Abstract
Social experiences carry tremendous weight in our decision-making, even when social partners are not present. To determine mechanisms, we trained female mice to respond for two food reinforcers. Then, one food was paired with a novel conspecific. Mice later favored the conspecific-associated food, even in the absence of the conspecific. Chemogenetically silencing projections from the prelimbic subregion (PL) of the medial prefrontal cortex to the basolateral amygdala (BLA) obstructed this preference while leaving social discrimination intact, indicating that these projections are necessary for socially driven choice. Further, mice that performed the task had greater densities of dendritic spines on excitatory BLA neurons relative to mice that did not. We next induced chemogenetic receptors in cells active during social interactions—when mice were encoding information that impacted later behavior. BLA neurons stimulated by social experience were necessary for mice to later favor rewards associated with social conspecifics but not make other choices. This profile contrasted with that of PL neurons stimulated by social experience, which were necessary for choice behavior in social and nonsocial contexts alike. The PL may convey a generalized signal allowing mice to favor particular rewards, while units in the BLA process more specialized information, together supporting choice motivated by social information.
Significance Statement
Social experiences color choices, even when social partners are not present. For instance, we might favor a restaurant because it is where we had a memorable first date. We demonstrate that projections from the prelimbic cortex (PL) to basolateral amygdala (BLA) are required for socially driven decisions—that is, for mice to favor a familiar food previously associated with a conspecific. Neuronal ensembles in the BLA stimulated by social experience appear specialized to socially driven choices, while ensembles in the PL are not. The BLA may form social memories, which, with PL input, enable organisms to prioritize social information in later action.
Introduction
Day-to-day choices often involve social interactions, and our prior social experiences can carry tremendous weight in our decision-making. For instance, we might favor a particular restaurant because it is where we had a memorable first date. How the brain translates social information into action selection strategies is unclear. Most studies in this vein utilize nonhuman primates to investigate so-called social decision-making, due to their translational appeal and robust set of complex social behaviors (Gangopadhyay et al., 2021). However, rodents offer a much more readily accessible genetic toolbox, with several tasks revealing the capacity to reliably test social cognition (Ben-Ami Bartal et al., 2014; Burkett et al., 2016; Scheggia et al., 2020; Kietzman and Gourley, 2023). Our group recently developed one such task, social incentivization of future choice (SIFC), in which rodents utilize social information to favor one familiar reward over another (Kietzman et al., 2022).
The prelimbic subregion of the prefrontal cortex (PL) serves as a central hub for cognitive processes that integrate social content (Prounis and Ophir, 2020; Yizhar and Levy, 2021). The PL is highly interconnected with the basolateral amygdala (BLA; Vertes, 2004; Hoover and Vertes, 2007), which, broadly, controls behavioral plasticity in response to highly salient events. “Top-down” processing by the prefrontal cortex is thought to help integrate information about goals, intentions, and internal states into behavior (Miller and Cohen, 2001). PL→BLA connections promote conditioned fear responses and avoidance behavior, “self-interest,” feeding, and even inhibit certain social behaviors—adaptive reactions to threatening or negative stimuli (Land et al., 2014; Marek et al., 2019; Song et al., 2019; Diehl et al., 2020; Huang et al., 2020; Scheggia et al., 2022). Thus, one might predict that they might mitigate the propensity of mice to favor rewards associated with social conspecifics, allowing them to instead explore other options. On the other hand, we recently demonstrated that BLA→PL interactions facilitate this socially based choice (Kietzman et al., 2022), raising the possibility that PL→BLA projections may instead play a similar role. We thus aimed to clarify the role of information traveling from the PL to the BLA during social incentivization of choice, particularly to determine whether PL→BLA projections contrast or complement the prosocial role of BLA→PL projections.
Here, we first found that PL→BLA projections were necessary for mice to favor rewards associated with social conspecifics. We then investigated the possibility that neuronal ensembles in the BLA serve as social memory engrams, enabling SIFC. Engrams are defined as cell populations that (1) are activated by learning, (2) display enduring cellular changes, and (3) are reactivated by a part of the original stimulus for memory retrieval (Josselyn and Tonegawa, 2020). We find that BLA neurons (1) are stimulated by social experiences (memories of which later modulate behavior) and (2) undergo structural plasticity during SIFC and (3) that cells stimulated by social experience are necessary for memory retrieval in the SIFC task, but not other tasks. BLA neurons are distinctive from PL neurons, which appear to play a more general function in action selection. Overall, our findings suggest that neuronal ensembles in the BLA allow social information to guide choice via PL input.
Materials and Methods
Subjects
Procedures were approved by the Emory University IACUC. Mice were 2–6 months of age. Wild-type mice, Fos2A–iCreER (TRAP2) transgenic mice (DeNardo et al., 2019), or YFP-expressing mice, expressed under control of the Thy1 gene (H line; Feng et al., 2000) were used. Stimulus mice were novel same-strain conspecifics within 1 month of age. Females were used throughout to avoid aggression between unfamiliar male mice. Experimental mice were group housed until behavioral testing, at which point they were singly housed. Stimulus mice remained in group housing throughout. All mice were bred from Jackson Laboratory stock on a C57BL/6 background, maintained on a 12 h light cycle (0700 on), and provided food and water ad libitum unless otherwise noted.
Stereotaxic surgery and viral vectors
Mice were anesthetized with ketamine/dexmedetomidine (100 mg/kg/0.5 mg/kg, i.p.) and placed in a digitized stereotaxic frame (Stoelting). Small holes were drilled in the skull, and viral vectors were infused at AP +1.7 mm, ML ±0.17, DV −2.5 (PL) or AP −1.4, ML ±3.0, DV −4.9 (BLA). Viral vectors were infused over 5 min in a volume of 0.5 μl. The viral vectors used in the current report were as follows: rgAAV(pENN)-hSyn-HI-eGFP-Cre-WPRE (for PL→BLA inactivation studies) and AAV5-hSyn-DIO-mCherry hM4Di (for PL→BLA inactivation studies and TRAP2 studies). Syringes were left in place for ≥5 min for PL infusions, or ≥8 min for BLA infusions, prior to removal and suturing. Mice were revived with antisedan (3 mg/kg, i.p.) and left undisturbed for at least 3 weeks prior to behavioral experiments.
Instrumental response training
First, mice were food restricted to ∼90% of their free-feeding body weight to motivate food-reinforced responding. Mice were then trained to nose poke for two distinct food reinforcers (20 mg Bio-Serv Dustless Precision Pellets, purified grain and chocolate, selected because one is not systematically preferred over the other; Kietzman et al., 2022) in Med Associates operant conditioning chambers equipped with two nose poke apertures and a separate food magazine. Responding on each aperture was reinforced using a fixed ratio 1 (FR1) schedule of reinforcement, such that 30 pellets were available for responding on each aperture. Sessions ended at 70 min, or 60 pellets acquired. Training ended at seven sessions, or when mice acquired all 60 pellets within 70 min (if mice required >7 sessions). Response acquisition curves represent responses/min during the last seven training sessions, with no significant side preferences throughout.
Social incentivization of future choice
Mice were trained to nose poke for food as above. Then, social conditioning occurred. First, experimental mice were habituated to testing chambers (large, clean, empty cages) by allowing them to explore under low light with no stimuli present for 30–60 min. The conditioning phase of the task consisted of two sessions (one/day), which were counterbalanced for order. In one session, the experimental mouse was placed in the chamber with a novel conspecific and 2 g of either the grain or chocolate pellet. The novel conspecific was not food restricted and had never been exposed to the pellet prior to the social conditioning phase and so averaged <1 instance of eating in a 60 min session. In the other session, the mouse was placed in the same chamber with the other pellet and a novel object, a 50 ml falcon tube. Pellets were weighed before and after each session to ensure that a minimum of 0.2 g of pellets were consumed. If 0.2 g were not consumed, the pairing session was repeated the following day. For the “no association” condition in one experiment, 2 g of standard vivarium chow replaced the pellets; otherwise, procedures were identical.
Social conditioning sessions lasted 60 min, always under low light. The flavor of pellet that was paired with a novel conspecific was the mouse's less earned pellet, as determined by counting all pellets earned during the last 7 d of instrumental training. The purpose was to bias against any individual pellet preferences (or side preferences within the chambers), in which case, social experiences would have to overturn said preferences for mice to display SIFC (Kietzman et al., 2022).
To determine whether social experience motivated later instrumental responding, we returned mice to the instrumental conditioning chambers for a 15 min probe test conducted the following day in extinction. A socially motivated mouse would preferentially engage the action predictive of a reward associated with social experience, while a failure to differentiate between actions reflects a failure of social experience to incentivize responding. Preference scores were calculated for before and after social conditioning. Pretest preference scores were calculated by dividing the responses for the less earned pellet (the “to be social” pellet) by the responses for the more earned pellet (the “to be nonsocial” pellet) averaged over the last 7 d of training. Any mouse that exhibited a pretest score of <0.5 (an extreme preference for one of the pellets) was excluded from testing. Post-test scores were calculated by dividing the responses for the “social” pellet over the “nonsocial” pellet during the probe test.
In cases when SIFC was tested twice, the 3 d procedure was repeated, and the pellet associated with a novel social conspecific was reversed so mice could not rely on prior experience to guide choice.
Three-chamber social interaction and social discrimination test
Mice were given 10 min to habituate to the middle section of a large chamber (40.5 × 60 × 25 cm; MazeEngineers) divided into three sections. Dividers between each section were sliding doors to allow for free movement of the animals once opened. After the habituation period, the social interaction test commenced. Sliding doors were removed to reveal two lateral chambers containing an acrylic cage (10 cm diameter, 20 cm height) holding either a novel conspecific or nothing. Once the mouse left the middle section for the first time, 10 min was recorded and scored by a blinded rater for time in each chamber. A typical mouse will spend more time in the chamber with a novel conspecific than an empty cage, indicating sociality. Following the social interaction test, the experimental mouse was returned to the middle section and the doors were closed.
The social discrimination test occurred immediately after the social interaction test. An unfamiliar novel conspecific was placed in the previously empty cage, and doors were removed for another 10 min. Time spent in each chamber was again quantified. A typical mouse will spend more time in the chamber the unfamiliar conspecific than the familiar conspecific, evidence of social discrimination. All videos were analyzed using JWatcher v1.0 software.
Familiar versus novel conspecific interaction
We aimed to assess immediate-early gene (IEG) levels following interaction with a familiar versus novel conspecific. Three 60 min sessions (one/day) were conducted in large clean cages under low light. Half of the mice were placed individually in their cages with the same conspecific each day. The other half were placed in their cages daily, but a novel conspecific was included only on the third day. Mice were killed as described below 90 min following the start of the third session.
Contingency modification test
We used a task in which mice must make choices based on action–outcome contingency, rather than social information, as a comparator to SIFC. This task was used because flexible responding requires the PL and BLA (Zimmermann et al., 2017; Sequeira et al., 2023). Following SIFC testing, responding on both nose poke recesses was reinstated by retraining mice for two sessions using the same training conditions as above. Next, one of the two nose poke apertures was occluded, and responding was reinforced, as during training, for 25 min. As such, the predictive relationship between a nose poke and pellet delivery remained intact. The following day, the other aperture was available for a 25 min “nonreinforced” session, during which pellets were delivered noncontingently at a rate matched to each animal's reinforcement rate from the previous session. Under these conditions, only ∼7% of pellets are delivered coincidentally with a nose poke (Butkovich et al., 2015). Thus, this nose poke action becomes significantly less predictive of reinforcement than the other. The aperture at which the contingency was reinforced versus nonreinforced, as well as the order of the reinforced and nonreinforced sessions, were randomized.
Finally, both apertures were made available during a 15 min probe test conducted in extinction the following day. Sensitivity to causal relationships between actions and outcomes is reflected by preferential engagement of the aperture available during the “reinforced” session—that is, the response that remained predictive of reward. Meanwhile, engaging both responses equivalently reflects a failure to utilize causal knowledge to guide responding. The 3 d procedure was conducted twice, with reversal of the associations between first and second test so mice could not rely on prior information for choice.
Activity-dependent cell labeling
Following recovery from surgery, TRAP2 mutant mice were singly housed in their home cages for 1 week without disturbance except for standard laboratory care. Next, mice were habituated to a testing room for 30 min and finally, placed for 5 min in a clean cage with a novel, same-age, same-sex conspecific. Mice remained in the quiet, dim testing room for 90 min, at which point they were injected with 4-OHT (Sigma; 50 mg/kg, i.p., in 5% DMSO and saline/2% Tween-80, 1 ml/100 g), which induces Cre recombinase (Cre) in c-Fos+ cells. They then remained without disturbance for at least 4 more hours to minimize IEG expression unrelated to the novel conspecific. Mice were then returned to the animal vivarium for at least 1 week before behavioral training or 3 weeks before electrophysiological testing.
Clozapine N-oxide (CNO)
All mice in a given experiment received CNO (Sigma; 1 mg/kg, i.p., in 2% DMSO and saline), regardless of viral vector condition, to equally expose mice to any unintended consequences of CNO (Gomez et al., 2017). Notably, the present dose is not obviously subject to reverse metabolism to its parent compound, clozapine (Manvich et al., 2018). CNO was administered 30 min prior to testing. In experiments when all mice received CNO in conjunction with one test session and vehicle in conjunction with another test session, the order was counterbalanced.
Euthanasia, histology, and immunostaining
Mice were transcardially perfused under deep anesthesia with ketamine/xylazine (120 mg/kg/10 mg/kg, i.p., 1 ml/100 g), prior to decapitation and brain extraction and incubation in 4% paraformaldehyde. Brains were next transferred to 30% w/v sucrose solution and then sectioned on a Leica Microtome held at −15°C. Fifty-μm-thick coronal sections were mounted and cover slipped for histology, immunohistochemistry, or dendritic spine analysis.
For detection of viral vector spread, mCherry or GFP was visualized using a Nikon 4550s SMZ18 microscope or a Keyence BZ-X710 microscope. Immunohistochemistry for mCherry (mouse, 1:1,000, Takara; goat anti-mouse-Alexa594, 1:500, Invitrogen) was used to amplify the mCherry signal and thoroughly delineate viral vector spread, as needed. Viral vector spread for all mice is summarized in the first figure.
Free-floating immunohistochemistry was used to visualize c-Fos (rabbit, 1:1,000, Abcam; goat anti-rabbit-Alexa647, 1:500, Invitrogen). Z-stacks of five serial images were collected using a Keyence BZ-X710 microscope at 20× with a 0.8 µm step size. Fluorescence was quantified by a blinded rater who counted c-Fos+ cells within a defined ROI held constant across all images using NIH ImageJ. Two or three images per mouse contributed to analysis.
Dendritic spine imaging
Brains were collected and prepared as described above 30 min following the probe test. Unobstructed dendritic segments expressing YFP were imaged on a spinning disk confocal (VisiTech International) on a Leica DM 5500B microscope. Z-stacks were collected with a 100× 1.4NA objective using a 0.1 μm step size, sampling above and below the dendrite. After imaging, we confirmed at 10× that the image was collected from the BLA.
Dendritic spines were reconstructed in 3D using Imaris software using previously described methods (Swanger et al., 2011; Gourley et al., 2013) by a single blinded rater. Dendritic spines on Thy1+ excitatory BLA neurons were classified according to previously described parameters (Cahill et al., 2016; Zhang et al., 2019). If spine head diameter was greater than or equal to 0.3 μm, and the head:neck diameter ratio was greater than 1.1, the dendritic spine was classified as mushroom shaped. If the head:neck diameter ratio was <1.1, and the spine length:head diameter ratio was >2, the dendritic spine was classified as thin shaped. Spines that did not meet these criteria were classified as stubby. For each mouse, 4–6 independent basal dendritic segments, located 25–150 μm from the soma, were imaged and scored. Each mouse contributed one value/spine type, reflecting the average of its dendrites, to avoid over-representation of any given mouse.
Electrophysiology
Electrophysiological recording was conducted as described previously (Kietzman et al., 2022). Briefly, TRAP2 mice were killed, and brains were extracted and sectioned. Sections containing the BLA were incubated in 95%O2/5%CO2 oxygenated 32°C sectioning artificial cerebrospinal fluid (ACSF) for 1 h before transferring to a recording chamber mounted on the stage of a Leica microscope and perfused with recording ACSF. BLA neurons were identified by the expression of viral-induced mCherry using a Leica STP6000 epifluorescence microscope. Whole-cell patch-clamp recordings were performed using a MultiClamp 700B amplifier, a Digidata 1550 digitizer, and pClamp 10.6 software (Molecular Devices). Recording pipettes and patch solution were prepared as before (Kietzman et al., 2022). We used 0.3% biocytin to stain and localize patched BLA neurons. Membrane potential was current clamped at −60 mV for baseline recordings, and input resistance was monitored by observing membrane responses to hyperpolarization current injections (100–200 pA, 0.5 s). We recorded membrane potential depolarization before and during CNO (10 μM) application. Analysis was performed offline using Clampfit 10.6 (Molecular Devices).
Experimental design and statistical analysis
All statistics were performed using GraphPad Prism v9.0.0 and SPSS v.27. Repeated measures ANOVAs were used to compare food intake, response rates, and preference scores (before vs after social conditioning) in behavioral experiments. Response rates for the pellet that would later be paired with a social conspecific were also compared with response rates after social conditioning by repeated measures ANOVA. Pie charts were used to visualize preference: They represent the percentage of mice in each group that displayed a response preference, defined as ≥1 responses for the food pellet associated with a novel conspecific versus the pellet associated with a novel object.
In the absence of repeating measures, ANOVA or unpaired t tests was applied. Two-sample t tests were used to compare preference in the social interaction task and dendritic spine subtype densities. In electrophysiological experiments, a one-sample t test was applied to evaluate the degree of hyperpolarization against a baseline of 0.
Throughout, Tukey's post hoc tests or paired t tests were applied following significant interactions or main effects between >2 groups and are indicated in the figures. All comparisons were two-tailed, apart from the c-Fos comparison, given the specific a priori hypothesis for that experiment. p < 0.05 was considered significant. One trend at p = 0.055 is noted.
A small number of values were >2 standard deviations above the mean and considered outliers and excluded (Aguinis et al., 2013) from primary behavioral end points, resulting in the exclusion of two mice in the initial SIFC test, one mouse in the social discrimination test, one preference score in the penultimate figure, and two mice in the final behavioral experiment. Any mice with missed viral vector placement were also excluded. Sample sizes were based on prior experiments and power analyses; n's for each experiment are reported in the figure legends.
Results
PL→BLA projections are required for mice to favor conspecific-associated rewards
We first tested whether PL→BLA projections are necessary for rodents to use social information to guide their decision-making. Retrograde adeno-associated viral vectors expressing Cre+ green fluorescent protein (rAAV-Cre-GFP) were infused into the BLA, and Cre-dependent Gi-coupled Designer Receptor Exclusively Activated by Designer Drugs (Gi-DREADDs) were placed in the PL (Fig. 1a). Cre will permit Gi-DREADDs to be transcribed in PL→BLA projections, hyperpolarizing transduced projections in the presence of CNO (Kietzman et al., 2022). Control mice received a viral vector expressing a Cre-dependent fluorophore in the absence of DREADDs, and all mice received CNO.
PL→BLA projections are necessary for social information to guide behavior. a, Representative viral vector infusion in the PL and BLA. Red depicts Cre-dependent Gi-DREADDs + mCherry, and green depicts retrograde-Cre-GFP. White arrowhead indicates colocalization of Cre-dependent mCherry with GFP. mCherry is not present in the absence of Cre. Scale bars: 1 mm, 50 μm (inset, above) 200 μm (inset, below). Overlaid vectors depicting brain regions are from the Allen Brain Atlas (Lein et al., 2007). b, Schematic of the task. Mice were trained to nose poke for two food reinforcers. Next, mice were placed in a chamber with either a novel conspecific or object and with one of the two reinforcers. The next day, each mouse received the other stimulus and pellet. Then, choice was assessed in a probe test. CNO was administered at this time (yellow syringe). c, Response acquisition. Responses on both ports are collapsed for simplicity. d, Pellet consumption did not differ between the conspecific versus object phases of social conditioning. e, Following social conditioning, mice in the control groups responded more for the food associated with the social experience, while PL→BLA inactivation abolished this response. Instead, mice responded more for the pellet that they had earned most during training, which was paired with the novel object (nonsocial condition). f, The same data can be converted into response preferences (responses for the social/nonsocial pellet). Scores prior to social conditioning were ∼1, reflecting no preference, and preference developed only in the control group following social conditioning. Dashed line at one depicts no preference. Pie charts represent the percentage of mice in each group that preferentially responded for the food that was paired with the novel conspecific. g, Summary of viral vector infusions in the PL, or (h) BLA throughout this report, transposed onto images from the Mouse Brain Library (Rosen et al., 2000). Left, Images from Allen Brain Atlas (Lein et al., 2007) depict the PL (purple) or BLA (green). Distance from bregma is indicated. Bars and lines in (c) represent means (±SEM), while gray lines represent individual mice. *p < 0.05. n = 7–10 mice/group.
Mice were first trained in instrumental conditioning chambers to nose poke at two apertures for two distinct food pellets (Fig. 1b), with no systematic preference for one pellet over the other. Control and DREADDs groups did not differ in response acquisition, with mice increasing responding over time [main effect of day (F(6,90) = 26.880, p < 0.0001); no day*group interaction (F(6,90) = 1.354, p = 0.2421); no main effect of group (F(1,15) = 0.275, p = 0.6076); Fig. 1c].
Next, one pellet was paired with an opportunity to investigate a novel same-age, same-sex conspecific, while the other pellet was paired with a novel object. The flavor of pellet that was paired with a novel conspecific was the mouse's less earned pellet during training. The purpose was to bias against any individual pellet preferences (or side preferences within the chambers), in which case, social experiences would have to overturn said preferences for mice to display SIFC. Pellet consumption in the presence of the novel conspecific versus the novel object did not differ between conditions [no main effect of phase (F(1,15) = 1.341, p = 0.265); no group*phase interaction (F(1,15) = 0.042, p = 0.841); no main effect of group (F(1,15) = 2.052, p = 0.173); Fig. 1d] here or in any comparisons. The following day, CNO was delivered, and mice were returned to the instrumental conditioning chambers. Control mice responded more for the “social” pellet following social conditioning, while PL→BLA inactivation ablated SIFC, such that mice maintained their preferences from training, thus responding more for the pellet associated with the object [group*response interaction (F(2,26) = 5.162, p = 0.0129); no main effect of group (F(1,13) = 3.006, p = 0.1066); main effect of response (F(2,26) = 10.580, p = 0.0004); Fig. 1e].
Response preference scores were calculated as nose pokes for the “social” pellet over the “nonsocial” pellet and compared with preference for the same pellet prior to social conditioning. Values >1 reflect preference for the “social” pellet. Again, control mice preferred the “social” pellet, while PL→BLA inactivation ablated this preference [group*time interaction (F(1,15) = 4.982, p = 0.0413); main effect of group (F(1,15) = 5.948, p = 0.0276); no main effect of time (F(1,15) = 2.947, p = 0.1066); Fig. 1f]. Thus, PL→BLA projections are required for mice to favor a reward associated with a social conspecific.
PL→BLA projections are not required for sociality or social discrimination
We next tested whether PL→BLA projections were necessary for sociality, defined as attending to and prioritizing interactions with a conspecific. Viral vector infusion sites were consistent with Figure 1 and are summarized for the entire manuscript in that figure (Fig. 1g,h). In the three-chamber social interaction task, mice were habituated to a chamber and then allowed to investigate a novel conspecific or empty cup; typical mice will preferentially engage with the novel conspecific, reflecting sociality (Fig. 2a). Inactivation of PL→BLA projections did not affect sociality, indicated by comparable time in the chamber containing the conspecific versus that containing the empty cup between groups [main effect of chamber (F(1,19) = 46.649, p < 0.0001); no other main effects or interactions (Fs < 1); Fig. 2b]. Accordingly, groups generated similar preference scores, referring to the time spent in the chamber with the novel conspecific divided by the time spent in the chamber with the empty cup (t(19) = 0.1695, p = 0.8672; Fig. 2c).
PL→BLA projections are not necessary for sociality or social discrimination. a, Schematic of the three-chamber social interaction task. Following habituation, two chambers are revealed, containing either a novel conspecific or an empty cup. b, Silencing PL→BLA projections did not affect investigation time with the novel conspecific or (c) preference for the novel conspecific over an empty cup. d, Schematic of social discrimination phase, in which a novel conspecific was placed under the cup that was previously empty. e, Chemogenetic silencing of PL→BLA projections had no effect on the duration of time investigating the unfamiliar mouse or (f) the ratio of time spent investigating the unfamiliar mouse versus familiar mouse. Bars represent means; symbols/lines represent individual mice. **p < 0.001. n = 10–11 mice/group.
In the second phase of this task, a new, unfamiliar conspecific replaces the cup; typical mice will spend more time investigating an unfamiliar conspecific, discriminating it from the familiar conspecific (Fig. 2d). Inactivation of PL→BLA projections again did not affect the duration of time spent in a chamber with a familiar conspecific versus the unfamiliar conspecific [main effect of chamber (F(1,18) = 16.521, p = 0.0007); no other main effects or interactions (Fs < 1); Fig. 2e]. Groups generated similar preference scores, as well (t(18) = 0.2701, p = 0.7901; Fig. 2f). Thus, PL→BLA projections are not necessary for attending to a conspecific or for distinguishing between conspecifics.
SIFC alters the microarchitecture of BLA neurons
Excitatory BLA neurons can be visualized in mutant mice expressing Thy1-driven Yellow Fluorescent Protein (YFP; Feng et al., 2000). These mice were trained and tested as above and compared with a “no association” group, in which case, the pellets in the social conditioning phase were replaced with familiar chow (Fig. 3a). As such, these mice experienced handling, instrumental conditioning, interaction with a novel conspecific, etc., but not the opportunity to associate a social experience with food reinforcers. Dendrites were visualized following the choice test. Stubby- (t(7) = 3.180, p = 0.015) and mushroom-shaped (t(7) = 2.297, p = 0.055) dendritic spine densities were elevated in mice that formed social associations. No differences in thin-shaped spines were detected (t(7) = 0.9450, p = 0.376; Fig. 3b–d). Thus, SIFC is accompanied by greater densities of dendritic spines on excitatory BLA neurons.
Social association formation increases dendritic spine densities in the BLA. a, SIFC was conducted as above, or in the “no association” cohort, chow replaced the pellet reinforcers on each pairing day. b, Coronal sections from Thy1-YFP mice used to examine dendritic spine architecture on BLA neurons. Scale bars: 1 mm, 50 μm (inset). c, Representative dendrites, with their associated 3D reconstructions. Scale bar, 2 μm. d, The density of mushroom-shaped and stubby-shaped spines was elevated in the social association group. Bars represent means; symbols represent individual mice. *p ≤ 0.055. n = 4–5 mice/group.
BLA neurons stimulated by social interaction control later social decision-making behavior
Mice acquire information during social interactions that influences later choice—as shown in both the SIFC and social interaction tests here. We hypothesized that social interactions stimulate plasticity in BLA neurons that is necessary for later choice. We initially compared IEG levels in the BLA following interaction with a novel or familiar conspecific, relevant because novel conspecifics most effectively induce SIFC (Kietzman et al., 2022). Novel conspecifics triggered more c-Fos in the BLA than familiar conspecifics (t(35) = 1.829, p = 0.0380; Fig. 4a,b). We next utilized TRAP2 transgenic mice to induce Gi-DREADDs in c-Fos-expressing cells (Fig. 4c,d). CNO hyperpolarized transduced neurons (t(6) = 5.297, p = 0.0018; Fig. 4e,f), validating the Gi-DREADDs and allowing us to investigate cell function in decision-making behavior.
BLA neurons stimulated by social interaction control social, but not nonsocial, decision-making behaviors. a, c-Fos in the BLA was assessed following exposure to a familiar (fam.) or novel conspecific. b, A novel conspecific increased c-Fos puncta in the BLA. c, TRAP constructs allow for the “trapping,” or labeling, of cells expressing c-Fos. When 4-OHT (green) is administered, c-Fos+ cells produce Cre (white), which in this case induces mCherry-Gi-DREADDs (pink). d, Representative BLA neurons expressing mCherry. Scale bar, 100 μm. e, In a Gi-DREADD(+) neuron, bath application of CNO (10 μM) induced fast membrane hyperpolarization and decreased input resistance (top trace). In a Gi-DREADD(−) neuron, CNO did not induce changes of membrane potential and input resistance (bottom trace). f, CNO-induced hyperpolarization in Gi-DREADD(+) neurons. g, Response acquisition. h, Following social conditioning, vehicle-treated mice preferentially responded for the social pellet, as expected. However, CNO-mediated inactivation of neurons that had been stimulated by social experience ablated responses for the social-associated pellet, with mice responding as if they had never had a social experience. Dashed line at one depicts no preference. White and yellow syringes reflect vehicle and CNO injections. i, Next, mice were tested in a task assessing response flexibility based on reward probability rather than social experiences. Specifically, one trained response ceases to be reinforced, and typical mice will prefer the other, reinforced response. Following vehicle, mice preferred the more reward-predictive response, as expected. When the neurons previously stimulated by social experience were silenced via CNO, preference was unchanged. Dashed line at one depicts no preference. White and yellow syringes reflect vehicle and CNO injections. Bars and lines in g represent means, with SEMs indicated, while other lines represent individual mice. *p < 0.05, **p < 0.001. n = 6–7 mice/group.
Mice were trained to respond for food reinforcers as before (main effect of day, F(6,48) = 13.16, p < 0.0001; Fig. 4g). SIFC was tested twice, with either vehicle or CNO on-board during the choice test. Mice responded more for the “social” pellet following vehicle, as expected, but reverted back to previous preferences when BLA neurons previously stimulated by social interaction were inactivated [group*response interaction (F(2,24) = 4.382, p = 0.0239); no main effect of group (F(1,12) = 1.065, p = 0.3224); main effect of response (F(2,24) = 5.005, p = 0.0153)]. This pattern was further reflected in preference scores: Mice demonstrated social preference following vehicle, while inactivation of BLA neurons that had been previously stimulated by social interaction ablated this preference. Mice instead behaved as if they had never had a social experience (main effect of condition, F(2,10) = 7.172, p = 0.0117; Fig. 4h). Thus, BLA neurons stimulated by social experiences are necessary for mice to favor rewards associated with social experience.
To determine whether these social interaction-sensitive BLA neurons are particularly tuned to social contexts, we reinstated food-reinforced responding. Then, the predictive relationship between one action (e.g., the left nose poke response) and its outcome (e.g., the chocolate-flavored pellet) was degraded by providing the pellet noncontingently, or “for free.” Meanwhile, the other action remained reinforced. A typical mouse will prefer an action predictive of reward, a process that is BLA dependent (Balleine et al., 2003). Nevertheless, inactivation of BLA neurons that had been activated by a social interaction did not impact choice [main effect of condition F(2,10) = 6.910, p = 0.0130); Fig. 4i). These findings suggest that BLA neurons activated by social interaction are at least somewhat specialized to social choices.
We next performed the same series of experiments, except Gi-DREADDs were induced in the PL (Fig. 5a). Mice were trained to respond for food reinforcers as before (main effect of day, F(6,90) = 5.022, p = 0.0002; Fig. 5b). Mice demonstrated social preference during SIFC following vehicle, as expected, while inactivation of PL neurons that had been previously stimulated by social interaction ablated this preference (main effect of condition, F(2,24) = 11.878, p = 0.0003; Fig. 5c,d).
PL neurons stimulated by social interaction control social and nonsocial decision-making behaviors alike. a, We activity dependently induced mCherry-Gi-DREADDs in the PL following a social interaction. Representative mCherry in a coronal section. Scale bars: 1 mm, 0.5 mm (inset). b, Response acquisition. c, SIFC was tested twice, with vehicle or CNO, represented by white and yellow syringes. Then the contingency modification (CM) task was also conducted twice, with vehicle or CNO, represented by white and yellow syringes. d, Following social conditioning, vehicle-treated mice preferentially responded for the food associated with a social experience, as expected. However, inactivation of neurons that had been stimulated by social experience ablated preference, with mice responding as if they had never had a social experience. Dashed line at one depicts no preference. White and yellow syringes reflect vehicle and CNO injections. e, Inactivation also ablated preference based on action–outcome contingency in a task with no social content. Dashed line at one depicts no preference. Bars and lines in b represent means ±SEMs, and gray lines represent individual mice. *p < 0.05, **p < 0.001. n = 13.
We next inactivated these neurons during the nonsocial decision-making task. Inactivation of PL neurons that had been activated by a social interaction ablated sensitivity to reward contingency (main effect of condition, F(2,20) = 10.809, p = 0.0007; Fig. 5e). These findings suggest that PL neurons activated by social interaction are not specialized to social contexts, in that they control choice behavior with and without social content.
Discussion
Here, we first investigated PL→BLA projection neurons, inactivating them when mice were given the opportunity to choose rewards that were or were not previously associated with a conspecific. Projection inactivation blocked preference for a conspecific-associated food, but not social discrimination, suggesting that this projection retrieves memories regarding the association between a conspecific and an external reward, rather than memory for the conspecific itself. BLA neurons remodeled in response to social conditioning and those activated by a novel social experience were necessary for mice to make a “social choice,” but not a choice lacking social content. Meanwhile, PL neurons displayed no such specificity. The PL has long been associated with learning goal information (Hart et al., 2014; Woon et al., 2020). It may be that the PL conveys a generalized signal allowing mice to favor particular rewards based on any number of features (e.g., value, contingency, etc.), while units in the BLA process more specialized information unique to the reward, together supporting adaptive choice motivated by social information.
PL→BLA projections are necessary for SIFC
Here, we trained mice to respond for two food pellets of different flavors and then paired one with a novel conspecific, while the other was paired with a novel object. Typical mice will preferentially respond later for the “social” pellet—that is, the pellet that had been paired with the conspecific—evidence that the social experience influences later action (Kietzman et al., 2022). This SIFC behavior requires memory retrieval—recalling that one pellet is associated with a novel conspecific. Retrieval of fear-related memories involves postsynaptic strengthening of the PL→BLA pathway, suggesting that PL→BLA connections orchestrate associative memory retrieval in some contexts (Marek et al., 2013; Arruda-Carvalho and Clem, 2015). PL→BLA projections appear to similarly be required for memory retrieval of social associations, given that silencing these projections during the choice test here ablated social preference. The impact of PL→BLA projection inactivation could not obviously be attributable to effects on motivation, food preference, or memory for the social conspecific: Projection inactivation did not affect overall nose poke rates or innate food preference, since inactivation mice displayed preference for the object-associated pellet, which was the pellet that they earned most prior to social conditioning. It also did not appear to impact memory for a social conspecific in a traditional social discrimination task, though a caveat is that this task requires only short-term memory.
PL neurons project to inhibitory and excitatory neurons in the BLA (Hübner et al., 2014), including those that project back to the PFC (Little and Carter, 2013; McGarry and Carter, 2017). We recently demonstrated that BLA→PL projections are also required for SIFC and speculated that these projections might convey a salience signal (Kietzman et al., 2022). Conceivably, PL→BLA projections are involved in memory retrieval, which then triggers BLA projections back to the PL, reinforcing the salience of a given memory or action.
PL→BLA projections were required for SIFC, but not sociality, indicating that the ability of a social experience to motivate responding for an external reward is neurobiologically dissociable from the ability to attend to and prioritize a conspecific. Broadly, the PL controls social interaction (Yizhar et al., 2011; Lee et al., 2016), but different projections likely have distinct functions. And some appear to modulate behavior but are not necessary for that behavior. For instance, activation of PL→BLA projections decreases sociability (Huang et al., 2020), while inactivation has no impact (Fig. 2). Similarly, activation of PL→NAc projections also disrupts sociability, while their inhibition spares it (Murugan et al., 2017; Kietzman et al., 2022). An important topic for future investigations will be to identify factors (physiological, molecular, genetic) within specific corticolimbic circuits that control social information processing.
The formation of social associations modulates the structure of BLA neurons
Dendritic spines are the primary sites of excitatory plasticity in the brain. Transformation from immature to mature morphologies is linked with some forms of learning and memory (Whyte et al., 2019; Li et al., 2022), with mature spines characterized by a mushroom shape and synaptic AMPA receptors (Harris and Stevens, 1989; Matsuzaki et al., 2001; Noguchi et al., 2005). The capacity of cues to motivate food-seeking behavior in a task termed “pavlovian-to-instrumental transfer” (PIT) requires AMPA receptor activity in the BLA (Malvaez et al., 2015). SIFC resembles PIT in many ways, as mice use pavlovian associations between conspecifics and specific food outcomes to motivate later responding for that precise food. We found that mice performing SIFC had more mature, mushroom-type spines than those subject to all task parameters except for the selective association of a specific food with a conspecific. Stubby-type spines, which are immature and have the capacity for maturation, were also elevated. Neuronal structural changes in mice demonstrating SIFC may (1) support local plasticity in the BLA necessary for choice flexibility and (2) provide sites of input for excitatory PL projections.
Social interactions induce IEGs in the BLA (Ferri et al., 2016; Rogers et al., 2017). Comparisons in existing literature are typically between mice exposed to an object versus social conspecific. In contrast, we questioned whether novel versus familiar conspecifics stimulated BLA neurons, relevant because social conspecifics must be novel to induce SIFC, regardless of their sex, sexual history, or history of stressor exposure (Kietzman et al., 2022). Our interpretation is that the opportunity to gather novel social information, writ large, is salient and/or valuable, which is transferred to external rewards. According to this logic, we might anticipate higher IEG levels in the BLA following interaction with novel conspecifics, which was indeed the case. Interestingly, ∼20% of cells expressing c-Fos following interaction with a novel conspecific also expressed Thy1, the cell membrane protein used to drive YFP expression in our dendritic spine investigations (data not shown). These Thy1+ neurons are so-called “Fear-off” neurons necessary for conditioned fear extinction (Jasnow et al., 2013; McCullough et al., 2016), raising the possibility that their recruitment helps to mitigate a fear reaction to the potential threat of any given social interaction and thereby support investigation.
BLA ensembles stimulated by social interaction guide social but not nonsocial decisions
Next, we induced Gi-DREADDs in BLA neurons that were active following an interaction with a novel conspecific (a procedure referred to as “trapping”) and found that later inactivation of those neurons ablated SIFC. Thus, the same neuron population that is stimulated by a social experience is required for mice to later seek rewards associated with social interaction. The BLA is thought to form detailed, cue-specific representations to guide responding in both appetitive (Keefer and Petrovich, 2022) and aversive learning (Sengupta et al., 2018), so we next hypothesized that this function may be stimulus-specific, such that the neuronal ensembles stimulated by a social stimulus would be necessary for social, but not nonsocial, decision-making. Indeed, “trapped” BLA neurons were not necessary for mice to select actions based on reward likelihood—that is, to make choices in the absence of social influence—even though as-yet-undefined BLA populations are necessary for this kind of choice (Balleine et al., 2003). This profile contrasted neurons in the PL, which were necessary for mice to select actions based on social history and reward likelihood alike. This outcome aligns with the “common currency” hypothesis of social decision-making (Ruff and Fehr, 2014), which posits that at least some neuron ensembles use a common neural code to signal the values of specific options, regardless of social content.
Conclusions
Engrams are cell populations that (1) are activated by learning, (2) display enduring cellular changes, and (3) are reactivated by a part of the original stimulus for memory retrieval (Josselyn and Tonegawa, 2020). BLA neurons here appear to fulfill these criteria, in that (1) they are stimulated by social experiences that mice evidently remember because they guide later investigation and choice, (2) they undergo structural plasticity during SIFC, and (3) cells stimulated by social experience are necessary for memory retrieval in the SIFC task.
What might the putative engram contain? The BLA is thought to process several kinds of information, which can be generally categorized as reward prediction error, fear and reward, and salience and details of valued outcomes (Vázquez et al., 2022). We think the first two options are unlikely: We induced chemogenetic constructs in BLA neurons that were active in the presence of a conspecific and little else; thus, it seems unlikely that ensembles were responding to any specific prediction error-type signal. And we previously found that social interactions that motivate SIFC need not be affiliative (i.e., rewarding), per se, and even interactions that could be conceived as negative (i.e., fear-inducing) cause SIFC (Kietzman et al., 2022)—thus, it is also unlikely that the BLA supports SIFC simply because a given social interaction is rewarding or aversive.
The BLA is considered a site of convergence processing the sensory properties of salient stimuli (Hart et al., 2014). It is critical for adaptive behavior when detailed representations are needed. In instrumental conditioning scenarios, it is necessary to motivate responding for a specific food in the presence of a cue related to that food (like ordering a pizza after seeing an advertisement for pizza; Wassum, 2022), but not for cue-induced food seeking in general (Corbit and Balleine, 2005). In social domains, inactivation of the BLA causes rats to fail to prefer the urine of conspecifics over nonsocial olfactory stimuli, even though they can distinguish between the smells (Song et al., 2021)—evidence that the BLA processes socially salient sensory information that could impact later decision-making. We imagine that BLA ensembles perform a similar function here, encoding the sensory features of interacting with novel social partners and then retrieving this information when it is relevant to choice. Together with PL input, these memories then enable organisms to pursue conspecific-associated reward.
Data Availability
Data will be archived and available through the Emory University Dataverse.
Footnotes
Work in the S.L.G. lab is supported in part by the National Institutes of Health (NIH) F30 MH117878, R01 MH133740, and P50 MH100023. The Emory National Primate Research Center is supported by NIH P51 OD011132.
- Correspondence should be addressed to Shannon L. Gourley at shannon.l.gourley{at}emory.edu.
