, by constant shaking for 24 hours at room temperature in a container protected from light. Individual aliquots (1 ml each) were stored at 4 °C. Tamoxifen induction was performed via intraperitoneal (i.p.) injections at a dose of 0.1 mg/g of body weight, once every 2 days for 3 times. For virally driven gene expression, the induction was performed one day after surgery. All mice were closely monitored throughout the course of tamoxifen injections and the post-injection period.

Immunohistochemistry

Immunohistochemistry experiments were performed following standard procedures. Briefly, mice were anesthetized with Euthasol (0.2 ml; Virbac, Fort Worth, Texas, USA) and transcardially perfused with 30 ml of PBS, followed by 30 ml of 4% paraformaldehyde (PFA) in PBS. Brains were extracted and further fixed in 4% PFA overnight followed by cryoprotection in a 30% PBS-buffered sucrose solution for 36 h at 4 °C. Coronal sections (50 μm in thickness) were cut using a freezing microtome (Leica SM 2010R, Leica). Brain sections were first washed in PBS (3 x 5 min), incubated in PBST (0.3% Triton X-100 in PBS) for 30 min at room temperature (RT) and then washed with PBS (3 x 5 min). Next, sections were blocked in 5% normal goat serum in PBST for 30 min at RT and then incubated with the primary antibody overnight at 4 °C. Sections were washed with PBS (5 x 15 min) and incubated with the fluorescent secondary antibody at RT for 2 h. After washing with PBS (5 x 15 min), sections were mounted onto slides with Fluoromount-G (eBioscience, San Diego, California, USA). Images were taken using an LSM 780 laser-scanning confocal microscope (Carl Zeiss, Oberkochen, Germany).

The primary antibodies used in this study were: rabbit anti-CaMKII (1:500; Abcam, catalog number ab52476, lot number GR312537-6), rabbit anti-GABA (1:500; Sigma, St. Louis, MO, USA; catalog number A2052), chicken anti-GFP (1:1000; Aves Labs, Catalog number GFP1020, lot number GFP697986), rabbit anti-RFP (1:1000; Rockland, Catalog number 600-401-379, lot number 35868), rabbit anti-Substance P (1:1000; ImmunoStar, Catalog number 20064, lot number 1531001), rabbit anti-HA-Tag (1:1000; Cell Signaling, Catalog number 3724S), rabbit anti-mCherry (1:1000; Abcam, catalog number ab167453, lot number GR3213077-3) and rabbit anti-DARPP32 (1:1000; Abcam, Catalog number ab40801). The fluorophore-conjugated secondary antibodies used were Alexa Fluor® 488 donkey anti-chicken IgG (H+L) (1:1000; Catalog number 703-545-155, Jackson ImmunoResearch), Alexa Fluor® 488 goat anti-rabbit IgG (H+L) (1:1000; Catalog number A-11008, Invitrogen) and Alexa Fluor® 555 goat anti-rabbit IgG (H+L) (1:1000; Catalog number A-21428; Life Technologies, Carlsbad, California, USA).

Fluorescent in situ hybridization

Single molecule fluorescent in situ hybridization (smFISH) (RNAscope, ACDBio) was used to detect the expression of Rspo2, Pppr1b1, Fezf2 and Flp mRNAs in the BLA of wild-type, Fezf2-CreER or Fezf2-Flp mice. We used the Fezf2-CreER;H2B-GFP mice, in which Fezf2BLa neurons can be readily recognized by their expression of the nuclear GFP three weeks after tamoxifen induction. For tissue preparation, mice were first anesthetized under isoflurane and then decapitated. Their brain tissue was first embedded in cryomolds (Sakura Finetek, Catalog number 4566) filled with M-1 Embedding Matrix (Thermo Scientific, Catalog number 1310) then quickly fresh-frozen on dry ice. The tissue was stored at −80 °C until it was sectioned with a cryostat. Cryostat-cut sections (16-μm) containing the entire BLA were collected rostro-caudally in a series of four slides, and quickly stored at −80 °C until processed. Hybridization was carried out using the RNAscope kit (ACDBio).

The day of the experiment, frozen sections were post-fixed in 4% PFA in RNA-free PBS (hereafter referred to as PBS) at room temperature (RT) for 15 min, then washed in PBS, dehydrated using increasing concentrations of ethanol in water (50%, once; 70%, once; 100%, twice; 5 min each). Sections were then dried at RT and incubated with Protease IV for 30 min at RT. Sections were washed in PBS three times (5 min each) at RT, then hybridized. Probes against Ppp1r1b (Catalog number 405901-C3, dilution 1:50), Rspo2 (Catalog number 402001-C1, dilution 1:50), Fezf2 (Catalog number 888591-C2, dilution 1:50) and Flp (Catalog number 448191-C1, dilution 1:50) were applied to the BLA sections. Hybridization was carried out for 2 h at 40 °C. After that, sections were washed twice in 1x Wash Buffer (Catalog number 310091; 2 min each) at RT, then incubated with the amplification reagents for three consecutive rounds (30 min, 15 min and 30 min, at 40 °C). After each amplification step, sections were washed twice in 1x Wash Buffer (2 min each) at RT. Finally, fluorescence detection was carried out for 15 min at 40 °C. The green channel was kept available for the intrinsic nuclear GFP signals, which did not require amplification. Sections were then washed twice in 1x Wash Buffer (2 min each) , incubated with DAPI for 2 min, washed twice in 1x Wash Buffer (2 min each), then mounted with coverslip using mounting medium. Images were acquired using an LSM780 confocal microscope equipped with 20x and 40x lenses, and visualized and processed using ImageJ and Adobe Illustrator.

Behavioral tasks

The active approach and avoidance (AAA) task.

This task was designed to train mice to actively seek reward and actively avoid punishment. Mice were subjected to water deprivation starting from 23 h before the training, during which they were head restrained using custom-made clamps and metal head-bars. Each mouse was first habituated to the head-restraint for 2-4 days, one session (1 h) per day. Unpredicted drops of water (5 μl) were delivered through a metal spout placed next to the mouth of the mice during habituation. Once animals had learned how to obtain water by licking the spout, they were subjected to conditioning that included different types of trials. In the reward trials, a 1-s 4-kHz tone (CSR) was presented, followed by a 2-s decision window. If mice licked the spout at least one time during the decision window, water (5 μl) would be delivered through the spout immediately after this window. Otherwise no water would be delivered. In the punishment trials, a 1-s 12-kHz tone (CSP) was presented, followed by a 2-s decision window. If mice ran above a threshold speed (10 cm/s) during the decision window, they would avoid an unpleasant air-puff (40 psi, 100 ms) blowing to the face, in an area close to the eye. Otherwise mice would receive the air-puff immediately after the decision window.

For the behavioral training in GCaMP6 imaging or photometry experiments, we added neutral trials, in which a 1-s 8-kHz tone or white noise (CSN) was presented. The CSN was followed by nothing and served as a control. We recorded neuronal activities during both the early and the late training stages for all the mice used in these experiments. The early stage was the first training session. The late stage was defined as when the animals reached a threshold performance level (80%) in both the reward and the punishment trials.

Animals were trained one session per day, with each session consisting of 200 trials if the neutral trials were not included, or 300 trials if the neutral trials were included. The different types of trials were randomly interleaved. The inter-trial interval was randomly variable between 10 to 20 s. To reduce photobleaching, each imaging session consisted of 60 trials in total, with 20 trials being allocated for each of the three trial types.

Real-time place preference or aversion (RTPP/RTPA) test.

Freely moving mice were initially habituated to a two-sided chamber (23 × 33 × 25 cm for each chamber; made from Plexiglas) for 10 min, during which their baseline preference for the left or right side of the chamber was assessed. During the first test session (10 min), we assigned one side of the chamber (counterbalanced across mice) as the photo-stimulation side, and placed the mice in the non-stimulation side to start the experiment. Once the mouse entered the stimulation side, photo-stimulation (5-ms pulses, 30 Hz, 10 mW (measured at the tip of optic fibres)), generated by a 473-nm laser (OEM Laser Systems Inc., Bluffdale, Utah, USA), was immediately turned on, and was turned off as soon as the mouse exited the stimulation side. In the second test session (10 min) we repeated this procedure but assigned the other side of the chamber as the stimulation side. The behavior of the mice were videotaped with a CCD camera (C930, Logitech) interfaced with Ethovision software (Noldus Information Technologies), which was also used to control the laser stimulation and extract behavioral parameters (position, time, distance and velocity). We used a preference score as a measure of the effect of laser stimulation, which was calculated as: (time in stimulation side – time in non-stimulation side)Stimulation – (time in stimulation side – time in non-stimulation side)Baseline, that is, the difference in time spent in the stimulation side and non-stimulation side during the stimulation session, subtracted by the difference in time spent in the two corresponding sides during the baseline session (to correct baseline bias). A positive score indicates preference for the stimulation side.

Intracranial self-stimulation test.

Mice were tested in a chamber equipped with two ports. In each test session (1 h in duration), mice were allowed to freely explore and poke into both ports. Poking into one of the ports (the active port) would trigger a laser and thus lead to photo-stimulation in the OT for 2 s (5-ms pulses, 30 Hz, 10 mW; λ = 473 nm). Poking into the other port (the inactive port) would not trigger the photo-stimulation. 24 h later the test was repeated, with a different port being designated as the active port. The poking events during each session were recorded and analyzed by custom script written in MATLAB (The MathWorks, Inc., Natick, Massachusetts, USA).

In vivo chemogenetic inhibition

To test the behavioral effects of inhibiting Fezf2BLa neurons in the AAA task, we expressed KORD (a DREADD derived from the kappa-opioid receptor)[40] or GFP (as the control) into these neurons in mice with the respective viruses, and treated these mice with the KORD agonist salvinorin B (SALB)[40] (10 mg/kg of body weight, subcutaneous (s.c.) injection) 15 min before behavioral testing.

In vivo optogenetic inhibition and activation

For optogenetic inhibition in the BLa during learning in the AAA task, laser stimulation (4-s square pulse, λ = 473 nm, 10 mW measured at the tip of optical fiber) was delivered following the onset of CS presentation in each trail, to cover both CS and US presentation. For optogenetic inhibition after learning in this task, the laser pulse (3 s in duration) was delivered following the onset of CS presentation to cover CS presentation and decision window, in a randomly selected subset (~50%) of trials.

To determine whether activation of Fezf2BLa→NAc neurons or their axon terminals could drive negative reinforcement learning, we used the same design as that for the punishment trials in the AAA task, but replaced the air-puff with a 2-s train of optogenetic stimulation (5-ms pulses, 30 Hz, 10 mW; λ = 473 nm) of these neurons. Mice were trained one session (100 trials) per day, with each trial beginning with a 1-s 2-kHz tone, followed by a 2-s decision window. The inter-trial interval was randomly variable between 5 to 10 s.

Behavioral data collection and analysis

An open source Bpod behavioral control system (Sanworks, Stony Brook, NY,USA) was used to control real-time animal behavior and optogenetic manipulation. Custom-written scripts in MATLAB based on Bpod commands were used to control the delivery of CS and US and record behavioral responses, including licking events and running velocity. Pure tones (70 db; the CS) with different frequencies were generated as sine wave. The tones were uploaded to the audio adaptor board using Bpod control system. The amount of water (the reward US) was controlled by fast solenoid valves (Lee Company). A metal spout was placed in front of the mouth of an animal for water delivery. The spout also served as part of a custom “lickometer” circuit, which registered a lick event each time a mouse completed the circuit by licking the spout. The licking events were recorded and analyzed using custom scripts written in MATLAB. Air-puff delivery (the aversive US) was controlled by a solenoid valves (U.S. Solid) with a higher pressure operating limit. The air-puff (40 psi) was delivered through a metal tube pointing towards the area surrounding the eye. We used a rotary encoder (YUMO-E6B2-CWZ3E-1024; SparkFun Electronics) to detect and record real-time running velocity. The rotary encoder was attached to a running wheel and connected with a microcontroller (Arduino UNO R3; SparkFun Electronics). The running wheel (diameter, 14 cm; width, 8 cm) was made using a 3D printer (MakerBot Replicator 2; MakerBot). The microcontroller converted the digital inputs from the rotary encoder into analog signals reflecting running velocity, which was in turn recorded and analyzed with custom scripts written in MATLAB.

Pupil size measurement and analysis

To measure pupil size changes, we captured a video of an area surrounding the eye using an infrared-filter mounted camera (20 Hz, FL3-U3-13S2C-CS, Point Grey) under lighting with infrared light-emitted diodes. The camera was controlled by an open source Bonsai software (Bonsai). Offline video analysis was conducted using EthoVision XT software (Noldus; Wageningen, The Netherlands). To measure the size of a pupil, we manually selected a region of interest (ROI) surrounding the pupil. Pixels corresponding to the pupil were assigned as those that were darker than the surrounding background within the ROI. To quantify the changes in pupil size (ΔA), we computed ΔA/A0(t) = (A(t)-A0)/A0, where A0 is the mean size of the area corresponding to the pupil during the baseline before laser stimulation and A(t) is the pupil size in each picture frame, using a custom script written in MATLAB.

In vivo calcium imaging and data analysis

We performed calcium imaging using a custom-built wide-filed one-photon fluorescence microscope equipped with an objective lens (10x, NA 0.3, MPlanFLN, Olympus) and an LED fluorescent light source (pE-100, CoolLED). The light was bandpass filtered (FF02-482/18-25, Semrock) to obtain the desired excitation wavelength, and reflected by a dichroic mirror (FF409/493/573/652-Di01, Semrock) to be collected by the objective through the back aperture, and was relayed by the implanted GRIN lens to reach neurons of interest. The emission fluorescence signals coming back through the same GRIN lens were collected by the objective and passed through an emission filter (FF01-520/35-25, Semrock) before they were recorded by a CCD camera (PCO. Pixelfly, PCO-Tech Inc.), which was controlled by software written in LabView (National Instruments).

All imaging experiments were conducted on awake behaving mice under head-restraint in a dark, sound attenuated box. We imaged neuronal activities before and throughout behavioral training. To reliably detect stimulus-driven responses while minimizing photobleaching, we typically imaged neuronal responses in the same condition in 20 trials, with the imaging duration for each trial being 20 s to cover baseline, CS and/or US responses. During the pre-learning habituation, we imaged the responses to either CSs or USs, which were presented randomly interleaved. During conditioning, we carefully adjusted the camera to image neurons in the same focal plane as that in the pre-learning sessions.

For imaging data processing and analysis, we first used Mosaic software (version 1.0.0b; Inscopix, Palo Alto, CA) to combine all the video frames from one imaging session into a single image stack (in TIFF format). The image stack was then spatially down sampled by a factor of 4. Motion artifacts were corrected using image registration techniques in Mosaic environment to align all movie frames to a single target image and generate a stable, motion-corrected movie. Briefly, there were three steps in the motion correction process: (1) generating the target image, (2) running the image registration routine, and (3) removing the post-registration border.

The first step in the motion correction process was to generate a target image for the image registration routine. For this purpose, we chose a single frame from a movie and extracted the frame using the “extract frame” App in Mosaic, which was subsequently used as the target image for correcting the motion artifacts in this movie. The target image was chosen carefully such that it was representative of the entire movie and did not exhibit any unusual fluorescence signals.

The second step in the motion correction process was to set up and run the image registration algorithm using the “apply motion correction” App in Mosaic. This step included adjusting the contrast of the target image and selecting a reference region that had high contrast pixels and vasculature areas. The image registration algorithm aligns each frame in the movie to the reference region in the target image and outputs a new, motion-corrected movie.

The third step was to remove the post-registration border. The movies generated in the second step usually contain sections of black pixels around the edges of individual image frames. These black border areas are generated after the registration process as the image registration algorithm fills the pixels that fall outside of the acquired frame with null value.

For more detailed information about the motion correction process, see: https://support.inscopix.com/sites/default/files/sphinx/1547/html/tutorials/standardWorkflow.html

Next, we applied the extended constrained non-negative matrix factorization (CNMF-E)[58-60], which is optimized for one-photon imaging, to demix neural signals and get their denoised and deconvolved temporal activity (see Supplementary Video 1), termed ΔF[59, 60]. The CNMF-E method was carried out using a custom Matlab algorithm (see reference[59] for a detailed description of this method). The contours of neurons in Fig. 2a and Supplementary video 1 were generated by the CNMF algorithm on the basis of values 8 standard deviations (s.d.) above the mean in a spatial component matrix, which are pixel-by-pixel weighted fluorescence values.

To determine whether a neuron was significantly (P < 0.05) excited or inhibited by a stimulus, and thus can be classified as being “responsive” to the stimulus, we performed Wilcoxon signed-rank test on the data acquired from 20 trials per stimulus, which compared the mean ΔF values in two 3-s time windows using all these trials, with one window immediately preceding stimulus onset and the other immediately following stimulus onset. To obtain the temporal z-scores for a neuron, we first obtained the mean activity trace for the neuron by averaging the fluorescence signals (ΔF) at each time point across all trials, and then computed the z-scores as (F(t) – Fm)/SD, where F(t) is the ΔF value at time t, Fm and SD are the mean and standard deviation, respectively, of the ΔF values over the baseline period. For the trial-by-trial analyses, the z-scores were computed for each trial using the same method (but without the averaging across trials).

Cell registration

To identify the same individual cells from images acquired from different imaging sessions, we performed cell registration using a probabilistic method that automatically registers cells across multiple imaging sessions and estimates the registration confidence for each registered cell [61]. Briefly, as we previously described [58], we first used CNMF-E analysis to generate the spatial footprints for all cells imaged in the pre-learning session. We then repeated this process for the cells imaged in the post-learning sessions. We used the footprints from the pre-learning session as a reference map, and aligned with this map the footprints from the post-learning sessions by correcting for translation and rotation differences between different sessions. We subsequently calculated the probability of a given pair of cells, each from one of two imaging sessions, to be the same cell (Psame) based on their spatial correlation and centroid distance. A pair of cells is considered to have the same identity if Psame > 0.5. The centroid distance between a pair of cells deemed to have the same identity is generally small (≤ 6 μm).

Classification of neurons with clustering analysis

To classify neurons based on their CS response profiles after learning, we performed principal component analysis (PCA) on the z-scores representing the CS responses of these neurons, as previously described[58]. We subsequently applied hierarchical clustering analysis to the first three principal components (PCs) using a correlation distance metric and complete agglomeration methods.

Decoding analysis

We used the support vector machine (SVM) algorithm (fitcsvm) in MATLAB to examine whether different trial types could be predicted on the basis of populational CS responses of Fezf2BLa neurons acquired in each mouse at different stages of learning. We used z-scores to represent the trial-by-trial responses of each neuron to presentations of CSR, CSN or CSP during the early or late stage of learning. For each mouse, we captured the trial-by-trial population CS responses by a vector containing the responses of all neurons from that mouse, with each neuron representing one dimension in a multidimensional population space. We applied PCA on the multidimensional data from each mouse and used the first two PCs to represent the population response in each trial. We then used the low dimensional trial-by-trial population responses from each mouse to train and test an SVM decoder with a Gaussian kernel to distinguish between reward, punishment and neutral trials. Specifically, we used the responses from 60 trials (20 CSR, 20 CSN and 20 CSP) for each mouse during either the early or late training stage to train and test a classifier. For decoder training and testing, we used a 10-fold cross-validation procedure, in which all datasets were randomly partitioned to 10 equal-sized subsamples, with each subsample containing equal number of responses from a given class (i.e., CSR, CSN or CSP). We used 9 of the 10 subsamples for training and determining the boundary that optimally separated reward, neutral and punishment trials. The remaining 1 subsample was used for validating the decoder. We repeated this process 10 times such that each of the 10 subsamples was used exactly once as the validation data. The percentage of accurate classification incidence in the 10 times was reported as the final classification accuracy.

In vivo fiber photometry and data analysis

To record the activities of Fezf2BLa neurons projecting to the NAc or OT in vivo in behaving animals, we used a commercial fiber photometry system (Neurophotometrics Ltd., San Diego, CA, USA) to measure GCaMP6 signals in these neurons through an optical fiber (Fiber core diameter, 200 μm; Fiber length, 5.0 mm; NA, 0.37; Inper, Hangzhou, China) implanted in the BLa. A patch cord (fiber core diameter, 200 μm; Doric Lenses) was used to connect the photometry system with the implanted optical fiber. The intensity of the blue light (λ = 470 nm) for excitation was adjusted to ~20 μW at the tip of the patch cord. Emitted GCaMP6f fluorescence was bandpass filtered and focused on the sensor of a CCD camera. Photometry signals and behavioral events were aligned based on an analogue TTL signal generated by the Bpod. Mean values of signals from a region of interest were calculated and saved by using Bonsai software (Bonsai), and were exported to MATLAB for further analysis.

To correct potential photobleaching of fluorescence signals, we subtracted the slow fluctuations in baseline according to F_raw_correction(t) = F_raw(t) – F_norm(t) + F_average, where F_raw(t) is the raw fluorescence at time t, F_norm(t) is the average of F_raw over 100 second sliding window around time t and F_average is the averaged fluorescence of F_raw over the entire recording time. The ΔF/F0 was calculated by (F_raw_correction(t) – F0) / F0, where F0 is the averaged F_raw_correction during baseline period.

In vitro electrophysiology

Patch clamp recording was performed as previously described[62]. Briefly, mice were anaesthetized with isoflurane (4%) before they were decapitated. The brains were then dissected out and placed in ice-chilled dissection buffer (110 mM choline chloride, 25 mM NaHCO3, 1.25 mM NaH2PO4, 2.5 mM KCl, 0.5 mM CaCl2, 7.0 mM MgCl2, 25.0 mM glucose, 11.6 mM ascorbic acid and 3.1 mM pyruvic acid, gassed with 95% O2 and 5% CO2). An HM650 Vibrating-blade Microtome (Thermo Fisher Scientific) was then used to cut 300 μm thick coronal sections that contained the NAc and OT. These slices were subsequently transferred to a storage chamber that contained oxygenated artificial cerebrospinal fluid (ACSF) (118 mM NaCl, 2.5 mM KCl, 26.2 mM NaHCO3, 1 mM NaH2PO4, 20 mM glucose, 2 mM MgCl2 and 2 mM CaCl2, at 34 °C, pH 7.4, gassed with 95% O2 and 5% CO2). Following 40 min of recovery time, slices were transferred to RT (20–24 °C), where they were continuously bathed in the ACSF.

Whole-cell patch clamp recording from NAc or OT neurons was obtained with Multiclamp 700B amplifiers and pCLAMP 10 software (Molecular Devices, Sunnyvale, California, USA), and was visually guided using an Olympus BX51 microscope equipped with both transmitted and epifluorescence light sources (Olympus Corporation, Shinjuku, Tokyo, Japan). The external solution was ACSF with tetrodotoxin (TTX) (1 μM) and 4-aminopyridine (4-AP) (200 μM) added. The internal solution contained 115 mM cesium methanesulfonate, 20 mM CsCl, 10 mM HEPES, 2.5 mM MgCl2, 4 mM Na2ATP, 0.4 mM Na3GTP, 10 mM sodium phosphocreatine and 0.6 mM EGTA (pH 7.2).

As the acute slices were prepared from mice in which Fezf2BLa neurons were infected with AAV expressing ChR2-YFP, to evoke synaptic transmission onto either NAc or OT neurons driven by Fezf2BLa neurons, a blue light was used to stimulate ChR2-expressing axons originating from Fezf2BLa neurons. The light source was a single-wavelength LED system (λ = 470 nm; http://www.coolled.com/) connected to the epifluorescence port of the Olympus BX51 microscope. A light pulse of 1 ms, triggered by a TTL signal from the Clampex software, was delivered every 10 seconds to drive synaptic responses. Synaptic responses were recorded at a holding potential of −85 mV, low-pass filtered at 1 kHz and were analyzed using pCLAMP 10 software. Comparable number of neurons were recorded from the NAc or OT from each slice.

Mapping monosynaptic inputs with pseudotyped rabies virus

To map the brain-wide monosynaptic inputs onto Fezf2BLa→NAc and Fezf2BLa→OT neurons, we used a pathway- and cell-specific tracing strategy[35] with an optimized rabies virus system[36]. We injected the NAc or OT of Fezf2-CreER mice with the retrograde CAV-Flex-FlpO, which allows the FlpO to be specifically expressed in Fezf2BLa→NAc or Fezf2BLa→OT neurons, respectively, flowing tamoxifen induction. A small amount of blue fluorescent beads was added to the CAV-Flex-FlpO viral solution to mark the injection site for histologic verification. Subsequently, in the same surgery, we injected the BLa of these mice with AAV-fDIO-TVA-mCherry and AAV-fDIO-oG (1:1 in volume, for a total of 0.3–0.4 μl) that express the following components in a Flp-dependent manner: a fluorescent reporter mCherry, TVA (which is a receptor for the avian virus envelope protein EnvA), and the rabies envelope glycoprotein (oG). Tamoxifen induction was performed in these mice one day after the surgery. Three weeks after the first injection, mice were injected in the BLa with Rbv-CVS-N2c-dG-GFP (0.5 μl), a rabies virus that is pseudotyped with EnvA, lacks the envelope glycoprotein, and expresses GFP. This rabies strain has been shown to have enhanced retrograde trans-synaptic transfer and reduced neurotoxicity[36]. Brain tissue was prepared one week after the rabies virus injection for histological examination. This method ensures that the rabies virus exclusively infects the cells that express TVA, i.e., the Fezf2BLa→NAc or Fezf2BLa→OT neurons. Furthermore, complementation of the modified rabies virus with the envelope glycoprotein in the TVA-expressing cells allows the generation of infectious particles, which can then trans-synaptically infect presynaptic neurons.

We followed a commonly used method to analyze the rabies tracing data[34, 35, 38, 39, 63-65]. Briefly, 50-μm coronal sections were obtained across the anteroposterior axis of the brain. We counted the number of GFP-labeled (that is, rabies-labeled) input cells in each brain structure throughout the entire brain, using every third brain section for the counting. Brain regions were identified on the basis of anatomical landmarks and defined according to Allen Brain Atlas. Data are presented as the ratio between the number of GFP-labeled cells in a given brain area and the total number of GFP-labeled cells in the entire brain.

Data availability

All data are contained in the main text, extended data or supplementary data. Source data can be downloaded at https://figshare.com/articles/dataset/NN-A72265C/15130017

Code availability

Source code can be downloaded at https://figshare.com/articles/software/code_for_NN-A72265C/15157614

Statistics and data presentation

All statistics are indicated where used. Statistical analyses were performed with Matlab (MathWorks, Natick, MA) or GraphPad Prism Software (GraphPad Software, Inc., La Jolla, CA). We tested the normality of all data with Kolmogorov-Smirnov normality test in Matlab, and used nonparametric statistical tests for nonnormally distributed datasets, as indicated where used. All statistical tests were two-sided and adjustments were made for multiple comparisons. Data collection and analysis were not performed blind to the conditions of the experiments. All behavioral experiments were controlled by computer systems, and data were collected and analyzed in an automated and unbiased way. No statistical methods were used to pre-determine sample sizes but our sample sizes are similar to those reported in previous publications[34, 38, 58]. An animal was excluded if the viral injection location (as indicated by a virally expressed fluorescent protein or co-injected blue beads) was outside of the target area, such as the BLa, NAc or OT. If the tip of an implanted GRIN lens or optical fiber was outside of the target area, or if the behavioral performance could not reach threshold in the imaging and fiber photometry experiments, the mouse was also excluded. No other mice or data points were excluded.

Supplementary Video 1. Extraction of the spatial and temporal components of GCaMP6 signals from individual neurons with CNMF-E. Shown are videos of GCaMP6 signals from Fezf2+ BLA neurons imaged through a GRIN lens in a representative mouse (the same mouse as that in Fig. 2a) receiving water and air-puff presentations. These videos are short clips of movies stitched together, with each clip representing a trial (200 frames). These videos are played in synchrony, and represent the following contents: “raw data”, GCaMP6 signals without any processing; “denoised”, the remaining signals after subtraction of the background from the raw data; “demixed”, the deconvolved spatiotemporal activity for each neuron using CNMF-E. In the “demixed” video, the contours of 4 representative neurons are traced and numbered. The temporal activities of these 4 neurons are displayed in the bottom panel, in which the moving line indicates the passage of time synchronized with all the videos, and the dashed green and red lines denote the onset of water and air-puff, respectively. Neurons #1 & 2 are water-responsive neurons, whereas neurons #3 & 4 are air-puff-responsive neurons. The scale bar beside each of the videos or response traces denotes ΔF values. These 4 neurons are also identified and labeled in Fig. 2a. See Methods for more information.