An Infralimbic Cortex Engram Encoded during Learning Attenuates Fear Generalization

Abstract

Generalization allows previous experience to adaptively guide behavior when conditions change. The infralimbic (IL) subregion of the ventromedial prefrontal cortex plays a known role in generalization processes, although mechanisms remain unclear. A basic physical unit of memory storage and expression in the brain is a sparse, distributed group of neurons known as an engram. Here, we set out to determine whether an engram established during learning contributes to generalized responses in IL. Generalization was tested in male and female mice by presenting a novel, ambiguous, tone generalization stimulus following pavlovian defensive (fear) conditioning. The first experiment was designed to test a global role for IL in generalization using chemogenetic manipulations. Results show IL regulates defensive behavior in response to ambiguous stimuli. IL silencing led to a switch in defensive state, from vigilant scanning to generalized freezing, while IL stimulation reduced freezing in favor of scanning. Leveraging activity-dependent “tagging” technology (ArcCreER^T2 × eYFP system), an engram, preferentially located in IL Layer 2/3, was associated with the generalization stimulus. Remarkably, in the identical discrete location, fewer reactivated neurons were associated with the generalization stimulus at the remote timepoint (30 d) following learning. When an IL engram established during learning was selectively chemogenetically silenced, freezing increased. Conversely, IL engram stimulation reduced freezing, suggesting attenuated fear generalization. Overall, these data identify a crucial role for IL in suppressing generalized conditioned responses. Further, an IL engram formed during learning functions to later attenuate a conditioned response in the presence of ambiguous threat stimuli.

Significance Statement

Generalization refers to the ability of organisms to use previous experience to guide behavior when environmental conditions change. Despite the immense importance of generalization in adaptive behavior, the precise brain mechanisms remain unknown. Here, we identified a small population of neurons, known as an engram, in a discrete region of the frontal cortex that was associated with the expression of generalization related to a threatening situation. When these cells were turned off, generalization increased. When they were turned on, generalization decreased. Considering that overgeneralization of threatening stimuli is a known fundamental dimension of both anxiety and posttraumatic stress disorders, these findings have implications for our understanding of not only intrinsic generalization processes but also highly prevalent clinical disorders.

Introduction

When confronted with a clear and present threat, organisms must respond appropriately for defense. Sometimes threats are unclear, or ambiguous, which might call for a more generalized response, based on past experience (Richards and Frankland, 2017). In the field of psychology, generalization refers to the transfer of conditioned responses to stimuli that are similar, but not identical, to the original conditioned stimulus (Guttman and Kalish, 1956). Generalization was first described over a century ago (Watson and Rayner, 1920; Pavlov, 1927), is highly conserved across species (Orr and Lukowiak, 2008; Dymond et al., 2015), and has even been proposed as one of the only universal laws in the field of psychology (Shepard, 1987). While generalization is evolutionarily adaptive, overgeneralization of defensive behavior is maladaptive and relevant to understanding posttraumatic stress disorder (PTSD) and anxiety-related disorders (Dunsmoor and Paz, 2015; Cooper et al., 2022). The fundamental nature of stimulus generalization, in both the study of intrinsic memory processes and clinical disorders, makes the investigation of generalization a central topic in the field of neuroscience (Sangha et al., 2020).

The ventral medial prefrontal cortex (vmPFC), and infralimbic (IL) subregion, is an identified hub in the defensive conditioning circuit (Sotres-Bayon et al., 2006). Functionally, the role of IL in extinction processes is canonical (Giustino and Maren, 2015). A growing number of studies have also indicated IL functionality in generalization (Sangha et al., 2014; Corches et al., 2019; Scarlata et al., 2019; Bayer and Bertoglio, 2020; Day et al., 2020; Kreutzmann et al., 2020; Kreutzmann and Fendt, 2020; Ng et al., 2023). The consensus from these studies is that IL functionality mediates memory specificity by attenuating generalization.

The term “engram” refers to a sparse, distributed, group of neurons that forms a physical substrate of memory in the brain (Josselyn et al., 2015). An approach for studying engrams is the use of transgenic systems for the selective expression (or “tagging”) of fluorescent molecules, or optogenetic and chemogenetic actuators, during learning for later visualization and manipulation during memory retrieval (DeNardo and Luo, 2017). One such system is the ArcCreER^T2 × eYFP mouse line. In this system, activity-regulated cytoskeletal-associated (Arc) gene transcription, a key component in learning- and memory-induced synaptic plasticity, leads to the expression of a CreER^T2 fusion protein which, upon activation with a synthetic estrogen receptor modulator (i.e., 4-OHT), leads to Cre-induced recombination and expression of a fluorescent molecular tag (Denny et al., 2014).

Generalization has long been proposed to be established during learning (Hull, 1943), although this hypothesis has never been directly tested. Here, we designed a set of experiments to determine whether an IL engram, encoding an inhibitory process, is established during pavlovian defensive conditioning. We hypothesize that after learning, this engram contributes to memory specificity by reducing conditioned responses to generalization stimuli. The ArcCreER^T2 × eYFP transgenic mouse system is particularly advantageous for addressing this question because neurons activated during learning can be later manipulated under conditions that promote generalization.

To first establish a role for IL in cued fear memory generalization, we used a designer receptor exclusively activated by designer drug (DREADD) system and CaMKII promoter to bidirectionally control IL excitability during the expression of generalization. In the second set of experiments, we took advantage of the ArcCreER^T2 × eYFP system to visualize and measure IL engrams associated with generalization at recent and remote timepoints following learning. In the final set of experiments, a double-floxed inverse open reading frame (DIO)-DREADD system and hSyn promoter in combination with ArcCreER^T2 × eYFP transgenic mice was used to transfect DREADD constructs in Arc-expressing neurons during learning for later synthetic chemical manipulation during memory retrieval.

Materials and Methods

Animals

Adult male and female ArcCreER^T2 × eYFP mice were used in all experiments. B6.Cg-Tg(Arc-cre/ERT2)MRhn/CdnyJ (ArcCreER^T2) mice were crossed with B6.129X1-Gt(ROSA)26Sor^tm1(eYFP)Cos/J (eYFP) mice (The Jackson Laboratory strain #022357 and strain #006148, respectively) to produce double transgenic mice (hereafter referred to ArcCreER^T2× eYFP mice). ArcCreER^T2 × eYFP mice were bred on-site over multiple generations. At the memory retrieval stage of all experiments, adults ranged from 71 to 150 d old (mean ± SD  = 112.5 ± 27.8; female weight, 16.4–32.9 g; median weight, 22.1; male weight, 21.9–42.8 g; median weight, 30.4). Same-sex mice were group-housed (2–5/cage) in individual vented standard cages, except when singly housed after surgery. There were three types of enrichment in each cage, including a wood gnawing block, nestlet, and EnviroPAK. The vivarium temperature (23–25°C), humidity (35–37%), and 12 h light/dark cycle (lights on 0600) were controlled throughout. Food and water were available ad libitum, and cages were changed once per week. Mice were randomly assigned to experimental groups based on litter before the start of each experiment. All experimental procedures were conducted in accordance with the National Institutes of Health guidelines on the Care and Use of Animals in Research and approved by the Vassar College Institutional Animal Care and Use Committee (IACUC). Disclosure of animal housing, husbandry, and experimental procedures follows principles for transparent reporting and reproducibility in behavioral neuroscience (Prager et al., 2011, 2018).

Genotyping protocol

Tissue biopsy was performed by tail snip under brief isoflurane anesthesia. DNA was extracted using DirectPCR Lysis Reagent (Viagen Biotech) following manufacturer protocol. Amplification of Cre and R26R was performed using the following primer sets: Cre, 5′-GCC TGC ATT ACC GGT CGA TGC AAC G-3′; 5′-AAA TCC ATC GCT CGA CCA GTT TAG TTA CCC-3′; R26R (for eYFP mice), 5′-GGA GCG GGA GAA ATG GAT ATG-3′; 5′-AAA GTC GCT CTG AGT TGT TAT-3′; 5′-AAG ACC GCG AAG AGT TTG TC-3′, following recommended cycling protocols from The Jackson Laboratory and Denny et al. (2014).

General behavioral experimental procedures

All experiments were performed during the light cycle. A background strain of the ArcCreER^T2 × eYFP mouse is the C57BL/6J (B6) mouse, which is an age-related hearing decline model (Ison et al., 2007). Hearing in B6 mice declines with age. To address this, no mice over 150 d old were tested. We also evaluated peripheral auditory brainstem responses to relevant frequencies (see below). All mice were habituated to a holding room 30–45 min prior to conditioning and testing. To reduce contextual (background) freezing, the training context (hereafter referred to as “Context A”) was disguised from the testing context (hereafter referred to as “Context B”) using several manipulations. Context A was an unmodified fear conditioning chamber (Coulbourn Instruments), and a 70% EtOH solution was used to clean the chambers between mice. For Context B, (1) mice were transferred from the vivarium to the holding room using distinctive cages, carts, and covering, (2) the ambient lighting and background noise of the holding and testing room were changed using different illuminance and a fan, (3) a white plexiglass floorboard sprinkled with clean bedding was used to cover the shock bars, (4) the testing chamber walls were disguised with black and white striping, and (5) the chambers were cleaned with a 1% acetic acid solution between mice (Bergstrom, 2020). Each day, prior to training and testing, the decibel (dB) level for the auditory tone frequency was measured in each chamber using a sound level meter (R8050, REED Instruments) and calibrated to 70–75 dBC. All conditioning was conducted in commercial chambers (20 × 30 × 18 cm) in sound-dampening cabinets (58 × 61 × 45 cm; Coulbourn Instruments). FreezeFrame 4 software was used for controlling and delivering the tone and footshock stimuli (ActiMetrics). All acoustic stimuli were delivered by a speaker mounted on the upper center of one wall, and footshock stimuli were delivered via a stainless-steel rod floor (0.8 cm distance between rods, 1.0 cm rod diameter).

Fear conditioning

Mice were placed in the fear conditioning chamber (Context A) 180 s prior to three pairings of an auditory tone CS (20 s, 5 kHz, 70–75 dBA) that coterminated with an electric footshock US (0.5 s, 0.5 mA). The CS/US pairings were separated by variable intertrial intervals (ITI; 20 and 80 s). Mice were removed from the chamber 60 s after the final CS/US pairing. The total conditioning time was 400 s.

Context B pre-exposure

To reduce “background” generalized contextual freezing (Jacobs et al., 2010), mice were placed into Context B for 15 min and left to explore on the day prior to the cued generalization test (Bergstrom, 2020).

Generalization test

Either 6 d (recent group) or 30 d (remote group) following training (1 d following Context B pre-exposure) mice were placed in Context B prior to three presentations either the CS (5 kHz, 70–75 dB, 20 s) or a novel tone generalization stimulus (GS; 3 kHz, 70–75 dB, 20 s). A 3 kHz tone GS was used because in previous experiments, following conditioning with a 5 kHz CS, it produced the greatest degree of generalization over time in comparison with alternate tone frequencies (Pollack et al., 2018). Mice were removed from the chamber 60 s after the final stimulus presentation and returned to the colony room (400 s total test time; ITIs 80 and 20 s). Mice in the “no-tone” control group were allowed to explore Context B for 400 s.

Genetic labeling procedures

To reduce nonspecific genetic labeling, mice were dark-housed the night before and 3 d following the 4-OHT injection. These methods have been previously validated (Denny et al., 2014; Cazzulino et al., 2016). For dark housing procedures, mice were placed in a separate housing room with lights off, limited noise, and no handling. ArcCreER^T2 × eYFP mice were intraperitoneally (i.p.) injected with 4-OHT exactly 5 h prior to fear conditioning. This timeframe has been previously validated (McGowan et al., 2024).

Drugs

4-Hydroxytamoxifen (4-OHT)

Cre-mediated recombination in ArcCreER^T2 × eYFP mice was induced using 4-OHT (Hello Bio; SKU, HB6040). 4-OHT was made fresh prior to each injection (i.p.). 4-OHT was dissolved by water bath sonication in a 10% EtOH/90% corn oil solution at 10 mg/ml. The final dose was 55 mg/kg.

DREADD actuators

Clozapine N-oxide (CNO)

For the CaMKII-DREADD experiments, CNO was used as the chemical actuator (Hello Bio; SKU, HB6149). It was made fresh prior to intraperitoneal injection, dissolved in 0.9% saline (1 mg/ml), and injected intraperitoneally (5.0 mg/kg) exactly 45 min prior to the memory test.

Deschloroclozapine dihydrochloride (DCZ)

For DIO-DREADD experiments, DCZ was used as the chemical actuator (Nagai et al., 2020; Hello Bio; SKU, HB9126). It was made fresh immediately prior to intraperitoneal injection, dissolved in 0.9% saline (1 mg/ml), and injected intraperitoneally at 100 µg/kg exactly 15 min prior to the memory test.

Immunohistochemistry

Tissue collection

Exactly 90min following the fear memory retrieval test, all mice were injected with a ketamine/xylazine cocktail (100:10 mg/ml) and transcardially perfused with ice-cold 1× PBS (7.2–7.4 pH), followed by ice-cold 4% paraformaldehyde (PFA) in 1× PBS (7.2–7.4 pH). The time point for perfusion following the retrieval test was based on several previous reports (Ploski et al., 2008; Maddox and Schafe, 2011). Brains were extracted and placed into 4% PFA overnight at 4°C and then transferred to 1× PBS and kept at 4°C until sectioning. Coronal brain sections (40 μm thick) were cut through the mPFC using a vibrating blade microtome (VT1200, Leica Biosystems). Every other section (to avoid double counting) was collected in a well plate containing 1× PBS (7.4 pH) for free-floating immunohistochemistry. Sections were rinsed first in 1× PBS (3 × 10 m), blocked in a 1× PBS/1% bovine serum albumin (BSA)/0.2% Triton X-100 solution for 1 h, and incubated in antigen-specific antibodies (see below). All incubations were performed on orbital shakers.

Arc and eYFP primaries

After blocking, sections were incubated overnight in anti-Arc rabbit polyclonal antibody (1:5,000; catalog #156003, Synaptic Systems) and anti-GFP chicken polyclonal antibody (1:10,000; catalog #13970, Abcam) at 4°C. The next day, sections were washed in 1× PBS (3 × 10 m) before a 1 h incubation in Alexa Fluor 594 donkey anti-rabbit IgG (1:1,000; catalog #A32754, Invitrogen) and Alexa Fluor 488 goat anti-chicken IgG (1:500; catalog #A32931, Invitrogen) at room temperature.

c-Fos

To test the efficacy of CNO to activate the DREADD system (either hM4Di or hM3Dq), a subset of mice, having completed all behavioral testing, received intraperitoneal injections of CNO at 5.0 mg/kg dosage and were placed into the Context B recall test after 45 min (peak expression time; Campbell and Marchant, 2018). Exactly 135 min after CNO injection and 90 min after recall (peak c-Fos expression; Barros et al., 2015), mice were killed for IHC. All procedures for IHC were identical to those described above except that sections were then incubated for 24 h with anti-c-Fos rabbit polyclonal antibody (1:500; catalog #RPCA-c-Fos, RRID: AB_2572236, EnCor Biotechnology) at room temperature. For the secondary antibody, subjects in either the pAAV-CaMKIIa-hM4Di-mCherry or pAAV-CaMKIIa-hM3Dq-mCherry groups were incubated with Alexa Fluor 488 goat anti-rabbit IgG (H + L; lot #1981125, REF: A11008, Invitrogen), and pAAV-CaMKIIa-EGFP subjects were incubated with Alexa Fluor 594 donkey anti-rabbit IgG (H + L; lot #WD319534, REF: A32754, Invitrogen) for 1 h at room temperature.

DIO-DREADD Arc

To test the efficacy of DCZ to activate the DIO-DREADD system (either DIO-hM4Di or DIO-hM3Dq) and modify Arc expression in preferentially transfected cell populations, mice were injected with DCZ 105 min prior to fear conditioning and processing for Arc IHC. The timeline for DCZ injection was based on high DCZ brain level concentrations at 15 min (Nagai et al., 2020) and peak Arc expression at 90 min following fear conditioning (Ploski et al., 2008; Maddox and Schafe, 2011). The experimental procedures for IHC were identical to those described above for Arc IHC except the secondary antibody for Arc was donkey anti-rabbit IgG (Alexa Fluor 647, 1:1,000, lot #YF374181, REF: A32795, Invitrogen).

Tissue mounting

Following all incubations, sections were rinsed a final time in 1× PBS (3 × 10 m) and then mounted onto gel-coated slides in 0.05 M PB. Sections were cover slipped (#1 or #1.5 thickness) using Fluoromount-G mounting medium with DAPI (#00-4959-52, Invitrogen) and sealed with nail polish. Slides were stored at 4°C in darkness until imaging.

Image acquisition and analysis

Images were acquired using a Leica TCS SP5 II laser scanning confocal with the Leica Microsystems LAS AF software (Version 2.6.0.7266). The objective used was a Leica HCX PL APO CS 20×/0.70 dry. 3D images were acquired by taking a z-stack of 20–30 slices with 1.14 µm spacing and a pixel dimension of 760 × 760 nm. Images for the DIO-DREADD experiment were captured using the Leica Stellaris 8 FALCON laser scanning confocal platform.

The experimenter was blind to experimental conditions throughout all engram quantification procedures. Cell counts were acquired by sampling across six IL regions/subject. Both hemispheres were included in the analysis. IL sampling regions for data collection were chosen based on the quality of the staining and visibility of anatomical landmarks for localization of the counting frame (see below). Labeled cells were quantified using FIJI. Background subtraction was applied across all channels. For c-Fos microscopy, images were captured under a fluorescent microscope (Nikon Eclipse 50i, Nikon Instruments). c-Fos⁺ cells were counted manually (FIJI–ImageJ open source) in the mPFC using a counting frame (250 × 250 µm) with AAV⁺ and adjacent AAV⁻ expression area and six locations/subject (mCherry⁺ n = 6, eYFP⁺ n = 6).

To locate IL and PL for placement of the counting frame, the lateral ventricle (LV) was used as an anatomical landmark, as it is readily identifiable and located in a consistent position ventral to the IL and aligned with DP, for a majority of the longitudinal axis. The midline and the corpus callosum were also used as stable anatomical landmarks to identify the location of the IL and PL. For the IL, the counting frame (250 × 250 µm) was positioned approximately 600 µm dorsal from the LV. For the PL, the counting frame was positioned 1,250 µm dorsal from LV. For layer measurements, the counting frame was centered 300 µm from midline for L2/3 (shallow layers) and 500 µm (deep) from midline for L5/6. Cells in the IL were counted between rostrocaudal levels of bregma at 1.93 and 1.53 (Paxinos and Franklin, 2019). DAPI cells were counted using the 3D object counter in FIJI. Fluor 488+, Fluor 594+, and colabeled cells were counted manually within the counting frame. Colocalized eYFP⁺:Arc⁺ neurons were first analyzed as a percentage of the number of DAPI⁺ neurons. Chance rate neuronal colocalization was calculated as follows: (eYFP⁺/DAPI⁺) × (Arc⁺/DAPI⁺) × 100.

DREADDs

Viral constructs

All viral constructs were purchased from Addgene. In the first set of experiments, pAAV-CaMKIIa-hM4Di-mCherry (AAV5), pAAV-CaMKIIa-hM3Dq-mCherry (AAV8), or a fluorophore-only control AAV (pAAV-CaMKIIa-EGFP; AAV5) was used. In the DIO-DREADD experiments, pAAV-hSyn-DIO-hM4D(Gi)-mCherry (AAV8), pAAV-hSyn-DIO-hM3D(Gq)-mCherry (AAV8), or fluorophore control AAV (pAAV-hSyn-DIO- mCherry (AAV8) was used. All viral vectors were aliquoted and stored at −80°C until use.

Surgery

Prior to surgery, mice received an injection of carprofen (5.0 mg/kg, s.c.) and an intradermal injection of bupivacaine (0.05 ml) at the craniotomy site. Inhaled isoflurane levels were maintained between 1.25% and 2.5% throughout. Mice were bilaterally microinjected (100–150 nl/hemisphere) with either the hM4Di, hM3Dq, or fluorophore control AAV targeting the IL (stereotaxic coordinates: AP +1.8, ML +0.3/−0.3, DV −2.8). Throughout chemogenetic experimentation, stereotaxic AAV transfection was preferentially targeted to bias away from dmPFC (caudate and PL) in favor of vmPFC (IL and DP). Microinjections were conducted using a 2.5 µl glass syringe and a 32-gauge needle (Hamilton). Following surgery, a dietary supplement (DietGel Recovery Purified) and carprofen (5.0 mg/kg, s.c.) were provided as needed. Subjects were singly housed for 2 weeks and group-housed, if possible, prior to behavioral testing. Fear conditioning occurred no less than 2 weeks following stereotaxic surgery. All fear conditioning and generalization experimental parameters were identical to those described above.

On generalization test day, mice were injected (i.p.) with CNO 45 min prior to testing. The following day, mice were injected (i.p.) with saline (same volume as CNO) using procedures identical to those described above for CNO and were run through the generalization test protocol again.

DIO-DREADD methods

To manipulate engram reactivation, a Cre-dependent viral construct expressing either an excitatory DREADD (pAAV-hSyn-DIO-hM3D(Gq)-mCherry), inhibitory DREADD (pAAV-hSyn-DIO-hM4D(Gi)-mCherry), or a fluorophore control (pAAV-hSyn-DIO-mCherry) was injected into the IL of ArcCreER^T2 × eYFP mice. After at least 2 weeks, on the fear conditioning day, all mice were injected with 4-OHT to drive Cre-recombination and DIO-DREADD expression in neurons with high levels of Arc. This permits specific expression of DREADD constructs in activated cellular populations for later reactivation or silencing using the actuator DCZ.

Either 6 d (recent group) or 30 d (remote group) following training (1 d following Context B pre-exposure), mice were injected with DCZ 15 min prior to the generalization test. All mice were sacrified for Arc IHC 90 min following testing (see above).

Behavioral analysis

Freezing

An overhead camera recorded digital video of the fear conditioning chamber. Freezing behavior was automatically quantified using FreezeFrame 4.0 (ActiMetrics). Freezing was defined as the lack of movement except for respiration for >1 s (FreezeFrame threshold = 5). Percentage freezing data was calculated by scoring freezing during the CS or GS presentations (20 s), pre-CS/GS period (habituation), and intertrial intervals (ITIs).

Scanning

Scanning was operationally defined as a side-to-side head and front paw movement while the tail base remained motionless. Scanning behavior was recorded immediately upon initiation and halted when the mouse (1) froze >1 s or (2) initiated full movement. Movement was defined as a larger category of behaviors, including four-paw locomotion, grooming, rearing, and small, and jerk-like body and head movements, not identified as freezing or scanning.

Pose estimation for each video was created using DeepLabCut (DLC) Version 2.3.10 (Mathis et al., 2018). Each video was 400 s long and recorded at 8 frames per second (fps), resulting in 3,000 frames/video. Fifty randomly sampled frames from each video were manually labeled, with the nose and tail base serving as the labeling points.

To train the ResNet-50 network, 95% of the labeled frames were used, while the remaining 5% were reserved for testing the neural network's performance. Each video underwent over 10,000 training iterations, yielding a training error of 8.02 pixels and a test error of 8.61 pixels. By applying a p-value cutoff of 0.4, the training error was reduced to 7.22 pixels, though the test error remained unchanged.

A separate Python script was developed to extract freezing and scanning behaviors from DeepLabCut (DLC) output, which included pose estimations of nose and tail base. This script averaged the coordinates over eight frames to determine the coordinates per second. It then calculated the velocity (pixels/s) of nose and tail base/s. The script identified freezing and scanning behaviors using criteria established through extensive comparisons between test results, hand-scored, and FreezeFrame 4.0 results (ActiMetrics). In DLC, freezing behavior was defined as the velocity of both the nose and tail base being <6 pixels/s with immobility lasting >1 s. Scanning behavior was defined as the nose velocity >10 pixels/s and tail base velocity <10 pixels/s.

Auditory brainstem responses

To determine auditory thresholds, a set of 5 ms tone bursts (1 ms Blackman–Harris gating) was generated in SigGenRZ (v 5.6.0). We generated stimuli at nine frequencies (1, 2, 2.5, 3, 3.15, 4, 8, 10, and 12.5 kHz) that spanned the frequency range of stimuli used in the generalization experiments. We also generated broadband clicks that were periodically presented to the animals to track physiological stability across the course of the experiment. Peak-to-peak equivalent stimuli levels were determined with a Larson Davis LxT sound level meter and a long-duration 1 kHz tone. We calibrated the frequency response of the speaker with long-duration tones in one-third octave bands with a Larson Davis LxT sound level meter (fast, z-weighting). Frequency-specific output levels were adjusted using the gain function in SigGenRZ until a flat frequency response was achieved (± 1 dB).

All tests were performed in a 1.8 m × 1.9 m × 2 m IAC Acoustics audiology booth lined with pyramidal acoustic foam to provide sound deadening. We used electrode placement and stimulus presentation rates previously used. Each subject was anesthetized with an injection (i.p.) of ketamine (90 mg/kg) and xylazine (10 mg/kg) solution. The subject was placed on a heating pad covered with surgical towels. When the animal no longer responded to a toe-pinch, we cleaned the skin with 70% isopropyl alcohol, and three 27-gauge 12 mm subdermal needles (Rochester Electro-Medical) were inserted: one noninverting (active) electrode at the vertex of the head, one inverting (reference) electrode directly below the auditory meatus of the right ear, and one grounding electrode directly below the auditory meatus of the left ear. The electrode leads were connected to a Tucker-Davis Technologies RA4LI head stage and RA4PA preamp, which were then fed into a TDT RZ6 processor via a fiber-optic cable. After placing the electrodes, the impedance of the electrode was checked and repositioned if necessary to maintain an impedance at or below 5 kΩ. During the experiment, layers of surgical towels were periodically added or removed to maintain body temperature.

Stimulus presentation and evoked potential recordings were coordinated by BioSigRZ (v 5.6.0), a POE5 signal processing card, and the RZ6 processor. Stimuli were presented at a rate of 31.1 stimuli s-1 from an Orb Mod2 satellite speaker (Orb Audio; frequency response, 0.12–15 kHz) positioned 10 cm from the right ear of the subject. Different stimulus amplitude intervals were used depending on known thresholds for each frequency to assess thresholds more rapidly. Typically, larger steps (10 or 20 dB) were used farther away from the threshold, and smaller steps (5 dB) were used near the threshold. The exact set of stimuli amplitudes varied by frequency. Two sets of 400 stimuli were played in alternating phases for each combination of stimulus frequency and amplitude. Between each set of frequencies, we also assessed the response to a pair of clicks presented at 80 dB to monitor the physiological stability of the subject. Evoked responses were notch-filtered at 60 Hz and bandpass filtered between 0.03 and 3 kHz.

Auditory thresholds at each frequency were determined using visual detection, where two trained observers independently identified the lowest stimulus amplitude evoking a response. Thresholds were estimated as the sound pressure level halfway between that of the last detectable response and the next quietest stimulus. Since stimulus intensities in threshold regions differed by 5 dB, ABR thresholds were defined as the intensity 2.5 dB below the lowest stimuli amplitude at which a response could be visually detected.

Experimental design and statistical analysis

For all behavioral data, a mixed ANOVA was used to compare the conditioned freezing response across CS/GS and ITI presentations (within-subject variable) across groups. Male and female mice were included in all experiments in a full factorial experimental design. All data were first checked for normality using Mauchly's test for sphericity. Violation of the assumption of sphericity was addressed by adjusting the degrees of freedom using the Greenhouse–Geisser correction. For some analyses, a generalization index was calculated and compared across groups. The generalization index was calculated by dividing mean CS by the sum of CS and GS [CS/(CS + GS)]. Although the generalization index range is 1 to 0, a value of 1 indicates no generalization, and a value of 0.5 indicates complete generalization.

For the ArcCreER^T2 × eYFP tagging experiments, a multivariate analysis of variance (MANOVA) was used to test the statistical relationship among eYFP⁺, Arc⁺, or eYFP⁺:Arc⁺ colabeled cells in L2/3 and L5/6 of the PL and IL at different frequencies [0 (control), 3, and 5 kHz] and time points (recent or remote) following learning. Statistics were run on the number of eYFP⁺/DAPI⁺, Arc⁺/DAPI⁺, and eYFP⁺:Arc⁺/DAPI⁺ neurons. Box’s M test was used to test the equality of variance–covariance matrices. Violation of the homogeneity assumption was followed up by a rank order transformation. A significant value for the conservative Pillai’s trace test statistic was only followed up by Bonferroni-corrected univariate ANOVAs. Follow-up ANOVAs were checked for the assumption of equality of covariance matrices using Levene’s test. The Welch test was used in the case of a significant Levene’s test. A significant ANOVA was followed up with a Scheffe post hoc test. Prior to analysis, outliers were determined by calculating the interquartile range for each group. Any values >1.5 steps beyond the interquartile range were considered outliers and removed from the analysis. These are reported in the results. Effect size (partial η²) is reported for all nonsignificant sex interactions and post hoc analysis of IL and DP functionality. For all statistics, significance was set at p < 0.05. All data are represented as the mean ± standard error of the mean (SEM). All group sizes can also be found in the figure captions. Group sizes were based on previous studies (Pollack et al., 2018; Scarlata et al., 2019). Statistics were run on SPSS (International Business Machines, v. 26).

Results

Experiment 1: bidirectional IL chemogenetic control during fear generalization

DREADD system efficacy

Microinjections of the stimulatory DREADD (hM3Dq), inhibitory DREADD (hM4Di), or fluorophore control were administered into IL at least 2 weeks prior to behavioral testing or c-Fos analysis (Fig. 1A). To assess the efficacy of the DREADD system, we analyzed mPFC c-Fos⁺ cell density after both hM3Dq and hM4Di CNO-induced stimulation. Results showed robust CNO-induced effects across DREADD constructs (F_(2,13) = 38.8; p < 0.001), with increased (p < 0.001), and decreased (p < 0.01), c-Fos⁺ cell density relative to the fluorophore control (Fig. 1B,C). All injections were stereotaxically biased toward vmPFC coordinates to avoid transfection in the dorsal medial prefrontal cortex (dmPFC). After exclusion of mice with dmPFC transfection, histological analysis mapping the extent of viral transfection across individually aligned brains in stereotaxic group space confirmed AAV transfection predominantly located to IL, with some expression in dorsal peduncular cortex (DP; Fig. 1D).

Figure 1.

Bidirectional chemogenetic IL manipulation during fear generalization expression. A, Experimental design. B, Representative confocal micrographs showing AAV⁺ expression (left panel), c-Fos⁺ FIH (middle panel), and AAV⁺/c-Fos⁺/DAPI⁺ coexpression (20×/0.8 NA air). Scale bar, 25 µm. C, There was a twofold increase in c-Fos⁺ expression in hM3Dq mice (n = 4) and nearly 0.5-fold decrease in c-Fos⁺ expression in hM4Di mice (n = 6) versus controls (n = 6). D, The extent of AAV expression in the mPFC across five coronal planes is depicted. Relatively darker shaded regions show greater overlap across individuals. Transfection was predominantly located in the IL, with some expression in DP. Mouse brain atlas images modified from (Paxinos and Franklin, 2004). E, IL stimulation (hM3Dq; n = 10; female n = 2; male n = 8) reduced freezing to the context and GS stimulus compared with hM4Di mice (n = 8; female n = 5, male n = 3) and fluorophore controls (n = 22; female n = 10, male n = 12). F, In a separate group of mice, there were no CNO-induced differences between IL stimulation (hM3Dq n = 5; female n = 4, male n = 1) and control (n = 12; female n = 8, male n = 4) mice in response to CS presentation. The next day, after vehicle injection, there was reduced freezing in the hM3Dq group. G, An unbiased k-means clustering algorithm run on all mice produced three prominent clusters (low, mid, and high). Chi-square analysis indicated all mice in the DIO-hM4Di were clustered in the “high” and “mid” group and most mice in the DIO-hM3Dq groups clustered in the “low” group. H, Analysis of the generalization index showed greater values (less generalization) in the hM3Dq group versus the hM4Di group and controls. I, DeepLabCut analysis of scanning behavior. Top panel, Markers for pose estimation on the tail base and nose. Bottom panel, Representative path analysis depicting low activity in tail base and high activity in the nose (scanning). J, Prior to GS presentation, there were no differences in scanning behavior between groups. Upon presentation of the GS, scanning collapsed in the hM4Di group but was maintained in the hM3Dq group. Bottom, Donut plots depict percentage freezing, scanning, and moving during the mean GS presentations. n = 8–22/group, *p < 0.05, **p < 0.01, ***p < 0.001. n.s., nonsignificant.

Fear generalization test during IL manipulation

CNO was injected prior to a cued fear memory generalization test to activate the DREADD construct. Results revealed a main effect of the DREADD manipulation (F_{(2, 7)} = 18.9; p < 0.001). In the hM3Dq group, GS-elicited freezing was reduced compared with the control (p < 0.001) and hM4Di group (p < 0.001; Fig. 1E,F). There was also a smaller, but significant, effect of the DREADD manipulation on pre-CS freezing (F_(2,37)] = 4.1; p = 0.024). hM3Dq mice froze less than hM4Di (p = 0.027). There were no sex interactions (p = 0.37, partial η² = 0.09). These data indicate IL excitatory stimulation suppresses the expression of auditory and context conditioned generalized freezing. On the next day, without CNO (vehicle injection), there was no difference in tone-elicited freezing between the control and either hM4Di (p = 0.54) or hM3Dq groups (p = 0.14), confirming the efficacy of the DREADD system and supporting a role for IL activity in modifying generalized freezing responses.

Because IL stimulation consistently decreased freezing in response to the GS, an important consideration is the generality of IL function in suppressing conditioned freezing responses. To test this question, we replicated the generalization protocol described above in a separate set of hM3Dq and fluorophore control mice, but rather than presenting the GS, we presented the CS. Results showed no difference in CS-eliciting freezing between hM3Dq and the control group when administered CNO (p = 0.29; Fig. 1F). However, on the following day, the same hM3Dq mice injected with vehicle showed a significant decrease in CS-elicited freezing as compared with the fluorophore control (p = 0.005), suggesting facilitation of extinction consolidation.

To address the localization of IL/DP expression with respect to the behavioral results, we compared mice with transfection localized in IL with IL/DP. For the hM4Di data, those in the IL/DP group (n = 3) showed increased freezing (68%) relative to the IL group (n = 5; 60%); however, it was not statistically significant (p = 0.70; partial η² = 0.026). For the hM3Dq data, again, those in the IL/DP group (n = 8) showed more freezing (19.5%) relative to the IL group (n = 2; 7.7%), although it was not statistically significant (p = 0.42; partial η² = 0.082). There were no mice with expression localized preferentially in DP but not IL.

IL stimulation, conditioned freezing, and extinction

In our paradigm, and especially in these initial experiments, presentation of GS produced relatively high levels of generalized freezing (60.5%) and relatively high variance (SD, 23.5). This finding led us to examine individual differences in generalized responses. An unbiased k-means clustering algorithm was applied to mean GS freezing levels across all mice. The analysis segregated three “high, mid, and low” clusters (Fig. 1G). A reanalysis of the fear conditioning acquisition data showed no differences between low, mid, and high groups, indicating this behavioral phenotype is not present during learning. A chi-square test was performed to test the relationship between experimental groups and clusters. Results showed a significant effect χ² (4, N = 40) = 20.9, p < 0.001, with mice in the hM4Di group preferentially clustered in a rightward skewed direction in “high” and “mid” generalization groups with none in the “low” group (Fig. 1G). Mice in the hM3Dq group clustered leftward almost exclusively in the “low” generalization group (two mice clustered in the “mid” group) with none in the “high.” This contrasts with control mice that displayed a normal-shaped distribution across the “low,” “mid,” and “high” groups. This unbiased analysis further confirms IL silencing drives more generalized freezing responses, while IL stimulation promotes less generalized freezing. In a final analysis of the generalization index, results showed a significant effect of the DREADD manipulation (F_(2,34) = 4.7; p < 0.001; Fig. 1H). Mice in the hM3Dq groups exhibited greater scores (less generalization) relative to the hM4Di (p < 0.001) and control group (p < 0.001). As a further control, we analyzed data obtained from mice with unilateral DREADD expression in IL. Surprisingly, we found no effect of unilateral hM3Dq or DREADD activation on generalization expression.

IL stimulation and scanning behavior

Next, we asked the question, if hM3Dq mice exhibited reduced freezing to the GS, how were they behaving? We speculated that IL stimulation might promote movement classified under “threat detection” (Blanchard et al., 2011) or “pre-encounter threat responses” (Roelofs and Dayan, 2022), such as scanning behavior (Choy et al., 2012). To test this hypothesis, scanning behavior was analyzed using machine learning technology (DeepLabCut 2.0; Mathis et al., 2018; Nath et al., 2019) and custom code (Fig. 1I). RMANOVA showed a time–group interaction (F_(2,35) = 3.7; p = 0.036). There were no differences in scanning behavior prior to the presentation of the GS across all groups (Fig. 1J). However, upon GS presentation, scanning behavior collapsed in the hM4Di (p = 0.01) and control group (p = 0.014) but was maintained in the hM3Dq (F_(2,35)] = 6.5; p = 0.004; Fig. 1J). These data indicate a role for IL in suppressing freezing and maintaining scanning. Overall, IL excitatory activity is sufficient to suppress post-encounter defensive responses (freezing) in favor of pre-encounter defensive (scanning) and nondefensive (movement) behaviors in response to an “ambiguous” threat stimulus.

Brainstem auditory evoked potentials

A fundamental question in the study of stimulus generalization is the consideration that generalized responses may represent a failure in perceptual (sensory) discrimination (Dunsmoor and Paz, 2015), rather than mnemonic processes per se (Zaman et al., 2021). This might confound the present results indicating generalization at the level of forebrain plasticity. To begin to address this question, auditory thresholds using auditory brainstem responses were tested in a subset of ArcCreER^T2 × eYFP mice (n = 5). While thresholds cannot determine whether two tones are discriminable from one another, they can address whether tones are likely to be detectable. Evoked potential threshold estimates tend to underestimate behavioral thresholds, so stimuli presented above the AEP threshold should be of sufficient amplitude to evoke behavioral responses (Dent et al., 2018). The threshold by frequency response we found is consistent with the patterns in C57BL/6J (B6) mice. Generally, thresholds decreased (i.e., sensitivity improved) as stimulus frequency increased, although the rate of change increased with each subsequent octave. Thresholds at 3 kHz ranged from 57.5 to 67.5 dB SPL, with an average of 62.5 dB SPL. These thresholds are lower than the amplitude of the stimulus used in the generalization experiments, suggesting the ArcCreER^T2 × eYFP mice should be able to detect this tone. Thresholds do decrease between 4 and 8 kHz, suggesting that the 5 kHz tone may have a greater sensation level than a 3 kHz tone.

Estimates of auditory filter bandwidth using auditory evoked potentials or psychophysical frequency discrimination limens would further address whether individuals from this strain of mice differ in their ability to discriminate between these tones. C57BL/6J (B6) mice are capable of discriminating frequency changes of 250 Hz or less at 8 kHz, with frequency discrimination decreasing with increasing frequency (Kulig and Willott, 1984). Absolute frequency discrimination limens typically decrease with decreasing frequency (Weber's law), suggesting mice should have no difficulty discriminating the 3 kHz tone from a 5 kHz tone (Dent et al., 2018). In fact, frequency difference limens from 3 to 5 kHz are less than 1 kHz for all rodent species that have been tested (multiple mouse strains, chinchilla, guinea pig, gerbil, and multiple rat strains; Dent et al., 2018). Although sensitivity is lower at 3 kHz than at 5 kHz, our results suggest that tones of both frequencies should be detectable at 70–75 dBA in our subjects.

Experiment 2: cued fear memory generalization and the passage of time

Here, we tested whether cued fear memory generalization increases over time in ArcCreER^T2 × eYFP transgenic mice, a finding previously shown in the C57BL/6N substrain (Pollack et al., 2018). All mice were fear conditioned with the 5 kHz CS. Then, either 7 d (recent) or 30 d (remote) later, mice were presented with either the CS again or the novel tone GS (Fig. 2A). A “no-tone” control group was included as a baseline measure in which all procedures were identical to the experimental groups, except that on the test day, mice were not presented with a tone but instead left to explore the context for the same amount of time as the tone stimulus groups. At the recent time point, there was less freezing in response to the GS (3 kHz) compared with the CS (5 kHz; F_(1,40) = 13.4; p < 0.001; Fig. 2B). At the remote time point, there were no differences between the GS and CS groups (p = 0.58; Fig. 2C). Analysis of the generalization index showed a modest, but significant, reduction over time (F_(1,28) = 4.3; p = 0.045; Fig. 2D), indicating enhanced generalization with the passage of time, a finding congruent with previous work (Pollack et al., 2018). At either recent (p = 0.12; partial η² = 0.09) or remote (p = 0.22; partial η² = 0.19) timepoints, sex interactions were not detected.

Figure 2.

Cued fear memory generalization over time. A, Schematic of the experimental design. B, At the recent timepoint following learning (7 d), mice in the 3 kHz group exhibited less freezing than the 5 kHz group. C, At the remote time point (30 d), freezing was equivalent between 3 and 5 kHz groups. D, The generalization index reduced over time, indicating greater generalization at the remote compared with the recent timepoint. N = 9–21/group (recent, 0 kHz female n = 6, male n = 4; 3 kHz female n = 8, male n = 13; 5 kHz female n = 9, male = 12; remote, 0 kHz female n = 6, male n = 3; 3 kHz female n = 4, male n = 5; 5 kHz female n = 5, male = 4). *p < 0.05, ***p < 0.001.

Experiment 3: mPFC engram quantitative measures

In this experiment, a new cohort of mice underwent experiments described in Experiment 2, but 90 min after generalization tests, brain tissue was harvested and processed for Arc immunohistochemistry (IHC). mPFC engrams were tagged using the ArcCreERT2 × eYFP system in which 4-OHT administration induces Cre-mediated recombination and expression of the eYFP reporter in activated neurons (Fig. 3B–D). Neurons were counted in PL and IL (Fig. 3E).

Figure 3.

mPFC engram quantitative measures. A, Experimental design. B, ArcCreER^T2 × eYFP system genetic design. C, Representative photomicrograph depicting the efficacy of 4-OHT to drive eYFP expression after 4-OHT injection versus vehicle (oil) control. 4-OHT (n = 6) selectively induces eYFP expression (58-fold increase) versus vehicle (oil; n = 6) control in ArcCreER^T2 × eYFP mice. D, Representative photomicrographs depicting GFP and Arc immunofluorescent labeled cells. The arrowhead indicates a double-labeled cell. E, Depiction showing the location of counting frames (250 × 250 microns) in the mPFC. The arrowhead indicates the anatomical location of the lateral ventricle reference point. F, Recent timepoint. Atlas image showing the location of identified engram across PL and IL L2/3 and L5/6. There were no differences in GFP⁺, Arc⁺, or colocalized cells across kHz groups in the PL and IL deep layers. In IL L2/3 (highlighted in red box), there was no difference in eYFP⁺ or Arc⁺ expression across groups. However, there was a significant increase in the number of colocalized cells in the 3 kHz group versus the no-tone control and the CS. G, Remote timepoint. Atlas image showing the location of identified engram across PL and IL L2/3 and L5/6. There were no differences in GFP⁺, Arc⁺, or colocalized cells across kHz groups in the PL and IL deep layers. In IL L2/3 (highlighted in red box), there was no difference in eYFP⁺ or Arc⁺ expression across groups. However, there was a significant decrease in the number of colocalized cells in the 3 kHz group and an increased number of colocalized cells in the no-tone group. N = 8–12/group (recent, 0 kHz female n = 6, male n = 2; 3 kHz female n = 4, male n = 8; 5 kHz female n = 4, male = 2; remote, 0 kHz female n = 6, male n = 3; 3 kHz female n = 4, male n = 5; 5 kHz female n = 6, male = 5). Hashed lines indicate the level of chance calculated as (eYFP⁺/ DAPI⁺) *(Arc⁺/ DAPI⁺) *100. *p < 0.05 and **p < 0.01. n.s., nonsignificant.

To assess whether 4-OHT itself might impact fear learning and memory, a separate group of male and female mice was run through the identical behavioral paradigm described above, but instead of 4-OHT, they were injected with vehicle (corn oil) control. These mice were compared with mice injected with 4-OHT. Results showed no group or sex interactions during either learning or retrieval of the CS. These data indicate that 4-OHT does not impact fear memory consolidation.

MANOVA conducted on eYFP⁺ and Arc⁺ cell counts revealed no differences across groups at either time point (Fig. 3F–G). However, MANOVA conducted on coactivated cells revealed a time–kHz interaction (V = 0.49, F_(4,40) = 4.99; p = 0.002). There were no main effects of sex (p = 0.70; partial η² = 0.05), sex–time (p = 0.95; partial η² = 0.02), or sex–kHz–time interactions (p = 0.80; partial η² = 0.05). Follow-up Bonferroni-corrected ANOVAs revealed a kHz–time interaction in L2/3 IL only (F_(2,54) = 8.0; p  =  0.001). At the recent time point (Fig. 3F), there was a main effect of kHz (F_(2,23)  =  4.7; p = 0.02). There was a greater number of coactivated neurons in the GS group versus control (p = 0.04) and no difference between the CS group and controls. Strikingly, a main effect of kHz was also observed at the remote timepoint in L2/3 IL only (F_(2,25) = 6.5; p = 0.005; Fig. 3G). However, in this group, there were fewer coactivated neurons in the GS group versus control (p = 0.004) and fewer coactive neurons in the CS group versus control (p = 0.023). Overall, these data indicate that, when generalization is low, there are more IL L2/3 coactivated neurons but, when generalization increases over time, there are fewer coactivated neurons in the identical location.

Experiment 4: bidirectional IL chemogenetic engram control during the expression of cue fear memory generalization

In this experiment, a Cre-dependent DIO-hM4Di, DIO-hM3Dq, or DIO-mCherry control was injected into IL to selectively express DIO-hM4Di, DIO-hM3Dq, or DIO-mCherry in ArcCreER^T2 × eYFP mice (Fig. 4A). Subjects underwent 4-OHT-dependent tagging during fear conditioning and later retrieval induced engram Arc reactivation was induced using DCZ (Fig. 4C).

Figure 4.

Bidirectional chemogenetic IL engram control during fear generalization over time. A, Representative photomicrographs showing (a) DIO-hM3Dq AAV transfection in mPFC, (b) GFP⁺-labeled cells, (c) Arc⁺-labeled cells, (d) image merge (arrowheads indicate double-labeled GFP⁺:Arc⁺ cells), (e) DIO-hM4Di AAV transfection in mPFC, (f) GFP⁺-labeled cells, (g) Arc⁺-labeled cells, and (h) image merge (20×, 0.8 NA, 1.5 zoom). B, DCZ injection in the DIO-hM3Dq system (n = 6) increased the number of colocalized cells relative to DIO-hM4Di (n = 7) mice and fluorophore controls (n = 4). There were no differences across GFP⁺ or Arc⁺ cells. C, Schematic depicting DIO-DREADD × ArcCreER^T2 system design and experimental timeline. D, Heatmaps depicting transfection localized in vmPFC. The extent of AAV expression in the mPFC across four coronal planes is depicted. Relatively darker shaded regions show greater overlap across individuals. Transfection was predominantly located in the IL, with some expression in DP. Mouse brain atlas images modified from (Paxinos and Franklin, 2004). E, Left panel, Mice in the DIO-hM4Di (recent) group exhibited greater freezing and less scanning during presentation of the GS compared with fluorophore control. Right panel, Mice in the DIO-hM3Dq (remote) group showed less freezing and more scanning during presentation of the GS compared with fluorophore control. F, Donut plots depict percentage freezing, scanning, and moving during the mean GS presentations for recent (left panel) and remote (right panel) groups. G, Left panel, Unbiased k-means cluster analysis on mean GS freezing for all mice produced two clusters (low and high) in the recent group. Subsequent frequency distribution analysis illustrates a rightward shift in the distribution from low to high freezing in the DIO-hM4Di compared with control. Right panel, Unbiased k-means cluster analysis on mean GS freezing for all mice produced two clusters (low and high) in the remote group. Subsequent frequency distribution analysis illustrates a leftward shift in the distribution from high to low freezing in the DIO-hM3Dq compared with control. N = 7–9/group (recent, control female n = 1, male n = 7; DIO-hM4Di female n = 6, male n = 1; remote, control female n = 3, male n = 5; DIO-hM3Dq female n = 6, male n = 3). *p < 0.05, **p < 0.01. n.s., nonsignificant. Scale bar, 50 µm.

The efficacy of DCZ to modulate neuronal activity in DIO-DREADD expressing cells was assayed using IHC against eYFP and Arc. Results revealed no differences in eYFP⁺ or Arc⁺ neuronal expression across groups (Fig. 4B). When the number of eYFP⁺:Arc⁺ colocalized cells was analyzed, we found a difference across groups (F_(2,14) = 15.2; p < 0.001). There were a greater number of coactivated cells in the DIO-hM3Dq group compared with the control (p = 0.015) and DIO-hM4Di group (p < 0.001; Fig. 4B). These data indicate the combination of DCZ and the DIO-DREADD system preferentially drives Arc⁺ expression in eYFP-labeled cells. Analysis plotting the spread of viral transfection across brains showed clear AAV transfection in IL and also in DP (Fig. 4D). We grouped both bilateral and unilateral transfection into the analysis because there was a high degree of laterality in the expression of eYFP which may have anatomically biased transfection. In addition, there were no statistical differences between bilateral and unilateral groups.

By chemogenetically manipulating the activity of an engram established during learning, here we test its contribution to the expression of generalization at recent and remote time points following learning. At the recent time point, mice in the DIO-hM4Di group showed greater freezing (p = 0.026) and less scanning (p = 0.032) compared with a time-matched (recent) fluorophore control group (Fig. 4E,F, left panel). Like the previous global DREADD analysis, an unbiased k-means clustering algorithm was applied to mean GS freezing levels across all mice in the recent group. The analysis segregated two “high and low” clusters (Fig. 4G). A chi-square test was performed to test the relationship between experimental groups and clusters. Results showed a significant effect χ² (1, N = 13) = 3.9, p = 0.048. A predominant number of mice in the control group clustered in the “low freezing” group which shifted to a “high freezing” cluster in the hM3Di group. Overall, these data indicate a necessary role for an IL engram in suppressing conditioned freezing in response to a novel tone stimulus.

At the remote time point, mice in the DIO-hM3Dq showed less GS-elicited freezing (p = 0.004) and greater scanning (p = 0.01) compared with a time-matched (remote) fluorophore control group (Fig. 4E,F, right panel). Unbiased k-means clustering revealed two clusters (“high and low”; Fig. 4F). A chi-square test was performed to test the relationship between experimental groups and clusters. Results showed a significant effect χ² (1, N = 17) = 5.13, p = 0.024, with a greater number of mice in the control group clustered in the “high freezing” group which shifted to a “low freezing” cluster in the DIO-hM4Dq group. These results provide an unbiased confirmation that an IL engram established during learning is sufficient to suppress conditioned freezing in response to a novel tone stimulus. Together, these data suggest IL engram functionality in modulating fear generalization. A caveat to this interpretation is that because CS-elicited freezing was not measured, the extent of generalization and discrimination among groups cannot be determined.

In the control groups, there was increased freezing, albeit nonsignificant (p = 0.12), over time. Given DP/IL transfection in a few of our subjects (n = 4; Fig. 4D), interpretation of these results cannot rule out DP functionality in generalization. At the recent time point, sex interactions could not be determined as there was no sufficient sample size. At the remote time point, a sex interaction was not detected (p = 0.067; partial η² = 0.25).

Discussion

In this set of experiments, ArcCreER^T2 × eYFP transgenic mice were used to genetically “tag” prefrontal cortex neurons activated during pavlovian fear conditioning for later visualization and manipulation during the expression of generalization. Results showed that following learning, an engram located in IL L2/3 was associated with a novel tone stimulus, and silencing an IL engram resulted in more freezing. These data suggest an IL engram encoding an inhibitory process for attenuating generalization was formed in the IL during initial memory consolidation and expressed in the presence of the ambiguous GS. These empirical findings support a long-standing theory proposing generalization as a learning process (Hull, 1943). There is also supporting empirical evidence showing that blocking the output of the nucleus reunions prior to learning inhibited memory specificity, but had no effect when inhibited after learning (Xu and Sudhof, 2013). In a more recent study, IL stimulation (picrotoxin) soon after learning improved later extinction performance, and protein synthesis inhibition eliminated the effect, suggesting memory consolidation processes for extinction are established at the time of learning (Bayer et al., 2024). We theorize that a similar “opponent process,” as that theorized by Bayer et al. (2024), may have been formed in the IL during learning to later modulate generalization.

Presentation of the CS did not activate an IL engram, but a novel tone stimulus did. This raises an important theoretical question as to the exact nature of IL engram functionality. One model of global IL function proposes that IL biases behavior based on cues present, rather than past associations and contingencies (Nett and LaLumiere, 2021). Consistent with this model, we hypothesize the identified IL engram acted to suppress conditioned freezing, in favor of scanning, when conditions changed. Without IL engram reactivation, behavior output defaulted to the original learned association, resulting in generalization. This supports the idea that one function of IL, and perhaps IL engrams, is to inhibit past associative behaviors to promote alternate adaptive behaviors in the face of change. This hypothesis requires further study (Gräff, 2024).

Memory precision and a theoretical basis for dynamic changes in IL ensemble activity over time

Following learning, presentation of a novel GS promoted IL engram reactivation and memory specificity, presumably because memory for the original CS was intact. Over time, generalization was not associated with L2/3 IL engram activity, and stimulating an IL engram resulted in discrimination and the maintenance of scanning behavior. A key outstanding question is whether the same IL engram identified shortly after learning was the same engram showing no reactivation later after learning. There are several plasticity-related mechanistic possibilities that might account for this phenomenon: (1) IL gradually disengaged its obligatory top-down inhibitory role in the defensive memory circuit over time to support increased generalization, in a systems consolidation-like process (Bergstrom, 2016), (2) IL activity or IL engram reactivation may have been suppressed over time via newly recruited inhibitory mechanisms (Morrison et al., 2016), or (3) IL engram activity may have weakened over time through a synapse destabilization mechanism, in a process akin to forgetting (Ryan and Frankland, 2022).

With respect to Possibilities 2 and 3, one proposal is that neuronal networks employing mechanisms of instability, such as synaptic decay, elimination, or suppression, would promote a greater degree of generalization (Richards and Frankland, 2017). A weakening of synapses (destabilization) may provide a “forgetting” mechanism to promote brain states that reflect the past. That is, in the face of threat ambiguity and over time, it may be advantageous to promote or bias the original CR to enhance the chance of survival in “a better safe than sorry” strategy (Eilam et al., 2011). The exact nature of the suppression of IL L2/3 engram reactivation with respect to generalization processes remains an open question.

Risk assessment, switching defensive states, and IL functionality

In the presence of a potential threat, organisms must first detect the threat and then respond, with the most appropriate defensive state. Drawing from predatory imminence continuum theory (Fanselow, 1994), we hypothesized that mice presented with the GS and exhibiting low levels of freezing might engage in alternate defensive responses based on the relative ambiguity of perceived threat. Vigilant scanning behavior was identified as a distinguishable, and highly prominent, defensive behavior that aligns with a pre-encounter threat response to the ambiguous tone (Blanchard et al., 2011). When IL was stimulated, scanning behavior made up a substantial portion of all movement. Conversely, freezing (generalization) was the predominant response when the IL was inhibited. These data indicate IL functionality, and IL engrams, in switching defensive states between post- and pre-encounter; IL inhibition resulted in postencounter, and IL stimulation promoted pre-encounter, defensive states in response to the GS, potentially via downstream signaling with central nucleus of the amygdala (Moscarello and Penzo, 2022). These findings support other work indicating IL in switching behavioral defensive states to promote movement (Halladay and Blair, 2017) and also theoretical work suggesting a role for IL in promoting behaviors that align with the most rigid reading of the associative relationship among cues present (Nett and LaLumiere, 2021). While freezing and scanning are categorized as risk assessment behaviors, they are differentiated by the actual or perceived proximity or ambiguity of the threat.

Cued fear memory expression, extinction, and IL functionality

The cued freezing response was spared during IL stimulation but decreased the next day. We speculate that the brief CS exposure schedule (three presentations) may have acted as a “weak” extinction training regimen that, in combination with IL chemogenetic stimulation, promoted stronger extinction retention the next day. This finding supports IL functionality in fear extinction consolidation rather than learning (Bukalo et al., 2015; Bloodgood et al., 2018; Bayer and Bertoglio, 2020) and another study showing that increasing IL activity (using picrotoxin) normalized extinction retrieval in an extinction-impaired mouse strain (Fitzgerald et al., 2014).

Ventral medial prefrontal cortex and dorsal peduncular (DP) prefrontal cortex functionality in fear generalization

The vmPFC encompasses IL and also the dorsal peduncular cortex (DP; Botterill et al., 2024). DP has been implicated in promoting defensive states (Borkar et al., 2024; Botterill et al., 2024; Campos-Cardoso et al., 2024), a function in apparent opposition to IL. Precise stereotaxic space alignment and mapping of transfection spread revealed the majority of DREADD transfection localized to IL, but also to some extent in DP (ventral to IL; Figs. 1D and 4D). Analyses comparing IL versus IL/DP transfection patterns showed greater, albeit nonsignificant, conditioned freezing with IL/DP expression, a finding that may support DP functionality in biasing defensive behavior. The sole contribution of DP functionality to cued fear memory generalization expression remains an open question.

Clinical relevance

The use of novel “ambiguous” stimuli in an auditory cued pavlovian defensive conditioning paradigm to study brain functionality underlying generalization processes has preclinical utility for the study of PTSD (Dunsmoor and Paz, 2015) and anxiety disorders (Lissek et al., 2006). In the field of psychology, a “weak situation” refers to a context in which environmental stimuli are less predictive of threat imminence. The ambiguous nature of the environment tends to produce more individual difference variability and has been proposed as a model for studying anxiety (Lissek et al., 2006). In our paradigm, both the contextual elements and auditory tone frequency were shifted, which might account for wide variation in individual differences (Fig. 1). Considering the negative valence domain within the Research Domain Criteria (RDoC) framework, the study of potential, but ambiguous, threat has implications for the study of anxiety that can be differentiated from acute threat (Fanselow and Hoffman, 2024). The vmPFC functions in the expression of generalization in humans (Spalding, 2017) and vmPFC dysfunction has been associated with PTSD and anxiety-related disorders (Alexandra et al., 2022). The consensus of data presented here indicates vmPFC functionality, and perhaps vmPFC engrams, in “downshifting” defensive behavioral states, from high to low. Rodent IL is considered a putative homolog of the human subgenual anterior cingulate (Area 25), which regulates mood and emotion (Alexander et al., 2019). Thus, the methodology and data in this set of experiments may add preclinical value for understanding Area 25 dysfunctionality and overgeneralization phenomenon seen in anxiety and trauma-related disorders.