Second-order fear conditioning involves formation of competing stimulus-danger and stimulus-safety associations

Abstract

How do animals process experiences that provide contradictory information? The present study addressed this question using second-order fear conditioning in rats. In second-order conditioning, rats are conditioned to fear a stimulus, S1, through its pairings with foot-shock (stage 1); and some days later, a second stimulus, S2, through its pairings with the already-conditioned S1 (stage 2). However, as foot-shock is never presented during conditioning to S2, we hypothesized that S2 simultaneously encodes 2 contradictory associations: one that drives fear to S2 (S2-danger) and another that reflects the absence of the expected unconditioned stimulus and partially masks that fear (e.g. S2-safety). We tested this hypothesis by manipulating the substrates of danger and safety learning in the brain (using a chemogenetic approach) and assessing the consequences for second-order fear to S2. Critically, silencing activity in the basolateral amygdala (important for danger learning) reduced fear to S2, whereas silencing activity in the infralimbic cortex (important for safety learning) enhanced fear to S2. These bidirectional changes are consistent with our hypothesis that second-order fear conditioning involves the formation of competing S2-danger and S2-safety associations. More generally, they show that a single set of experiences can produce contradictory associations and that the brain resolves the contradiction by encoding these associations in distinct brain regions.

associative learning, higher-order conditioning, chemogenetics, infralimbic cortex, basolateral amygdala

Introduction

Pavlovian fear conditioning in rats is widely used to study the substrates of learning and memory in the mammalian brain. In a standard protocol, rats are exposed to pairings of an initially innocuous stimulus (e.g. a sound; denoted S1) and an innate source of danger, such as brief but aversive foot shock (unconditioned stimulus, US). After these pairings, subsequent presentations of the now conditioned S1 elicit a range of responses indicative of fear in people, including changes in heart rate, blood pressure, freezing, and potentiated startle. In a second protocol, rats are exposed to S1-US pairings as described (stage 1) and then to pairings of a second stimulus (e.g. a light; denoted S2) and the already conditioned S1 (stage 2). After this training, presentations of either S1 or S2 elicit fear responses (Parkes and Westbrook 2010; Holmes et al. 2014). Importantly, we and others have shown that fear to S2 is associatively mediated as it is not displayed by controls exposed to explicitly unpaired presentations of the relevant stimuli in either stage of training (Holmes et al. 2014; Michalscheck et al. 2021; Parkes and Westbrook 2010; Rizley and Rescorla 1972; see also Fig. S1, see online Supplementary Material for a color version of this figure). Pavlov (1927) termed fear to S2 “second-order” to distinguish it from fear to S1, which he termed “first-order.”

First- and second-order conditioned fears both require neuronal activity in the basolateral amygdala (BLA) for their acquisition and expression (Gewirtz and Davis 1997; Parkes and Westbrook 2010; Holmes et al. 2013; Lingawi et al. 2021) and are similarly responsive to variations in the intensity and contiguity of stimuli presented in training (Rizley and Rescorla 1972; Rescorla and Furrow 1977). However, an important difference between the 2 protocols is that, whereas S1 maintains its ability to elicit fear across repeated pairings with the US, S2 loses its fear-eliciting ability across repeated pairings with the S1 (Herendeen and Anderson 1968; Rescorla 1972; Yin et al. 1994). This reduction in fear has been identified with subjects learning that S2 is in fact safe (as it is perfectly correlated with the absence of the US) which then opposes their learning that S2 signals danger (the fear conditioned S1). However, little is known about how these contrasting sources of information are encoded with respect to each other. In particular, it is unknown whether subjects solve this problem sequentially, by first learning that S2 signals danger and then gradually learning that it is in fact safe. Alternatively, they encode both types of associations on each second-order conditioning trial but, because it is better to be safe than sorry, use the danger information to control their behavior until they are more certain that S2 is safe.

The present study addressed this gap in knowledge by exploiting what is known about the substrates of danger and safety learning in the brain. This approach was necessary as any behavioral test designed to detect the S2-safety association in second-order conditioning (e.g. summation or retardation tests for inhibitory learning) will be contaminated by the S2-danger association. Hence, the only way to determine whether the 2 associations form together is to target their distinct substrates in the brain and examines the consequences of this targeting for expression of second-order fear. Formation of CS-US associations in standard Pavlovian conditioning protocols has been shown to require neuronal activity in the BLA (Wilensky et al. 1999; LeDoux 2000; Goosens and Maren 2001; Maren et al. 2001; Sevelinges et al. 2009), whereas safety learning (e.g. formation of CS-no US associations during extinction of conditioned fear) is commonly identified with activity in the infralimbic cortex (IL; Awad et al. 2015; Giustino and Maren 2015; Laurent and Westbrook 2009; Lay et al. 2020; Lingawi et al. 2018, 2019). The IL has also been found to regulate conditioned inhibition using both aversive (Kreutzmann et al. 2020) and appetitive USs (Rhodes and Killcross 2007; Meyer and Bucci 2014). Thus, we examined whether second-order conditioning involves simultaneous encoding of S2-danger and S2-safety associations by examining the effects of chemogenetically silencing neuronal activity in the BLA or IL on acquisition and expression of second-order conditioned fear. If the 2 types of associations are encoded together, we reasoned that the test level of fear responses (freezing) to S2 would be reduced among rats whose BLA was silenced during second-order conditioning (as this impairs formation of the S2-danger association; Experiment 1) but enhanced among rats whose IL was silenced during second-order conditioning or testing (as this impairs formation/retrieval of the S2-safety association; Experiments 1 and 2). We also reasoned that these effects of silencing the IL on test levels of freezing to S2 would be specific to second-order conditioning by reversing the order of stages 1 and 2: that is, by first exposing rats to S2-S1 pairings and then to pairings of S1 and shock. This sensory preconditioning protocol also imbues S2 with the ability to elicit freezing responses at test but, in contrast to the second-order protocol, does not result in any safety learning to the S2. Hence, silencing the IL in this protocol should not affect the level of responding to a sensory preconditioned stimulus at the time of testing.

Materials and method

Subjects

Subjects were 247 experimentally naïve adult female Long Evans rats (8–14 weeks old) obtained from the Rat Breeding Facility at the University of New South Wales (Sydney, Australia). Rats were housed in groups of 4 in plastic ventilated cages located in a climate-controlled holding room which was maintained on a 12-h light/dark cycle (lights on at 7 am). Food and water were available ad libitum. Rats were handled daily for 5 days prior to the commencement of procedures. All experimental procedures were approved by the Animal Care and Ethics Committee at the University of New South Wales and carried out in accordance with the guidelines provided by the Australian National Health and Medical Research Council.

Viral vectors

Adeno-associated viral vectors encoding the inhibitory Designer Receptor Exclusively Activity by Designer Drug hM4Di (AAV5-hsyn-hM4D (Gi)-mCherry) or the fluorophore mCherry (AAV5-hSyn-mCherry) were used. Plasmids were obtained from Addgene (hM4Di: #50475, 7 × 10¹² vg/mL; mCherry: #114472, 7 × 10¹² vg/mL) and packaged by the Vector and Genome Engineering Facility (VGEF, Children and Medical Research Institute, Westmead, Australia).

Surgery

Rats were anesthetized with isoflurane (5% for induction and 2.5–2% for maintenance, Cenvet Australia) and positioned on a stereotaxic apparatus (Kopf Instruments, Tujunga, CA). After site incision and bilateral craniotomy above targeted brain regions, 0.5 μL of the hM4Di or mCherry virus was infused at a rate of 0.1 μL/min using a 1-μL Hamilton glass syringe. The syringe needle was left in place for 10 min after infusion to allow for complete diffusion of the viral vectors. The stereotaxic coordinates in mm relative to bregma (Paxinos and Watson 2006) were: for BLA, −2.52 AP, ± 5.1 ML, −9.2 DV; for IL, +3.24 AP, ± 0.7 ML, −5.4 DV. After surgery, rats received prophylactic injections (subcutaneous) of procaine penicillin and benzathine penicillin (Duplocillin: 0.1 mL/kg of a 150-mg/mL solution), and the anti-inflammatory carprofen (Rimadyl: 0.1 mL/kg of a 50-mg/mL solution).

Apparatus

Behavioral procedures took place in 8 Med Associates operant chambers each located in a sound- and light-attenuating cabinet. The floor of the chambers consisted of stainless-steel rods (3.8 mm in diameter, spaced 16 mm apart) which were connected to a constant current generator that delivered a 0.5-s duration × 0.8-mA intensity shock US. Each chamber was equipped with a house light (3 W, 24 V) and speaker and infra-red cameras for recording of the rat’s behavior. The stimuli were a 2-Hz flashing house light and a 3-kHz pure tone (80 dB SPL), counterbalanced for their roles as S1 and S2. S1 was 10 s in duration, while S2 was 30 s in duration. These durations were selected to match those used in prior studies of second-order conditioning and sensory preconditioning (Parkes and Westbrook 2010; Holmes et al. 2014; Wong et al. 2019; Kikas et al. 2021). A computer located in another room in the laboratory controlled the delivery of stimuli via Med-PC V Software Suite.

Drugs

Clozapine (CZ; Sigma Aldrich Australia) was dissolved in 0.9% (w/v) nonpyrogenic saline containing 13% dimethyl sulfoxide (DMSO; Sigma Aldrich Australia) to obtain a dose of 0.3 mg/kg. This dose has been shown to be effective for activating the hM4Di receptor without producing off-target effects (Jendryka et al. 2019). Nonpyrogenic saline containing 13% DMSO was used for vehicle control injections. All injections were administered intraperitoneally (i.p.) 7 min before behavioral procedures commenced. In Experiment 1, injections occurred prior to S2-S1 pairings, while in Experiments 2 and 3, injections occurred prior to Test 1.

Behavioral procedures

All behavioral procedures occurred a minimum of 4–5 weeks after surgery to allow for sufficient viral expression. Second-order conditioning (Experiments 1–2) and sensory preconditioning (Experiment 3) were conducted as described in our previous demonstrations of these phenomena (Parkes and Westbrook 2010; Holmes et al. 2013; Lay et al. 2018). These two protocols differ in the order in which rats acquire the S2-S1 association relative to the S1-US association. In second-order conditioning, S2-S1 pairings follow S1-US pairings, while in sensory preconditioning, S2-S1 pairings precede S1-US pairings. For both protocols, rats receive context exposure prior to training and context extinction following S1-US pairings. In Figs. S1 and S2 (see online Supplementary Material for a color version of this figure), we show that second-order and sensory preconditioned fear to S2 are associatively mediated and not due to generalization from S1.

Context exposure

All rats were exposed to the chambers in the absence of any scheduled events for 20 min twice daily, once in the morning and again in the afternoon, for 2 days. This was done to familiarize the rats with chambers and prevent any neophobic responses that could interfere with the detection of conditioning.

S1-shock pairings

Rats received 4 presentations of S1, each of which co-terminated with the shock. The first S1 presentation occurred 5 min after rats were placed in the conditioning chambers and the interval between S1 presentations (offset to onset) was 5 min. Rats remained in the chambers for 2 min following the final conditioning trial.

S2-S1 pairings

Rats received 8 presentations of S2 each of which co-terminated with the onset of the S1. The first S2 presentation occurred 5 min after rats were placed in the conditioning chambers and the interval between S2 and S1 pairings (S1 offset to S2 onset) was 5 min. Rats remained in the chambers for 2 min following the final pairing.

Context extinction

Rats were placed in the chambers for 20 min for 2 sessions (separated by 3 h). No stimuli were presented. For sensory preconditioning (Experiment 3), rats were given a brief session of context extinction (10 min) 1 h before Test.

Test

Rats were tested with S2 in 2 sessions. In each session, S2 was presented 8 times in the absence of any other stimuli. The first S2 presentation occurred 2 min after rats were placed in the conditioning chambers and the interval between S2 presentations (offset to onset) was 2 min. S2 was tested twice to identify acute versus long lasting effects of neuronal silencing via clozapine injections. For Experiments 2 and 3, Test 1 was conducted following clozapine injections, while Test 2 was conducted drug-free, allowing effects of IL or BLA silencing on the expression of fear (which would be evident acutely at Test 1 and not during the drug-free Test 2) to be differentiated from the effects on the underlying associative mechanisms (which would be evident at both tests).

Histology: immunofluorescence and electrophysiology

Two approaches were used to verify the effects of the chemogenetic manipulations: immunofluorescent staining of mitogen-activated protein kinase/extracellular signal-related kinase (pERK) as a marker of neuronal activity, and whole-cell ex-vivo electrophysiology. For pERK staining, rats from Experiment 3 were re-exposed to S1 to induce activity in the IL or BLA. S1 was presented 8 times in the absence of the US, with an interval between presentations of 2.5 min. They were then removed from the chambers, euthanized, and perfused for tissue analysis.

Transcardial perfusions

Rats were deeply anesthetized with sodium pentobarbital (500 mg/kg, i.p.) and transcardially perfused with ice-cold 4% paraformaldehyde in 0.1 M sodium phosphate buffer (PFA). Brains were extracted, postfixed in PFA solution at 4 °C overnight, and sectioned coronally at 40-μm thickness on a vibratome (Leica Microsystems VT1000). Sections were then stored in cryoprotective solution (30% ethylene glycol, 30% glycerol, 0.1 M sodium phosphate buffer) at −20 °C until further processing.

pERK immunofluorescence

Free-floating sections were rinsed in Tris-buffered saline (TBS) 3 times (10 min each) and incubated for 2 h at room temperature in TBS with 0.2%Triton X-100 for membrane permeabilization. Sections were then rinsed 3 times in TBS (10 min each) and incubated in monoclonal rabbit anti-phospho-p44/42MAPK (Erk1/2) primary antibody (1: 300, #4370 Cell Signaling Technology, diluted in TBS) for 24 h at 4 °C on a rocking platform. Sections were then rinsed 3 times with TBS (10 min each) and incubated in donkey anti-rabbit Alexa Fluor-488 secondary antibody (1, 400, #A21206, Invitrogen, diluted in TBS) for 1 h at room temperature. Finally, sections were again rinsed 3 times in TBS (10 min each), mounted onto Superfrost Plus-coated slides (Thermo Fisher Scientific), dried, and cover-slipped with Vectashield hard-set mounting medium (Vector Laboratories).

Verification of viral injection placements

Sections were rinsed in TBS 3 times (10 min each), mounted onto Superfrost Plus-coated slides (Thermo Fisher Scientific), dried, and cover-slipped with Vectashield Hardset mounting medium (Vector Laboratories). Placement of viral infection was mapped at its widest point onto a common template to indicate viral spread. Rats that had misplaced viral injections were excluded from statistical analysis.

Brain slice preparation

Eight rats from Experiment 3 were euthanized under deep anesthesia (isoflurane 4% in air), and their brains were rapidly removed and cut on a vibratome in ice-cold oxygenated sucrose buffer containing (in mM): 241 sucrose, 28 NaHCO3, 11 glucose, 1.4 NaH2PO4, 3.3 KCl, 0.2 CaCl2, 7 MgCl2. Coronal brain slices (300-μm thick) containing the IL or BLA were sampled and maintained at 33 °C in a submerged chamber containing physiological saline with composition (in mM): 126 NaCl, 2.5 KCl, 1.4 NaH2PO4, 1.2 MgCl2, 2.4 CaCl2, 11 glucose and 25 NaHCO3, and equilibrated with 95% O2 and 5% CO2.

Electrophysiology

After equilibrating for 1 h, slices were transferred to a recording chamber and visualized under an upright microscope (Olympus BX50WI) using differential interference contrast (DIC) Dodt tube optics, and superfused continuously (1.5 mL/min) with oxygenated physiological saline at 33 °C. Whole-cell patch-clamp recordings were made using electrodes (2–5MΩ) containing internal solution consisting of the following (in mM): 115 K gluconate, 20 NaCl, 1 MgCl2, 10 HEPES, 11EGTA, 5 Mg-ATP, and 0.33 Na-GTP, pH 7.3, osmolarity 285–290 mOsm/L. Biocytin (0.1%; Sigma-Aldrich) was added to the internal solution for marking the sampled neurons during whole-cell recording. Data acquisition was performed with a Multiclamp 700B amplifier (Molecular Devices), connected to a Macintosh computer and interface ITC-18 (Instrutech). Voltage recordings under whole-cell configuration were sampled at 5 kHz (low-pass filter 2 kHz; Axograph X, Molecular Devices).

Immediately after physiological recording, brain slices containing biocytin-filled neurons were fixed overnight in 4% paraformaldehyde/0.16 M phosphate buffer (PB) solution and then placed in 0.3% Triton X-100/PB for 3 day to permeabilize cells. Biocytin staining was revealed by incubation in Alexa Fluor 488-conjugated Streptavidin (1:1000; Life Technologies) for 2 h. Stained slices were rinsed in PB 3 times for 10 min each and mounted with Fluoromount-G mounting medium (Southern Biotech). Neurons were imaged under a confocal microscope (Fluoview FV1000 and BX61WI, Olympus).

Experimental design and analysis

Design

All experiments used chemogenetic silencing in a 2 × 2 design, where rats received microinjections of the inhibitory hM4Di virus or the mCherry control virus into a brain region (factor 1), and i.p. injections of clozapine or vehicle (factor 2) prior to S2-S1 training (Experiment 1) or testing with S2 (Experiments 2 and 3). For Experiments 1 and 3, the identity of the brain region was an additional design factor: IL or BLA. In these experiments, rats that received the control virus in the BLA or IL were combined to form a single control group, as there were no statistically significant differences in their conditioned responding across acquisition or testing (Fs < 0.9).

Analysis of behavioral data

Freezing was the index of conditioned fear. It was defined as the absence of all movement except those related to breathing and characterized by a rigid posture distinct from resting or sleeping (Fanselow 1980). Each rat was observed every 2 s during presentations of S1 in first-order conditioning, S2 and S1 in second-order, and S2 on test and scored as either freezing or not freezing by 2 observers, one of whom was naïve to the group allocations. The correlation between the scores of the 2 observers was high (r > 0.9). Any discrepancies between their scores were resolved in favor of the naïve observer. The levels of freezing are reported as a percentage in all figures.

Analysis of immunofluorescence

Tissues were analyzed for co-labeling of pERK and mCherry. For each rat, 4 sections were selected to encompass the IL or BLA along the anterior–posterior axis according to boundaries defined by Paxinos and Watson (2006). All fluorescent imaging was acquired with a confocal microscope (Olympus BX16WI) using a 10× objective, where Alexa Fluor 488 (pERK) and mCherry fluorescence were scanned sequentially. Image files were randomly renumbered using an MS Excel plug-in (Bio-excel2007 by Romain Bouju, France) prior to quantification. Open-source ImageJ software (MacBiophotonics upgrade v. 1.43u, Wayne Rasband, National Institutes of Health, Bethesda, MD) was used to determine cell counts. First, regions of interest (ROIs) were manually selected corresponding to the brain ROI. Next, pERK-immunoreactive cells within the ROI were marked, quantified and averaged across the 4 sections for each rat.

Statistical methods

All data were analyzed in SPSS (IBM) using ANOVA. For the 2 test sessions, data were analyzed using a mixed-model ANOVA, with test session as a repeated-measures factor. Any significant interactions were followed up using simple effects contrasts to establish the source of the interactions. The criterion for rejection of the null hypothesis was set at 0.05 (2-tailed). Effect sizes are calculated as partial eta squared.

Results

Experiment 1: S2-danger and S2-safety associations are encoded together in second-order conditioning

Experiment 1 examined whether S2-danger and S2-safety associations form together across second-order conditioning trials. The inhibitory hM4Di or mCherry control virus was expressed in the IL or BLA 5 weeks prior to the start of experimental procedures (Fig. 1C). In stage 1, rats received first-order conditioning which consisted in 4 presentations of S1 each of which co-terminated in an aversive US (foot shock). In stage 2, rats received second-order conditioning which consisted in 8 nonshocked presentations of S1 each of which was preceded by a presentation of S2. Prior to stage 2, rats received a systemic injection of clozapine or vehicle. Thus, the IL or the BLA was silenced across second-order conditioning for rats that had been infused with the hM4Di virus and injected with clozapine but activity in these regions was normal in all other rats. Finally, rats were tested for freezing to S2 alone. There were 2 such tests, each of which involved 8 presentations of S2 alone (Test 1 and Test 2). If second-order conditioning involves the formation of both S2-danger and S2-safety associations, chemogenetic silencing of the IL or BLA would have opposing effects on the test levels of freezing to S2. Specifically, relative to controls, chemogenetic silencing of the BLA before the second-order conditioning session would reduce the test levels of freezing to S2 (as it impairs formation of the S2-danger association), whereas chemogenetic silencing of the IL before that session would increase the test levels of freezing to S2 (as it impairs formation of the S2-safety association).

Fig. 1

The IL and BLA encode opposing information during acquisition of second order conditioning. A) Representative images of control and hM4di expression in the IL (top) and BLA (bottom); scale bars = 400 μm. B) Placement of viral injections; numbers indicate anterior–posterior coordinates relative to bregma. C) Second-order conditioning procedure used in Experiment 1; systemic injections of vehicle (Veh) or clozapine (CZ) were administered prior to stage 2 of training. D) Freezing to stimuli during stage 1 (left) and stage 2 (right) of training in Experiment 1. E) Freezing to S2 across Test 1 (left) and Test 2 (right) in Experiment 1. All error bars indicate ±1 SEM. ^*P < 0.05; ^**P ≤ 0.01.

Rats that had the control virus expressed in the IL or BLA were combined to form a single control group as there were no statistically significant differences in the levels of freezing across acquisition and testing (Fs < 0.8). After 6 rats were excluded due to misplaced viral injections, the final group sizes were: Control-Vehicle (Ctrl-Veh), n = 14; Control-Clozapine (Ctrl-CZ), n = 17; IL-Vehicle (IL-Veh), n = 8; IL-Clozapine (IL-CZ), n = 11; BLA-Vehicle (BLA-Veh), n = 11; and BLA-Clozapine (BLA-CZ), n = 12. Representative images of viral expression and the extent of this expression are shown in Fig. 1A and B. Given the collapsing of the control groups, the freezing data from each stage of training and testing were analyzed using a 3 (Ctrl, IL, BLA) × 2 (injection: vehicle vs. clozapine) ANOVA with an additional within subject factor of trials for stage 1, or blocks of trials for stage 2 of training.

Figure 1D shows freezing across first- and second-order conditioning (left and right panels, respectively) in Experiment 1. All rats acquired freezing to S1, which increased across its pairings with shock in stage 1 (F_1,67 = 117.869, P < 0.001, η_p² = 0.638; Fig. 1D, left). There were no between-group differences in the overall level of freezing or the rate at which freezing increased to S1 (Fs < 1.8). All rats acquired freezing to S2 across its pairings with the conditioned S1. During the initial trials of training, freezing to S2 was low, but increased significantly across subsequent S2-S1 pairings. This was confirmed by significant linear and quadratic trends in the freezing across trials (linear: F_1,67 = 15.950, P < 0.001, η_p² = 0.192; quadratic: F_1,67 = 33.590, P < 0.001, η_p² = 0.334). There were no between-group differences in the overall level of freezing to S2 (Fs < 1.5). Freezing to S1 decreased across the S2-S1 pairings (F_1,67 = 14.746, P < 0.001, η_p² = 0.180). There was a significant group × injection interaction: F_2,67 = 3.889, P = 0.025, η_p² = 0.104). Follow-up contrasts revealed that silencing the BLA produced a significant reduction in freezing to S1 (Group BLA-CZ vs. Group BLA-Veh: F_1,67 = 8.894, P = 0.004, η_p² = 0.117), while clozapine injections had no effect in the IL groups or for control animals (F_s < 0.5).

Fig. 2

The IL regulates the expression of second-order conditioned fear. A) Representative images of control and hM4Di expression in the IL (left) and placement of viral injections across groups in Experiment 2 (right); scale bars = 400 μm, distances indicate anterior–posterior coordinates relative to bregma. B) Second-order conditioning procedure used in Experiment 2; systemic injections of vehicle (Veh) or clozapine (CZ) were administered prior to Test 1. C) Freezing to stimuli during stage 1 (left) and stage 2 of training (right). D) IL silencing produced an acute enhancement in the expression of second-order conditioned fear. All error bars indicate ±1 SEM. ^*P < 0.05; ^**P ≤ 0.01.

The data from the 2 consecutive tests of S2 are shown in Fig. 1E (see also Fig. S3). All rats showed a decline in freezing from Test 1 to Test 2 (significant main effect of test: F_2,67 = 56.39, P < 0.001, η_p² = 0.46). There was also a significant group × injection × test interaction (F_2,67 = 4.74, P = 0.01, η_p² = 0.12), suggesting that silencing the IL or BLA during second-order conditioning had opposing effects across the test sessions. This was confirmed by follow-up contrasts which revealed that silencing the IL in stage 2 increased the levels of freezing to S2 on both Test 1 (F_1,67 = 5.961, P = 0.017, η_p² = 0.082) and Test 2 (F_1,67 = 6.606, P = 0.012, η_p² = 0.090). In contrast, silencing the BLA in stage 2 decreased the levels of freezing to S2 on Test 1 (F_1,67 = 8.718, P = 0.004, η_p² = 0.115) but not Test 2 (F = 1.110), likely due to floor effects. For control animals, clozapine injections had no effect on freezing to S2 in either test session (F_s < 1.7). In addition, there were no differences between vehicle injected animals and the 2 control groups (Supplementary Materials).

These results show that silencing the IL or BLA during second-order conditioning has contrasting effects on freezing to S2 at test: silencing the IL increased while silencing the BLA decreased that freezing. They also replicate previous findings that neuronal activity in the BLA is necessary for the acquisition of second-order conditioned fear (Gewirtz and Davis 1997; Parkes and Westbrook 2010; Holmes et al. 2013) and show for the first time that the role of the IL stands in opposition to that of the BLA. In sum, these results are consistent with the hypothesis that, during second-order conditioning: (i) the BLA encodes an S2-danger association that is expressed in freezing to S2, (ii) the IL encodes an association between S2 and the expected-but-absent shock (S2-safety), and (iii) the subsequent test level of freezing elicited by S2 reflects a balance of the 2 associations with the danger association dominating.

Experiment 2: The IL retrieves the S2-safety association that forms during second-order conditioning

If rats encode opposing associations in second-order conditioning—an S2-danger association in the BLA and an S2-safety association in the IL—then suppressing retrieval of the safety association at test should remove its opposition to the danger association, thereby increasing the levels of freezing. Experiment 2 tested this prediction by silencing the IL at the S2 test. The protocol consisted in expressing the inhibitory hM4Di or mCherry virus in the IL and repeating the first- and second-order procedures used in Experiment 1 (Fig. 2B). Briefly, rats were exposed to 4 S1-shock pairings in stage 1 and 8 S2-S1 pairings in stage 2. They were then tested for freezing to S2 across 2 sessions. Test 1 was preceded by a systemic injection of clozapine or vehicle, while Test 2 was conducted drug-free. We predicted that chemogenetic silencing of the IL prior to testing would impair retrieval of the S2-safety association that had been encoded there during second-order conditioning. This would remove its opposition to the S2-danger association and thereby result in S2 eliciting more freezing at test.

Two rats were excluded from analyses due to misplaced viral injections, resulting in the following group sizes: Control-Vehicle (Ctrl-Veh), n = 8; Control-Clozapine (Ctrl-CZ), n = 9; IL-Vehicle (IL-Veh), n = 10; and IL-Clozapine (IL-CZ), n = 10. Representative images of viral expression and the extent of its expression are shown in Fig. 2A and B.

Figure 2C shows freezing in stages 1 and 2 in Experiment 2. All rats acquired freezing to S1 across its pairings with shock in stage 1 (Fig. 2C, left). There was a significant linear trend (F_1,33 = 113.662, P < 0.001, η_p² = 0.775) but no significant between-group differences in the overall level of freezing or trial × group interaction (F_s < 1). During second-order conditioning (Fig. 2C, right), freezing to S2 was low during the initial trials and increased significantly across subsequent S2-S1 pairings: there were significant linear and quadratic trends in the freezing across trials (linear: F_1,33 = 8.415, P = 0.007, η_p² = 0.203; quadratic: F_1,33 = 21.296, P < 0.001, η_p² = 0.391). Unexpectedly, there was a significant main effect of virus condition (F_1,33 = 4.21, P = 0.050, η_p² = 0.110): rats which had the hM4Di receptor expression froze less than mCherry controls overall. This, however, did not interact with the injection condition (F < 0.299) and there was no significant difference in freezing between rats in Groups CZ and Veh (i.e. no main effect of injection condition; F < 1). Freezing to S1 declined across the S2-S1 pairings. There was a significant linear trend across trials (F_1,33 = 13.866, P = 0.001, η_p² = 0.296), but there were no between-group differences in freezing to S1 or trend × group interaction (F_s < 2.5).

Prior to Test 1 (Fig. 2D, left), half the rats in each virus condition received an injection of clozapine while remaining rats received an injection of vehicle. There were no injections prior to Test 2 (Fig. 2D, right). All rats showed a decline in freezing across testing (significant main effect of test: F_1,33 = 7.303, P = 0.011, η_p² = 0.181; see also Fig. S3). There was a significant 3-way interaction between test, virus condition, and injection condition (F_1,33 = 5.784, P = 0.022, η_p² = 0.149). Follow-up analyses revealed that rats in Group IL-CZ froze more than those in Group IL-Veh at Test 1 (F_1,33 = 10.369, P = 0.003, η_p² = 0.239) and that rats in these groups froze at an equivalent low level in Test 2 (F < 1). In contrast, rats in Groups Ctrl-CZ and Ctrl-Veh exhibited an equivalent level of freezing to S2 in both test sessions (F_s < 1.2) and there was no significant difference in freezing between these 2 groups.

Experiment 2 shows that silencing neuronal activity in the IL increases the test levels of freezing elicited by the second-order conditioned S2. This result is consistent with the hypothesis that the IL encodes an S2-safety association during second-order conditioning and that activity in the IL is necessary for retrieval and expression of this association at the time of testing. Consequently, when the IL is silenced, the S2-safety association cannot be retrieved, resulting in unopposed (and thereby, enhanced) expression of the S2-danger association that is encoded in the BLA (Gewirtz and Davis 1997; Parkes and Westbrook 2010; Holmes et al. 2013; Lay et al. 2018).

It is worth noting that enhanced freezing to S2 was observed when the IL was silenced in Test 1 but not when it was functioning normally in Test 2: here, levels of freezing were equivalent and low in IL rats injected with either clozapine or vehicle. We take this result to mean that, even though the IL was silenced in Test 1, rats were able to extinguish freezing to S2 via substrates in the BLA. This is consistent with previous work showing that neuronal activity in the BLA is necessary for extinction of second-order conditioned freezing (Parkes and Westbrook 2010; Holmes et al. 2013).

Experiment 3: The effect of silencing the IL at test is specific to the second-order conditioned S2 as silencing the IL fails to affect a sensory preconditioned S2

The results obtained in Experiments 1 and 2 are consistent with the hypothesis that, during second-order conditioning, S2-danger and S2-safety associations form together and are encoded in the BLA and IL, respectively. An extension of this hypothesis is that the IL should fail to enhance the test levels of freezing to an S2 that had never signaled the absence of shock. One protocol that generates such an S2 is sensory preconditioning. This protocol is identical to second-order conditioning except that the order of the S1-shock and S2-S1 pairings is reversed: that is, subjects are exposed to S2-S1 pairings in stage 1 and then to S1-shock pairings in stage 2. As in second-order conditioning, test presentations of S2 alone elicit freezing, which we have shown to be associatively mediated (Holmes et al. 2013; Holmes and Westbrook 2017; Kikas et al. 2021; Parkes and Westbrook 2010; Wong et al. 2019, see also Fig. S2, see online Supplementary Material for a color version of this figure). However, unlike second-order conditioning, a sensory preconditioned S2 is not associated with the absence of the shock (basis of the safety signal) as the S2-S1 pairings in stage 1 precede S1-shock pairings in stage 2. Accordingly, Experiment 3 had 2 aims. The first was to assess if the IL regulates freezing responses to a sensory preconditioned S2. If the IL selectively regulates freezing to an S2 that has signaled the absence of shock, the effects of silencing the IL in this experiment should differ from those observed in the previous ones: specifically, silencing the IL should have no effect on freezing to the sensory preconditioned S2. The second aim of Experiment 3 was to replicate our previous finding that silencing neuronal activity in the BLA disrupts expression of freezing to a sensory preconditioned S2 (Parkes and Westbrook 2010; Holmes et al. 2013). Achieving these 2 aims would allow us to interpret the predicted null effect of silencing neuronal activity in the IL on freezing to the sensory preconditioned S2.

The inhibitory hM4Di or mCherry control virus was expressed in the IL or BLA 5 weeks prior to the start of experimental procedures. Rats were exposed to 8 S2-S1 pairings in stage 1 followed by 4 S1-US pairings in stage 2. They were then presented with S2 alone 8 times during Test 1, which was preceded by a systemic injection of clozapine or vehicle. Rats were tested again with 8 S2 alone presentations the next day in the absence of any injection (Test 2). The 2 groups of rats that had the control virus expressed in the IL and BLA were combined to form a single control group as there were no statistically significant differences in their levels of freezing across S1-shock pairings or testing of S2 (F_s < 0.9). After 4 rats were excluded due to misplaced viral injections, the group sizes were: Control-Vehicle (Ctrl-Veh), n = 16; Control-Clozapine (Ctrl-CZ), n = 15; IL-Vehicle (IL-Veh), n = 10; IL-Clozapine (IL-CZ), n = 10; BLA-Vehicle (BLA-Veh), n = 10; and BLA-Clozapine (BLA-CZ), n = 10). Representative images of viral expression and the extent of this expression are shown in Fig. 3A and B.

Fig. 3

The BLA, but not the IL, regulates the expression of sensory preconditioning. A) Representative images of control and hM4Di expression in the IL (top) and BLA (bottom); scale bars = 400 μm. B) Placement of viral injections; distances indicate anterior–posterior coordinates relative to bregma. C) Sensory preconditioning procedure used in Experiment 3; systemic injections of vehicle (Veh) or clozapine (CZ) were administered prior to Test 1. D) Freezing to S1 across stage 2 of training. E) Freezing to S2 across Test 1 (left) and Test 2 (right) in Experiment 3. All error bars indicate ±1 SEM. ^*P < 0.05; ^**P ≤ 0.01.

All rats acquired freezing to S1 across its pairings with shock in stage 2 of sensory preconditioning. There was a significant linear trend in freezing across the pairings (F_1,65 = 319.035, P < 0.001, η_p² = 0.831; Fig. 3D) but no significant between-group differences in the overall levels of freezing or trend × group interactions (F_s < 1.1). Rats received an injection of clozapine or vehicle before being tested across S2 alone presentations (Test 1; Fig. 3E, left) and were tested again the next day without any prior injection (Test 2; Fig. 3E, right). There was a significant main effect of test, indicating a decline in freezing across testing (F_1,65 = 4.636, P = 0.035, η_p² = 0.067; see also Fig. S3.). There was also a significant 3-way interaction between test, virus condition (Ctrl, IL, and BLA), and injection condition (F_2,65 = 6.748, P = 0.002, η_p² = 0.173). Follow-up analyses revealed that silencing neuronal activity in the BLA reduced freezing to S2 at Test 1 (F_1,65 = 10.777, P = 0.002, η_p² = 0.142) but increased freezing to S2 at Test 2 (F_1,65 = 5.081, P = 0.028, η_p² = 0.072), indicating that silencing the BLA had not only reduced freezing but had also impaired the extinction learning produced by the S2 alone presentations. In contrast, among rats that received the mCherry control virus or the inhibitory hM4Di virus in the IL, clozapine injections had no detectable effects on freezing in either test session (F_s < 1).

This experiment has confirmed that neuronal activity in the BLA is necessary for the expression and extinction of sensory preconditioned fear: chemogenetic silencing of the BLA prior to Test 1 acutely reduced freezing to the sensory preconditioned S2 but increased freezing to S2 during the final, drug-free Test 2 (Parkes and Westbrook 2010; Holmes et al. 2013). It has also shown that chemogenetic silencing of the IL had no effect on expression of sensory preconditioned fear: relative to controls, rats for which the IL was silenced showed just as much freezing to S2 in Test 1, indicating that sensory preconditioned fear was intact; and just as little freezing to S2 in Test 2, indicating successful extinction of the sensory preconditioned S2. This latter result shows that, in contrast to the BLA, which is critical for the expression of freezing responses to both a second-order and a sensory preconditioned S2, the role of the IL is specific to the expression of freezing responses to the second-order S2 (Experiments 1 and 2). More generally, these results add to the evidence that the BLA and IL play dissociable roles in the encoding of S2-danger and S2-safety associations: the BLA encodes S2-danger associations and regulates fear conditioning independently of the training protocol, whereas the IL encodes S2-safety associations and selectively regulates second-order conditioning, where, uniquely, S2-danger and S2-safety associations form together.

Validation of chemogenetic silencing in the IL and BLA

We validated our chemogenetic approach in 2 ways: first, with immunofluorescent staining of pERK as a marker of neuronal activity (Fig. 4A and B); and second, with ex-vivo electrophysiology (Fig. 4C and D). For pERK immunofluorescent staining, rats from Experiment 3 (n = 68) were re-exposed to 8 S1 extinction trials to induce neuronal activity in the IL and BLA, removed from the chambers, and perfused. Prior to S1 exposure, rats received a systemic injection of clozapine or vehicle. Representative confocal images (Fig. 4A) show reduced expression of pERK in the rats that received the combination of hM4Di receptor expression and clozapine injections in both the IL (top) and BLA (bottom). This was confirmed by statistical analyses (Fig. 4B), where for both the IL and BLA, there was a significant group virus condition × injection interaction (IL: F_1,28 = 6.096, P = 0.019, η_p2 = 0.169; BLA: F_1,26 = 5.498, P = 0.027, η_p2 = 0.175). Follow-up simple effects analyses showed that IL-CZ and BLA-CZ rats had less expression of pERK than vehicle-injected counterparts (IL-CZ vs. IL-Veh: F_1,28 = 12.239, P = 0.001, η_p2 = 0.290; BLA-CZ vs. BLA-Veh: F_1,26 = 8.154, P = 0.008, η_p2 = 0.239). Clozapine injections had no effect on pERK expression in mCherry controls (F < 1). A similar result was obtained when we quantified the co-expression of pERK and mCherry (Fig. S4, see online Supplementary Material for a color version of this figure).

Fig. 4

Verification of chemogenetic inhibition. A) Reduced expression of pERK following clozapine injections in hM4Di rats only: representative immunofluorescence for pERK in the IL (top) and BLA (bottom). Scale bars indicate 50 μm. B) Quantification of pERK expressed relative to the area of regions of the IL (left) and BLA (right). C and D) Clozapine reduced neuronal excitability in hM4Di-infected cells only: representative electrophysiological recording traces (left) from IL cells (C) and BLA cells (D) infected with control mCherry (top) and hM4Di (bottom), change in neuronal firing following application of clozapine relative to baseline (right). Insets show confocal images of biocytin-filled cells. BL = baseline; CZ = clozapine. All error bars indicate ±1 SEM. ^*P < 0.05; ^**P ≤ 0.01.

For the ex-vivo electrophysiological recordings (Fig. 4C and D), the change in neuronal excitability following clozapine application was analyzed relative to baseline firing. There was a significant interaction between group and change in neuronal activity in both the IL (F_1,8 = 7.360, P = 0.027, η_p² = 0.479; Fig. 4C) and BLA (F_1,9 = 7.364, P = 0.024, η_p² = 0.450; Fig. 4D). Follow-up simple effects analyses indicated that the application of clozapine reduced the firing of IL and BLA neurons transfected with the hM4Di virus (IL: F_1,5 = 20.610, P = 0.006, η_p² = 0.805; BLA: F_1,5 = 8.855, P = 0.031, η_p² = 0.639) but had no effect in IL and BLA neurons transfected with the mCherry control virus (F_s < 1).

Across 2 validation methods, we found chemogenetic silencing to be effective. In rats that had hM4Di receptor expression, clozapine injections reduced neuronal activity in response to S1 extinction trials as indexed by pERK immunofluorescence in both the IL and BLA. Similarly, bath application of clozapine during ex-vivo electrophysiological recordings from hM4Di-infected cells in IL and BLA produced a reduction in the number of action potentials fired. Critically, across both validation methods, clozapine had no effect on cells infected with the mCherry control virus. This confirms that the observed group differences in conditioned responding to S2 were due to neuronal silencing of the IL and BLA.

Discussion

This study tested the hypothesis that S2-danger and S2-safety associations form together in second-order conditioning, when the previously shocked S1 is preceded by S2 but not followed by shock. It did so by exploiting what is known about where these associations are encoded in the brain. We predicted that the S2-danger association is encoded in the BLA and that silencing this region during second-order conditioning would therefore reduce the expression of second-order conditioned responding (freezing) at test. We also predicted that the S2-safety association is encoded in the IL and that silencing this region would therefore increase the expression of second-order conditioned freezing at test. Both predictions were confirmed. In Experiment 1, rats were exposed to S1-shock pairings in stage 1 and S2-S1 pairings in stage 2. Relative to controls, rats for which the BLA was silenced prior to S2-S1 pairings froze significantly less when tested with S2, while rats for which the IL was silenced prior to S2-S1 pairings froze significantly more when tested with S2. Notably, rats for which the BLA were silenced also froze less to S1 across the S2-S1 pairings, suggesting that part of the impairment in second-order conditioning could relate to disrupted retrieval of the first-order conditioned S1, the learned source of danger that conditions S2 in the second-order protocol. In contrast, there were no acute effects of silencing the IL on freezing to S1 across the S2-S1 pairings, suggesting that the increased freezing to S2 at test was not due to increased retrieval/expression of the first-order conditioned S1. Instead, these results support our hypothesis that second-order conditioning involves the formation of both S2-danger and S2-safety associations that exert contrasting effects on responding to S2: the former is encoded in the BLA and promotes responding to S2, whereas the latter is encoded (or consolidated) in the IL and opposes responding to S2. There was also no acute effect of IL silencing on freezing to S2, which is consistent with previous work demonstrating effects of IL manipulation on extinction learning at test in the absence of acute effects during extinction itself (Laurent and Westbrook 2008, 2009). Such findings have been taken to imply that the IL regulates consolidation and retrieval of S2-no US associations rather than their acquisition (Quirk and Mueller 2008; Laurent and Westbrook 2009). It is likely that the IL plays a similar role in the development the S2-safety association during second-order conditioning.

Having established that second-order conditioning involves encoding of both S2-fear and S2-safety associations, the next experiments examined whether the IL regulates retrieval of the S2-safety association at test. In Experiment 2, after establishing second-order conditioned freezing to S2, activity in the IL was silenced prior to test presentations of S2 alone. Relative to controls, rats subjected to this treatment froze significantly more to S2, indicating that retrieval of the S2-safety association was disrupted. Surprisingly, increased freezing to S2 was only observed when these rats were tested under acute silencing of the IL (Test 1), with no effect of IL silencing on freezing to S2 in the subsequent drug-free test (Test 2). This suggests that silencing the IL had spared any long-term extinction learning produced by the S2 alone presentations on Test 1.

The specificity of the IL in encoding the S2-safety association was confirmed in Experiment 3, which examined whether the IL is involved in expressing conditioned freezing to a sensory preconditioned S2; a stimulus whose history had involved an association with S1 but not with the absence of shock. Here, after rats had been exposed to S2-S1 pairings in stage 1 and S1-shock pairings in stage 2, the IL was again silenced prior to test presentations of S2 alone. Rats subjected to this treatment exhibited the same high level of freezing as controls in the initial test session, showing that the expression of preconditioned fear was unaffected, and the same low level of freezing as controls in the subsequent test session, suggesting that silencing the IL had failed to affect any extinction learning produced by the S2 alone presentation on Test 1. This failure was in contrast to the effect of silencing the BLA at Test 1: relative to controls, rats for which the BLA was silenced prior to Test 1 exhibited less freezing to S2 during this test but more freezing to S2 in the final drug-free test. These results replicate previous findings that the BLA is necessary for expressing and extinguishing sensory preconditioned freezing responses (Parkes and Westbrook 2010; Holmes et al. 2013). Together with the results of the previous experiments, these findings have 2 major implications. First, the role of the IL in fear expression is specific to an S2 that has signaled the absence of a shock US: silencing the IL enhances fear to a second-order S2 which has signaled the absence of such a US but does not enhance fear to a sensory preconditioned S2 which has never signaled the absence of a shock US. Second, the IL is not necessary for extinction of a second-order or sensory preconditioned CS.

It is important to note that, while extinction can occur in the absence of a functioning IL, as shown here, it is likely to be regulated by a normally functioning IL (Milad and Quirk 2002). That is, we suppose that the IL is ordinarily involved in both acquisition and extinction of second-order fear and, furthermore, that the IL interacts with the BLA to regulate these effects, just as the extinction of first-order fear has been shown to require such interactions (Sierra-Mercado et al. 2011; Orsini and Maren 2012; Cho et al. 2013; Asede et al. 2015; Bloodgood et al. 2018). Briefly, extinction of first-order fear has been associated with IL projections to inhibitory interneurons in the lateral and basal amygdala nuclei and the intercalated cell masses (Herry et al. 2008; Quirk and Mueller 2008; Tovote et al. 2015). These interneurons act as an interface between the BLA, which receives thalamic and cortical afferents that convey information about the auditory or visual S1, and downstream circuits that regulate different types of fear responses (e.g. projections from the CeA to regions of the midbrain and hypothalamus; Ehrlich et al. 2009). When these interneurons are activated by the IL, they disrupt neurotransmission from the BLA to CeA, and thereby, an S1’s capacity to elicit fear responses (Likhtik et al. 2008; Herry et al. 2010; Tovote et al. 2015). We predict that the S2-safety association that forms in second-order conditioning is encoded in this same circuitry and, hence, that the level of responding to S2 reflects a balance between BLA-dependent processes that drive fear expression and interactions between the BLA and IL that have the opposite effect. Future work will examine whether BLA and IL interactions regulate the S2-safety association that forms in second-order conditioning, and whether IL-BLA and BLA-IL pathways are selectively involved in its encoding and/or expression. For example, the BLA-IL pathway may be critical for encoding the S2-safety association as the BLA provides the IL with information about the absent US; the IL-BLA pathway may be critical for the retrieval/expression of the S2-safety association as the amygdala coordinates activity in midbrain regions that regulate fear responses.

Inhibitory learning in second-order conditioning protocols increases with the number of S2-S1 pairings in stage 2 (Herendeen and Anderson 1968; Holland and Rescorla 1975; Muñiz-Diez et al. 2021; Rescorla 1973; Stout et al. 2004; Yin et al. 1994). That is, the S2-danger association evident after a few S2-S1 pairings ceases to be evident as the number of S2-S1 pairings increases, and with further S2-S1 pairings, the S2 passes various tests of its inhibitory properties (Yin et al. 1994; Muñiz-Diez et al. 2021). Such findings have been taken to suggest that subjects in these experiments (animal and human) eventually detect the omission of the expected US, triggering formation of S2-no US associations (Rescorla 1972, 1973; Lee and Livesey 2012). Instead, the present findings suggest that S2-danger and S2-safety associations form together in second-order conditioning: i.e. subjects detect the omission of the expected US while continuing to treat the S1 stimulus as dangerous. Future work will examine how omission of the expected US is detected early in second-order conditioning, and whether the S2-danger and S2-safety associations in second-order conditioning can be dissociated by other means. For example, inhibitory associations accrued to a stimulus fail to be retrieved when it is presented in a different physical context or after the lapse of time. Hence, testing an S2 in a different context to the one in which it was conditioned, or after the lapse of time, may block retrieval of the S2-safety association, and thereby, increase fear to S2 in the same way that silencing neuronal activity in the IL does. Future work will also seek to reproduce these results using male rats, as the current study used female rats only. While we cannot rule out potential sex differences, our results here replicate several findings from our previous experiments which were obtained using male rats (Parkes and Westbrook 2010; Holmes et al. 2013, 2014; Wong et al. 2019).

In summary, this study has shown that S2-danger and S2-safety associations form together in second-order fear conditioning. This was revealed by targeting the neural substrates of these types of associations in the BLA and IL. The BLA encodes the S2-danger association that drives second-order fear, whereas the IL encodes an S2-safety association that opposes expression of that fear. The study additionally showed that the IL remains critical for the retrieval/expression of the S2-safety association at the time of testing and, furthermore, that its role in regulating fear expression is specific to an S2 that was correlated with the absence of the US. Future work will examine how the BLA and IL cooperate to determine conditioned fear: specifically, how omission of the US in second-order conditioning triggers the formation of the S2-safety association in the IL, while animals acquire and maintain fear to S2 and S1, respectively. This work will also examine whether the S2-danger and S2-safety associations formed in second-order conditioning can be dissociated by testing the S2 in a different physical or temporal context to the one in which it was conditioned: such manipulations should selectively disrupt retrieval of the S2-safety association and, thereby, enhance the expression of second-order fear.

Acknowledgments

The authors would like to acknowledge Anne Rowan and Jennifer Strempel for animal care and husbandry.

Funding

This work was supported by a project grant (number 1146999) from the National Health and Medical Research Council, Australia.

Conflict of interest statement: The authors declare no conflict of interest.

Author notes

Vincent Laurent and Nathan M. Holmes contributed equally to this work.