Synaptic organization and associative learning-related activity of basolateral amygdala interneurons

SUMMARY

Basolateral amygdala (BLA) GABAergic interneurons (INs) are critical for associative learning; however, the comparative synaptic organization and learning-related activity of BLA IN types have not been systematically evaluated. Here, we show somatostatin (SOM) INs provide inhibition onto, and are excited by, local pyramidal cells (PCs), whereas vasoactive intestinal peptide (VIP) INs are driven by extrinsic afferents. Parvalbumin (PV) INs inhibit PCs and are activated by local and extrinsic inputs. Thus, SOM and VIP INs exhibit complementary roles in feedback and feedforward inhibition, respectively, while PV INs contribute to both microcircuit motifs. Functionally, each IN subtype exhibits unique activity patterns across fear and extinction learning, with SOM and VIP INs showing most divergent characteristics, and PV INs display an intermediate phenotype. Finally, SOM and PV INs dynamically track behavioral-state transitions across learning. These data provide insight into the synaptic microcircuit organization and activity of distinct BLA IN classes across aversive associative learning.

In brief

Báldi et al. show differential synaptic organization of BLA interneurons subserving participation in feedback and feedforward inhibition for SOM and VIP interneurons, respectively. Different interneuron types exhibited distinct patterns of experience-dependent plasticity ex vivo and in vivo across fear conditioning and extinction.

Graphical Abstract:

INTRODUCTION

Associative learning facilitates survival by optimizing behavioral responses to changing environmental contexts and cues.^1,2 These learned behavioral responses promote both the securing of resources needed for survival and reproduction and the avoidance of real or potential threats. Importantly, the expression of learned defensive responses to environmental cues must be balanced with the physiological extinction of these responses when the associations between environmental cues and danger no longer exist.³ Failure to appropriately extinguish conditioned defensive responses may compromise behaviors aimed at resource and reward acquisition. Extinction impairment also serves as a hallmark of several neuropsychiatric disorders, including posttraumatic stress disorder (PTSD).^4,5 Thus, elucidating the mechanisms governing aversive associative learning processes could provide valuable insight into the pathophysiology of trauma and stressor-related disorders.

The amygdala is well established as a key site for associative learning and synaptic plasticity within the circuits contributing to aversive associative learning and extinction.^6–8 Indeed, distinct basolateral amygdala (BLA) pyramidal cell (PC) populations defined by projection target or molecular phenotype have been implicated in both acquisition and extinction of conditioned fear behavior using rodent models.^9–12 More recently, the role of specific interneuron (IN) types has been investigated in aversive associative learning. Similar to other cortical-like structures, the BLA contains largely non-overlapping populations of GABAergic INs expressing somatostatin (SOM), parvalbumin (PV), and vasoactive intestinal peptide (VIP), which have been implicated in the regulation of associative learning.^13–16 For instance, using an auditory fear conditioning (FC) paradigm, investigators observe decreased SOM IN activity following conditioned stimulus (CS) presentation, contributing to enhanced cue-shock associations.¹⁷ Conversely, increased SOM IN activity is associated with reduced fear behavior and theta frequency oscillations in the presence of learned safety signals.¹⁸ VIP INs are activated by aversive unconditioned stimulus (US; shock) and strongly modulated by expectation, and they mediate adaptive disinhibitory gating supporting associative learning.¹⁹ Similarly, PV INs are US responsive in an expectancy-dependent manner across CS-US pairings,²⁰ although another study found suppression following US deliveries.¹⁷ Furthermore, enhanced CS-induced PV IN activity could promote fear learning and retrieval via the inhibition of SOM INs¹⁷ and disinhibition of PCs. From a synaptic connectivity perspective, VIP INs are reported to primarily inhibit the activity of PV and SOM INs in the BLA, while SOM and PV INs exert inhibition over PCs.¹⁹ These data highlight critical roles for BLA INs in the regulation of aversive associative learning.

Despite the aforementioned data, a variety of critical questions related to the roles and organization of BLA INs remain unanswered. These include how distinct INs are regulated by extrinsic long-range inputs and intrinsic BLA PCs, how INs gate extrinsic afferent-induced activation of BLA PCs, how different IN types represent US and CS information during fear acquisition and extinction, and how IN activity relates to the expression of defensive freezing behavior. Here, we show that distinct IN types contribute to different synaptic microcircuit motifs, with SOM and VIP INs participating in feedback and feedforward inhibition, respectively, and PV INs contributing to both motifs. In vivo, we identify divergent activity patterns for SOM and VIP INs across FC and extinction, with PV INs displaying intermediate phenotypes with properties of both SOM and VIP INs. Moreover, we find that SOM and PV INs dynamically track behavioral-state transitions between motion and defensive freezing across fear acquisition and extinction. These data provide insight into the comparative synaptic organization of distinct BLA IN types and their activity patterns across acquisition and extinction of aversive associative learning and defensive behavioral-state transitions.

RESULTS

Differential long-range targeting of BLA IN types

To begin to understand potential differential synaptic drive onto BLA INs, we first examined spontaneous excitatory postsynaptic currents (sEPSCs) onto BLA (but not lateral amygdala) PCs and SOM, PV, and VIP INs using whole-cell voltage-clamp recordings (Figures S1A and S1B). SOM and PV INs exhibited higher sEPSC frequency than PCs, and sEPSC amplitude was also larger in SOM and PV INs relative to PCs. Furthermore, varying differences in inhibitory transmission were observed between cell types (Figures S1C and S1D). Analysis of miniature excitatory postsynaptic currents (mEPSCs) and inhibitory postsynaptic currents (IPSCs) in the presence of tetrodotoxin citrate (TTX) revealed increases in mEPSC frequency onto PV INs relative to PCs and, similar to our spontaneous data, increases in the amplitude of mEPSCs in SOM and PV INs relative to PCs (Figures S1E–S1G). These data suggest differences in synaptic regulation of distinct IN types prompting a more detailed analysis of differences in synaptic excitation onto distinct IN types.

Using SOM, PV, and VIP reporter mice, we first expressed an excitatory opsin in areas implicated in associative learning^21–23 (dorsal medial prefrontal cortex [dmPFC], dorsal midline thalamus [DMT], or lateral entorhinal cortex [LEC]) and conducted whole-cell current-clamp recordings from pairs of identified INs and neighboring PCs (Figures 1A and 1B). Anatomical distribution of fibers from the dmPFC, DMT, and LEC, and different IN types within the BLA, are shown in Figure S2. Excitatory postsynaptic potentials (EPSPs) in PCs were monophasic, while SOM and PV INs often showed multiphasic responses with two identified peaks (Figure 1C). Specifically, upon dmPFC stimulation, all SOM and PV, while only 13% of VIP INs, showed a second long-latency response (Figures 1D and 1E). Importantly, long-latency responses were generally larger in SOM and PV INs than short-latency EPSPs (Figures 1D and 1F). Application of TTX+4AP confirmed short-latency first peaks in SOM, PV, and VIP INs to be monosynaptic as no significant change in amplitude was noted relative to baseline EPSP amplitudes, while the large-amplitude long-latency responses were abolished, supporting their putative disynaptic nature (see Figures 1C and 1F). Following DMT terminal stimulation, 100% of SOM, 57% of PV, and 40% of VIP INs showed putative disynaptic long-latency responses, with disynaptic responses being larger in SOM INs (Figures 1G and 1H). Short-latency EPSPs in SOM and PV INs remained comparable after TTX+4AP application with no second peaks present, validating a monosynaptic response (Figure 1I). Upon LEC stimulation, 91% of SOM, but only 20% of PV and 22% of VIP INs, exhibited disynaptic responses, and, again, these disynaptic peaks were significantly larger than the monosynaptic EPSPs in SOM INs and were eliminated after the application of TTX+4AP (Figures 1J–1L). These data demonstrate that dmPFC, DMT, and LEC afferents can excite all IN subtypes in the BLA and can trigger disynaptic responses to varying degrees depending on the input stimulated and the IN type. Importantly, SOM INs appear to receive more robust disynaptic excitation, both in terms of number of cells demonstrating disynaptic responses and the overall amplitude of these disynaptic EPSPs, relative to PV and VIP INs. In contrast, VIP INs generally show the smallest degree of disynaptic responses and receive overall the strongest monosynaptic inputs from the dmPFC and DMT. PV INs show an intermediate or mixed phenotype, resembling SOM INs upon dmPFC stimulation but VIP INs upon DMT stimulation, for instance. To further validate these findings, we conducted voltage-clamp recordings upon dmPFC-IN synapses and found similar results (Figure S3). Specifically, multiphasic responses were observed in SOM and PV INs but not VIP INs or PCs. After TTX-4AP application, only short-latency smaller-amplitude responses remained in SOM and PV INs. Latencies for the first and second peaks (relative to that of PC peak) were not statistically different for current-clamp and voltage-clamp experiments, with the exception of the second peak in SOM INs, which was longer using the voltage-clamp approach (p < 0.01); this may be related to greater sensitivity of voltage-clamp recording technique and of the analysis (see STAR Methods). These data align with our current-clamp recordings and provide additional support for the notion that BLA SOM and PV INs receive smaller monosynaptic excitation from the dmPFC and larger disynaptic excitation, while PCs and VIP INs show only monosynaptic responses to dmPFC stimulation.

Figure 1. Distinct long-range connectivity patterns across BLA IN types.

(A and B) Diagrams of experimental design. An excitatory opsin-encoding AAV was injected into the dmPFC, DMT, or LEC of SOM-, PV-, or VIPxAi14 mice to perform whole-cell current-clamp recordings of adjacent BLA PC and IN pairs.

(C) Representative traces from PCs, SOM, PV, and VIP INs upon single stimulation at baseline, after TTX, and after TTX+4AP drug applications.

(D–F) Data derived from single optogenetic dmPFC input stimulation. (D) x-y plot of EPSP peak amplitude and EPSP onset time differences relative to the paired PC of individual SOM, PV, and VIP INs. Data show SOM and PV INs have long latency responses while VIP cells generally do not. (E) Proportion of cells exhibiting long-latency disynaptic responses per IN types. (F) Quantification of observed putative monosynaptic (_1st), polysynaptic (_2nd) EPSP peaks, and confirmed monosynaptic (TTX+4AP) EPSP peaks for each IN type.

(G–I) Data parallel to (D)–(F) upon DMT terminal stimulation.

(J–L) Data parallel to (D)–(F) following LEC input stimulation. Sample sizes of PC-IN pairs (afferent: IN type [n = cell pairs, number of mice]). dmPFC: SOM (10,4), PV (5,3), VIP (15,6). DMT: SOM (5,2), PV (7,3), VIP (9,4). LEC: SOM (11,5), PV (5,2), VIP (9,4).

(F), (I), and (L) Statistical analysis and p values via repeated-measures two-way ANOVA followed by Tukey’s post hoc analysis (see Table S1). Data are represented as mean ± SEM. Scale bar in DIC image in (B), 250 μm.

Differential excitation of BLA INs from extrinsic and intrinsic inputs

Our data suggest SOM and PV INs receive substantial disynaptic excitation upon long-range afferent stimulation, relative to VIP INs. Since the only source of local excitation to BLA INs are BLA PCs, these findings imply that SOM and PV INs receive more robust internal excitation from local PCs than from extrinsic sites. To test this hypothesis, we examined the ability of extrinsic afferents, from the dmPFC, DMT, and LEC, as well as intrinsic inputs from local BLA PCs, to elicit action potential (AP) firing of BLA INs and neighboring PCs. To isolate extrinsic afferents, we injected an excitatory opsin-encoding AAV into the dmPFC, DMT, or LEC of SOM, PV, or VIP IN reporter mice and performed paired recordings of INs and adjacent PCs (Figure 2A). To isolate local intrinsic inputs to BLA INs, we injected an Flpo-recombinase encoding retrograde-AAV into the dmPFC or the central amygdala (CeL) of SOM-, PV-, or VIP-Ai14 mice, in addition to the combination of Cre-off/Flp-on ChRmine and DIO-eYFP-encoding viruses into the BLA to perform electrophysiological recordings from non-labeled PCs and eYFP-labeled INs (Figure 2B, see STAR Methods). The number of neurons showing AP firing in response to extrinsic and intrinsic inputs varied between IN populations, with SOM and PV INs activated by both extrinsic and intrinsic inputs, and VIP INs almost exclusively activated by extrinsic afferents (Figures 2C–2E). Quantitatively, we found that AP probabilities and the number of APs generated in PCs upon a single stimulation was generally greater following the activation of extrinsic over intrinsic inputs (Figures 2F and 2G). On the other hand, SOM INs did not show overall differences in AP probability between extrinsic and intrinsic inputs, although the number of APs evoked by a single intrinsic stimulation was significantly higher relative to extrinsic inputs (Figures 2H and 2I). Analogous findings were observed for PV INs (Figures 2J and 2K). It is important to note that extrinsic stimulation was still able to trigger APs in both SOM and PV INs but to a smaller degree than local BLA PC stimulation, which in many cases resulted in burst-like firing (see voltage trace in Figure 2B). In contrast to SOM and PV INs, but similar to PCs, AP probability and number in VIP INs were significantly lower upon intrinsic, relative to extrinsic, stimulation (Figures 2L and 2M). Interestingly, LEC inputs to PCs and VIP INs were less effective at driving APs than other extrinsic inputs. These data confirm that SOM and PV INs are more robustly activated by local BLA PCs than by extrinsic afferents, while VIP INs are almost exclusively activated by extrinsic (non-LEC) inputs and not by local BLA PCs. Furthermore, analysis of evoked AP/EPSP peak amplitude latency differences between extrinsic and intrinsic stimulation for different INs, relative to the paired PCs, revealed longer peak latencies for extrinsic than for intrinsic inputs for SOM but not for PV or VIP INs (Figures 2N–2P and traces in Figures 2A and 2B), in accordance with their predominantly large disynaptic response to long-range afferents described earlier.

Figure 2. Differential effects of external versus internal stimulation on action-potential firing of BLA IN types.

(A and B) Diagrams of experimental approaches to isolate external and internal afferents to BLA cell types. For external stimulation, excitatory opsins were injected in the dmPFC, DMT, or LEC, and dual patch-clamp recordings from paired PCs and INs were conducted for analysis of AP probability and number in response to a single light simulation. For internal stimulation, two populations of BLA PCs were transfected with excitatory opsin, one projecting to the CeL (PC_BLA->CeL; small gray cell) and the other projecting to the dmPFC (PC_BLA->dmPFC; small red cell) via use of retrograde AAV injected into the CeL or dmPFC. Dual patch-clamp recordings from unlabeled BLA PCs (large dark gray cell) and INs were conducted for analysis of AP probability and number in response to a single light simulation of either CeL-projecting or dmPFC-projecting BLA PCs.

(C–E) Proportion of cells exhibiting APs among the recorded pairs in response to extrinsic afferent (dmPFC, DMT, LEC) or local PC (PC_BLA->dmPFC, PC_BLA->CeL) stimulation. Data show proportion of cells with responses observed in INs only, INs + PCs, PCs only, or cells with neither response.

(F and G) AP probability and number triggered in PCs in response to external or internal stimulation.

(H and I) AP probability and number triggered in SOM INs in response to external or internal stimulation.

(J and K) AP probability and number triggered in PV INs in response to external or internal stimulation.

(L and M) AP probability and number triggered in VIP INs in response to external or internal stimulation.

(N–P) Time differences between peak EPSP/AP recorded from SOM (N), PV (O), and VIP (P) INs relative to the peak EPSP/AP of paired PCs upon stimulation of local PC_BLA->dmPFC or PC_BLA->CeL (top) or extrinsic (dmPFC, DMT, or LEC) inputs (bottom).

All data derived from PC-IN pairs (input: n = cell pairs, number of mice): SOM (dmPFC: 21, 12), (DMT: 12, 5), (LEC: 27, 11), (PC_BLA->dmPFC: 8, 3), (PC_BLA->CeL: 11, 4); PV (PL: 23, 11), (DMT: 12, 6), (LEC: 19, 7), (PC_BLA->dmPFC: 8, 4), (PC_BLA->CeL: 6, 3); VIP (dmPFC: 24, 8), (DMT: 9, 4), (LEC: 18, 6), (PC_BLA->dmPFC: 8, 3), (PC_BLA->CeL: 10, 4). (F)–(P) Statistical analysis and p values via two-way ANOVA followed by Sidak post hoc tests (see Table S1). Data are represented as mean (black or white line) with full data distributions.

Next, we examined effects of intrinsic BLA stimulation on distinct IN types and neighboring PCs using voltage-clamp recordings and analysis of EPSC/IPSC amplitudes elicited by stimulation of local dmPFC- or CeL-projecting BLA PCs (Figure S4A). Isolated EPSCs and IPSCs were recorded from the same neurons at −70 mV and +10 mV, respectively, and application of ionotropic glutamate receptor blockers additionally confirmed the disynaptic nature of IPSCs (Figure S4B). Upon stimulation of either CeL- or dmPFC-projecting BLA PCs, SOM INs showed significantly smaller putative disynaptic IPSC but comparable EPSC amplitudes and an overall increase in the excitation/inhibition (E/I) ratio relative to PCs (Figures S4C and S4D). PV INs showed both greater EPSC amplitude and smaller putative disynaptic IPSC amplitude in response to CeL- and dmPFC-projecting PC stimulation and, like SOM INs, exhibited overall higher E/I ratios relative to PCs (Figures S4E and S4F). In contrast, although VIP INs also showed significantly lower IPSC amplitude in comparison to BLA PCs, the E/I ratio remained comparable to PCs, likely due to concomitantly lower EPSC amplitudes (Figures S4G and S4H). Taken together, these voltage-clamp and current-clamp experiments suggest SOM and PV INs are strongly activated by intrinsic BLA inputs and thus likely participate in lateral and/or feedback inhibition of BLA PCs, while VIP INs get almost exclusively extrinsic afferents and thus likely participate in feedforward inhibition. All IN types receive significantly less feedback inhibition upon intrinsic BLA PC stimulation compared to adjacent PCs.

Overall, these data indicate that SOM and PV INs are preferentially suited for feedback inhibition. However, their contributions to feedforward inhibition cannot be excluded given both extrinsic and intrinsic inputs can generate APs, albeit to different degrees, in each IN. To explicitly test a role for SOM and PV INs in feedforward inhibition, we recorded from BLA PCs while optically stimulating dmPFC inputs (ChrimsonR, 617-nm light) during Cre-dependent kappa-opioid receptor DREADD (KORD)-mediated inhibition of SOM or PV INs in addition to Cre-dependent ChR2 expression to validate the strength of KORD inhibition (Figure S5A). Bath application of Salvinorin B caused suppression of direct SOM- and PV-mediated (455-nm light) GABAergic IPSCs onto PCs as expected but only inhibited feedforward IPSCs elicited by dmPFC stimulation in PV-Ai14 mice (Figures S5B and S5C). Indeed, a direct comparison of the efficacy of Salvinorin B on feedforward inhibition, relative to direct IN to PC inhibition, revealed a significantly greater effect for PV over SOM INs (Figure S5D). These data further confirm that PV INs participate in feedforward inhibition, while SOM INs do not.

BLA IN types exert differential inhibitory control over PCs

Our data suggest distinct IN types differentially participate in microcircuit motifs to regulate the coordinated activity of the BLA; however, the degree to which these INs gate extrinsic afferent-induced activation of BLA PCs remains unclear. To address this, we first determined the IN classes providing direct inhibition to BLA PCs via the expression of ChR2 in SOM, PV, or VIP INs and analysis of evoked inhibitory postsynaptic potentials (IPSPs) in BLA PCs (Figure S6A). The majority of PC IPSPs were GABA_A receptor dependent (Figure S6B) and showed voltage dependence with reversal potentials of −83.5 mV for PV and −96.2 mV for SOM IN activation (Figures S6C and S6D). Average IPSP amplitudes recorded from PCs were highest for SOM > PV INs with VIP activation showing minimal responses even at depolarized potentials (Figures S6C and S6E). Onset times were not different between SOM and PV cells, but decay time was prolonged in SOM over PV cells (Figures S6F and S6G). These data indicate that SOM and PV INs inhibit BLA PCs, while VIP INs generally do not, consistent with previous results.¹⁹

We next wanted to examine the gating efficacy and temporal determinants of SOM and PV IN-evoked inhibition on long-range afferent-induced excitation received by BLA PCs. To this end, we expressed ChrimsonR in the dmPFC, DMT, or LEC of SOM- and PV-Ai14 mice, to allow for red-light stimulation of long-range afferents to the BLA. Additionally, DIO-ChR2 was injected into the BLA to allow for blue-light activation of different IN types (see Figures 3A and 3B). Using this approach, we found that simultaneous (ΔT = 0 ms), but not separated (ΔT = 300 ms), blue- and red-light stimulation suppressed afferent stimulation-induced EPSP amplitude (see Figures 3C and 3D for example of EPSP suppression for ΔT = 0 ms and ΔT = 300 ms). Upon dmPFC stimulation, SOM and PV IN activation reduced the EPSP amplitudes, with SOM INs exerting greater suppression than PV cells (Figures 3E and 3F). AP probability was also suppressed by SOM and PV activation following suprathreshold dmPFC stimulation (Figures 3G and 3H). Similarly, in response to DMT stimulation, both SOM and PV activation were able to reduce the EPSP amplitudes, with SOM INs exerting greater suppression than PV INs (Figures 3I and 3J). AP probability was likewise suppressed by SOM and PV activation in the presence of suprathreshold DMT stimulation (Figures 3K and 3L). In response to LEC stimulation, SOM and PV activation reached the biggest reduction in PC EPSP amplitudes with similar effectiveness (Figures 3M and 3N). PCs AP probability was suppressed by both IN types paired with suprathreshold LEC stimulation (Figures 3O and 3P). Additionally, post hoc analysis revealed that significant differences between SOM- and PV-induced suppression of EPSPs occurred at longer latencies, specifically at 100–200 ms ahead of DMT and LEC afferent stimulation. This is consistent with our findings showing similar peak times but longer decay times of SOM- relative to PV-evoked IPSPs on PCs (Figures S6F and S6G). Control experiments confirmed that dual-opsin-based approaches were not contaminated by light cross-over activation at light intensities used, and that, in the absence of ChR2 expression in INs, ChrismonR-triggered afferent EPSPs were not suppressed by blue-light stimulation (Figures S6H–S6L). These data indicate that SOM and PV INs can inhibit BLA PC activity and gate sub- and suprathreshold activation of BLA PCs from long-range afferents over long timescales.

Figure 3. Suppression of extrinsic afferent excitation-induced activation of BLA PCs by different IN types.

(A and B) Diagram of experimental design. The excitatory opsin ChrimsonR-encoding AAV was injected into the dmPFC, DMT, or LEC and a Cre-dependent ChR2 was injected into the BLA of SOM-, PV-, or VIPxAi14 mice. Whole-cell current-clamp recordings were obtained from BLA PCs.

(C and D) Representative traces of dual-opsin experiments showing extrinsic dmPFC afferent-induced EPSP triggered by red light, PV activation-induced IPSPs triggered by blue light, and mixed responses to blue- and red-light combinations given simultaneously (ΔT = 0 ms or 300 ms apart, ΔT = 300 ms). Data show that simultaneous activation of dmPFC inputs and PV INs strongly suppresses EPSP amplitude (vertical arrow), which is not seen when stimulation times are separated by 300 ms.

(E and F) Quantification of effects of SOM and PV IN activation on dmPFC-triggered postsynaptic potentials (PSPs) as a function of blue-red-light stimulation interval relative to dmPFC stimulation, and maximal PSP inhibition at ΔT = 0 ms for SOM and PV INs.

(G and H) Quantification of effects of SOM and PV IN activation on dmPFC-triggered AP probability in response to suprathreshold red-light activation as a function of blue-red-light stimulation interval relative to dmPFC stimulation, and maximal AP probability differences between ΔT = 300 ms (used as control [Co]) and ΔT = 0 ms for SOM and PV INs.

(I–L) Same as (E)–(H) for DMT-induced PSP and APs in BLA PCs.

(M–P) Same as (E)–(H) for LEC-induced PSP and APs in BLA PCs.

Sample sizes (n = cells, number of mice); (E and F) SOM (12,2), PV (10,2); (G and H) SOM (11,2), PV (8,2); (I and J) SOM (14,4), PV (9,2); (K and L) SOM (11,2), PV (7,2); (M and N) SOM (13,3), PV (14,3); (O and P), SOM (6,2), PV (9,3). Statistical analysis and p values for (E), (G), (I), (K), (M), and (O) via two-way ANOVA followed by Sidak post hoc test comparing IN types; (F), (J), and (N) via unpaired t test; (H), (L), and (P) via two-way ANOVA followed by Sidak post hoc test comparing AP probabilities under control (Co) and ΔT = 0 ms conditions (see Table S1). Data are represented as mean (black lines) with full data distributions.

BLA IN types exhibit distinct adaptations in intrinsic excitability across auditory cue FC

Given the substantial differences observed in extrinsic and intrinsic connectivity between specific BLA cell types as well as the diverse inhibitory effects of these INs onto PCs, we reasoned that different IN classes would display distinct adaptations during associative learning critically dependent on BLA function.^13–16 To this end, we examined the intrinsic excitability of SOM, PV, and VIP INs across FC. Three groups of mice were used: a basal group subjected to tone-only conditioning and recall, a fear group exposed to six tone-shock pairings, and fear recall during which five recall tones were presented. An extinction group went through conditioning and 2 days of extinction, where 14 tones were presented on each day, and a recall day during which eight tones were presented. Each group was scarified 30 min after their last behavioral testing session and used for slice electrophysiological recordings. No differences were found across any intrinsic membrane property or excitability in SOM neurons between groups (Figures S7A–S7F). PV neurons showed increased intrinsic excitability after extinction relative to basal and fear-conditioned groups without differences in membrane properties (Figures S7G–S7K). For VIP neurons, we found a more depolarized membrane potential, increased rheobase, and decreased intrinsic excitability after fear-conditioned relative to basal and extinction groups (Figures S7L–S7P). These data suggest cell-type-specific adaptations are detectable ex vivo across FC and extinction and reveal experience-dependent plasticity of IN types in the BLA, most notably, a reduction in excitability of VIP neurons after FC, which reverts after extinction training.

BLA IN types exhibit unique activity patterns across auditory cue FC

While our ex vivo data suggest dynamic changes in intrinsic excitability across FC, whether these changes reflect adaptations in the in vivo activation dynamics of INs is not known. Moreover, adaptations may occur in vivo that are not detectable in the ex vivo preparation. Therefore, we next examined the activity of INs across auditory cue FC using in vivo fiber-photometry approaches. Mice were injected with Cre-dependent GCaMP7f encoding AAV followed by optical-fiber implantation into the BLA to measure the activity of SOM, PV, or VIP INs during an auditory cue FC paradigm (Figures 4A, 4B, and S8A). During conditioning, SOM INs showed progressive increases in both CS+- and US-evoked responses across conditioning (Figures 4C and 4D). Comparisons of peak Z score and area under the curve (AUC) for onset (0–5 s) and duration (5–29 s) of CS+ presentations showed progressive increases in activity between the first and last CS-US pairing, with no signals detectable following the first tone (Figure 4E). US-evoked responses exhibited a similar sensitizing pattern with higher persistent activity, strikingly lasting up to 30 s, following the brief 1-s shock (Figures 4C, 4D, 4F, and 4G). In contrast to SOM neurons, PV INs displayed consistent CS+-evoked signals across conditioning trials (Figures 4H–4J). PV US responses, on the contrary, showed a significant reduction in peak Z score and AUC from early to late presentations (Figure 4K), mostly limited to the 5 s following shock onset (Figure 4L). VIP INs showed minimal CS+-related activity (Figures 4M–4O), but their robust US response followed a similar pattern to PV INs: decaying phasic activity across conditioning trials (Figures 4P and 4Q). Averaged data across all conditioning trials from individual male and female mice support the reliability of responses across subjects (Figures S8B–S8E). YFP-expressing control mice showed no consistent CS+ or US-evoked activity across conditioning; although peak responses to shock 5 were greater than shock 1 on PV YFP mice, the AUCs were not different and showed downward deflections in contrast to GCaMP7f mice (Figures S8F–S8L). These data indicate differential fear-conditioning-associated changes in IN activity across learning, with SOM INs displaying progressive and long-lasting increases in CS+- and US-evoked activity across conditioning. In contrast, VIP INs showed primarily sharp, short-lived, US-evoked signals that habituate across CS-US pairings. Interestingly, PV INs exhibit some features of both VIP (sharp habituating US response) and SOM (late CS+ response) INs.

Figure 4. Activity of distinct IN types across FC.

(A) Schematic diagram of experimental design. In vivo calcium levels were monitored via Cre-dependent expression of GCaMP7f using bulk fiber photometry recordings of SOM, PV, and VIP BLA INs during FC.

(B) Experimental design: habituation day consisted of eight CS− presentations, conditioning consisted of five CS+-US (shock 1 s, 0.45 mA) pairings, and extinction days consisted of four CS− presentations and 14 CS+ presentations without shock. Represented data are derived from the FC behavior.

(C) Freezing data and heatmap of averaged Z score values across all mice for the five CS-US pairings for SOM INs.

(D) Average Z score traces of SOM activity during CS-US pairing number 1 (left) and 5 (right). Data were averaged across mice per the number of CS+-shock presentation.

(E) Peak Z score and AUC early (0–5 s) and late (5–29 s) responses to CS+ no. 1 and 5.

(F) Peak Z score and AUC responses to US (shock) no. 1 and 5.

(G) AUC distribution of neuronal activity following shock presentations over 30 s averaged into 5-s bins in SOM INs.

(H–L) Same as (C)–(G) for PV INs.

(M–Q) Same as (C)–(G) for VIP INs.

Sample size (mice; sex): SOM (n = 11; 6 male [M]/5 female [F]), PV (n = 13; 7M/6F), VIP (n = 7; 3M/4F). Statistical analysis and p values via two-tailed paired t test throughout. Data are represented as mean (bar height) with all individual data points superimposed ([E and F], [J and K], and [O and P]), and mean ± SEM for ([C and D], [H and I], [M and N], and [G, L, and Q]).

BLA IN subtypes exhibit unique activity patterns across conditioned fear extinction

During extinction training, another set of unique activity patterns emerged among the IN types (Figures 5A and 5B). SOM INs responded with sharp, phasic (0–3 s) increases in activity following CS+ presentations throughout extinction days (early day 1 vs. late day 5) (Figures 5C and 5D). In contrast, a slower, long-lasting reduction in calcium signal was detected over the second half of CS+ presentation (15–30 s) during the early phase of extinction, which attenuated with extinction learning (Figures 5C and 5D). Additionally, upon CS+ offset, another sharp phasic response was detected in SOM INs during the early, but not late, phase of extinction (Figures 5C inset and 5E). Interestingly, the AUC analysis of slow reductions (15–30 s) showed indirect correlations with the overall CS+-induced freezing values, with larger reductions at higher freezing levels and smaller reductions at lower freezing levels. (Figure 5F). The phasic CS+ offset-dependent increases, on the other hand, showed direct correlations with the tone-induced freezing and completely attenuated with extinction training, when the freezing lessened (Figure 5G). PV INs exhibited similar activity patterns to SOM neurons on early extinction day for both the phasic CS+ onset rise and the late long-lasting decay in calcium signals during tone presentations (Figure 5H). Both, however, attenuated by late extinction trials (Figure 5I). As in SOM INs, late CS+ decline in PV IN activity indirectly correlated with freezing levels, signifying increased PV neuronal activity with the development of extinction learning compared to fear state (Figure 5K). CS+ offset response, although small, revealed a direct correlation with freezing, both decreased by late extinction (Figure 5L). Analysis of VIP INs demonstrated solely a phasic increase upon CS+ presentation that weakened with the development of extinction (Figures 5M and 5N), and CS+ offset yielded to no changes in calcium activity in these cells (Figure 5O). CS+ onset peak showed direct correlation with freezing level, both attenuated with extinction (Figure 5P), and lack of late-onset CS+ shift (15–30 s) or CS+ offset response meant no correlations with freezing for VIP INs (Figure 5Q). These data indicate differential and dynamic changes to CS+ presentation across extinction training in SOM, PV, and VIP INs. Analogous to the distinction in activity patterns during FC, SOM and VIP INs displayed markedly distinct activity patterns and correlations with freezing levels, whereas PV INs showed an intermediate phenotype with properties of both SOM (late CS+ slow reductions) and VIP INs (habituating CS+ onset response).

Figure 5. Activity of distinct IN types across fear extinction.

(A and B) Schematic diagram of experimental design. In vivo calcium levels were monitored via Cre-dependent GCaMP7f expression using fiber photometry recordings of SOM, PV, and VIP INs during fear extinction days when mice were presented with four CS− and 14 CS+ tones. Only data related to CS+ presentation are shown.

(C) Freezing reduction across extinction (day 1 early vs. day 5 late) and average Z score traces of SOM IN activity during early (day 1 early; dark line) and late (day 5 late, white line) extinction days upon CS+ onset and offset (insert). Z score traces of CS+ onset and offset were first averaged per session, then across mice; traces represent mean ± SEM.

(D) AUC following CS+ presentation (0–3 s and 15–30 s).

(E) AUC (0–3 s) for CS+ offset.

(F and G) (F) Linear regression for CS+ AUC (0–3s) vs. freezing, CS+ AUC (15–30 s) vs. freezing, and (G) CS+ offset AUC (0–3 s) vs. freezing.

(H–L) Same as (C)–(G) for PV INs.

(M–Q) Same as (C)–(G) for VIP INs. For correlations, only data from day 1 (early) and day 5 (late) extinction are included.

Sample size (mice; sex): SOM (n = 11; 6M/5F), PV (n = 13; 7M/6F), VIP (n = 7; 3M/4F). Statistical analysis via two-tailed paired t test (D, E, I–J, N, and O) and via Pearson r correlation and simple linear regression (F, G, K, L, P, and Q). Data are represented as mean (bar height) with all individual data points superimposed and mean ± SEM for GCaMP traces. Linear regressions show 95% confidence intervals in shaded color.

BLA IN subtypes show differential responsivity to behavioral-state transitions

While our analysis above demonstrates discrete changes in calcium activity accompanying CS+ presentation and reveals correlations to overall CS+ freezing levels, the close relationship between IN activity and freezing behavior remained unclear in that evaluation, as mice exhibit multiple transitions between freezing and non-freezing states during the 30-s CS+ presentations and freezing in between tones (inter-tone-intervals). To address the question of whether IN activity is directly related to behavioral transitions in addition to sensory cue presentations, we measured SOM, PV, and VIP IN activity specifically during freezing and moving states, as well as at the transitions from moving to freezing and freezing to moving states across our FC and extinction protocol (Figure 6A). Our analysis excluded the first 5 s after CS+ onset, shock onset, and CS+ offset periods to reduce confounding from sensory signals (see above) and distinguished between tone and inter-tone intervals.

Figure 6. Behavioral-state-transition-related activity of distinct IN types during CS presentation.

(A) Diagram of task days during which behavioral-state-transition data were analyzed (during CS− presentation on habituation day and CS+ presentation during conditioning and extinction). We excluded analysis of freezing-associated calcium signals occurring during the first 5 s of tone onset or shock onset to avoid confounds caused by sensory stimuli onset.

(B) Averaged ΔF/F values during bouts of motion (M) and freezing (F) on habituation (H), FC, and early (E_early) and late extinction (E_late) days recorded from SOM (brown), PV (green), and VIP (pink) INs.

(C) Averaged ΔF/F values during bouts of motion during CS presentations compared across behavioral days for SOM, PV, and VIP populations.

(D) Averaged ΔF/F values during bouts of freezing compared across behavioral days for SOM, PV, and VIP populations.

(E) Diagram of behavioral-state-transition analysis depicting dissection of transitions where mice start moving (sM) or start freezing (sF). Data analysis for habituation day was omitted due to low number of transitions per mouse during habituation.

(F) AUC (0–3 s) values of sM and sF transitions during CS+ presentations compared within IN types on separate days for SOM, PV, and VIP populations.

(G) Averaged Z score traces during behavioral-state transitions for SOM, PV, and VIP populations on FC and early and late extinction days. Z score traces of transitions were first averaged per session and then across mice; traces represent mean ± SEM. Data were baselined to 1 s prior to behavioral-state transition as shown in black.

(H) AUC (0–3 s) values of sM transitions during CS+ presentations compared across task days for SOM, PV, and VIP populations, separately.

(I) AUC (0–3 s) values following sF during CS+ presentations compared across days for SOM, PV, and VIP INs.

Sample size (n = mice; sex); SOM (n = 11; 6M/5F), PV (n = 13; 7M/6F), VIP (n = 7; 3M/4F).(B and G) p values and analysis via paired t test comparing M to F and sM to sF for each IN type; (C, D, H, and I) analysis via one-way ANOVA followed by Tukey post hoc test comparison across task days (see Table S1). Data are represented as mean (bar height) with all individual data points superimposed for (B) and (F) and mean ± SEM for (C), (D), and (G–I).

We started by quantifying the calcium-signal dynamics of distinct IN populations as a function of motion and freezing during habituation (H), FC, and extinction days (E_early and E_late) during CS+ presentation. The additional inclusion of a habituation day dataset allowed us to assess data from a full range of fear states (low-high-low). ΔF/F values were averaged during motion and freezing behaviors across sessions and compared within mice. Analysis revealed a divergent pattern in calcium signals with generally higher ΔF/F values during movement and lower values during freezing, depending on the day of behavioral task and cell type. Specifically, on habituation day, only SOM INs exhibited lower ΔF/F values during freezing relative to motion throughout CS− presentations (an auditory stimulus only present in the contexts of H and E days) (Figure 6B). On the FC day, remarkably, each IN type revealed a split between overall calcium signals derived from freezing and motion bouts during CS+ presentations; albeit small for VIP, the split remained notable in SOM and PV INs on the following day (E_early), when mice expressed the highest level of fear (Figure 6B). After the development of fear extinction (E_late) SOM INs (and to a lesser degree VIP INs) still showed attenuated signals during freezing relative to motion (Figure 6B). Direct comparisons of movement-derived ΔF/F values across the days within IN classes revealed increased SOM IN activity on the day of highest fear expression (E_early), relative to other task days, while VIP IN activity indicated slight separation between FC (higher) and late extinction day (lower) (Figure 6C). Freezing-associated SOM IN calcium signals showed comparable attenuations across days, whereas PV INs showed attenuations on days associated with high fear expression (FC and E_early) relative to days with low fear expression (H and E_late) (Figure 6D). VIP IN calcium signals displayed no distinctions across sessions (Figure 6D).

To explore whether these changes in calcium signals during freezing and movement were associated with discrete time-locked events, we examined the transitions between movement and freezing behaviors during conditioning and extinction (Figure 6E). We analyzed events with a minimum of 1-s stable behavioral states both before and after transitions to evaluate well-defined transitions, and, due to minimal freezing expression, the habituation day dataset could not be assessed. This analysis revealed a sharp increase in calcium activity accompanying the freezing-to-movement transitions, whereas a sharp decrease coincided with the movement-to-freezing transitions. These phenotypes were most notable in SOM INs followed by PV INs, with minimal distinctions in VIP cells (Figure 6F). AUC (0–3 s) values of Z scored calcium traces were used for quantification at behavioral transitions and averaged across each session per mouse during CS+ or between CS+ presentations (no-tone periods). AUC values showed marked separations between transition types (start moving [sM]/start freezing [sF]) on each day for SOM neurons, on FC and fear recall day (E_early) for PV cells, and only on fear recall (E_early) for VIP cells (Figure 6G). Direct comparisons of freezing to moving (sM) transitions across days within IN classes revealed larger increase in SOM neuronal activity on early vs. late extinction day, higher activities on both FC and early extinction vs. late extinction for PV cells, and no significant changes in VIP neurons (Figure 6H). Comparisons of moving to freezing (sF) transitions showed separation only between FC and extinguished fear association (E_late) days for PV cells (Figure 6I). Similar effects were observed in all measures between CS+ presentations, i.e., during no tone periods (Figures S9B–S9D and S9G–S9I). YFP-expressing control mice displayed minimal changes in activity during times of motion and freezing with only a very small reduction in SOM activity during early extinction noted (Figure S9A) and with very small changes at transition points in the opposite direction of GCaMP7f mice (Figure S9F). Taken together, these data illustrate that both SOM and PV INs undergo marked behavioral-state-transition-related activity changes with sharp increase and decrease at the start of movement and freezing, respectively. Interestingly, these signals were present on each day in SOM neurons but only emerged on days with heightened fear expression in PV neurons (and to a much lesser extent in VIP neurons). Additionally, these behavioral-state-associated activity changes were independent of CS+ presentation, underscoring their independence from sensory-driven activity.

DISCUSSION

While the amygdala is a well-established hub critical for the acquisition and extinction of conditioned fear responses,^1,3,5,7 a comprehensive analysis of how specific BLA GABAergic INs contribute to different circuit motifs and the functional properties of IN classes across conditioned fear acquisition and extinction learning is lacking. Moreover, learning-associated plasticity within the IN classes and their potential differential contributions to the representation of conditioned freezing behavior and related behavioral-state transitions has not been explored. Here, we address these open questions via investigation of functional synaptic connectivity and microcircuit organization of BLA SOM, PV, and VIP INs and their responses to sensory cues and behavioral-state transitions across fear acquisition and extinction.

Our electrophysiological analysis of IN synaptic connectivity and microcircuit organization support a model whereby SOM INs prominently participate in feedback inhibition of BLA PCs, VIP INs contribute primarily to feedforward disinhibition of PCs, and PV INs engage in both feedforward and feedback inhibition in the BLA (Figure S10). This model is supported by multiple pieces of convergent data. First, we observed using current-clamp experiments direct monosynaptic and putative disynaptic excitation of SOM INs upon stimulation of all extrinsic inputs tested, whereas VIP INs received almost exclusively monosynaptic long-range excitation, and PV INs displayed both strong monosynaptic and disynaptic excitation depending on the afferent input. These results were confirmed using voltage-clamp experiments examining dmPFC inputs to BLA IN types. Second, direct analysis of the probability and number of APs generated in response to intrinsic versus extrinsic stimulation shows that, while both intrinsic and extrinsic sources can readily trigger APs in SOM and PV neurons, activation of local PCs evokes a higher number of APs in both IN types, and SOM neurons fire after PCs in response to extrinsic sources. VIP INs could not be triggered by intrinsic afferents, and their firing probabilities showed a similar profile to those of PCs. Third, voltage-clamp studies revealed that both SOM and PV INs exhibit higher E/I ratios upon stimulation of local intrinsic BLA PCs (relative to paired PCs), but VIP INs do not. Fourth, unlike VIP INs, SOM and PV cells provide effective and long-lasting inhibition of PCs. Finally, our chemogenetic data indicate that PV, but not SOM, INs participate in feedforward inhibition of BLA principal cells driven by extrinsic afferents, supporting the heterogeneity within the PV population in supporting feedback and feedforward inhibition.

This model aligns with several previous findings. Monosynaptic rabies tracing suggests robust local innervations of SOM INs,¹⁹ supporting our data that SOM INs preferentially participate in feedback inhibition of BLA PCs, similar to their role in the neocortex.^24,25 With regard to BLA PV INs, research in cortical areas reinforces the roles for PV INs in both feedback and feedforward inhibition,^24,25 and anatomical and functional BLA studies have revealed strong excitatory inputs from local PCs on PV INs,^26–29 supporting the mixed phenotype we observed depending on the examined terminals. Anatomical data have also revealed BLA VIP INs receiving VGluT2 (likely of thalamic origin) and VGluT1 cortical synaptic contacts,³⁰ consistent with our electrophysiology data showing prominent monosynaptic excitation by extrinsic afferents only. Finally, previous studies corroborate our findings that, while SOM and PV INs can provide powerful inhibition to BLA PCs,^19,27 VIP INs generally do not.^19,30 This suggests that VIP INs are primarily involved in feedforward inhibition onto other INs,¹⁹ mediating feedforward disinhibition of PCs, as similarly described in other cortical-like structures.²⁵

In partial contrast to our results, a recent study suggested that a fast-spiking subtype of SOM, but not PV, INs mediate feedforward inhibition onto BLA PCs upon LEC stimulation.²⁸ We found that, although SOM INs could be activated by LEC inputs, disynaptic responses were more robust. These discrepancies could be due to the methodological differences. For example, our data derive from optical stimulation of the fibers originating from the mid-posterior LEC, while the earlier study utilized electrical stimulation in horizontal slices. Specific recording sites within the basolateral amygdala complex could also account for these differing results. In support of this contention, another study using electrical stimulation in coronal sections found divergent roles for PV INs in feedforward inhibition in the LA relative to BLA.³¹ This is consistent with our finding that PV INs show varying anatomical substrates for feedforward inhibition depending on the afferent source. Furthermore, a recent study reported prelimbic-induced increase in BLA SOM IN activity contributing to the discrimination of non-threatening stimuli.¹⁸ dmPFC-induced activation of SOM INs is apparent in our data, although the strongest activation results from disynaptic excitation with delayed outputs compared to adjacent PCs.

We next examined plasticity of intrinsic excitability of INs across conditioning and extinction and found two notable adaptations. First, PV INs demonstrated increased current-driven AP firing after extinction relative to the basal and FC. Second, VIP INs demonstrated increased current-driven AP firing after FC relative to basal and extinction groups. These data suggest BLA INs may undergo experience-dependent plasticity across associative learning and extinction, which could contribute to changes in in vivo activity upon threat-predictive cure presentation or behavioral expression.

To gain insight into the functional properties of distinct IN types in vivo, we utilized fiber-photometry-based calcium measurements from SOM, PV, and VIP INs across a conditioned fear acquisition and extinction learning task. During FC, CS+ calcium responses showed sensitization in SOM INs, remained stable in PV, and were minimal in VIP INs. Upon US deliveries, responses again showed sensitization in SOM INs but displayed habituation in both PV and VIP INs. Indeed, the reduction in intrinsic excitability of VIP INs observed ex vivo after conditioning may contribute to the habituating US responses observed in vivo. Interestingly, the CS+- and US-evoked calcium activity in the SOM IN population lasted substantially longer than in PV or VIP neurons. These data highlight divergent response and plasticity profiles of specific IN populations, with SOM and VIP cells displaying the most distinct phenotypes and PV INs exhibiting a mixed representation of SOM and VIP INs, namely CS+ responsivity and habituative US responses, respectively. These data are consistent with previous reports that habituating US responses in VIP cells may represent attenuating US salience as a function of predictability,¹⁹ and studies demonstrating dynamic activity patterns of subpopulations of BLA SOM and PV INs in response to CS+ and US presentations.^17,19 Lastly, sensitizing increases in SOM IN activity across FC have also been reported in the prefrontal cortex,³² raising the intriguing possibility that these learning-associated changes may represent a uniform property of SOM INs across brain regions.

After conditioning, mice underwent extended extinction training, which resulted in reduced freezing accompanied by distinct changes in activity of different IN types. In the early phase of extinction, when freezing levels were high, we discerned three types of activity patterns associated with CS+ presentation. First, a phasic spike upon CS+ onset, which was observed in all three IN types. Second, a subsequent slow reduction in activity for the duration of the tone found in SOM and PV INs. Third, a CS+ offset spike, present most prominently in SOM INs. Interestingly, dynamic changes in these activity patterns were observed between early and late extinction trials. The first, a phasic CS+ onset spike, remained stable in SOM but habituated in PV and VIP INs. The second, a protracted decreasing signal during CS+ presentation, was attenuated across early to late extinction in both SOM and PV INs. The third, a CS+ offset peak, became absent in SOM neurons by the late phase of extinction. Yet again, SOM and VIP INs displayed the most distinct dynamic activity patterns with PV INs exhibiting a mixed profile of both SOM INs (slow CS+-related reduction in activity that habituated across extinction) and VIP INs (sharp CS+ phasic signal that habituated across extinction). In addition, examination of the relationships between these calcium activity patterns and overall freezing levels revealed a direct correlation upon CS+ onset in VIP INs and a negative correlation during tones in SOM and PV INs, when their calcium signals displayed varying degrees of reduction. In other words, VIP IN activity decreased while SOM and PV activity increased (i.e., less reduction in calcium signal) with the development of extinction learning. Furthermore, a direct correlation was observed at CS+ offset peak in both SOM and PV INs between freezing level and calcium activity.

Next, we explored the activity of specific IN types as a function of freezing and moving and during transitions between these states. We started out by assessing the overall activity during times of motion and freezing and, depending on the trial and cell type, found a distinct separation in activity levels, with higher signals during movement and lower signals during freezing. Notably, in SOM neurons, this segregation was present across all days, even during habituation when freezing levels were minimal. The degree of divergence between moving/freezing calcium signals, however, markedly increased with the expression of fear, during FC and early extinction, with a similar separation also present in PV INs. VIP INs, on the other hand, showed small changes that did not systematically vary with the behavioral training. To explore whether these activity patterns originated from discrete time-locked events, we examined the transitions between movement and freezing behaviors. Indeed, sharp increase accompanied the freezing-to-movement transitions, and sharp decrease coincided with the movement-to-freezing transitions on each day in SOM INs, during FC and early extinction in PV INs, and during early extinction in VIP INs. These responses were present not only during but in between tone periods, indicating they are not related to sensory input. Interestingly, a recent study reported congruent mobility-related activity changes of these INs in a tail-suspension test, and other work has demonstrated increased SOM IN activity during presentation of safety-predicting cues associated with suppressed freezing.^18,33 Moreover, chemogenetic inhibition of BLA SOM INs increases conditioned freezing behavior,³⁴ in line with our data revealing low SOM IN activity during freezing.

Various functional implications of our data are worth mentioning. First, studies have highlighted the presence of valence-coding ensembles of BLA PCs likely utilizing subpopulations of GABAergic INs to generate lateral inhibition and support functional antagonism between such ensembles.^12,35 Given that SOM and PV INs are strongly activated by local PCs, they could contribute to valence coding through lateral inhibition if they preferentially target opposing valence ensembles. Determining whether these INs are anatomically positioned to promote functional antagonism of opposing ensembles requires future investigations. Second, that intrinsic inputs can evoke multiple APs in SOM and PV INs, combined with their ability to markedly modulate the effectiveness of extrinsic afferent-induced activation of BLA PCs (up to 200 ms), highlights the ability of these INs to dramatically constrain BLA PC activity and potentially act as low-pass filters in the face of robust extrinsic excitation. Third, we show that SOM and PV INs respond to both sensory cues as well as behavioral-state transitions suggesting multidimensional coding properties for subtypes of BLA INs, as has been previously reported for BLA PCs^36,37; however, adequate testing of this hypothesis will require detailed single-cell-level analysis of IN activity. Fourth, while we detected clear fluctuations in IN activity as a function of behavioral state (i.e., freezing or moving), the magnitude of these changes was dynamic across conditioning and extinction for some cell types. One interpretation of these data is that the magnitude of these transitions reflects an internal state, such as fear, not captured by a binary behavioral classification, consistent with recent studies demonstrating state-dependent coding within BLA PCs.^38,39 In other words, if alterations in IN activity reflect only or purely behavioral transitions, the magnitude of these changes is not expected to vary across conditioning and subsequent extinction, as fear states decline. Lastly, our data demonstrate that SOM IN activity (and, to a lesser degree, PV activity) tightly reflects freezing onset and offset with decreases and increases, respectively. Similarly, increases in SOM IN activity have been reported during learned safety cues causing desynchronization of BLA PC firing and, consequently, blocking theta oscillation phase reset.¹⁸ Thus, it is plausible that suppression of theta power could correspond to increased motion and SOM activity, while enhancement of theta power would yield decreased motion and SOM activity. Further studies are necessary to dissect the causal relationships between IN activity, synchronization, and behavior across fear acquisition and extinction.

In summary, our data reveal a synaptic/circuit organization diversity among BLA SOM, PV, and VIP INs. Specifically, we show that SOM INs mediate feedback inhibition and exhibit the greatest degree of learning-induced plasticity changes displaying multiplexed representations of both salient sensory stimuli and behavioral-state transitions that are dynamic across acquisition and extinction of conditioned fear. VIP INs, which mediate feedforward disinhibition, respond primarily to salient sensory cues, while PV INs participate in both feedback and feedforward inhibition and exhibit functional properties of both SOM and VIP INs throughout our behavioral assay. Considering the subpopulations of PV cells reported,^40–42 it is possible that distinct subpopulations contribute to the mixed functional phenotypes of PV INs observed herein. For example, PV INs mediating feedback inhibition could exhibit in vivo activity patterns more akin to SOM INs (e.g., responses to behavioral-state transitions, and CS+-related slow reductions during fear recall), while PV cells conveying feedforward inhibition could appear much like the activity of VIP INs (e.g., habituating US responses, and lack of freezing onset/offset-associated changes). Additionally, whether the examined microcircuit organization-function relationships persist across other brain regions or other forms of learning remains to be assessed.

Limitations of the study

A primary limitation of the work is the lack of clear relationship between synaptic organization and in vivo functional properties of BLA IN types across auditory FC. Specifically, while we show distinct INs participate in different synaptic organizational motifs, which sets of feedback or feedforward circuits are engaged during particular behaviors or sensory responses in vivo is not resolved, as our fiber photometry experiments analyzed IN types based on genetic identity not specific circuit participation per se. Related, our recordings focused on the BLA that receives delayed auditory inputs rather than the LA that receives direct auditory inputs from the thalamus. Another limitation is that neither our current-clamp nor voltage-clamp recordings can control for electrotonic differences between cell types; thus, interpretations of differences in current and potential amplitudes between cell types may be confounded and subject to additional voltage-gated current contamination. Additionally, the ability to evoke an AP using maximal optical stimulation as a measure of synaptic integration and output is limited, as the number of inputs from any source (e.g., dmPFC) that would be active during behavior is unknown and thus whether the cell reaches threshold in vivo cannot be determined by the ex vivo electrophysiological approaches taken. Lastly, in our examination of internal excitation of BLA INs, we activated two populations of BLA output neurons (projecting to the dmPFC and CeL) while recording from unlabeled PCs and INs. This was required to gain optical access to a subpopulation of PCs to test our hypotheses regarding differential feedforward and feedback inhibition motifs as a function of IN type. Whether our results on intrinsic IN activation will generalize to BLA PCs that project to other targets, such as the ventral hippocampus or nucleus accumbens, needs to be determined.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Sachin Patel (sachin.patel@northwestern.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• All data reported in this paper will be shared by the lead contact upon request.

• Custom codes developed for this study are available on GitHub at https://github.com/sunilmut/NumPPy (https://doi.org/10.5281/zenodo.16750159).

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

All experiments were approved either by the Vanderbilt University Institutional Animal Care and Use Committees or by the Northwestern University Animal Care and Use Committee and were conducted in accordance with the National Institute of Health guidelines for the Care and Use of Laboratory Animals. Adult (10+ weeks) male mice were used for electrophysiology experiments and were housed on a 12:12 light/dark cycle (Vanderbilt University) and adult mice of both sexes were used for fiber photometry behavioral experiments housed on a 14:10h light/dark cycle (Northwestern University). Sex differences were not a primary analysis variable in this study and no overt sex differences were observed, thus all data were pooled from both sexes. All mice were group housed by sex (2–5 mice per cage). Only data depicted in Figure S7 contained mice single housed for 7 days prior to the start of the behavioral procedure, because of the potential confounding effects between the experimental groups. Animal housing took place in a temperature- and humidity-controlled environment with ad libitum access to food and water; and all experiments were performed during the light cycle.

All homozygous transgenic lines were purchased from The Jackson Laboratory. SOM-IRES-Cre (JAX stock #013044), PV-IRES-Cre (JAX stock #017320) and VIP-IRES-Cre (JAX stock #010908) mice were crossed in-house with Ai14 (JAX stock #007914) to obtain SOM:Ai14, PV:Ai14 and VIP:Ai14, respectively. This allowed for Cre-driven expression of the red fluorophore, tdTomato, in the specific IN types. Part of the local excitation dataset (Figures 2 and S4) was acquired from SOM-IRES-FLPo heterozygous mice (JAX stock #031629 crossed in-house with wild-type C57BL/6J). For the sake of simplicity, the schematic figures contain the viral strategies used in all three Ai14 crossed IN-Cre lines. See the SOM-FLPo viral strategy in the section of Viral vectors below. SOM data from these two sources were pooled.

METHOD DETAILS

Viral vectors

The following viruses were used during stereotaxic surgeries. Gifts from Karl Deisseroth: AAV5-CaMKIIa-hChR2(H134R)-EYFP (Addgene plasmid #26969); AAV5-EF1a-double floxed-hChR2(H134R)-EYFP-WPRE-HGHpA (#20298); AAV5-Ef1a-DIO EYFP (#27056); rgAAV-EF1a-mCherry-IRES-Flpo (#55634) and rgAAV-EF1a-Flpo (#55637) were used combined. AAV8-nEF-Coff/Fon-ChRmine-oScarlet was a gift from Karl Deisseroth & INTERSECT 2.0 Project (#137160); AAV5-Syn-ChrimsonR-tdT was a gift from Edward Boyden (#59171); AAV8-hSyn-dF-HA-KORD-IRES-mCitrine (#65417) was gift from Bryan Roth. In SOM-FLPo heterozygous mice the following viral combination was used for local pyramidal cell excitation in addition to SOM:Ai14 approach labeled on Figures 2 and S4: rgAAV-hSyn-Cre-P2A-dTomato (#107738; in dmPFC or CeL), a gift from Rylan Larsen; AAV-Ef1a-fDIO EYFP (#55641, in BLA) a gift from Karl Deisseroth and AAV5-Syn-FLEX-rc[ChrimsonR-tdTomato] (#62723, in BLA), a gift from Edward Boyden. AAV9-syn-FLEX-jGCaMP7f-WPRE (#104492) was a gift from Douglas Kim & GENIE Project.

Stereotaxic surgery

At 6+ weeks of age, mice were anesthetized with 5% isoflurane. The hair over the incision cite was trimmed and the skin was prepped with alcohol and iodine. The animal was transferred to a stereotaxic frame (Kopf Instruments, Tujunga, CA) and kept under 1–2% isoflurane anesthesia. The skull surface was exposed via a midline sagittal incision and treated with the local anesthetic benzocaine (Medline Industries, Brentwood, TN). For each surgery, a 10uL microinjection syringe (Hamilton Co., Reno, NV) with a Micro4 pump controller (World Precision Instruments, Sarasota, FL) was guided by a motorized digital software (Neurostar; Stoelting Co) to each injection coordinate. The following coordinates were used relative to bregma, unless noted otherwise (in mm): dmPFC (AP: 2.25, ML: ±0.40, DV: 2.09), DMT (AP: −1.00, ML: ±0.5, DV: 3.40), CeL (AP: −1.18, ML: ±2.77, DV: 4.80), BLA (AP: −1.35, ML: ±3.2–3.34, DV: 5.04–5.14), LEC (AP: 0.10, ML: ±4.28, DV: 4.42; relative to Lambda). For fiber photometry experiments, a fiber optic (Doric, 400 μm core, 0.66 NA) was implanted over the BLA injection site right after viral injection. All subjects received a 10 mg/kg ketoprofen (AlliVet, St. Hialeah, FL) injection as a perioperative analgesic, and additional post-operative treatment with ketoprofen was maintained for 48 h post-surgery.

Ex vivo electrophysiology

Coronal brain sections were collected at 250μm using standard procedures. Mice were anesthetized using isoflurane, and transcardially perfused in an ice-cold/oxygenated (95% v/v O2, 5% v/v CO2) cutting solution consisting of (in mM): 93 N-Methyl-D-glucamine (NMDG), 2.5 KCl, 20 HEPES, 10 MgSO4. 7H2O, 1.2 NaH2PO4, 30 NaHCO3, 0.5 CaCl2. 2H2O, 25 glucose, 3 Na+-pyruvate, 5 Na+-ascorbate, and 5 N-acetylcysteine. The brain was subsequently dissected, hemisected, and sectioned using a vibrating LeicaVT1000S microtome (Leica Microsytems, Bannockburn, IL). The brain slices were then transferred to an oxygenated 34°C chamber filled with the same cutting solution for a 10 min recovery period. Slices were then transferred to a holding chamber containing a buffered solution consisting of (in mM): 92 NaCl, 2.5 KCl, 20 HEPES, 2 MgSO4 7H2O, 1.2 NaH2PO4, 30 NaHCO3, 2 CaCl2 2H2O, 25 glucose, 3 Na-pyruvate, 5 Na-ascorbate, 5 N-acetylcysteine and were allowed to recover for ≥30 min. For recording, slices were placed into a perfusion chamber where they were constantly exposed to oxygenated artificial cerebrospinal fluid (ACSF; 31°C–33°C) consisting of (in mM): 113 NaCl, 2.5 KCl, 1.2 MgSO4. 7H2O, 2.5 CaCl2. 2H2O, 1 NaH2PO4, 26 NaHCO3, 20 glucose, 3 Na+-pyruvate, 1 Na+-ascorbate, at a flow rate of 2.5–3mL/min.

Cells were visually identified from Ai14 reporter lines or virally injected animals under illumination from a series 120Q X-cite lamp at 40× magnification using an immersion objective in coordination with differential interference contrast microscopy (DIC). BLA neurons were either current clamped or voltage clamped in whole cell configuration using borosilicate glass pipettes (3–6MΩ) filled with intracellular solution, containing (in mM) either: 125 K + -gluconate, 4 NaCl, 10 HEPES, 4 Mg-ATP, 0.3 Na-GTP, and 10 Na-phosphocreatine, or 120 CsMeSO3, 2.8 NaCl, 5 TEA-Cl, 20 HEPES, 2.5 Mg-ATP and 0.25 Na-GTP (pH 7.30–7.35), respectively. In most cases pairs of neighboring pyramidal cell and interneuron were recorded simultaneously. Excitatory postsynaptic currents (EPSC) and inhibitory postsynaptic currents (IPSC) were recorded at −70mV and +10mV, respectively, using the CS-based internal solution. Experiments were generally conducted in a drug free ACSF. The following drugs have been applied when noted: Tetrodotoxin citrate (TTX, 1μM, Tocris Bioscience), 4-Aminopyridine (4AP, 100μM, Sigma-Aldrich), Gabazine (10μM, Abcam), Picrotoxin (50μM, Abcam), CNQX (20μM, Tocris Bioscience), D(−)-2-Amino-5-Phosphonopentatonic acid (D-AP5, 50μM, Tocris Bioscience), CGP-54626 (2μM, Tocris Bioscience), Salvinorin B (SalB, 0.3μM, Tocris Bioscience).

For current-clamp recordings in Figure 1, inflection points were defined manually. we defined the 1^st peak as the maximal amplitude of the EPSP before the 2^nd inflection point, and the 2^nd peak was defined as the maximum amplitude of the evoked EPSP. For voltage-clamp recordings in Figure S3, peaks and onset times were identified using numerical differentiation methods in MATLAB. Traces were smoothened via application of Gaussian filtering. EPSC onset was determined as the inflection point where second derivative changed sign from negative to positive. Identified inflection points were visually verified. Peaks were defined as local maxima of signal within the onset and the next inflection point.

All data was acquired using a Multiclamp 700B amplifier, Digidata 1440A A/D converter, Clampex version 10.6 software (Axon Instruments, Union City, CA), was sampled at 20kHz and low pass filtered at 1 kHz and analyzed in ClampFit 10 software (Molecular Devices, San Jose, CA). Optical stimulation was achieved by using a Mightex LED controller system with a 455nm blue LED and a 617nm red LED receiving and conveying TTL signals from Clampex protocols.

Fear conditioning and extinction training

Behavioral fiber photometry experiments took place at least four weeks after stereotaxic surgeries allowing for GCaMP7f expression in the targeted BLA IN populations. Mice underwent extensive handling prior to training in order to eliminate basal fear expression. Two distinct contexts were used during the training protocol (Context A and B). Context A was used exclusively on fear conditioning day, and it consisted of vanilla scent, bright light, rectangular chamber with metal wiring floor for electric shock deliveries in a brightly lit room. Although Context B was presented in the same chamber as Context A, we used contrasting stimuli in order to help the differentiation between environments. Context B consisted of few drops of 2% acetic acid for olfactory cue, dark chamber with dim string lights, curved space created with a patterned plastic wall insert, white floor with clean bedding, within a dark room. Behavioral chamber was inside a Med Associates sound isolating box and Coulbourn Instruments was used for shock and sound deliveries using TTL signals initiated by FreezeFrame software.

Training started on day 1 in Context B referred to as habituation day (H). After 2-min baseline mice were presented with eight 30s auditory tone (CS−: 400Hz, 75dB) unique to Context B with varying >30s inter-tone intervals (no tone). Mice displayed minimal freezing throughout the session (data not shown). The following day (day 2) fear conditioning took place in Context A. After 2-min baseline 5 (initial SOMxAi14 cohort) to 5 auditory tone-shock pairing was presented. Each tone (conditioned stimulus, CS+: 4000Hz, 75dB) lasted 30s and co-ended with a 1s electric foot shock (unconditioned stimulus, US: 0.45mA), with inter-tone intervals of 90s. For fear extinction training (day 3–5 or 6 if extinction has not developed faster) mice were placed in Context B and a combination of CS− (4–5; to test for tone-response specificity) and CS+ (8–14) tones were presented, each for 30s with varying >30s inter-tone intervals. For electrophysiology experiments in Figure S7, mice were sacrificed 30 min after fear recall day, after 2 days of extinction and a recall day, or on recall day after tone-only presentations on conditioning and recall days.

Fiber photometry and data processing

Behavioral fiber photometry experiments took place at least four weeks after stereotaxic surgeries allowing for GCaMP7f expression in the targeted BLA IN populations. Mice were attached to a fiber optic patch cord (Doric, 400 μm core, 0.66 NA) and gently placed in the behavioral apparatus. Fiber photometry data was collected throughout the entire behavioral session using a rig with optical components from Doric lenses controlled by a real-time processor from Tucker Davis Technologies (TDT, RZ10x). TDT Synapse software was used for data acquisition. 465nm and 415nm (isosbestic control) LEDs were modulated at 210Hz and 330Hz, respectively. LED currents were adjusted to return a voltage between 75 and 85mV for each signal, were offset by 5mA, and were demodulated using a 10Hz lowpass frequency filter. For the analysis of the raw data, GuPPy,⁴³ an open-source Python-based photometry data analysis pipeline, was used in order to visualize Z score traces and quantify peak Z score and area-under-the-curve (AUC) shown on figures using PSTH analysis. For CS+ - and shock-related measurements 3s baseline prior to event was used, for all behavioral (start/stop freezing) transitions 1s baseline was used due to the high time-sensitivity of these events. Longer baselines shift the transients of single events resulting in less accurate data. Timestamps of start/stop freezing was extracted from FreezeFrame software (used for H-FC-E training) generated ‘Index files’ following careful manual determination of freezing thresholds for each mouse and inspection of the accuracy of chosen thresholds. Index files consisted of a string of 0/1 (motion/freezing) values assigned to time points at sampling frequency. Timestamps of 01 (start freezing) and 10 (start moving) transitions were extracted using our custom code (binary.py) through its user interface with adjustable settings described below. Due to separate manual starts of the behavioral (FreezeFrame) and photometry (Synapse) software, care needed to be taken to align the extracted index file timestamps to the photometry dataset. This time shift value could be assessed through the presence of known TTL signals (tone, shock) initiated by FreezeFrame to Synapse software, and applied at our code interface. Additionally, a minimum duration criterion was applied for the 0/1 strings prior and after the transitions. We chose 1s–1s duration criteria, which meant that only transitions where the animal was moving/freezing for at least 1s before it started freezing/moving for at least 1s were extracted. The reason for 1s is to eliminate potential artifacts yet to keep most of the stable transitions in the analysis. Moreover, the timestamps were separated into CS−, CS+, and inter-tone interval (no tone) periods, and our analysis excluded the first 3–5 s of tone- and shock onset, -offset periods (binariy.py) to eliminate confounding signals resulting from formerly uncovered phasic responses. These timestamps were visualized and analyzed by GuPPy. Another custom code (dff.py) was used for the calculation of overall ΔF/F values during motion and freezing separately, where data from CS (Figures 6B–6D) and no tone (Figures S9A–S9D) periods were further differentiated after the elimination of the first 3–5 s of tone- and shock onset/offset. The code uses GuPPy generated ΔF/F and timestamp files together with the time-shifted FreezeFrame index. To confirm viral expression and fiber placement, mice were perfused, and brains were sectioned for histological verification using a fluorescent microscope (Keyence BZ-X800).

QUANTIFICATION AND STATISTICAL ANALYSIS

Datasets were organized and quantified in Microsoft Excel and then transferred to GraphPad Prism 10 for generation of graphs and statistical analyses. For analysis of two groups, an unpaired or paired Student’s t test was used. For analysis of three or more groups across a single independent variable, an ordinary one-way ANOVA was used with Tukey post hoc multiple comparisons test between groups as noted in the figure legends. For analysis between two or more groups across two or more independent variables, a two-way ANOVA was used with a Sidak post hoc multiple comparisons test between groups as noted in the figure legends. Sample sizes were derived empirically and based on our previous experience with these assays. Outliers were identified via ROUT test and removed from statistical analysis. Significance was determined as p < 0.05 in all datasets. Data are represented as mean ± SEM, excepted paired data showing individual values.

Supplementary Material

SUPPLEMENTAL INFORMATION

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2025.116295.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
AAV5-CaMKIIa-hChR2(H134R)-EYFP	Gift from Karl Deisseroth PubMed 20473285	Addgene Cat # 26969-AAV5; RRID: Addgene_26969
AAV5-EF1a-DIO-hChR2(H134R)-EYFP-WPRE-HGHpA	Gift from Karl Deisseroth	Addgene Cat # 20298 -AAV5; RRID: Addgene_20298
AAV5-Ef1a-DIO EYFP	Gift from Karl Deisseroth	Addgene Cat # 27056 -AAV5; RRID: Addgene_27056
rgAAV-EF1a-mCherry-IRES-Flpo	Gift from Karl Deisseroth PubMed 24908100	Addgene Cat # 55634 -AAVrg; RRID: Addgene_55634
rgAAV-EF1a-Flpo	Gift from Karl Deisseroth PubMed 24908100	Addgene Cat # 55637-AAVrg; RRID: Addgene_55637
AAV8-nEF-Coff/Fon-ChRmine-oScarlet	Gift from Karl Deisseroth & INTERSECT 2.0 Project; PubMed 32574559	Addgene Cat # 137160-AAV8; RRID: Addgene_137160
AAV5-Syn-ChrimsonR-tdT	Gift from Edward Boyden; PubMed 24509633	Addgene Cat # 59171-AAV5; RRID: Addgene_59171
AAV8-hSyn-dF-HA-KORD-IRES-mCitrine	Gift from Bryan Roth; PubMed 25937170	Addgene Cat # 65417-AAV8; RRID: Addgene_65417
rgAAV-hSyn-Cre-P2A-dTomato	Gift from Rylan Larsen	Addgene Cat # 107738-AAVrg; RRID: Addgene_107738
AAV5-Ef1a-fDIO EYFP	Gift from Karl Deisseroth; PubMed 24908100	Addgene Cat # 55641-AAV5; RRID:Addgene_55641
AAV5-Syn-FLEX-rc[ChrimsonR-tdTomato]	Gift from Edward Boyden; PubMed 24509633	Addgene Cat # 62723-AAV5; RRID: Addgene_62723
AAV9-syn-FLEX-jGCaMP7f-WPRE	Gift from Douglas Kim & GENIE Project; PubMed 31209382	Addgene Cat # 104492-AAV9; RRID: Addgene_104492
Chemicals, peptides, and recombinant proteins
Tetrodotoxin citrate	Tocris Bioscience	Cat# 1069
4-Aminopyridine	Sigma-Aldrich	Cat# A0152
Gabazine (SR95531)	Abcam	Cat# ab120042
Picrotoxin	Abcam	Cat# ab120315
CNQX	Tocris Bioscience	Cat# 1045
D-AP5	Tocris Bioscience	Cat# 0106
CGP-54626	Tocris Bioscience	Cat# 1088
Salvinorin B	Tocris Bioscience	Cat# 5611
Experimental models: Organisms/strains
SOM-IRES-Cre mice	Jackson Laboratory	RRID:IMSR_JAX:013044
PV-IRES-Cre mice	Jackson Laboratory	RRID: IMSR_JAX:017320
VIP-IRES-Cre mice	Jackson Laboratory	RRID: IMSR_JAX:010908
Ai14 mice	Jackson Laboratory	RRID: IMSR_JAX:007914
SOM-IRES-FLPo mice	Jackson Laboratory	RRID: IMSR_JAX:031629
C57 WT mice	Jackson Laboratory	RRID: IMSR_JAX:000664
Software and algorithms
Prism 10	GraphPad	https://www.graphpad.com
pClamp 10	Molecular Devices	https://www.moleculardevices.com
MATLAB	MathWorks	https://www.mathworks.com/products/matlab.html
FreezeFrame	Actimetrix	https://actimetrics.com/products/freezeframe/
Synapse	Tucker-Davis Technologies	https://www.tdt.com/product/synapse-software/
GuPPy (Guided Photometry analysis in Python, open source tool)43	GitHub/LernerLab	https://github.com/LernerLab/GuPPy
NumPPy: binary.py and dff.py (open source tools)	GitHub	https://github.com/sunilmut/NumPPy

Highlights.

• SOM and PV BLA interneurons receive strong local excitation from BLA principal cells

• VIP interneurons receive primarily extrinsic excitation from long-range inputs

• BLA GABA interneuron types differentially participate in feedback and feedforward inhibition

• BLA interneuron types exhibit differential plasticity across fear conditioning and extinction